What are the connecting points between political science, economics, and statistics? For Jonathan Katz, "interdisciplinary" is less a buzzword and more a research way of life in order to understand human behavior and its implications for our governing structures and economy. Katz's work ranges from the pursuit of fundamental social science questions to research applications that have powerful implications in public policy and industry, and he has been involved in the creation of several start-up companies and has served as a consultant for work relating to election law.
In the discussions below, Katz reflects on his family origins in Manhattan and his educational trajectory from MIT, UC San Diego, and his faculty positions at Caltech and the University of Chicago. He provides valuable insight on the unique role of the Division of Humanities and Social Sciences at Caltech, and its historical positioning as one of the true centers of innovation in its early embrace of quantitative measures and their application across a broad range of social science questions. In his previous role as division chair, Katz conveys his interests in building and modernizing the Division, and the institutional value it brings owing to the fact that scientific research is inextricably linked to basic questions of political economy.
Katz is an elected fellow of the American Academy of Arts and Sciences and an inaugural fellow of the Society for Political Methodology, and he serves in numerous editorial roles for leading journals in political science and economics.
DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Friday, May 27th, 2022. I am delighted to be here with Professor Jonathan N. Katz. Jonathan, thank you for having me into your office today.
JONATHAN KATZ: My pleasure.
ZIERLER: To start, Jonathan, would you tell me your title and affiliation here at Caltech?
KATZ: I am the Kay Sugahara Professor of Social Sciences and Statistics. I was previously the Chair of the Division of the Humanities and Social Sciences. Before that, I was the Executive Officer for the Social Sciences. I'm a member of the Caltech/MIT Voting Technology Project. I am also a member of the Information and Data Sciences Program, the new undergraduate option and minor. I helped design it.
ZIERLER: We'll start with the first part of the title. Who is or was Kay Sugahara?
KATZ: Kay Sugahara is still alive. I actually am unusual at Caltech; I actually raised my own chair.
ZIERLER: Oh, wow.
ZIERLER: I'm sure that was appreciated.
KATZ: Yeah! Kay is an alum from—actually, he's a really interesting guy, Kay Sugahara. He and his family were interned in Manzanar, a Japanese-American family. Kay served in World War II in the Navy. They run a very successful family business called Fairfield-Maxwell. Fairfield-Maxwell was actually founded by Kay Sugahara's father, who is also Kay Sugahara, so it's also a little confusing. I'm actually named for the father, not the son.
ZIERLER: The son who endowed the position?
KATZ: Right. He did so in his father's honor. Kay's father and his business partner, Max Kleinfeld formed this company, which does lots of things. Some of their stuff is actually in the Seismo Lab. Part of their business is they design and service deep-sea oil deposits. They have these specialized sonars. That's what the Seismo Lab has from them. They design special technology that allows them to map deep-sea oil. They also are in the ocean transport business. For example, they have the contract—basically every Nissan car comes to the United States is shipped by Fairfield-Maxwell. The name came about because—Sugahara, Kay's father and his partner Max Kleinfeld—a Japanese guy and a Jewish guy in the 1940s, that was not a good idea. [laughs]
KATZ: They chose the name of their company—Sugahara in Japanese means "beautiful view," and so they got "Fairview." Max Kleinfeld, they just took his first name Maxwell. So, a Jew and a Japanese guy basically created a firm that sounded like a nice, white-shoe, protestant [laughs]—
ZIERLER: I love it. I love it.
KATZ: It's a family-held business, still is. Kay wanted to honor—Kay was the CEO of the company, although he has younger twin brothers who both went to Harvard. They wanted to give money to Harvard. At the end of the day, the agreement was that they did both. They did a chair at Harvard and they did a chair at Caltech.
ZIERLER: What were the circumstances of you meeting the Sugahara family?
KATZ: When I became division chair in 2007, at the time the new president at Caltech was Jean-Lou Chameau. Caltech has been very fortunate that we—we've been too fortunate. We've had these incredible angel benefactors who have literally given these huge gifts. Most recently, it's the Moores. Before them was Arnold Beckman. What Caltech did from a fundraising point of view is that we had these great megastar people who have given literally hundreds of millions of dollars. Now, the new Resnick Institute is the same thing. But that's not how most universities raise money. [laughs] Most universities raise money by—you have this pyramid, small donors who you then move to bigger donors, and move up. Caltech had a teeny alumni base.
One of the things that Lean-Lou wanted division chairs to do was to create these advisory councils, which is a very typical thing to do. I was kind of the only division chair who took Jean-Lou's thing. The whole goal was to bring in people who we hadn't previously gotten money from, or not very seriously involved in Caltech, to become more involved. With my then-development person, we went about trying to identify people. Kay was one of the original members of the HSS Chair's Council. He was based in New York, and I would visit him. Then I got him to agree to join the board. That's how I got Kay. There have been others. Kay was identified as an alum who was successful but had really not engaged much with Caltech. I grew up in New York—we'll come to that at some point—so I went to New York quite frequently, and would meet with Kay in New York. Again, I convinced him to join. It's funny, because Kay is actually a very sharp man, but he says he wasn't really Caltech material. It took him seven or eight years to graduate from Caltech. [laughs]
ZIERLER: Are you in regular touch? Does he keep in contact with you and what's going on at Caltech?
KATZ: We do. COVID has been a little bit of a problem. Unfortunately, I would say he's one of the donors—there are a bunch of donors who have been unhappy about the renaming and the diminishing of Millikan's legacy.
ZIERLER: And Kay is part of that?
KATZ: He is unhappy about this.
ZIERLER: Even though he would be in the underrepresented group for which Millikan might not have been a big fan.
KATZ: Yeah, but that's—
ZIERLER: These are complicated issues.
KATZ: Complicated issues. Actually one of the hardest things about getting Kay to give money to Caltech and to think about HSS is that—he was fine with HSS, and he thinks it's an important thing. Part of the reason that he gave this money to HSS was his father, besides being a very successful businessman, was very involved in the Asia Society, trying to improve Asian-American and Asian politics. That's why he gave money in honor of his father to HSS and the social sciences. But yeah, it's a complicated issue. He thinks that Caltech is first and foremost a premier physics, chemistry—one of the things he laments is all of our smart students who go off and work for like Goldman-Sachs or other big banks. He thinks that's a societal waste, that that's not where Caltech should be producing. They should produce people like him who actually either build things and do things—take the applied engineering—or more importantly doing fundamental science. I think that was his connection and he felt that somehow, Millikan, who had built this amazing institution around this key idea, is somehow being diminished. But it's a complicated issue, and he's not the only donor I have dealt with of his—well, he's the oldest—but of that generation, and the half-generation after him, who are quite unhappy.
ZIERLER: Do you think he's correct in assessing those trend lines? Do we send too many quants to Wall Street these days?
KATZ: We send less now than we used to, I would say. It's hard for me to say. They would go there anyway. To be honest with you, some of these people have been incredibly generous. Paul Young, who is a Caltech alum, headed up until recently all of Goldman Sachs' quantitative trading, and he will be quite generous to Caltech over the years. So, I think those provide valuable things; probably, net, too many. It's allure; it's a lot of money. But the answer is, I don't know how you'd stop it. If people decide that they want to use their technical skills to make a lot of money, and they're bright, our students are going to do that. I don't really know that we can do much to dissuade them.
ZIERLER: Where do you see the fact that the overwhelming majority of Caltech undergraduates now are in computer science? Where do you see that generationally in terms of the kinds of fields that they go into, including fields like finance?
KATZ: This is not a recent phenomenon. I was an undergraduate at MIT. This was even true—not to the extent that it is now. Literally half the students come in saying they want to do computer science. My generation was probably about a third. So, this has been on trend. The positive spin on this is that computation and computer science has fundamentally changed all aspects of science. There's just things today we can do that we just could not, including stuff I do. There's things you could not do in the 1990s that given advances in computer science, advances in engineering and actual computing, that are possible. It is an interesting challenge for the Institute in the sense of I think it's good that we don't admit students to particular majors. We don't really do that. But it is a huge strain on the computer science group about how they train students. It's why there's now an IDS major.
I was a major in what's called 18C, Applied Mathematics, which is really theoretical computer science. Again, MIT at the time was trying to figure out how to bleed students from computer science. Caltech is doing the same sort of thing. It's a concern, and it's not a concern. I think computation and algorithmic thinking has become so important across all aspects of human existence—science, policy, government—that it's not a bad thing. But I think as a university, there are some issues just about how we make sure these students are trained and advised. How do we recruit faculty? Computer scientists are some of the hardest faculty to recruit as are economists. They have lots of outside options. You're playing this fine line about how you make this all work.
ZIERLER: To go back to your title—Social Science and Statistics, is that a dual appointment, or what is the meaning behind the duality in the title?
KATZ: Caltech, especially in HSS, we don't have departments. This is true across most, but HSS probably the most. The social sciences run effectively as one group. There's no departments. The economists don't caucus and have their own graduate program; the political scientists, the people who do statistics. In fact, I should point out that basically everyone who does applied statistics at Caltech is actually in HSS. [laughs]
ZIERLER: Even if you do seismology.
KATZ: We don't really have any statisticians who—I said applied statisticians, the people who actually—we have lots of people in mathematics and in ACM, we have lots of people who do very theoretical statistics or machine learning. But at most other universities, we would have a statistics department. Like when I was on the faculty at Chicago, I had joint appointments between Political Science, Economics, and Statistics. There's no statistics department. Everyone who does applied statistics is in HSS. There's four of us. In part, it's just a reflection of the research I do. About half my research—it varies over time—is in statistics, published in stats journals—and about half my research is then applying cutting-edge statistical tools to various applied questions, mostly in politics. But I've done some stuff in finance, some stuff in economics, some stuff in law. The title comes because I work across these areas. In HSS particularly, we're pretty flexible about what title people want to call themselves, again because we don't have this departmental structure—so, like my colleague Mike Alvarez is Professor of Computational Social Science. We do similar things in social science, because I actually do work across fields. That's the answer.
The Meaning of Interdisciplinary
ZIERLER: What works well about the idiosyncratic way that HSS and more generally Caltech is organized, both in terms of encouraging multidisciplinary collaboration and more broadly having a no-walls approach to the research?
KATZ: Interdisciplinary work has been the buzzword in higher ed since I think I was in graduate school, probably before then. Most universities want to talk about doing that. The university that probably gets the most credit for doing interdisciplinary work besides Caltech is probably the University of Chicago, where I was on the faculty. We take very different approaches. At Chicago, and the approach that most universities take, is that they still have traditional departmental structures—Department of Political Science, Department of History, Department of Economics, ad nauseam—and then how they try to foster this interdisciplinary approach is they have these separate—committees, they call them. Basically a group of faculty can get together and basically form a committee on almost anything they want. If they can convince deans and stuff that it's a valuable thing that they could get resources to bring in graduate students, to hold seminars, and the like. But ultimately, at the end of the day, when you're being evaluated—my primary appointment at Chicago was in the Political Science Department—I'm being evaluated by political scientists. Fields differ quite radically in what they expect in how you should be publishing, what are the venues of publishing, what counts. Even at Chicago, which had this structure which tried to create glue, it was basically building it on top of this very traditional American and, by the way, world these days, higher education disciplinary department-based things.
At Caltech, especially in HSS, because we were never going to be big—Caltech is not big, and HSS—we decided that that wasn't going to be the thing. So, literally in HSS, all the social scientists are together. They're all in Baxter Hall. My neighbor to the right is John O'Doherty. He's a card-carrying cognitive neuroscientist, as is the guy to the left, across the window there, Dean Mobbs. Across the way is the behavioral economist, Antonio Rangel. Michael Ewens in that corner is a business and economics guy. We're all in one department. We have one graduate program. Well, now, we have two, but we started out as having one graduate program. We meet as one single faculty. So, when we review people, when we hire people and when we review them, it is this whole group. My colleague I mentioned, Jean Ensminger, who is retiring, Jean is an anthropologist. We all sit in the same faculty meetings, we all review, we all get along. We also have norms.
Caltech is unusual in social sciences in that we have a one-hour seminar. Most academic places, an hour and a half has become the norm, partly because projects are big and people want time. But the view at Caltech has always been that basically anyone can take a seminar for one hour on any topic. [laughs] I think the difference by this no-walls, in part enforced by size—across the entire Institute, there's only 300-odd tenure track faculty members—that the only way that we can get critical mass and to think, is you're going to have to reach out. In HSS, you don't come to Caltech if you want to be with five other labor economists, or six other people working in American politics. It's just not going to happen here.
ZIERLER: That's true in Biology and Chemistry also.
KATZ: Yeah. The one exception at Caltech is Geology. [laughs] Our geologists are way outside, relative to [laughs]—we have a huge geology department. But yes, right, it's true. The core at Caltech is both within the divisions, there's not this sort of departmental siloing structure because we just don't have the size, so if you want to have someone to talk to about things, you're going to have to talk to a broad subset. It also happens across divisions, again in a way that happens much more fluidly than happens even at, like I said, Chicago, which is usually credited with being sort of the most interdisciplinary.
For example, in this division, we've had searches—Ralph Adolphs is joining Biology. He's a neuroscientist. We've hired recently and she unfortunately replaced someone we lost—we have joint hires with computer science, people that work at the intersection of computer science and economics. Again, this just works because we're a small place and we don't have as much bureaucratic structure. Caltech is both decentralized and centralized at the same time. It's true the division are very decentralized. The division chairs have a lot of authority over what goes on there. But budgetarily, everything flows from the Provost's Office. At most large universities, that's not how it works; everything is decentralized. If the dean of Social Sciences wants to do a joint hire with the dean of Engineering, there's this whole rigamarole over who's paying for this faculty line, whose budget is going to give up. Here, the faculty lines actually all report to the Provost's Office, so there's not the same sort of finagling over money, even across divisions, that happens at other universities.
ZIERLER: Is this effective for recruitment?
KATZ: It can be. Caltech's small size is always both a positive and a negative. If you're a young scholar coming here, especially in the social sciences or the humanities, it can be a little intimidating. I'm coming to a place where I have to be evaluated by not just people in my discipline. Disciplines can be quite broad, right? You're a historian of science, right? There's internalist versus externalist views of the history of science. But at least you're all quote-unquote "historians of science." But here, if you come to be a historian of science—Nicolás Wey Gόmez—you're doing work on Columbus and exploration and the history of science, but you're being evaluated by analytic philosophers and by literary scholars. That can be very intimidating as a junior faculty member, because we all got our PhDs in department-based places. I think for junior people, this can be—we just look different. [laughs]
The positive side is I think it does foster these cross-disciplinary adventures. For me, early in my career—for example, I did work with a colleague who we actually both started at Caltech the same time, Paolo Ghirardato. He's a mathematical decision theorist. We applied what at the time were cutting-edge mathematical tools thinking about ambiguity to problems of political science, voting. How did that happen? Paolo gave a seminar on this stuff, what's called ambiguity aversion, and I said, "Oh, this is interesting, and there's these papers that I think aren't right, and this might be a way to answer that question." Again, that just doesn't happen, even at a place like Chicago. So, the upside to it is that the open doors engender all this cross-disciplinary, interdisciplinary work. The down side, we're a small place and it means that if hiring, particularly junior faculty are worried about their tenure and what their future is going to be, it's like, "I don't know." Their advisors are often—"I don't know." Again, it's a bigger issue in the social sciences. The humanists, because unfortunately the job market is more difficult, it's a little easier to hire. But the social scientists have a much harder time hiring. Economists are still in very high demand. We actually have the highest faculty turnover, in fact, at the Institute.
ZIERLER: Historically up to today?
KATZ: Up to today.
ZIERLER: Staying on junior faculty, in other divisions, which historically have prided themselves, unlike perhaps Harvard or Stanford or other places where there is not necessarily the culture of promoting junior faculty, what has been—this is historically—if you were a physicist at Harvard, assistant—
KATZ: I've turned down offers twice at Harvard, as a junior faculty member. I know all about it.
ZIERLER: Because this idea of you're just not going to get tenure there, right?
ZIERLER: How has that played out at HSS, in the way that, in other divisions here, Caltech prides itself on, "We will give you the tools to succeed. We brought you here because we want you to succeed, we want you to achieve tenure." Does that translate well in HSS?
KATZ: I think the goal does. I think the practice is different. Tenure rates are much, much lower in HSS, and the reason is this, particularly in the social sciences: Think about hiring a biologist at Caltech, or a chemist. They typically have been a graduate student and then a postdoc for eight to ten years before you hire them. You know exactly what you're getting.
ZIERLER: They're seasoned.
KATZ: You just have a track record. It's never perfect. There are tenure dials in Chemistry and Biology. People can't really make that transition to being the head PI. But for the most part, we know a lot. We hire assistant professors in Biology; they're in their late thirties, early forties. We hire in social sciences out of graduate school. Typically they haven't even finished their—
ZIERLER: They're in their late twenties, early thirties.
KATZ: Yeah, I was 26 when I joined the faculty at Caltech. I had gotten my PhD in four years. That's a risky—so we're always going to have riskier bets just because we don't have this world where you take a postdoc for ten years before you're going to get a job.
ZIERLER: You're also a lot less mature at 26 than you are 36.
KATZ: Exactly! You'd like to think you are—and like I told you, I like to think I'm young; it's just not true anymore. [laughs] When I was a division chair and when I recruit people here, I always tell them, "Caltech's goal and my goal is for you to be successful here." It's a bit of a Faustian bargain. We're going to give you tons of resources. Relative to our peers, you don't teach that much. Caltech has been a place where always, if you can make the intellectual argument that you need something for your research, we will find a way to make it happen. But we expect incredible performance, that you are going to be a leader in what you do. That's just a harder thing to forecast for us than it is for a biologist.
ZIERLER: What are those resources for social scientists at Caltech, in the way that a biologist, they need their lab, they need their microscope? What are the resources that a social scientist needs?
KATZ: It really varies. For me when I came here, at the time—it's hard to imagine—Caltech bought me a $50,000 workstation, which my iPhone that's sitting on the table here is more powerful than.
KATZ: It was software that was at the time called S+, which was $1,200 a year. The ability for me to hire RAs. For our neuroscientists, literally the packages look very similar to what hiring a young biologist is. They tend to not use much wet lab space, but some do, but they need access to time and money to buy time on our MRI machines. They need lab space. They usually need initial commitments to bring in graduate students and bring in postdocs. You're talking millions of dollars. We just brought in Kirby, who's an experimental economist. For her, it's again—we have a full social science laboratory downstairs, so she has access to that. Initial funding to do experiments before she can start applying for external grants. Again, very similar to the rest. But then you could hire an economic theorist; literally what they need is like a whiteboard [laughs] and a pen and paper. There's much more heterogeneity in what start offers look like, because the costs of research and the costs of what people do varies because there's such heterogeneity even in the social sciences about what people do. But again, if there's a dataset someone needs to buy, that they don't have access to, we try to do that. Again, there's always working with especially junior faculty members to minimize their administrative burdens. I was actually telling someone we're thinking about recruiting—here once, at the last minute, this great opportunity came up for me to apply for this grant. It was basically a 24-hour turnaround time. At any other university, if I went to a grants manager and said, "I need to turn around this grant that you need to sign off on in 24 hours" they would just laugh. At Chicago, they would just laugh. If you do not submit your NSF grant 21 days before the deadline, they would just say, "Sorry, you're going to do it next time." That would never happen at Caltech.
ZIERLER: We're nimble enough.
KATZ: We're nimble enough. You have to turn what sometimes can be a deficit—small size—into a benefit. We're nimble. We don't have this deep structure. We're not USC with thousands of faculty and tens of thousands of staff. We just can do things.
ZIERLER: In the way that you identified social science as a group—we don't caucus individually within social science—is that generally true for the "H" as well, for the humanities?
KATZ: Yeah, it is. Yes, it is the case that sometimes the political scientists get together or the economists get together. There can be questions about hiring, or what they want to bring in. But we don't have formal political science faculty meetings. We have social science faculty meetings. Same on the H side. On the Humanities side, all the Humanities faculty get together at their faculty meeting. Hiring is a joint decision, so we don't say, "Oh, we need to hire a philosopher, so you philosophers, you just tell us who you want to hire, and we're just going to rubber-stamp it." That's not how it works.
ZIERLER: For better?
KATZ: Again, nothing's a panacea. I do think it forces this broader perspective, and it forces proponents of people to hire and for people who are coming in about why is what you're doing is important. Not important to someone—not to a historian of science. Why is this thing you're doing—why is Nicolás Wey Gόmez's work on Columbus—by the way, which there's like tons of work on Columbus. Literally, there's entire libraries on Columbus. How is this history of science? His proponents, and Nico when he was hired here, had to convince philosophers and other people why this is an important thing to have at Caltech. I think in the end, that makes us better at what we do. I'm a big proponent that I think often times academics, it becomes too easy to get caught up in our jargon and our inside baseball speak. Things like this force you to say, "Why is this person really good? Why is this research really important?" To someone who's intelligent, who's got a PhD, but is not from your tribe. [laughs]
Election Law and MIT
ZIERLER: The last aspect of your title and affiliation—the Caltech/MIT collaboration. What is that about and how did it get started?
KATZ: Although we still seem to be having nightmares with elections in the United States, this goes to what is arguably—we've had historically bad elections, too. But in 2000, we had Bush v. Gore, the election, and a large part of it came down to Florida, and we all learned about hanging chads. To refresh your memory, if you don't remember what hanging chads are—and I don't remember how old you are—at the time, much of Florida used basically what were computer punch cards. You voted by basically putting a pin through a hole and it would knock out a precut chad. Then when you put it through a counter, which was just basically a computer reader, it will tell you did you vote for Bush, did you vote for Gore, or whatever it was you were voting on.
The election was very close, and there was this big—there was a Supreme Court case. It literally came down to Florida, and there were admitted problems. At the time, the president of Caltech was David Baltimore, and the president of MIT was Chuck Vest. Shortly after the election, when things had been resolved but there was still lots of uproar—sounds familiar [laughs]—they said, "Oh, this is an engineering problem. We can just fix this." So they put together this group of mostly engineers and computer scientists, some very famous people—Ron Rivest of RSA was a part of this. Anyway, lots of people. They mostly viewed it as an engineering problem, that, "Let's just put the technology together. Elections are easy." Turns out the engineers learned that elections even on the aspect of making a voting machine are not so easy, because the problem is the goal of security versus anonymity are at odds, so it's very hard to make a voting machine that is verifiable and auditable that also maintains the secrecy of the ballot.
The other problem, which is where people like me and Mike Alvarez got involved, and our counterpart at MIT, Charles Stewart, is what the engineers forgot is that actually the major problem is actually not the technological, although that's not inconsequential; it's really the human part and the political part of how elections are administered in the United States. Effectively, elections are administered by 5,000 counties. They're called townships in New England In almost every state, those are the people who run elections, and they are incredibly heterogenous. I assume you live in L.A. County; you live in the largest electoral jurisdiction in the United States. There are ten million people, and there are about five or six million registered voters in Los Angeles. Logan, the registrar, is very professional. It's a huge staff. They're pretty effective at running elections, because it's a big operation. There's economies of scale. We also have Alpine County, which is at the very edge, on the mountains between California and Nevada; I think there are maybe 1,000 people in all of Alpine County, and the person who is the registrar also has a dozen other hats.
So the elections are quite heterogenous about how they vote, how they run them. There was not one size fits all. Then there's the political economy problem which is that people only want to spend money on elections when there's a problem. In the aftermath of Bush v. Gore, Congress passed the Help America Vote Act, which gave grants to states and to counties to buy new equipment, to do training. That was a one-time set of money. It's all gone. That was 20 years ago. Everyone who put in these fancy new electronic machines, they're all obsolete now. [laughs] We created the Electoral Systems Commission, the EAC, which ended up being quite partisan and controversial and basically non-effective. That's why the Caltech/MIT Voting Technology Project still exists, because as we can see from the last election, we really haven't solved the problems! If anything, things have gotten worse. We have now politicized the running of elections, which was not the case. The advantage of the Bush v. Gore era was that the people who ran elections, no one thought they were corrupt. There was just a question about—
ZIERLER: They were on the same playing field.
KATZ: They were on the same playing field, and it was just a question about how do you measure voter intent and what is the law and what's the role of the courts to be involved in elections, which is something I do a lot of. I've been an expert witness many times in court on various aspects. But we've now gotten worse. This past week, there were primaries where there were candidates running for secretaries of state that basically think that the 2020 election was stolen. No one who actually researches in a non-partisan way thinks that there was any notion of an election being stolen, but for a portion of the Republican Party, and for Trump's portion, that is still something they hold on to.
ZIERLER: To go back to 2000, when Caltech and MIT looked at these hanging chads and saw an engineering problem, what about some of the political problems in the way that the Supreme Court made its decision? Was that out of MIT/Caltech lane, or that was baked into the origins of the project also?
KATZ: That was the origins, but the project has never taken on the question about what the appropriate role of the courts would be. We have been much more focused on the engineering and on ways of measuring electoral performance. From an engineering point of view, to know if something works, you need to have metrics. How would I know that an election is working well? We didn't have standardized metrics for thinking about elections. We came up with things that are now systematically measured, like what are the number of over or under votes, those sorts of things. What are the amount of discarded ballots? Ways that we can think about whether or not the electoral system as a systems engineering point of view was actually working. We tried very hard to be, and still are, non-partisan. We're not about telling either party or the courts what they should be doing. The question is, can we approach this from a scientific—either social scientific or engineering—perspective about how to make elections better?
ZIERLER: Is it unavoidable, though, that as scientists and engineers who are after the truth—
ZIERLER: —You know where I'm going with this—you are unavoidably partisan, because there's only one party that currently still adheres to notions of the truth.
KATZ: Yeah, it's definitely getting quite—depressing. [laughs] I don't know of another word for it. To be fair, I don't think it's all Republicans. I think it's a subset. But yes, it's a problem. It has become partisan, where literally the type of stuff—now, if you work on anything which tries to improve elections in an objective way, as best we can say, there are fights. In part, it goes to the objectives, and this is why you can't ever—and that's why it's political science, that there's politics involved. It goes back not to just how to run elections, but political scientists often think, and democratic political theorists often think, that you should maximize enfranchisement, maximize people's ability to get to vote. This has historically happened, too. Different sides have different beliefs over what the electorate should be that would get them elected.
This is not new, by the way. This went on in the late 1800s with the Chinese Exclusion Act, with whether or not we would allow Irish and Italian immigrants, how quickly we would let them come to vote. In fact, actually, a lot of what we call the Progressive Era reforms in politics, they were sold as being the social scientific "let's make things better" and in some ways they did. That was the introduction of what we call the Australian ballot, the secret ballot, state-printed ballots. But as my colleague who I think you might have talked to already, Morgan Kousser—he wrote a book—which is that the progressives weren't in this for good politics. In the North, they wanted to disenfranchise these immigrants who were being bought off by these machine politicians, and in the South, they wanted to disenfranchise non-whites; and by the way, even some poor whites.
What looked like very good arguments today, which we like—Americans would find it hard to think about democracy without a secret ballot—well, a state-printed ballot is an implicit literacy test. Again, maybe in the modern U.S., where literacy rates are hopefully in the ninety-percents, it's not a big deal, but that wasn't the literacy rate in the 1890s and early 1900s when these measures were adopted. In fact, how people used to vote were these big—in the absence of state-printed ballots, party agents would print these elaborate and color-coded ballots, and not only could you not change who you voted for, so it was basically an entire ticket—in fact, states had laws that said you couldn't alter the ballot, or if you did, you had to spell every person's name perfectly, and their legal full name. So, you couldn't write down "Jon Katz." Not that I ever go by Jon, but that wouldn't be—or if you wrote down "Jack Kennedy" for John F. Kennedy, that wouldn't work. Now, we take that as a given. There are costs and benefits. Those were done politically. The reason the progressives put this in place was because they were unhappy with the Democratic machines in the Northeast.
ZIERLER: It almost seems quaint to think, in the 2000s, Bush v. Gore, that there was an engineering and scientific solution. Nowadays, is there even? Is there a way that we can have engineering and science suggest that there are two parties that agree on the same standards and ground rules, or are we past that at this point?
KATZ: I always want to remain optimistic, but yes, it seems at this point quite difficult to see how we get back to this world where there was agreement that we should have democratic principles. Now, there are some good things. We have seen now some Republicans actually concede their election losses and stuff. So that's better, but we'll see. I think the 2022 and 2024 general election—particularly the 2024 election with the presidency—will be very interesting about what happens. Again, it's also tied into the fact with—political compromises were made at the founding of this country—well, the re-founding with the creation of the Constitution—about the Electoral College and how geographic—there were always fights between urban and rural voters. That's in fact why we have the Senate. The concern was the rural residents of places like Rhode Island and North Carolina were concerned that Virginia, which was the behemoth, if we did everything by purely popular vote or purely by allocating congressional seats by population, they would just dominate everything. Now we know we're in this crazy world where California, with one fifth of the U.S. population, has two senators.
ZIERLER: Just like Wyoming.
KATZ: Just like Wyoming. No, better yet, just like North and South Dakota, which were in fact divided—the reason there's North and South Dakota was in fact to make sure that there would be four senators. Four Republican senators. [laughs]
ZIERLER: That has worked out well!
KATZ: That has worked out well. So, it's an interesting thing. By 2050, a majority of the Senate will be controlled by 30% of the population. That's fine if we all had reasonably similar policy preferences. That doesn't seem to be the case. So, I do think that there's trying times ahead for democracy. I like to think that maybe cooler heads will prevail, but we'll see.
ZIERLER: Whether it's Donald Trump or it is simply Trumpism, do you think that there is a solution for a shared standard so that both parties agree before the election, "Here's what we're looking at and we will respect the outcome if these things are met at the end?" Or are we already beyond that?
KATZ: I think if it's Trump, we're beyond it. Clearly Trump and his faction of the Republican Party are clearly outcome-based: "The system must be rigged if I didn't win." That's a tough—
ZIERLER: Even if he goes on record saying, "This is what I'll agree to" beforehand. Or he wouldn't ever do that?
KATZ: One is I don't think he would go on record, and I think even if he did, I don't think consistency is his thing. Now, to be honest with you, I don't think it's likely Trump is the nominee of the Republican Party.
ZIERLER: From an electoral perspective or a legal perspective?
KATZ: Both. He is getting much older. So is President Biden. He is in legal hot water. We've already seen that his power seems to be waning if you look at the effect of his endorsements. There's clearly a core of constituents in the Republican Party which are very much in favor of Trump, but there's lots of forces on the other side trying to minimize him. Then the usual things which you see is coopting, right? Governor DeSantis in Florida is definitely trying to take the mantle from Trump. Is he better?
ZIERLER: And outdo him in many ways.
KATZ: Outdo him in many ways. The question, though, is how he would sort the electoral He's being quite aggressive on the Florida congressional redistricting. There was another ruling today throwing out the district court's—so, we'll see. It's hard to say. I think Trump is a problem. Whether or not his successors are a problem, I think a lot will depend on who the Republican nominee is, and what happens. From those of us who study elections, it's hard to know how well, but if you were betting—and I do bet—Republicans should do pretty well in 2022, and probably 2024, both given the seats that are up and the state of the economy and COVID and other things.
ZIERLER: And China/Taiwan—.
KATZ: And China/Taiwan, and yeah, all this sort of stuff. So, we'll see. This is why I never say I'm a political scientist, because if you're on a plane, people all have their opinions about what politics means. This is all fun, and I'm happy to engage in this; this is great, but it's not what I do research on. What I do research on is actually asking, what are questions we can actually answer? How can we actually bring some scientific measurement to some of the things we care about? Could be redistricting. How do we think about measuring both the impact or fairness of redistricting? We can think about how to forecast election outcomes. That's another thing I care a lot about. Those are things which clearly have political uses and political interpretations, but at the core, what I do as a researcher is about these tools, and about these techniques, and then I'm happy over drinks to think about like the state of democracy and which party—we've been couching this conversation as the Republicans are bad, and they are. I would say they're worse in terms of election issues. But the Dems—
ZIERLER: They have their own problems.
KATZ: They've got problems, even problems like—I worked in legal cases, for political lawyers, which has now become a big thing. Election law used to not be a big thing. You've heard of Baker v. Carr and Wesberry v. Sanders, famous for what was at the time called "one man, one vote." One person, one vote. That's actually not why they were important. They basically undid a century of precedent which basically the Supreme Court kept saying, "That's a political question. We stay out of the political thicket. It's up to Congress and it's up to the state legislatures to decide. They're the judges in their own elections." It was really not until 1962 and 1964 that the courts actually got involved in cases at all, and their initial cases were pretty apolitical. In Wesberry v. Sanders, which was about legislative stuff, there was egregious malapportionment. In Illinois, the ratio of smallest to largest district was eleven to one.
ZIERLER: Wow. [laughs]
KATZ: This is not a little malapportionment. The good thing is that at least malapportionment was something relatively easy to measure. What's the population? We can do it. But in some sense that was the nose under the tent. Because then courts got into things which were much more difficult to measure. Then we move into the Voting Rights Act and the interpretation of the Voting Rights Act. The Voting Rights Act doesn't really become effective at changing anything with regard to redistricting until 1986. That's when the Supreme Court in this Gingles decision defines what it means for elections to be racially polarized. But it's a very odd standard.
The Voting Rights Act is very odd. I don't teach classes in democratic theory, but democracy is about a process. It makes sure that you and I, or anyone else, has sort of equal opportunity to advocate and to influence electoral politics and policy outcomes. That is what democracy is about. That is not the Supreme Court's ruling of the Voting Rights Act. It's outcome-based. The Gingles decision laid out a three-prong test. It said, if whites are politically cohesive—that is, they vote on average for the same people—I'm sorry, if the minority group, let's say African Americans, are cohesive in their voting preferences, and whites are also cohesive in their voting preferences at odds with the minority group—African Americans ability to be elected—such that those African American candidates of choice cannot be elected, and it's possible to draw a district where African Americans or whatever group you are suing for could constitute a majority of the district, you have to make that happen. So, it's really outcome-based. You have to make districts such that African Americans are basically reasonably assured of their opportunity to elect a candidate of choice. That's not really democratic theory; that's a quota system. That's fine, we can have quota systems, but that's what it is.
But now this becomes a much more difficult task, because—like I said, malapportionment was pretty easy to define. We can argue over how much malapportionment is too much, right? The Supreme Court early on in state legislative cases said, "Ten percent." There can be deviations of ten percent, which is still pretty big, but given technology at the time, that seemed reasonable. Now basically the accepted wisdom is the safe harbor is basically 5%, although most states now, given modern computing, they use 1%. If your map is not malapportioned to greater than 1%, you're presumptively presumed—that's not a reasonable challenge to the map. The argument for why Congress needs to be equally sized was not based on just on the—that argument was based on the 14th Amendment, equal protection; the law should treat all of us equally. The argument against malapportionment is that in districts that are larger, my vote counts less than your vote.
There's a second basis for equal size districts for congressional districts, and that's Article I, which says that Congress shall be apportioned according to the population of the states. In fact, the court ruled pretty early on that basically you can't have any deviations, to the point of absurdity. For example, in 2002, the Pennsylvania congressional map was thrown out with a population deviation of 17 persons in 650,000. The Census is not accurate to 17 persons in 650,000. [laughs] Today, actually something else I work on and am interested in, with differential privacy, the data is definitely not accurate to 17 persons. That's just an absurdity. That gave wiggle room. Voting Rights—now it became harder. What evidence do I need to show that groups are cohesive? Then if I show it's cohesive—but the Supreme Court has also said in countervailing claims that race can't be the predominant reason you draw districts. So in the 1980s, if you were a jurisdiction—a state, a county—drawing districts and you could show there was racially polarized voting and you drew a map such that at least the minority voters had at least some districts that they were likely to win, you were good. Safe haven. You couldn't be sued under Section II of the Voting Rights Act.
Fast forward to the cases in North Carolina in the 1990s, on the one hand they said, yes, the Voting Rights Act says that you can't basically do things to purposely disenfranchise, to dilute the voting power of protected minority groups. On the other hand, race can't be [laughs] the predominant reason you draw a map. So now, you're damned if you do, and you're damned if you don't. If you don't draw a map that has minority-minority districts, you're going to get sued by groups like the NAACP or MALDEF or the ACLU or the League of Women Voters. But if you don't do that because you're citing the cases that said, "Oh, race can't be the predominant reason we're drawing districts," then you could be sued by a group who didn't like the map who said that race was the predominant reason you drew the map. What does this do? It means that, first of all, judges now have a lot more freedom about how they factually find and therefore can rule in the case, and a lot more litigation. When I first started doing expert witness work in election law cases, there were cases around the federal Census, and then basically nothing. Now, literally there are hundreds of cases. Election law used to be something that large law firms did and it's basically a loss leader.
ZIERLER: As a public good, essentially.
KATZ: No. [laughs] As something they did for their clients. The law firms are partisan. Most of the Republican work is done by Jones Day and BakerHostetler, and most of the Democratic work was done by Perkins Coie. They still do some. And what was Perkins Coie, but Mark Elias—very famous now. In fact, he got too political and got too out there, so the rest of the partners—Perkins Coie is a very large law firm; it's a 2,000-person law firm—they actually negotiated a deal where he would carve off his election law group to basically now it's called the Elias Group that would separate from Perkins Coie.
ZIERLER: He was too partisan, basically?
KATZ: Yeah, because Marc—I know Marc—Marc became a lightning rod, and so some of the corporate clients didn't want to be represented by Perkins, because there was this de facto—Marc was the attorney for Hillary Clinton. He's out there. The other law firm that does a lot of Democratic work is Jenner & Block. But literally, Baker Hostetler, for example, on the Republican side]—Mark Braden was the person—I know Mark—was the Republican counsel to the Republicans in the House of Representatives. His partners at Baker Hostetler agreed—his practice didn't really—it probably broke even, but it wasn't a money maker. But when their corporate members needed to like meet a member of Congress, Mark's rolodex—they took Mark's call [laughs] and so he could do introductions. So really election law started out as, "This is a way to get into the axis of power for our clients," but now, it's a moneymaker. Even more so in California. California has turned election law into what's called plaintiffs' bar, basically ambulance chasers, because in the U.S., contrary to, for example, in the United Kingdom, we don't have a loser pay system. In the United States, if I sue you and lose, you pay your legal fees, I pay my legal fees plus whatever the judgment is. In California, if I sue a jurisdiction under the California Voting Rights Act and win, I'm automatically entitled to what are called "reasonable attorney's fees."
ZIERLER: This creates an incentive.
KATZ: Right. So Rex Parris—I don't know if you've ever driven out to Mojave down the 14, you go through Palmdale—there are giant signs for Rex Parris He's literally—his firm is an ambulance chaser, and he funds this guy who is a single practitioner in Santa Monica basically to go off and they sue. They won't actually—
ZIERLER: Winning is irrelevant, almost.
KATZ: Yeah. They basically press these small—so like the first case I was involved in, they were involved in, was the Upper San Gabriel Water District. You happen to be in the Upper San Gabriel Water District, as long as you're not in Pasadena. I don't know if you live in Pasadena.
ZIERLER: South Pas.
KATZ: Me too. So, you are in the Upper San Gabriel Water District. It's a board. You probably don't even know that you voted. Anyway, they were sued in 1999. Why? Kevin Shenkman was basically asking a demand letter of basically $50,000 for them to move to a districted map and to pay him off. How big can these get? Santa Monica was sued under the Voting Rights. Their city council was sued under the Voting Rights Act. They hired Gibson Dunn, one of the most storied—it's the biggest California law firm—to defend them. Santa Monica has lost, currently. They appealed to the state supreme court. If they lose there, they are currently on the hook for $11 million in legal fees.
KATZ: Not small business. [laughs]
ZIERLER: Where do you slot in?
KATZ: I'm interesting. I've worked both for Democrats and Republicans, because I view my role as an expert witness, which is actually what the federal rules of evidence suggest.
ZIERLER: That being non-partisan is important.
KATZ: Yeah. I analyze data. I understand that lawyers are going to spin it, and I often will be hired by a client, I'll do an analysis and I'll say, "You want to fire me." That's actually perfectly fine under the American legal system. Then I'm not a testifying witness; I'm just a consulting witness. Then my work is covered by what's called attorney work product. I work for both Democrats and Republicans. I analyze data. I tell them, "This is what the data says." But as I said, I'm unusual, because lawyers work either for Dems or for Reps, and most academic expert witnesses work only for Dems, or work only for one side. I think that's wrong, actually, but that's the world we're in.
ZIERLER: What are the IP or ethical considerations being a Caltech professor? How does that work?
KATZ: Oh, it's like everything. I am concerned about that. All of the cases I work on are disclosed. It's very clear—when I write an expert opinion, it is mentioned in my biography that I am a Caltech professor, but when I write the report, it doesn't say, "Jonathan Katz, Caltech." It just says my name, and my consulting group is called Katz Consulting. But you're right; it's a concern, and it's a growing concern. This has already been a concern for the biologists and chemists for a long time, founding companies, and it's the same sort of issues. I make sure I don't use Caltech equipment. I don't use Caltech resources. Now, for example, if I hire an RA, like a research assistant to help me with a case, I have to file a request with the Dean of Graduate Studies office to make sure that this is all on the up and up, what they're being paid, that it's not implicating them. So yeah, you have to be very careful about what you do. Like I said, I'm much more comfortable because I work for both sides. I actually think I'm doing a public service, which is, I analyze the data, and I tell the court what the data says. Clients obviously have interest in what they say, but we have an adversarial system. But for example, Morgan Kousser—Morgan does this, too—he only works for Dems. Only.
ZIERLER: Ideological reasons.
KATZ: Ideological reasons. I won't speculate on Morgan in particular, but my concern with people who have an ideological—basically have a horse in the race—they know the answer to the data before they ever have looked at it. It's a standard problem in all statistical analysis. The Twain quote—"There's lies, damn lies, and statistics." The more formal discussion in the statistics literature is called p-hacking. If you give me enough degrees of freedom based on what cases I include, how I do the model, I can get any result I want.
I want to give you a little aside. When I was a graduate student, I taught every summer at the University of Michigan. There's something called the ICPSR, which was actually the first big data repository but it does other things as well. They had a summer school to teach social science graduate students statistical methods, because a lot of places, especially in the 1990s, lots of places didn't have people like me. So I taught a course. The final problem set for the course was, I gave them data from the Alabama death penalty case, which was a case that went to the Supreme Court, about whether or not the death penalty was applied in a racially discriminatory manner. Were Blacks or non-white defendants more likely to receive the death penalty, all things equal? All things equal is always hard. In the actual data, in the data that was presented in court, there's pretty clear evidence that the death penalty in Alabama was applied in a racially discriminatory way. We had the actual data, but we modified it. The data set I gave the students, there was literally no statistical relationship between race of defendant and the likelihood of getting the death penalty. Over half the students turned in analyses which showed a statistically significant relationship between race of defendant and—
ZIERLER: Because of preconceived notions?
KATZ: Well, because what happened is, if you ran the model you think you should have run, you would get this null effect. They didn't believe it. So they would put in different terms. They would restrict some of the data. They would do anything to try and make sure that—to undo this. It's not that hard, as my students showed you. Even though I told them after the fact that in the data I gave them, there was in fact no relationship. I do the same thing in my IDS applied data class. You have to let the data speak. It's really hard. It's very tempting to say, "I have these strong views about what the right answer is." As a scientist, as a statistician analyzing data, that's terrible. Because if that's your goal, I can always make the outcome.
This in part goes to this huge replication crisis, first in psychology but in general. One of the things that has happened over my career—it's something that I have actually encouraged—Mike Alvarez and I were longtime editors of Political Analysis, which is the journal of the Society for Political Methodology. Way back in 2010, we actually required every published article to submit all code and data to an archive, called Dataverse—it's a publicly maintained archive—that we control, that require you to release all your data. We had an RA who would actually verify—it's not perfect. It's not saying that this verifies in some other set of data or in some other augmented thing, but at least, at the very least, we can say that we can replicate every figure and every finding in the paper we're publishing. The problem is that as people try to replicate, they find lots of these effects don't happen.
Why does it happen more often—? The physicists like to say, "Oh, that's just the social science world." It's not. The problem with social sciences is that most effects that we look at for policy or other types are pretty small. With small effects that are noisy, there's a cherry-picking problem. There's the standard history—scientific problem that it's very hard to publish null results. Essentially if I get a large statistically significant finding that is unexpected, that gets published. "Oh, that's interesting." Well, no. Chances are you were either unlucky in some sense, or this is cherry-picking. That's the problem you have. This is the general problem, that we have small effect sizes in social sciences, and so if you're not very disciplined and very careful about how you conduct your analysis and report those findings, it's very easy to run down this rabbit hole of getting any finding you want. Either it's interesting because it's so large and—this happens all the time in—a big thing now in behavioral economics are these nudges. If we do these small things, we can get you do these things. Literally, companies are founded on these. It turns out that these are all really modest effects. If I actually graphed out all of the findings, all the findings are statistically significant and positive. Because first of all, if you try to publish a negative finding, a null finding, all the people who believe this sort of jump down your thing, and the editors say, "Null results? Maybe it's just because you have an underpowered study." This is a fundamental problem in quantitative social sciences.
ZIERLER: Is AI a solution? Does it remove some of the bias that you see from people looking at the data?
KATZ: You'd like to think so, but no. Because as we now know, people here who work on it and others, there's bias—the problem with AI is the AI learned—most of AI is basically pattern recognition. The best example I know of is Amazon hiring. They had years and years of hiring decisions, and they machine-coded characteristics about you, like where you got your degree, what it was in, experience, and they put it in this giant database, and then they also had what's called label data, and the machine learned what they called label data. Did Amazon hire this person or not? Maybe they had some other ones. Now, we fit the model on this label data, and we now use it so when we get a new applicant for a new position, we can use this AI model to predict who's the best candidate. It turns out, shockingly, that Amazon basically hired no women. So, what did the AI learn? The AI learned that women don't make good Amazon employees. Now, is that true? I doubt it. So you have to worry about biases built in, because typically almost all AI applications of these things are based on labeled data. So you have to ask, what was the population the AI learned on?
ZIERLER: AI is only as unbiased as the people who make it do what it does.
KATZ: Right. That's the main known problem. Then there's even what question you're asking the AI to answer, which also has this issue. This is coming up right now. Mike Alvarez and I were tasked by Kevin Gilmartin, who is the VP for Student Affairs—Caltech currently has a moratorium on SATs, ACTs. We don't require them. This was a response to COVID, because people in many countries couldn't take SATs or ACTs, or even if they could, whether or not they wanted to sit indoors with—so lots of places undid it. MIT just now, they're going to go back to requiring standardized tests. The UC system has gone the opposite way. They say they're going to maintain the moratorium.
Caltech is now asking—so we actually analyzed the data. We got the last five years' worth of data, and we basically showed that SAT scores are actually a really modest predictor of anything about performance in the first year. That's the only thing we can measure easily. But in our conversations with other faculty who are on the Admissions Commission—even though, here's the analysis of the data. We've actually analyzed the data, and we've done so accurately—they're like, "No, that's an important piece of information." What are they missing? What we basically showed in our report, because it's the only thing we can currently analyze, is that the SAT is not very predictive of performance at Caltech. In some sense, why would you want something that's not very useful for performance? The benefit would be if it was a good predictor of performance at Caltech, at least the first year. What they don't focus on is what's the cost, which we really haven't measured directly. What we do know anecdotally is that in the last couple years, Caltech has gotten a lot more diverse applicants, and a much larger set of applicants since we've dropped the testing requirement. Maybe that's a good thing.
ZIERLER: It's a good thing depending on what your goals are.
KATZ: Right, what your goals are. I think it's potentially a good thing. But this is the tradeoff. But they are so focused on—in part, it's always what "I" did. I took the SAT, and I did really well on it. Why? I'm an upper middle class white guy who went to private schools my entire life. I've been taking standardized tests since I've been in kindergarten. The fact that I did well on the SAT was perfectly predictable. Turns out that actually my wife, who went to the University of Chicago and she went to a large public high school in Rhode Island. She in fact did so terribly on her SATs that the Admissions Committee admitted her to Chicago early but they just—she actually talked to the Admissions people—"We just assumed that you were like off on a number or something." It was so inconceivable given—she was like sixth in her class out of 700. She had had all these things. She had an incredibly strong record. And this SAT was just so terrible. Now, it turns out Natasha is just terrible at standardized tests. That happens. We have these biases. "This is what we did. We think it's a bit of information." In some sense, we don't want to throw out information. But you have to think, what's the flip side? Is that information valuable? And what does it cost you?
We already have a lot of self-selection in the application at Caltech. We know this. We definitely know this with women applicants and with people of color. We have a mindset about what it means to be a Caltech student. If you're going to make it costly for them to get here, they're not going to come. But this is interesting where my colleagues—they're all scientists—we've given them quantitative analyses that shows pretty definitively there's no real benefit, or no benefit unless they can tell me some other benefit for this test. Now, MIT's claim is that it helps them identify these diamonds in the rough—these, say, students from underserved—I was fortunate. I grew up in New York. I started taking math classes at NYU when I was in ninth grade, tenth grade. One of my advisees this year is a Hispanic guy from a suburb of El Paso. He did not have options to go take college classes at something like NYU's Courant Institute of Mathematics to show that he was good at mathematics. The MIT report, if you read it—it's actually worth a read—is that they claim that this is going to help them identify this. Jarrid Whitney, our VP for—I forget his title now—he used to be the Director of Admissions; he's now above that—it's not so clear that that's true. Again, all the testing biases that we know about. What makes you not likely to do well—if you're not a white guy who is from a native English speaking family with college educated parents, you're not likely to do well at the SAT, no matter how smart you are. There's a skill that's separate from intelligence to doing well on standardized tests.
ZIERLER: Created by the society that seeks to push forward a certain class of people.
KATZ: Right. Exactly. The SAT is like our AI example. The SAT is very good at doing what it does: identifying people like me. For me, I'm very happy to approximate answers, and I'm very happy to jump around. On the SAT, it's disastrous if you get caught up in a question; one question is not worth it. You should just move to the next question, and if you have time at the end, you can go back to it. Natasha, she's linear, and she wants to get problems right, so literally she doesn't finish the test. You're rewarded for that. You're more rewarded for that than getting a wrong answer. That's a risk-taking behavior, that's a learned behavior, that's—anyway.
The Science of Social Science
ZIERLER: The last topic I want to touch on today as it relates to disciplines and academic tribalism and methods and all of that—maybe at the most general level, the term "social science." There's a range, among social scientists, where on one they say, "The modifier is ridiculous because we do science. Look at what we do; that's science." On the other end, it's, "Of course it's not science in the way that biology and chemistry and physics are science." For what you do, where do you see yourself on that spectrum?
KATZ: I want to actually change your question. This is a little bit an issue. I would say Caltech, the social scientists, all of us, we definitely anchor the science—if you think about this as a continuum, we definitely anchor the science in it. What I mean by that is that we can think about generalizability. That is, from data, from theory, we can build generalizable statements. Now, I agree, and we all agree, it's not physics. It's not mechanistic. That, I agree. But in the social sciences—
ZIERLER: What do you mean by mechanistic?
KATZ: Deterministic is maybe a better word for it, where somehow I can—Asimov's Foundation series. It's not like I can predict history or predict what you're going to do. Let's go to elections, what I mean by it. Any particular election, I can't tell you exactly what is going to happen. There are idiosyncrasies, weird candidates run, scandals happen, a shooting happens in your district and you say stupid things before the election. I don't know.
ZIERLER: This is more chaotic than a scientific system?
KATZ: Well, no, it's just unmodeled. But I can tell you on average, if you ask me to predict how many seats the Dems are going to win in Congress, I can do that pretty well. I can tell you with some probability what's going to happen in every district, but there's going to be outliers. There's going to be things that are unmodeled that don't work. That's what I mean by it. So it's not mechanistic. It's not like even if I had perfect data, that I could somehow perfectly predict outcomes. But we can do pretty well.
Now, why I want to change your question—the alternative to the social science take on social sciences is sometimes called interpretivist. It really goes back to this idea that all you can do is thick description, that there's no generalizability. This pervades both the humanist disciplines and social sciences. Again, not at Caltech; at Caltech, we are all, like I told you, firmly anchored in the social science end of things. But at Chicago, for example, one of my dear friends is Lisa Wedeen. Lisa Wedeen is a scholar of Arabic politics. I'm actually bummed; she offered to arrange a trip to Yemen, and we at the last minute got cold feet, and now Yemen doesn't really exist anymore, so I'm kind of bummed about that. Lisa's view is that what I do is not philosophically possible, that I can't do anything as an observer of politics but richly describe what agents are doing. There's no notion that I can generalize.
Chicago happens to be ground zero for this. It varies quite a bit in disciplines. In fact, in anthropology, it actually caused a split in the field. Anthropology now is actually two separate departments now. There's cultural anthropology. They believe strongly that all you can do as an anthropologist is observation, ethnography, no generalization. Then there's physical anthropology, who are much more anchored in the scientific end of things. They talk about genetic development. They talk about physically why things develop. They're on the science end of things. Sociology, my wife's area—or was, when she was at graduate school—they are probably two thirds cultural. That is, they don't believe you can do social science And the one third in sociology that is typically very scientific are the demographers and others who study those sorts of questions. In political science, it's like 50/50. About half of political scientists think you can do social science, and half don't, or maybe 60/40. Economics is the other. All economists think that economics is a science, and there is much more homogeneity in a belief about what are appropriate methods, what are appropriate questions to ask, what are ways of doing it. I think that's the continuum.
Caltech, we have definitely chosen that we're on the social science side. In fact, I don't know if you've talked to Jean Ensminger yet? You should definitely talk to her. In fact, why Jean came to Caltech was because she is a cultural anthropologist who actually is on the science end of things. What she does, in fact, is for example run economic experiments in field sites in Kenya. The other thing that Jean will tell you is in anthropology—anthropology grad school is incredibly hard, because typically you study, and then you have to spend years on your field site, typically in remote and difficult places. In fact, most anthropologists almost never go back to their field sites, so they all become armchair theorists. Jean, and actually her predecessor here at Caltech, Ted Scudder—I don't know if you've ever talked to Ted; I don't know how Ted is doing—she and Ted have been going regularly back to their field sites for 30 or 40 years.
Now, Jean can't, because literally there was a contract taken out on her life, and the security system has gotten—her field site is in Kenya, but very near the Somali border. We were supposed to do this big project, between COVID and security, where we were going to—she has this great data—so she basically has the census, for the last 30 years—we know how everyone's family is connected, but we also know who they're connected to, so there's this interesting question about corruption, and there are some ways of measuring and these economic games to do corruption. One question we'd like to know is about how that flows in a network. But the problem almost always with network studies is that I don't know the social network you live in. The good thing here, because these are isolated, poor communities, is that we actually know exactly their network, to reasonable accuracy. This was going to be this cool project, and we got funding for it, we had it all set up, we were starting to interview some companies that could do surveys in these remote locations, but then the security situation got terrible in Kenya and then COVID hit, and now Jean's retiring. [laughs] Jean actually was a refugee here. In fact, she has told us that we should not hire any anthropologists, any cultural anthropologists. [laughs]
ZIERLER: [laughs] Within social science, either from your degrees or your methodology—we've covered so much already—there's political science, there's computer science, there's economics, there's finance. If I had to pigeonhole you, what's your home discipline, would you say?
KATZ: Technically my home discipline is political science, although like I said—
ZIERLER: Not statistics?
KATZ: No. Here's what is different about me, and I guess something we'll cover at some other point. I generally develop statistical tools because I have a question I want to answer and the tools don't exist to answer it. Whereas statisticians typically start with, "I'm just interested in this set of estimators" or "I'm interested in expanding this set of estimators," my interests are typically substantively drawn, not necessarily statistically drawn. That's why I would call it political science. It's the thing I know the most about. The substantive area that I have the most knowledge about are things about elections. I can tell you lots of things about American elections. I've done other things, but that's the substantive area that I know the most about.
ZIERLER: Statistics is a tool for you?
KATZ: Yeah. I'm unusual in the sense that I spend some of my time—once I get interested in the substantive question that got me interested in the statistical tools, I will then go down that rabbit hole [laughs] and literally build the tools I need. Most people don't do that. Most people don't have the mathematical background. It's a rare combination. Most people don't have—it requires a fair bit of technical background. I would say lots of people go to graduate school in the social sciences, particularly political science and sociology, because they didn't want to do math. My undergraduate degree is in math, applied math. They don't have the technical tools.
Although it's interesting; even that is changing. The technical tools that are available and developed in the social sciences are actually in certain areas outstripping what goes on in some of the other areas, again because typically they're motivated by the substantive area. Necessity is the mother of invention. Statisticians will often answer a question not because that's the question we want to answer, but that's the one we have the mathematical tools to answer. I'm trying to think of examples that I can tell you easily. Lots of people in statistics and econometrics, which is the development of these tools, you get very math-heavy, you get very excited. So, if I can prove this estimator is consistent, even if it's not what I really care about, that doesn't really matter. Or, I can weaken those assumptions. Typically to prove that a statistical estimator works, I write down a set of conditions that need to be met in my data. Now, in principle, it's often very difficult to tell, but a large set of things that statisticians do and econometricians do is they say, "All right, I proved that this statistical estimator has these properties under these sets of assumptions. Can I weaken them?" For example, when I was in graduate school, I learned—do you know what a Markov process is?
ZIERLER: Yeah, yeah.
KATZ: A Markov process is a form of time-dependence. It says that basically a Markov process of order p means that I only have to worry about p periods before. After that, I don't need to worry about those; I can ignore them. The system is totally reset if I know these p sets of data. The easiest one is like a random walk. A random walk is I just need to know where I was last time, and then with some positive probability I move to some other square. Now, there are other forms of dependence. That's a form of temporal dependence. There's another form of dependence, which you probably have never heard of, called epoch dependence. I'm not even going to define it for you. Epoch dependence is weaker than Markov dependence.
ZIERLER: What does "weaker" mean in this context?
KATZ: It's a mathematical notion of weaker. Anything that is Markov-dependent is epoch dependent, but not everything that is epoch dependent is Markov-dependent.
ZIERLER: One is a subset of the other.
KATZ: One is a strict subset of the other. Literally, in the 1990s, people would prove results about consistency of various estimators—Markov is easy. Basically what you need to often prove results is you need Central Limit Theorems, and laws of large numbers. Central Limit Theorem just says that if I add up a sum of random variables, a large enough number of them, their mean is going to be distributed as a normal distribution, Gaussian distribution. The law of large numbers just says, suppose that—the expected value of a random variable—that's like the population, that's what I expect it to be, or that's like its average—and what the law of large numbers says is if I take a sample from that population, with this data-generating process, with this expected value, that if I take the sample mean of my data, just take the average, the law of large numbers says that if I have a large enough sample, that my average will converge to the population expected value. These two theorems are the workhorse of proving most statistical results. Basically what it means is that I can say something about what's the distribution of my estimator, and whether or not my estimator is converging to the truth. Okay. But it turns out that in order to prove the law of large numbers—and it's going to be pretty obvious—laws of large numbers; let's leave the Central Limit Theorem out—laws of large numbers means the data can't be too dependent. Basically what the law of large numbers says is if I take a sample and they're independent enough, then I'm getting enough independent data that if you let me take a very large sample of it, I'm going to get the truth. I'm going to get the true average. But think about it if they were perfectly correlated. Suppose that after I took the first draw, the next one was correlated at one. It had to be exactly the same one. Well, then I could keep adding more and more data, but I would never know the true mean, because once I've drawn my first one, there's no new information to be drawn even if could draw an infinite samples. That's how you think about dependence. The question is, how much dependence can I allow such that the law of large numbers still holds? Like I said, there's all these—I can relax dependence, that is I can allow more and more dependence, to prove laws of large numbers.
Literally I had colleagues who their whole career was just like, "Let's prove this result under weaker and weaker assumptions." By the way, weaker and weaker assumptions which I could never verify. I've told you the difference between epoch—I haven't really told you the difference, but I've asserted that epoch dependence is weaker than Markov dependence. It is basically impossible in a finite data set—so a finite number of observations—say I ran an experiment or I had a random number generator—for me to tell you whether or not a data set meets epoch dependence versus Martingale dependence. No way. So yes, I can prove—it's a neat mathematical result, I get to show my chops as a mathematician and statistician, but I actually haven't improved my understanding of applied questions, because these are assumptions without distinction for the applied researcher. That's very much the norm in much of statistics and econometrics.
ZIERLER: Do you need to be or do you want to be constrained by theory that serve as anchor points for your data?
KATZ: That's another big believer at Caltech. We are definitely theory-based. This is where I disagree with the machine learning people. Machine learning people think that if you just give me enough data, I can answer any question. The answer is, going back to these biases, if you don't have a theory about how your data was created, like what people are doing, then you miss these biases. For example, yes, if you gave me a large enough sample of Amazon's hiring decisions and characteristics about people they chose to hire and not hire, I can get a machine learner to perfectly predict a future person, whether or not Amazon would hire them. But if I didn't have a theory about how Amazon chose to hire, and that they for example didn't basically hire any women, then yeah, I perfectly replicated this, but without a theory about how the world works, I wouldn't know that this is potentially a bad estimator.
I think it's very difficult to do social science and quantitative social science without having some understanding of the mechanism, because that's how you know the problems in your data, the selection biases, the endogeneity, which sort of plague—endogeneity is this fancy word for saying—that just means the situation matters. For example, one question I spent a lot of time early in my career working about is incumbency advantage. Political scientists don't often agree on a lot, but we agree that incumbents do better than non-incumbents in elections. In fact, in congressional elections, we agree on what that number is; it's currently about 3%. I could have lots of theories about why incumbents do better than non-incumbents—it could be that incumbents get resources from being members of Congress, they get staffs, they get to send mail home—that look like substitutes for campaign resources. It could be that maybe candidates who are successful are just better campaigners. Those have very different consequences for democracy. One is I'm giving a leg up to the incumbent who may or may not be better. The other is it's just like a contest, and these are better politicians.
It turns out, how might we think about—if you don't think about theory, I have this estimate, but whether or not this estimate is important, whether or not elections are not or are competitive, I can't answer that without having an understanding about the causal mechanism, and I can't have a causal mechanism if I don't have a theory about how the world works. I definitely have no hope of testing a causal mechanism if I don't have a theory about how the world works. Gary Cox was one of my advisors, and actually he's a Caltech alum both undergrad and grad. We actually looked at what was fortunate for us, unfortunate for the members—some members die in office. Presumably their choice not to run for reelection is uncorrelated with why they died.
KATZ: Now, it takes some work, because there's not many of them, but if you look at how the person who replaces them from their party does, turns out that incumbents, if you look in the 1950s, incumbents who die or who voluntarily decide to leave, like to spend more time with family, there's no real difference between how their successor party candidate does. But in the modern area—the 1980s, the 1990s—there's a huge loss. If you strategically decide to retire to spend more time with family, the person who replaces you does way worse than the equivalent candidate who chose to run as a Democrat, say, because the Democrat died in office. They've gotten better about reading tea leaves. Now, is that good or bad? Need a theory, but it at least suggests that something has changed that we need to understand. A large part of what social science at Caltech is, and to understand policy, is about all this endogeneity, that these things have unintended consequences. Sometimes, the people you're studying either consciously or unconsciously react to these things. Now, do congressman actually know? Well, yeah, the answer is they probably know their district pretty well, and they know what's going on, and they know—"I'm reading the tea leaves in my district and mmm, I'm outta here." Because you don't want to lose. That can be bad, costly.
Same thing with my colleagues in the rest of Caltech who are looking at this question about the SAT and they say, "Well, yeah, but what does it cost us?" Well, what it costs us, and you might have to ask, what does the applicant pool look like? How is that affected by this decision? If you don't have a theory about the world, how could I even ask that hypothetical? So, I don't have you do theory-free causal inference. [laughs] Even though the machine learners and the people at Google want to tell me that you can do this. [laughs]
ZIERLER: Last question for today, a generational question. Looking at the development of social science at Caltech in the late 1960s and early 1970s, by the time you joined the faculty here, did it feel like a mature, fully formed program, or was your hire still part of the growth and building mode of social science at Caltech?
KATZ: I would say it was pretty well developed in the sense that we had a brand. Caltech in the social sciences, we did three things: We did mathematical economics, we did experimental economics, and we did political economy. Or really what was called formal models of politics, that is applying game-theoretic models and mathematical models of politics. Caltech was known for these things. For experimental economics and for formal models of politics, we were literally one of the only—we were literally developing those tools and those subfields here. I knew that. I think my hiring and Mike's hiring was a foray—we had some empirical people, but they didn't last. Caltech, because it was a weird place—you talked to John Ferejohn. We had this great set of people who were more empirically minded, although Ferejohn was kind of a theorist at the time he was here. He did some empirical work, not so well. John's a good friend, so I can—he's a member of the National Academy of Sciences, so—no worries. [laughs] There was this early group—John Ferejohn was here. Mo Fiorina, John, Bruce Cain. And on the economics side, actually someone who was on my committee, Roger Noll, who actually had been an undergraduate at Caltech as well. There was this core of people who were doing empirical work, but they never stayed long. John Ferejohn moved to Stanford. Mo Fiorina moved to Harvard. Bruce Cain moves to Berkeley. They were here for a bit. The theorists were really dominating, and so it was kind of hard to be an empirical person here. [laughs]
I think my hiring and Mike's were this turning point where people who were mainly empirical stayed around for a while. Although I flirted—I left, not because I disliked the place here, but because of trying to solve a joint career issue. I had met my wife in graduate school. So, it was a turning point in the sense that Mike and I were the first people in political science, at least in political economy, who were doing really mostly empirical-based work. I would say we were the first extension of—if you think about these three basic areas that we were doing a lot of, that we were known for, and we were literally ground zero for these things, we were the next foray, by building off something we were already doing. We were already quantitatively minded, but this notion that you would do these highbrow, more difficult, statistical models, what becomes now known in social sciences as these more structural or theory-based estimators, that wasn't going on yet at Caltech.
ZIERLER: The brand that you talked about that suggested that there was a level of maturity to the program here, you coming here, that changed the brand, or you worked within the brand?
KATZ: I would say I worked within and extended, moving the fence line a little bit, as opposed to creating a brand-new area here. Again, the brand, especially in the social sciences, especially in political science, we were always in the science, political economy end of things, very theory-driven. Then Mike and I coming here, adding people who are more focused on empirical stuff.
ZIERLER: Has that been the trend line through to the present?
KATZ: Yeah. We definitely have hired more—we hired Gabriel Lopez-Moctezuma. He does fully what's called structural econometrics. People like my colleagues John Ledyard or Tom Palfrey or Charlie Plott, they would come up with these game theoretic models of social interaction, and they might use that, but they didn't actually go out do the empirics. Now what some people do is they then try to actually estimate these parameters of these things. These games are defined by a set of parameters, and these parameters then tell you what the equilibria—what you would expect to see. We'd like to know some principled way of estimating these parameters. They're actually typically very difficult.
People like Gabriel and Mike Gibilisco in Political Science and—I just zoned on my colleague's name; give me one second and I'll come back to it—in Industrial Organization, they actually then take these game-theoretic models, and they actually take data, like from auctions or from—Gabriel Lopez-Moctezuma works on, for example, data on deliberation. His job market paper was about the FOMC, basically the Federal Reserve's policymaking board. They set interest rates. Big deal now, right? One question that social scientists have is about learning, right? It turns out that the debate among the FOMC, the members, the people who actually set the policy, it's very structured, about who speaks when. The Fed Chair speaks last. How does that structure change where people vote and where they learn about what they think future inflation will be or what interest rates should be? We took these theoretical models and we got—but the other problem is more just a—in some sense, when we were starting out in the 1960s and 1970s and early 1980s doing these things, we were a niche. It was a niche. We were the only ones doing it. Now there's political economy. There's experimental economics. Behavioral economics was something a little later that came on. Colin Camerer, and now neuro…now like every big department does this. Stanford, Harvard, MIT. In some sense right now we're at a crossroads. We're teeny, and so I don't want to go head-to-head with MIT and Stanford and Harvard.
ZIERLER: Even though it must feel good that they're picking up on these things.
KATZ: Yeah. [laughs] It's great for our students. I have students who are on the faculty at Stanford. I have students on the faculty of Wash U, faculty at University of Rochester. That's all great. It's great for our students. Not so great as we strategically think of how we navigate the future.
ZIERLER: To maintain that Caltech brand.
KATZ: Yeah. Caltech's social science group has always punched above its weight. That's true across all of Caltech. How do you do that? You do that by doing—what's something you have a comparative advantage? What's something we can uniquely do at Caltech? When the social sciences were being founded here, mathematical economics, formal models of politics, no one was doing that. This was and we were this weird place that was willing to take a bet. Turned out to be right. It was an informed bet. This wasn't willy-nilly. That was true I would say with mathematical economics, and that was true with formal theories of politics. Experimental economics was really fortuitous.
Lance Davis, whose former office we're in now, Lance came from Purdue. His colleague at Purdue was a young Charlie Plott. Lance came here and said, "I have this really smart junior colleague; you've got to bring him here. He's doing all these fancy mathematical economics." So all of Charlie Plott's early stuff, he did work in formal political economy, he did work in mathematical decision-making. Then he was doing these experiments to teach kids about—and then figured out that we can actually learn something about—we can use this to test-bed these theories. And so experimental economics sort of happened because we just chose Charlie, and Charlie just pulled the thread. Experimental economics was more fortuitous than a plan, whereas math economics and formal theory in politics were conscious decisions about what we were going to do. We also always had—I don't want to leave out—economic history was also always a big thing, because again it was this natural—when Lance was hired, he was hired because when they decided to have a social science program, we already had some reasonable historians here, and so it was easier to bridge by bringing in someone who they spoke some common language.
ZIERLER: Economics was the bridge, with history.
KATZ: Right. Then by friendship and by admiration, Lance brings in Charlie. Then Charlie and Lance bring in people like John Ferejohn and Roger Noll and Mo Fiorina and Bruce Cain, Morgan Kousser. Morgan was an interesting guy in history. He was doing this quantitative history when no one was doing—it was not a big deal. Again, it's funny, because Morgan's career has really changed. Morgan early on was actually much more like me. He was developing some of these tools. We had this aggregate data, like voting data. How do we try to make any inferences about what individuals—Blacks, whites—are doing? He was one of the developers of things called double regressions. We don't do that anymore, but in the 1970s, this was a really big idea. That's how we got Morgan here. He was an economic historian, but he was doing it in politics. He was very quantitatively interested in these, in some sense, generalizations, which most historians are not necessarily known—economic historians are known for that, but generalization is generally not the dominant paradigm in the field, right? That's how we got Morgan. Then we hired Rod Kiewit. Rod was probably the first foray—Rod, Mo Fiorina were the sort of empirical people. But like I said, all of them except for Rod left.
ZIERLER: And the rest is history.
KATZ: And the rest is history.
ZIERLER: This has been a phenomenal conversation. Next time, we're going to go all the way back, and we'll go to New York, we'll learn about your family background, and work the story right up to the present.
KATZ: Sounds good.
[End of Recording]
ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Wednesday, August 24th, 2022. It is great to be back with Professor Jonathan Katz. Jonathan, thanks again for having me in your office.
KATZ: My pleasure.
ZIERLER: In our first discussion, we did a very wide-angle view of your approach to research and the trends in the field, some institutional history of HSS here at Caltech. Let's go all the way back to the beginning, the Old Country—
Lower East Side Roots
ZIERLER: —New York. Let's start with your parents. Tell me about them and where they're from.
KATZ: Both my parents were first-generation. They grew up on the Lower East Side of Manhattan. Their parents were both Jewish refugees fleeing the Holocaust. My mother's family is from Austria, Ashkenazi Jews. My father's parents were from what is now Poland, but for most of the time was part of the Prussian Empire, Hungarian Prussian Empire, German-speaking, Yiddish-speaking. My father's father were slightly better off. Actually, in the end, they owned three bakeries in Manhattan, he and his two brothers. My mother grew up quite poor. Literally, when she was a little kid, she lived in a tenement at the base of the Williamsburg Bridge. It doesn't exist anymore; it has been knocked down. As an infant, they couldn't afford a crib, and she actually slept in a dresser. They padded a dresser drawer. Then, they met as kids, in high school.
ZIERLER: All four of your grandparents survived the Holocaust?
KATZ: They did. Two came just when things were heating up in Europe, and Nazism was just coming to be. My father's father actually, he escaped—he was actually in a Polish work camp, but it was actually relatively early on, and it was actually freed by Resistance, and then they got him out of Europe. He never really liked to speak about the details. He had his number tattoo. He did live life like every day was a gift. He tells a story—I don't know if it's true or not, but it captivated me as a child—on the Lower East Side, a guy tried to mug him at knifepoint. He pointed to the number tattoo on his arm, and said, "They've already killed me. Get walking." [laughs]
ZIERLER: Wow. [laughs]
KATZ: His name was Max Katz. He was a great guy. My father's parents owned a series of bakeries with his uncles. My mom's parents ran a produce stand in the Delancey Markets. I don't know how much you know about the history of New York, but there were these stall markets where they were independent contractors. I was very young when their market was still open. Like all New York Jews who were quasi-successful, they retired to Florida when I was probably 10 or 11. Then we'd take family pilgrimages down to Florida.
ZIERLER: But it wasn't one stop to Long Island or New Jersey first? Straight from the Lower East Side to Florida?
KATZ: Straight from Lower East Side to—my paternal grandfather did move to the Island.
ZIERLER: What were their Jewish observances? What were their connections in the Lower East Side?
KATZ: My entire family was Modern Orthodox. My grandparents were actually religious. I grew up Modern Orthodox, but not because my parents were religious in the least. It was just that's how you raise kids.
ZIERLER: Your father's father?
ZIERLER: Tallis and tefillin every morning?
KATZ: Oh, yeah.
ZIERLER: Yarmulke all day every day?
KATZ: Every day.
ZIERLER: Do you know where he davened?
KATZ: I don't, on the Lower East Side. But he did weird things, like he kept kosher at home, but like a lot of Jews, he would go eat Chinese food. [laughs]
ZIERLER: Yeah, that's a generational thing, too.
KATZ: Yeah, it was totally generational. I had my fill of it. I hated Hebrew school, and I was thrown out of three of them.
ZIERLER: Your dad went to yeshiva or public schools?
KATZ: My parents both went to public schools, because neither set of parents could afford to do anything other than public schools. They went to Seward Park, again on the Lower East Side. They went to shul every day, but they went to public schools.
ZIERLER: Where did your parents meet?
KATZ: They met in high school. My father was a year or two older. They got together. He then started NYU. They got married. She did two years of college. It was one of the City Universities of New York. I don't know if it was Baruch, or I forget which one it was. They changed names and stuff around that time. Then, my father was starting a business, my mom was helping him, so she, like many women of her generation, left college sort of two years in. She actually had kids late. My brother and I were born when my mom was 27.
ZIERLER: You're twins?
KATZ: I'm a twin. A funny story—I'm one of the last generation of twins—they did not know my mother was having twins until she was in labor.
ZIERLER: [laughs] I love it.
KATZ: Because sonograms existed in 1967 or 1968, but they were rare, and they would only be used if there was some sort of high-risk thing. My mother was a healthy 27-year-old middle-class woman; there was no reason. We were small, but the conjecture from the obstetrician was that we were basically one back to one front, and so when he listened on her back, he would hear one heartbeat, and when he listened on the front, he would hear a different one, and he couldn't tell that there were two. There was no noticeable—and she never got really big. There is a picture of my mom in a bikini like six months pregnant in Puerto Rico. I'm not even convinced my mother really knew that she had twins, because this is again the era—we were born in the middle of the night, she was pretty drugged up. My father was acutely aware of it, because he then had to run out and get like a second crib, a second everything. [laughs] Although he had a little time. Because we ended up being very small—my brother and I were about 4.1 pounds—we were actually in warmers for a couple of weeks after we were born, because we just were too little.
ZIERLER: Did your father finish at NYU?
KATZ: He finished. He actually started law school but only lasted a year at NYU Law.
ZIERLER: Then you said he started a business?
KATZ: He worked in the textiles and womenswear business.
ZIERLER: The shmata business!
KATZ: The shmata business, exactly. It turns out later I learned his partners were the Gambino crime family.
KATZ: Yeah, his partner Tony—I forget Tony's last name—Tony was a made man in the Gambino crime family. Last I heard, he is serving life in prison. [laughs] But, I've been to his house. He lived in Fort Lee, New Jersey, right over the bridge. I remember this, because 1978 was a weird year in the city. Son of Sam was going on.
ZIERLER: Yeah, that's a gritty time in New York.
KATZ: It was a gritty time in New York, and in the summer, there was a massive power outage in Manhattan. It lasted three or four days. In fact, it got so bad that we actually went out to his business partner, Tony's, in Fort Lee, because it was just too hot. No air conditioning. We lived on the 18th floor. Anyway, my father was in the shmata business. My father is super smart.
ZIERLER: Like in a different time, he would have gone for his PhD or something like that?
KATZ: Maybe. He liked money, so I don't know if he would have ever done a PhD. We haven't talked about me, but when I was finishing up college at MIT, I had a job offer at Bain & Company, for like I think $65,000 a year. That's 1990.
ZIERLER: That's good money.
KATZ: That was real money. I also got into graduate school with an NSF Fellowship. My father thought I was nuts that I was going to go to graduate school and make $15,000 a year, when I could be making $65,000 a year as a consultant, basically living wherever I wanted. He had this amazing ability—he was a con man. That's what my father was really good at. He was very personable, super smart, super quick. He was an amazing painter. Not that he could paint original things; he could copy—again, in another era, he could have been a forger. He was really good. If you showed him a painting, he'd just like paint it. He was a character. I didn't realize it at the time, but my parents were very—my mother didn't realize how shady my father was until probably we were really young. They did a great job of hiding it from my brother and I. Apparently, for example, when I was five or six, my father almost went to prison.
ZIERLER: Oh, wow.
KATZ: He got off, and I know none of the technical details.
ZIERLER: You just know that it happened.
KATZ: Yeah. It happened. They did a lot of things. Like typical parents, they stayed together. When we were in college, after our first year, they divorced.
ZIERLER: So they kept it together until you were out of the house?
KATZ: Yeah, which was a very common thing, especially of that generation. You stay together for the kids. But my mother wanted nothing to do with him.
ZIERLER: Did he straighten out after that one run-in with the law?
KATZ: Never. Never, never, never. He actually—things got bad when they divorced. He actually defrauded me. I learned when I was in graduate school—we actually learned when my wife and I—we weren't married yet, but we were together, and I tried to get a car, and my credit rating was trashed. He had taken out all of this debt in my name, and defaulted on it. Not just me. His second wife.
ZIERLER: So you don't take it personally so much.
KATZ: I did take it personally. I didn't speak to him from graduate school, so that was 1991, until his death.
ZIERLER: You never patched things up?
KATZ: He tried to reach out a couple times, sent me a letter once. He actually contacted my wife, once.
ZIERLER: You didn't want it?
KATZ: Didn't want it. He broke the bonds. So, yes, I had a complicated relationship with my father. And then he was killed, suddenly. He was literally killed—that's why I never got to see him again—he was literally hit by a bus. He was getting out of a taxi in the Upper East Side of Manhattan, got out on the traffic side, wasn't paying attention, and an MTA bus literally hit him.
KATZ: So—yeah. My mom is still alive. She did eventually finish college. She has done lots of things. She has reinvented herself. For a long time, she did insurance. Then she worked in education, for vocational schools. She would do admissions and financial aid. She did that until she retired. When did she retire? Probably five or six years ago.
ZIERLER: Oh wow, she stayed at it.
KATZ: Well, she needed to. There was no money. It's not like my father saved money. [laughs]
ZIERLER: No insurance plan, right.
KATZ: So, yeah.
ZIERLER: What neighborhood did you grow up in?
KATZ: I grew up in the Village, in Greenwich Village. My mother still lives in the same apartment I grew up in. She moved into that apartment in 1966.
ZIERLER: Oh, wow.
KATZ: It was the first high-rise in Greenwich Village. It's still there.
ZIERLER: I know the Village. Where is it?
KATZ: It's on the corner of 8th Street and Broadway. If you walk by it, it's a weird building. It has a raised plaza and it has a fountain and giant white columns. Georgetown Plaza is the name of the building. Like I said, in 1966, it was the first high-rise. It's 34 stories. It was the first one to go up.
ZIERLER: This is a rental, or a co-op?
KATZ: When my parents moved in, it was a rental, and my mother is still a renter. It converted to condos when I was in college, when my parents got divorced, like the optimal bad time. I wish my mother could afford to buy it. They still offer—I think my mother literally is one of five left renters. I should tell you, this apartment is 2,000 square feet, three bedroom, corner apartment, 18th floor—
KATZ: —overlooks Grace Church. It's a very nice apartment. She pays $2,000 a month, I think.
KATZ: Not only is it rent-controlled, but because my mother is old enough now, New York State has a policy—no one wants to throw grandma out—so actually now any rent increases are actually paid for by a state fund, so literally her rent is absolutely fixed.
ZIERLER: Good for her.
KATZ: Yeah! But periodically they offer her money to leave. Never enough. I think the last offer was a few years ago, they offered her half a million dollars to move, or to buy the apartment for like $1.6 million. Every time Natasha and I look—we could swing it, but we're like, "My mom is going to live forever, and we're going to be paying out—it will eat all of our disposable income."
ZIERLER: Would that ever be a retirement pad for you?
KATZ: I doubt it. It's always just a little out of reach.
ZIERLER: That's not a primary house. That's something else.
KATZ: Exactly. We've talked about maybe getting a place in Manhattan. We actually know—I don't know if you've met here, Nico Wey Gómez?
ZIERLER: Sure, yeah.
KATZ: Nico is a very good, dear friend of ours. Particularly, he's a very dear friend of Natasha's. He has a friend, Olivia, who is a retired Stanford faculty member. They rent out their house in Palo Alto, and they're actually renting a giant apartment on Roosevelt Island, and it's basically a wash. We've talked about doing something like that. Maybe at some point when I retire, keeping this place but—we have a small house, but maybe renting it out, and just going to live somewhere else. Yeah, I wish I could afford an apartment in Manhattan, but, no.
ZIERLER: Growing up, as you said, Modern Orthodox, that means very different things to different people. What did that mean for you?
KATZ: It meant I went to Hebrew school every day. When I was Bar Mitzvahed, I davened twice a week.
ZIERLER: But not Shacharis every morning?
KATZ: No. I went Tuesday and Thursday mornings. Then I would go to services Friday night, Saturday, often Sunday morning. And I hated it.
ZIERLER: Where was your shul? Which one was it?
KATZ: Temple Emanuel. It doesn't exist anymore. It was on the corner of 14th Street and 1st Avenue. That's where we'd go. But when I was 16, I put down my foot and I said, "I'm done." Natasha, her father is a Lutheran pastor, now retired, so we both had our lifetime fill of religion.
ZIERLER: Do you think your father's shady ways provided some cognitive dissonance for you?
KATZ: No. My cognitive dissonance way predates knowing anything about my father's shady ways. It was just—I was a scientist. I viewed religion—especially because my shul was really conservative. Questioning anything was just unacceptable. It just drove me nuts. As I said, I was actually thrown out of two prior Hebrew schools. Teachers would make outlandish claims, from the Torah, or from the Talmud, but I would say, "Really!?" [laughs] So, yeah. I hated it. The other thing was that my parents weren't religious. It was just this was how you raised children. So, my brother and I were going to shul all the time; my father barely went. My mom would go. I would say my mom is spiritual. My mom still believes in God, and she goes occasionally to synagogue. Now, she goes to a reform synagogue, because she believes in God, not because—
ZIERLER: But it's more like a social cultural thing.
KATZ: Social cultural thing. But I was just taught, "These are the rules. You're just doing this because this is the rules." And that just did not sit very well with me.
ZIERLER: Did your dad make kiddush on Friday night? Did he know how to lead a seder and that kind of stuff?
KATZ: Oh, of course, yeah. My father grew up going to shul every day. My father could do everything. He would, on occasion. He was better than a High Holy Day Jew, but only marginally. I'd probably say twice a month, he'd go to Friday services with us. We had to go because we were going to be Bar Mitzvahed, and it was just like—he always found an excuse.
ZIERLER: What schools did you go to growing up?
KATZ: I went to secular schools. My first school I went to was Packer Collegiate, which is a private school in Brooklyn Heights, actually.
ZIERLER: So P.S. is out of the question? Your family could have—?
KATZ: I never went to public school.
ZIERLER: Your family could afford to send you, and that was always the better option?
KATZ: Yeah. And you have to remember, I was a kid in the 1970s. My father got into Stuyvesant—I forgot to mention that—for high school. Stuyvesant, at the time, was right near my house. I could walk to Stuyvesant from my apartment. But he wouldn't let me go, because even the magnet schools—Stuyvesant and Bronx Science—were really rough. Basically, any upper middle class kids went to private school.
ZIERLER: Or you moved to Long Island.
KATZ: Right, or you moved to Connecticut, or to Rye, New York. So, I went to Packer. My parents chose Packer for one reason, and one reason only: it was the only private school in Manhattan that had two kindergarten classes. Because my parents didn't want us going to different schools, but they didn't want us in the same classroom. They chose Packer because my brother and I could be in separate classes, but only have one dropoff. [laughs] I was at Packer from kindergarten—
ZIERLER: Is that like you were inseparable with your twin brother?
KATZ: We were never close. We played together as kids, but we were very different. My brother really struggled in school. He is pretty learning disabled. We were at Packer together through third grade. Then he went to something called the Churchill School, which is actually a school for kids with learning disabilities. I stayed at Packer for a little bit longer, although Packer was a really bad school for me—my third grade teacher, Mrs. Schwab—I still remember her name—she told my mother in front of me that I would never turn out to be anything. Because I really struggled learning to read. In fact, I didn't really functionally read until third grade, which for upper middle class New York, that was—rare. But, I stayed there through fifth grade.
In the sixth grade, I moved to a school called York Prep, which was not a great school. It was on the Upper East Side; now it's on the Upper West Side. Private school, modeled loosely after private schools in the United Kingdom. Our headmaster, Mr. Stewart was actually a former British barrister. He married into a very wealthy American family, and literally the school was basically their wedding present. He was a very famous barrister, actually. He got this really big case in 1970; he got off this very famous British mobster off on a murder charge, and it really—he knew the guy was guilty, but that was his job. But he couldn't live with himself, so he couldn't practice law anymore. So, he comes to the States, and he meets this woman, and they form the school.
So, I went there, and had a great teacher. I had Mrs. Schwam in third grade who said I would never turn out to be anything, and I had—Trudy Marks was my sixth-grade teacher. Although the school in seventh and higher was more traditional—you went to different classrooms—the sixth grade was a very small program. It was a one-classroom thing. Ms. Marks taught us everything. She figured out really quickly that I was really good at math. It was kind of hippy-dippy. She was hippy-dippy, not the school. It was self-paced. We did a lot of stuff on our own. She would come around. I think there were like 14 of us in the classroom? I basically did four years of math in sixth grade.
KATZ: It was a real virtuous cycle, that, "Oh, I'm not stupid; I'm really good at this stuff," so then I did well at everything. Then there was a question of what to do. As an aside, York Prep is famous for a really bad reason. A notorious classmate of mine, Robert Chambers, was the Preppy Murderer. I don't know if you remember hearing about this case? Actually it was the summer before I was going to college, this guy Rob Chambers killed a woman in Central Park. There was a bar we all went to—again, drinking age in New York City when I was in high school was 18 and unenforced, especially if you were wealthy, so we were going to bars and clubs, 14 on, and Dorrian's was a famous one, where all Upper East Side kids would go to drink. If the average age in the place was 16, I'd be surprised.
KATZ: So they got drunk. He alleges it was an accident. He was found guilty. This was all over the papers. That's why my school got famous, because it was this whole social circle. In fact, my parents had a decision to make. York Prep was a school where a lot of wealthy kids went who failed out of other places or weren't great students. The families were wealthy enough but couldn't get them into Dalton or Collegiate or Trinity, the top-tier equivalent of Polys in New York City. Fieldston, Riverdale. There was a cadre of 10 or 14 sort of elite private schools in New York City. So, there was a question—"What do we do?" My brother has gone to this other school. I'm doing really well. But do I stay at York or do I leave? So, I get into Collegiate.
From York to MIT
ZIERLER: What are the grades you can go up to at York?
KATZ: It went through high school, and I did in fact graduate from there. But there was this question, which is—I was clearly much better than my classmates, so the question is, should I stay at York? I applied, which is very difficult to do. I got in, as a freshman in high school, to Collegiate, which is the oldest private school in Manhattan. It was an all-boys school; it is now integrated, co-ed. Then my parents were like—do I become a small fish in a big pond, or do I stay the—? Of course, York Prep very much wanted me to stay, because I was doing really well. My brother had successfully gone through Churchill and had actually moved to York Prep and was doing pretty well. So they cut a deal with my father. I agreed; it was my choice. I could have gone to Collegiate. We decided to stay, because I was taking math—at that point, in ninth grade, I was taking math classes at NYU. I guess they gave my father a break on tuition for me and my brother, for me to stay.
ZIERLER: A package deal.
KATZ: A package deal, so that's what we did. It turned out to be the right choice. I got into every college I applied to. I got into MIT, and I went, so it was a good thing for me. But it would not be the obvious school you would have sent someone who was doing as well as I did. But again, early in my childhood student years, I was a terrible student.
ZIERLER: What clicked or what changed?
KATZ: I hate to say that it was one teacher, but it was Trudy Marks, who figured out, "Here's what you're really good at" and let me do—the other thing is how I learn. I learn by reading. I'm perfectly happy to sit and to read something. I read incredibly quickly, now. She let me do whatever I wanted. I got along well in high school, but I was a geek. I was this weird kid. Like on my chemistry final in it must have been tenth grade, I got a 98 on the final. The next highest grade was like a 68. Luckily, the teachers all knew not to like—[laughs]. I was not on the curve. I was the valedictorian. I was very involved in the student government. I was captain of the track team. So, I had a social life. For me, high school was easy.
ZIERLER: Why just math at NYU? Everything else you could get on your own?
KATZ: Math was what I was really, really good at. Everything else, I was above grade level, but I didn't need to go off and—and my father had graduated high school early, and my parents were very much against, for social reasons, me finishing high school early, although I could have. In hindsight, I think that was the exact right decision. I even see that here. I've had one freshman advisee who was like 16, and they're just not emotionally ready for college. They could be intellectually more than ready, but—so that's what we did.
ZIERLER: I almost always know the answer before I ask this, but for you, I genuinely don't know. In high school, did you see yourself, when you started thinking about college, more on the math and science track, or more on the English and history track?
KATZ: Oh, math. And computer science. That was what I was going to do. That's what I liked. I had other interests. I was a polyglot. I loved reading. I loved history.
ZIERLER: Just to foreshadow to today, did you follow politics? Were you into data? Did you like that kind of stuff?
KATZ: Yeah, I was always into data. I was always an empiricist. I loved politics. I actually made money in high school, my own money, and in college—I wrote software for tracking expenditures for political campaigns. I was always a political junkie, but what I was good at was math and computers. I had a Commodore 64.
KATZ: I had an IBM.
ZIERLER: Probably, what, a 286?
KATZ: No, the first one was a 186!
ZIERLER: A 186!
KATZ: I remember my friend had an Apple II. He got it upgraded from 32K of memory to 64, and we were like, "What are we going to do with all that extra memory?"
KATZ: My machine there, I have 128 gigabytes of memory. I always knew I was going to do STEM stuff. That's what I thought I would do. Math.
ZIERLER: Did you intuit that there were novel connections that could be made between social science stuff and mathematical stuff?
KATZ: Not in high school. That wasn't until college. I got to MIT.
ZIERLER: Before we get to MIT, where else did you look at? What was on your radar?
KATZ: I looked at Columbia, both Arts and Sciences, and Engineering. I applied to Cornell. I applied to MIT. My safety school was Syracuse. That was like the feeder—about a quarter of my class went to Syracuse, so there was this like pipeline. In fact, so many kids from my high school at the time went to Syracuse that the guidance counselor would basically sit down with the admissions director from Syracuse, and they would go through basically the entire class. The admissions director pulls my file, and the guidance counselor, who actually happens to be the wife of the headmaster, Mrs. Stewart, she's like, "He's not going to Syracuse." They're like, "But we'll give him a free ride, and we see his brother has applied too, and we'll give him a half-scholarship."
ZIERLER: [laughs] Another package deal.
KATZ: Another package deal. But we decided not to. Syracuse was my safety school. I applied to Lehigh. I applied to math and engineering schools.
ZIERLER: But all driving distance away. Stanford, Caltech, Berkeley, this was not—?
KATZ: No, I never applied to the West Coast. I applied to Lehigh. I applied to Carnegie Mellon, MIT, Cornell, Columbia. I think I applied to eight schools, which is I guess unheard of now. Everyone applies to like three million schools.
ZIERLER: Did you visit a lot? Did you have a good sense of where you would go?
KATZ: Oh, I applied to RPI, too. I knew when I got into MIT that's where I wanted to go.
ZIERLER: That was it?
KATZ: That was it.
ZIERLER: Just by reputation?
KATZ: Reputation. Because I didn't really know—none of my peers—MIT was not on the radar of anyone—that's not where—
ZIERLER: Certainly not your parents' radar.
KATZ: Not my parents' radar. But because I was around NYU and I was doing mathematics, they were like, "Oh yeah, that would be a great place for you." But I didn't apply to Harvard or Princeton or Yale.
ZIERLER: Because you were more technically focused?
KATZ: I was interested in STEM. I was interested in computer engineering, computer science, and mathematics.
ZIERLER: What year did you arrive at MIT?
KATZ: I graduated high school in June of 1986, so September of 1986.
ZIERLER: What were your early impressions when you got to campus? What was it like?
KATZ: It was great. I loved being free. It was a great campus. It's a weird system. It has changed now, but when I was there—MIT, like Caltech, ostensibly said that everyone has four years of Institute housing. That was a lie—it's now not a lie—because basically the only way they can make that happen was with these independent living houses, which were basically almost all fraternities. There was one non-affiliated one. So, you get to MIT, you're put up in a temporary dormitory—I was put up in Burton, which was actually the all-male dormitory—and you have rush. I end up joining a fraternity called Theta Chi. It's on Beacon Street and Mass Ave. So, it was like this weird thing. You're going to this really rigorous academic place, but the first thing you do is you rush a fraternity. [laughs] But, it was cool, and that was my environment, and I'm still very close. My roommate through much of my time at MIT, Eric Miller, he's an electrical engineer, now machine learning type person on faculty at Tufts.
It was a good environment. I loved MIT. I did very well at MIT. It is this rough transition for lots of kids. It's two transitions. One was the freedom. I already had that. I grew up in Manhattan. Literally I was going clubbing—like I was the only person who went to college and drank and did fewer drugs. [laughs] Because I had done that already. But all my classmates, who had been from these like little towns, they all went nuts. First semester is just insane. Then the other problem, but the one that is also at Caltech, is basically everyone in our class was top of their class. By definition, half of you are going to be in the lower half of your class. At places like MIT and Caltech, it's even more obvious, because the first year, you're taking all the same classes, essentially. Modulo your humanities or social science class, everyone takes math, physics, chemistry. I had a little experience of this. I had gone to Harvard Summer School the summer before. It was actually really funny; I took chemistry at Harvard. MIT is so full of—MIT would not give me credit for my—I took freshman general chemistry. So I had to take another chemistry class at MIT, because they wouldn't—
ZIERLER: They didn't accept Harvard's—?
KATZ: They didn't accept anyone's. But it's really funny because actually, as an MIT student, I can cross-register, at the time, in a Harvard class, and I would get credit for it. But as a student who had taken it as a summer school that counted as a transfer credit, that didn't count. I actually took solid state chemistry, which was fine. It was actually something I didn't know anything about. It was okay. If they had made me take general chemistry again, I would have been really pissed! [laughs]
ZIERLER: Was it two years of a general education before you specialized?
KATZ: Exactly. MIT and Caltech, until recently, had basically two years of course requirements of math, chemistry, lab sciences. MIT was a little more flexible. MIT was more like I would say a year and a half of requirements. Sophomore year, you could really start taking classes in your field, if it was outside of physics or mathematics. Caltech, until recently, until the last decade, it was two full years of basically Institute requirements. We've toned down. That's something we have done. MIT has done the same.
So, I get to MIT. It was challenging, but I graduated with a 4.0. Well, they called it a 5.0. There, the first year is pass/fail. I did do something which I think was smart on my part. What was a big change for me was I was at this teeny private school. There were 67 kids in my graduating class. MIT is huge. It's now more, but when I was there, there were over 5,000 undergraduates. My class was 1,000-plus students. I had never been in an environment like that. There were these two experimental first-year programs. One was called ESP, Experimental Study Program. One was Concourse. I actually did Concourse, and Concourse was kind of a neat idea. I don't know if it actually worked in the end. It worked for me in the sense that it was a smaller community. It was like 60 kids. You had to apply to get in. All the Concourse instructors were faculty meetings. Like I learned linear algebra from Strang. Strang still has the textbook. And Finney's calculus book is still probably one of the default calculus books used.
How it worked was that these instructors, these faculty members, would work as a team. The first couple of weeks, we spent a lot more time on mathematics, although I didn't need that, so that we'd have the right calculus, at the right level, to do the physics that they wanted us to do. Even in the humanities, we did history of science. It was all integrated. We spent basically this time in this one room. It's actually these rooms they built after World War II; they've now been condemned. We had this one classroom. We had a lounge, study area. All of the instructors knew where everyone was, who was struggling, what was going on. That worked really well for me, as a transition from my little high school to a 1,000-person class.
ZIERLER: A good halfway point.
KATZ: Yeah, so I did that. And, I did applied mathematics. That was what I majored in. I was going to do computer science. I should point out that applied mathematics, at the time, was called 18…everything at MIT is a number, including majors. The first time someone meets from MIT, they say, "What was your major?" and they give a number, and we all know the numbers. I did applied mathematics, which was at the time—it's called 18C. It was really theoretical computer science. It wasn't really applied mathematics. Well, it was a particular version of applied mathematics.
ZIERLER: Did they have a CS program at that point?
KATZ: Sure, they did. This was the 1980s, and places like MIT and Carnegie Mellon were being inundated, because everyone wanted to study computer science or electrical engineering, and the programs were overwhelmed. Still to this day; it's even worse now. Carnegie Mellon's solution was—at most universities, you can basically major in anything, once you get there. You're admitted to the university. That's not true at Carnegie Mellon. At Carnegie Mellon, if you want to do either CS or electrical engineering—
ZIERLER: It's an internal application process?
KATZ: It's an internal application, which you might not be accepted for. MIT didn't do that. That's actually the right policy. Because kids—I tell this to my freshman advisees—you never really know what you're going to major in. You think you know, but the exposure you have in high school, even a really good high school, is just—things don't exist at high school. No one comes to Caltech to study geology and planetary sciences, although we are like the world's best place for this, and it's an amazing thing to study. You don't get that. Bioengineering; well, people kind of know what it is, but you don't really know what it is. Anyway, I continued and I did applied math.
ZIERLER: I want to return to this key point. I asked if it was in high school, but you said it was in college—
KATZ: I'm getting to this.
ZIERLER: —the connection, where that happens.
KATZ: While I'm at MIT, I'm still doing this job where I'm writing software for political campaigns. I'm still a political junkie. I get very involved, actually, in university politics. The student government at MIT is called the Undergraduate Assembly, UA; I was the Undergraduate Assembly president. Even before I was that, I was very involved in committees, and I actually did things now that as a faculty meeting I disagree with, but I understand why I did them as a student. The faculty very much wanted to reform the first-year program at MIT, and I was successful—I had an intuitive understanding for politics, and I rolled the—
ZIERLER: That you realized at the time. You realized you had a sense for those things.
KATZ: Oh, yeah, I was good at it.
ZIERLER: Good at what specifically? The numbers behind the politics? Like where people were?
KATZ: It's what I got from my father, which is how to understand people, and how to understand what people want, and how to get what you want.
ZIERLER: With the mathematical skills.
KATZ: Well, data, because that's an important thing. But it was really just straight politics. There was nothing quantitative. I was very involved in politics personally, at the university level. I was still involved in it, because I was involved in these political campaigns, writing software for them. So, I'm doing applied mathematics, and I don't really like. It's not that I don't like it; I decided that that's not really what I want to do for my life, even though I'm really good at it. If I had been at any other place, and I had been exposed to modern statistics, I would have probably been a statistics major. Like Caltech, MIT has no statistics department.
ZIERLER: What is that about?
KATZ: It's just a history. Here's the problem. It's a disciplinary issue, which is that the big players at both MIT and Caltech are mathematics, and mathematicians poo-poo statistics, both because it's applied, and that somehow you're getting it dirty, and also the mathematics was well worked out in the 1800s. [laughs]
ZIERLER: There isn't cutting-edge scholarship.
KATZ: No. The cutting edge stuff was—have you heard of the Radon-Nikodym theorem?
KATZ: Right, so the Radon-Nikodym theorem was proved in 1890, right? This is basically the axioms that give you an existence of a measure, basically a probability measure. Mathematicians think statistics, from a mathematics point of view—and they're correct; from a mathematics point of view—it's passé. But they controlled the resources and they never wanted to hire in statistics. That's why when I said applied mathematics at MIT—I did take numerical analysis, which was the closest I got. I was programming in Fortran, how to solve Newton solvers and things.
ZIERLER: There was still Fortran when you were an undergraduate?
KATZ: Oh! In mathematics—there's still Fortran today. For the most part, the language I used to program in for statistics is R, and a lot of the infrastructure is built in C++ now, but there are still fundamental libraries which are maintained and done in Fortran.
ZIERLER: I'm thinking hardware; you must be referring to software.
KATZ: Fortran was a programming language.
ZIERLER: I'm thinking like the big—the floppies.
KATZ: Oh, no, that's not Fortran. Those were floppies. The first computer language was Assembly, which is basically machine language. Really good for computers; really a pain to program for humans, because basically what you move is bits. You can do addition, you can do subtraction, you can do division, but you have to think about moving bits in a register, literally. The first big improvement in computer science were higher-level languages which would then be compiled to generate machine languages. You didn't have to think the way the computer thinks; you can think the way a human thinks. There were a bunch of early languages that were written. BASIC was one. But in scientific computing, Fortran was written from the ground up with the understanding that you would be basically doing mathematical modeling. It's super-fast, and it's really hard to write code faster, in another language other than Fortran, for numerical analysis. Again, it's the same thing of a general purpose language versus a single purpose language. Fortran is not a general purpose programming language. I mean, it is, technically. It meets the general requirements. I can print things, you can write things to screen, but it was designed at its core to be blazingly fast at doing matrix calculations, and it does it really well. That's why Fortran still—but then the other language I learned at MIT—computer science at MIT was all about artificial intelligence.
One of the keys, the early computer scientists thought, to artificial intelligence, was the idea of recursion. Usually in a programming language, you can't refer to anything you haven't already defined. Suppose I want to do something to some variable X. Until you've declared that X is something, it doesn't know what to do. You can't build these recursive statements without knowing what it is, because recursion, by definition, it's kind of an inductive way of building a structure. LISP was the first language that—well, it wasn't the first; it was the most successful language—that had recursion built in as a fundamental thing in the design of the language. LISP was perfectly happy to let you write down functions that build other functions, that have variables that you haven't even defined yet, that are going to be defined later by some other program. That's perfectly okay. In most computer languages, that's not okay.
Anyway, we got on a tangent. Yes, so Fortran, I was doing all this stuff, but I figured out that while I was really good at math, theoretical computer sciences, which was what my major was, was not really what I wanted to do. Like I spent a long time proving—you probably know the P versus NP? We would prove that problems were NP. I learned proofs for how to bound an algorithm, like what's the order of operations. Is this going to be order N, or is this N squared? It wasn't really what I was into.
ZIERLER: It lacked the blood and flesh of politics.
KATZ: Yeah. I had always had this interest in politics. I also took some economics classes. I thought the economists were asking really interesting questions, using really interesting tools. In the 1980s, they were the ones who were really pushing—this was the development of econometrics, the quantitative study of economics, game theory. I thought they had really interesting tools. I thought they were asking really important questions. I just wasn't interested in the economy, per se, but because of my interest in politics, and I had taken politics classes at MIT, I thought that was really interesting. Then the real turning point is—the world is super small, and unfortunately he just passed away—Howard Rosenthal.
ZIERLER: Oh, yeah. Yeah.
KATZ: Howard Rosenthal, Jean-Laurent's father, actually got his PhD from MIT. In 1986, he was a super famous political scientist. The Political Science Department at MIT would not invite him to visit, because he did this quantitative mathy thing, which was a real tension in political science, so he was actually invited to be a visitor in economics. I knew none of this. He was offering this class in political economy. It sounded really interesting to me, so I signed up for it. I was the only student signed up for it. There were lots of people there. The other participants in this seminar—Howard Rosenthal was teaching it. Paul Joskow, who is, in his own right, a very famous economist who went on to be president of the Sloan Foundation. Jean Tirole, who will win the Nobel Prize in Economics. Alberto Alesina, who died young—he just died a few years ago—he was in it. It was this crazy group. In fact, I was freaked out the first day, because here it is, mostly faculty members and a few grad students taking a seminar, and I'm like—
ZIERLER: What pulled you in? What was compelling about it?
KATZ: The title looked good to me—"Political Economy." This was what I was interested in. Literally, I knew nothing more than [laughs]—
ZIERLER: Did you have a good working definition of political economy at that point?
KATZ: Probably not, but it looked interesting, and I needed another social science class, and it fit my schedule. Howard then begs me not to drop the class. If I drop the class—they can't have a class with no enrolled students.
KATZ: As it is, I was the only enrolled student. In fact, we talked about my background, so he hired me, and I was actually an undergraduate RA for him. This was the fall of my senior year. So, I was working with him, and he was then convincing me that I should apply to—"Oh, you'd be great. You should go to graduate school." In fact, he wanted me to come to graduate school at Caltech. So, I kind of got into that this was something I could do. I liked what he was doing. What he was famous for and what he had done at the time—he had done lots of things but—was this thing called NOMINATE. We talk about politics beyond left/right; suppose that I want to actually figure out—these are called ideal points—here's Bernie Sanders, here's Mitch McConnell. I can do that abstractly, right? I know there politics and I can put it on there. But what they wanted to know is, suppose I have this matrix of roll call votes. Is there some way to summarize that matrix of roll call votes as a series of locations in one or more dimensions of where these people's preferences were? The idea in the spatial model of politics is that I vote for the option that is closest to me spatially, usually Euclidian preferences. I compare two points by the Euclidian distance, but you can think of other distance measures. A bill, then, is a cut point. In one dimension, it's literally just a line. If I'm on the left of the line, I vote "yea" and if I'm on the right of this line, I vote "nay."
So, they did this. They invented this. At the time, in the 1980s, they had to use a supercomputer to actually do this, and it was coded in Fortran, by the way. [laughs] It turns out they actually reinvented something that had been discovered in the statistics of education; it's also how your SAT score is graded. The way you should think about it—so, what does the SAT score do? Let's think about just the math score, right? We have some easy questions: 2+2 = x. Almost everyone gets it right, unless you really know nothing about mathematics. That's called the difficulty of the question. What do we know if you get it right? We know that you're somewhere between here and here, but I don't know if you're genius level Field Medalist, or if you're on this side, and you know woefully little math. Then you can think about marching difficulty questions up. This is exactly how the SAT is graded. What they're trying to figure out is where you fit on this line. It's arbitrary what the scale of the line is. They do it 200 to 800; you could do it zero to one, it doesn't matter. It's called the item response theory, item response model. That was developed at Bell Laboratories, in educational testing—the College Board, independently. They both developed it in the 1980s. They applied it to roll call data. There are some subtle differences. Anyway, Howard was covering this stuff and doing other stuff. It was kind of cool; I could use my math skills, but I could also use politics stuff, which I really liked. He really wanted me to come to Caltech. I'm like, "Howard—" I think it was "Professor Rosenthal" at that point. I was at MIT. "I'm not going to Caltech for graduate school, so choose again." [laughs]
ZIERLER: [laughs] Are you contemporary with Jean-Laurent? Did you know him at the time or interact?
KATZ: No, I didn't know Jean-Laurent. I didn't know until after. Technically, Jean-Laurent is a couple years older than I am. I'm 54, and I think Jean-Laurent is 58 or 59. But no, I didn't even know he had a—I knew nothing. I was an undergraduate. Howard Rosenthal was this faculty member. I knew nothing about him personally. I said, "Choose again." He goes, "Well, then you should go to UC San Diego." Because at the time, my advisors, both who were PhDs from here—Gary Cox and Matt McCubbins—had moved there from the University of Texas.
ZIERLER: They were PhDs at Caltech, junior faculty at Texas.
KATZ: They were junior faculty at Texas and then were recruited as senior faculty to San Diego. Remember, again, even in the 1980s—I don't know if you've been down to UC San Diego at any time; it's now a huge place. When I was there—
ZIERLER: Still in growth mode.
KATZ: Literally my first office—it was a former military base—my first office was in a Quonset hut. [laughs]
ZIERLER: [laughs] That's awesome.
KATZ: Anyway, he goes, "Gary Cox and Matt McCubbins just moved a couple years ago to UC San Diego. It's a small place. You should go there."
ZIERLER: Looking back, their recruitment was more San Diego is already big-time, and this was evidence of that, or they were trying to get there?
KATZ: This is again disciplinary knowledge. Political economy gets started at Caltech, University of Rochester, and Washington University. But it's still in the early phases, and so this still isn't a big—this sort of really quantitative using game theoretic tools and high-end statistics was still really pretty novel in political science. If I had gone to Harvard for graduate school—
ZIERLER: They were not onto this yet.
KATZ: They were not. I would have had to take a foreign language requirement. You read books, and this is not their style. But San Diego, they were building this new program, and you come. He goes, "You should go there and work with them." "Okay." But I hedged my bets. I told you I also applied for jobs. I got a job offer at Bain & Company. I applied. So, I get in to a bunch of programs. San Diego was where Howard wanted me to go if I wasn't going to come to Caltech. I get the NSF Fellowship, and I decide I'm going to take it. I had actually never visited San Diego. There were some phone calls. In fact, actually I freaked out a little the first time I met Matt McCubbins. He said, "Why don't you come out early?" They're on quarters like Caltech is, so the term doesn't start until basically October 1st. "Come out in July, and I'll hire you, and you'll start working as an RA for me before the term starts." So I show up there in like mid-July. The first time I meet him—MIT is very much an East Coast school. Even though it's a technie school, it's very much an East Coast school. At the time—these are the 1980s—faculty members at least wore khakis and a dress shirt. Many wore a jacket and tie. Matt McCubbins—I come in and meet Matt—so I'm wearing like—I know it's San Diego, so I'm like wearing a pair of like khakis and a Polo shirt. Matt is dressed in like grubby sweats. Unfortunately, he's always injured. He was injured at the time. So he's like unshaven, and I'm like, "What did I get into?" But anyway, Matt was my advisor, and it was a great relationship.
ZIERLER: What was his research at the time? What was he working on?
KATZ: He does lots of things. At the time, he and Gary Cox were working on this book—because I worked on it—called Legislative Leviathan. This is going to sound weird, because now we think of politics as being hyper-partisan in America, but if you go back to the 1970s and 1980s, it was a one-party system. The Democrats controlled the House. They had for all but two years from World War II on. Candidates would rarely label themselves. You would have to dig deep in the weeds to find out like what party they were. They had this idea that parties actually matter. That is, that they are actually setting the agenda, that they are important things. Again, this was not obvious in the 1980s. It now is obvious, but it wasn't obvious then.
ZIERLER: That's great scholarship. [laughs]
KATZ: They were working on this, but they worked on lots of things. Gary Cox was also working on a book, which is super important—I still teach it—called Making Votes Count. Most of the quantitative and theoretically driven—politics was mostly about American politics. That was because all the guys who—and they were all guys—who were doing this, the only thing they knew was American politics. You model what you know! [laughs] But the question was, how do you push it out? Gary Cox wrote this book—it's a great book, actually; Gary is a great writer. Gary, I should say, was an undergrad at Caltech and a grad student at Caltech.
ZIERLER: Who was his advisor?
KATZ: Morgan Kousser. He was a history major. His PhD advisor was probably Ferejohn and Fiorina. Matt was funny. With Gary, everyone knew he was going to be a superstar. He's super smart. There are like 16 political scientists in the National Academy; Gary is in the National Academy. Gary had gone through Caltech. Matt was a poor guy from the Valley here. He went to Cal State Northridge before getting to Caltech. He doesn't have the sort of pedigree—at elite universities, we become so prestige-driven, I think badly, but that's a discussion for another day. In fact, John Ferejohn wrote a letter about Matt when he was going out on the job market, that Matt would either rise like a rocket or fall like a stone, and "at this point, I can't tell you which."
ZIERLER: Oh, wow.
KATZ: Matt, of course, with a letter like that, didn't get any job. The one guy who wanted to take a risk on him was this dean at Texas. This is the old days when you could just do this—the dean calls up John Ferejohn and says, "John, I really appreciate the honest letter, so I now know what I'm buying. Can you now write me a letter that I can actually get this guy appointed?" [laughs]
KATZ: Matt and Gary, they were there at Texas for a few years, and then they moved to San Diego. I move there, I meet them. They're working on this stuff at the intersection. Because it's a new program, they let me do whatever I want.
ZIERLER: How much of a risk did you think that was in terms of—?
KATZ: I had no idea.
ZIERLER: What about just name recognition of the school?
KATZ: Oh, no, it was a disaster. If Howard hadn't told me to go work with these two—by the way, they are both—Matt just passed away. Matt was 64. Matt is only ten years older than I am. Gary is maybe a couple years older than that. It turned out to be a magical time at UC San Diego. In fact, my cohort, not just in this area, did incredibly well. I have classmates who are tenured at UCLA, at University of Minnesota, at Harvard.
ZIERLER: With a similar trajectory, coming from MIT-type places, and they got specific advice to go there?
KATZ: No. Some did. Like John Carey, who's tenured at Dartmouth, he was a Harvard guy, but he would tell you he got into Harvard because he was a hockey player. [laughs] So, I would say we were definitely more—let's come back to this. So, I go there, and there's this total flexibility that I can do whatever I want. I basically recreated Caltech's program. I said, "All right, I'm going to do all the econ program too." In fact, I showed up to the Econ Department, and I said, "Oh, I'm going to take the graduate sequences in game theory and econometrics," and they were like, "Yeah, yeah, yeah, whatever." I remember the first class in microeconomics was taught by Joel Sobel, who's still there, who's a math guy, an economic theorist. It's basically optimization theory. Literally, it's mathematics. There's hardly any economics on it. There's like 18 of us in the class, and I said, "I'm going to take this class," and he goes, "Yeah, yeah, whatever." I take the midterm, and I get the second-highest store; a Chinese student beat me. He calls me in like, "Why aren't you in the Econ Department and how did you do so well?" I said, "I was an undergraduate in math at MIT; this is trivial. I did this before college. This is just first and second order conditions of maximizing a—yeah, I know how to do that." Taking some simple derivatives; not a problem. [laughs]
So, I did that, but it was a totally risky strategy. But it worked because it was so little; I didn't know any better. I was treated like a colleague almost from the beginning. I wrote a bunch of papers with—actually, I never wrote a paper with Matt until well after that time. Gary Cox and I wrote a paper. The real hit—I worked with—he was never my advisor—he was this guy named Neal Beck, who was interested in doing basically econometric statistics. I should say, intellectually, there are two approaches to doing quantitative data. There's econometricians, basically developed in economics, basically lean heavily on applying economic theory to the statistical modeling, and there are the statisticians who hate that, because they don't believe in models; they only believe in observables. But in the early days, what is called methodology in political science, which is mainly what I do, we were basically in catch-up. We were, for the most part, importing techniques from econometrics and applying them to political data. Not too many people were like me who were developing anything new.
Neal Beck was there. He had done some stuff. He was older. I was working with Neal, and he had some ideas, and we were playing around with some data. One way you evaluate models which is now very common is this idea of cross-validation. You basically leave something out. In the cases like macro political economy data, like suppose you're interested in knowing whether or not having left-wing governments lowers the economic growth rate. You get data on EOCD countries over the postwar period, and so you fit this regression model, pretty standard, and you include a bunch of economic controls, and then you include the variable you care about: having a left-wing government. So people would do this. It turns out that was really—so, Neal and I were working on this, and we were doing this sort of cross-validation, where when you have structured data like that, you don't want to just randomly take out data points, because there's dynamics going on, so you basically remove entire countries and basically forecast them. It turns out the data were being driven by two data points: Norway. What is different about Norway than any other country in Europe? It's economy is more like Saudi Arabia. Norway has the largest sovereign wealth fund in the world. It has a $1 trillion sovereign wealth fund. Because in the early 1970s, they found North Sea oil. So, they had these amazing growth rates. They also have incredibly left-wing governments, and they have an incredibly generous welfare state, because basically they can afford to pay for it. But it's an outlier; it basically drives all the results.
So, we were doing this cross-validation stuff, and then we were looking at the models. The technique that was used was this model developed by this statistician named David Parks at University of Washington. They run these cross-sectional regressions. When you first learn regression, you basically learn the IID case. That is, you think that every observation is independent of every other observation. But in lots of real-world settings, that's not true. Again, let's go back to this political economy example that I've been running through. When France has a good year, Germany also likely has a good year, even if you condition it on everything you know about them, because they trade a lot together, and they're linked economies. Conversely, when they have bad years, they both do. There's also heterogeneity, so like the German economy versus the Belgium economy. They're just different scales, so there's heterogeneity. All that is the workhorse of statistics, but it assumes that they are IID. Even if they aren't IID, the coefficients under certain circumstances are correct but they are inefficient. That is, you're not using all the information in the data, because you're not using this information that there's correlations across countries or over time. Parks, in 1968, proposed this paper, this idea that was called the feasible generalized least squares model, where you basically do this in two stages. You basically exploit this covariant structure. You do it in two stages. You fit a first-stage regression where you get residuals. Then we use those residuals to actually calculate like the correlation between the residual—the errors—in Germany and France, in the same year. This was implemented back in the day in SAS. Have you ever heard of the statistical language called SAS?
KATZ: SAS is now big. It still exists. It is not used in universities anymore, but it was back then. Now, it's big in healthcare and finance. It was designed for Big Data before we knew what Big Data was. It had this macro language that you could add on packages. So, this econometrician at North Carolina had a grad student write up Parks's model from 1968 in a SAS package. Unfortunately, the grad student was killed in a biking accident. But the code was released out there, and it had a very peculiar error. The people using it didn't really understand the methods. It would produce standard errors that were too small. If you want to know the technicalities, I can tell you why we can prove that they're too small. Neal and I were reexamining these papers, and we're looking at the standard errors, and they're like way too small. You've heard of p-values and t-statistics? The t-statistics are huge. In social scientific data, things are just small and noisy. You just don't get huge, precise effects. It just doesn't exist. But the software was generating these huge precise effects, not because the effects would be whatever they were, but because the standard errors were too small. Neal and I figured this out. In fact, we figured it out because it was based on a paper that Mike Alvarez, who is on faculty here, had written when he was in grad school at Duke, with Geoff Garrett, who is actually the dean of the Business School at USC. Garrett and a guy who went on to become provost of Duke. Anyway, they were just political scientists when they were there. They wrote this paper that was in ABSR; they used this method. They were nice enough to give us their data. This was before the era when people posted data. You literally had to email someone to get data. They were nice enough that they gave us the data, and we were like, "Something's fishy here." Because they were really nice to us, we didn't want to be jerks to them, so we emailed them and we said, "Listen, here's the problem, and why what you did is wrong. Here's how we can fix it." We all agreed—we wrote this—my first publication was this corrigendum because it all of us basically updating their previous article that was published in the flagship journal in political science. Neal and I went to write a follow-up paper which actually fleshed out how to do this right. That paper is the most cited paper in the American Political Science Review.
ZIERLER: Wow! [laughs]
KATZ: So, I got really lucky. My first paper was this. Then I wrote another paper in grad school, with my classmate Brian Sala who left the Academy, on committee assignments. Again, it was tying together theory and a little bit of a natural experiment. People think we've always run elections the way we do. Elections have actually radically changed. In the Progressive years—this is actually what Morgan Kousser wrote about—we moved to state-printed ballots. Prior to the 1890s, there was no Secretaries of States printing ballots. If you wanted to vote, the parties would literally print out colorful ballots that would list all their candidates, and you would just literally take the ballot and stick it in the ballot box. There was no notion of secret. Everyone knew exactly how you were voting, and in most states, they made it almost impossible for you to split the ticket. You couldn't take this pre-printed ballot and say, "It's the Democratic ballot, but I'm going to take the Republican candidate for governor." Not possible.
Research on Legislative Political Economy
ZIERLER: Straight down the line.
KATZ: This changes. This has implications. A fundamental working theory of political economy about legislatures is that they should design the rules—since you get to write the rules of how you organize—based on their incentives. When the system changed, this changed the incentives for members of Congress and made it worthwhile for them to pursue what is called a personal vote. Before that point, you're not getting elected to Congress because of who you are; you're getting elected to Congress because you're on the Republican ticket or the Democratic ticket, kind of like it is today. We showed that there was this change in organization, particularly on how committees were structured and how long people stayed on committees. We did what at the time was really fancy statistics. In fact, the models I ran, these models of how long people stayed on committees. Brian [sp] was a great writer. He knew a lot about the history of Congress. I was the stats guy, the math guy. I wrote this code that would do a model. Again, now it would be trivial to write and do. I had to trick SAS into doing it. I had to take a complicated optimization problem and then rewrite it as a non-linear optimization problem, non-linear least squares problem. On the server that we had access to, it took eight hours to run. It would take like two seconds to run on my computer. So, we wrote this paper that was also in the APSR. So, I went out on the job market. I got through grad school pretty quickly.
ZIERLER: Those two constituted your thesis, those two papers?
KATZ: I had another paper with Gary Cox that was on the incumbency advantage. That also eventually got shortly published thereafter in the AJPS, American Journal of Political Science. Those are the three papers in my dissertation.
ZIERLER: Before we get beyond the dissertation, what were the through-lines, as you saw them, between them?
KATZ: There were no through-lines. In economics, it had already become accepted, that you could just write three papers. You write three papers; that's good.
ZIERLER: By design, did you want them to be quite distinct?
KATZ: Even today I tell students it's better to have a defined reputation, like, who are you, and how do you fit into—. That wasn't me. I had a bunch of papers. I had other papers, too. I had lots of papers.
ZIERLER: So you could wear different academic hats as it suited you?
KATZ: Yeah. The hat I wore was a methods guy first, a stats guy, who knew game theory, but who knew—and I still know—a ton substantively about elections. That was how I sold myself. I was four years in grad school. Even back in the day, five and six years was pretty typical in political science. But Matt McCubbins, my main advisor, was going to go to Stanford for the year, and he was like, "You can come to Stanford for the year, or we can work something else out." I had already been in touch with and gotten to know Gary King, who was at Harvard. He was a superstar. He had just gotten tenure at Harvard. It was a big deal. The year that Gary King got tenure internally, no one get tenure internally at Harvard. In fact, a little aside—not to boast; it was never going to happen—I actually gave a job talk at Harvard two years into graduate school. I lost out to a guy who had been in grad school for 14 years.
KATZ: I had no business giving that job talk.
ZIERLER: All of this tells me that your graduate experience was really a postdoc.
KATZ: Yeah, I was a postdoc. I was treated as a postdoc. Anyway, Matt is going here; I'm like, "I don't need to go to—" I've been hanging out with Matt. I had this relationship, through Neal, with Gary King, who was a superstar who had, at the time, what was called the Data Center. He had a postdoc.
ZIERLER: You're a bachelor at this point? There's no two-body problem?
KATZ: Oh, that's right, we didn't even talk—so the other fortuitous thing is I meet Natasha the first week of graduate school. She had gone to Chicago as an undergraduate. She was kind of like me, didn't really know what to do. She had a faculty member who she liked in Sociology. She actually wanted to go to graduate school in anthropology, but she said, "Oh, but you're a sociology major, so you can't possibly go to grad school in anthropology." That's utter BS, but that was what she was told. But this guy Rick Biernacki, who was a superstar sociologist, gets recruited to UC San Diego. He goes, "You should come to San Diego with me." She applies. She has never been to San Diego. A classmate of mine, a political theorist who does like political philosophy, had gone to school with her. They weren't close friends, but they had friends in common. The GSA, the Graduate Student Association, was throwing the welcome party the first week of graduate school. He goes, "Oh, I'm going to this party. I'm going to meet up with this friend I went to Chicago with, who wants to come." So I went. I was totally infatuated with her. She couldn't stand me. But I'm very persistent. We started dating, and the rest, as they say, is history. She told me at first when we started dating I was going to be her two-week fling. September 17th will be 33 years.
ZIERLER: Oh, wow. Married in graduate school or after?
KATZ: We did get married in graduate school. We got married in 1993, so we had been in grad school three years. We lived together two years.
ZIERLER: Any Jewish customs in the wedding, just to see what you retained?
KATZ: None. Nothing. Neither of us. We actually eloped. My parents were decimated financially because of the divorce and my father's issues. Her father was a pastor. All of our friends were in grad school at all these other places.
ZIERLER: Quite sensible option, that way. Just keep it easy.
KATZ: In fact, it was very sweet. San Diego was a very small community at the time. There was this great couple. They were the first sort of superstars to move to San Diego in political science. Susan Shirk, she actually was on NPR the other day; she studies China, and she ran the China desk in the Clinton administration in the State Department. She is a very senior—she's got one of these crazy titles of deputy associate secretary of State, but she ran the China desk. And Sam Popkin. Because they moved to San Diego in 1970, they had this beautiful home on top of Mount Soledad. I don't know how much you know about San Diego, but it's like the prime—it's like ocean views. They live on top of the hill. Because they could afford it in 1970. Today, the houses—Billionaire Lane. They offered that we could have it at their place. I'm like, "Are we really going to fly our friends in for like some cheap champagne?" Her sister is five years younger, and was going to college at the time in San Francisco. We were going to up there already to visit her for the holidays, for Christmas time. So we have this stupid anniversary. Our wedding anniversary is December 27th.
KATZ: I didn't even know what I was doing. I arranged to get married at City Hall at like 10:00 a.m. or 10:30. I thought it was going to be like in a basement, like the guy in a polyester suit. No, this African American guy, very deep voice—it was even deeper because he had just gotten over a cold. We got married in the rotunda, with just random people walking by. He did this whole spiel on the sanctity of marriage. We got married. We went and had a beautiful lunch at this place called Boulevard, which is a restaurant that's still there. It was then a brand-new restaurant. A friend of ours got us in. We went to Napa for a couple days. That was when we got married. So, we were already together. Natasha is a classic—unfortunately it still happens today—I had a reputation as being a superstar in grad school, so her advisors, even though they were in Sociology, they knew who I was. Literally, when one of her advisors found out that she was engaged to me—like a normal person would say, "Congratulations"—he's like, "You're destroying your career! What are you doing?" In public, in the department lounge.
ZIERLER: Whoa. So the postdoc considerations that you had, and you balanced that with her.
KATZ: She had already passed candidacy. Gary was a good friend, even then. I had a postdoc and they arranged for her to be a visiting graduate student. She made money. She was an RA for a book—Brady, Schlozman, and Verba—Sid Verba is a great—did you ever hear that name?
KATZ: Unfortunately he passed. She was the RA on the last book Sid wrote, which was on political participation. She was an RA on that, and then she did some TA'ing. I was a postdoc in this political economy program, and I also got extra money—unfortunately, the technical person who ran the Data Center—not Gary; he was the faculty director—got a better job, so for a while I was running it. it was good for us, because we actually had money, because I was basically doing two jobs. This goes back to status. So, I'm there, and I'm on the job market, and I'm having a very successful job market. I was invited to Princeton, and I gave job talks at Caltech, Minnesota, Berkeley, Northwestern. I gave like 12 job talks. Literally, I kid you not, a Harvard grad student comes up to me and says, "You went to UC San Diego. Why are you getting all these talks?" I said, "Because I have technical skills and I have real publications!" [laughs] I was more like a postdoc. I was definitely a postdoc when I was at Harvard. We started a project, which we still work on together, Gary and I.
ZIERLER: What did you work on when you were at Harvard?
KATZ: I finished up the papers that I had done in grad school. As I told you, all of quantitative models of politics were about American politics. Well, what we do know about American politics? We basically get to model two parties. The nice thing there is that it's one-dimensional, because once I know the Democratic share of the vote, I know the Republican share of the vote, so you can basically use regression tools. Suppose I want to model British politics. And we'll even simplify it—like the three-party—at the time, the Tories, Labor, and the Liberals. How do I model that? What do I do? There were these techniques developed for that type of data. The problem is you have this built-in dependence. Why? The vote shares have to add up to one. Once I know two of the parties' vote shares, I automatically know the third. This creates a strong dependence structure, so I need a distribution that does this. The problem is there's exactly one distribution, essentially, that has this structure for multivariant data like that, and it's called the Dirichlet distribution. It's used extensively in physics. It has a really problematic issue, though, which is that it assumes that the correlations between all of the shares are negative. It could be zero, but it has to be zero or negative. It can't be positive. It can't be like this particular district leans liberal, and so the Liberals or Labor could do well, but the Conservative—and they all pull vote from the—you can't have that, which happens. I don't even know how we got on this topic, but we were interested. I started reading some stuff, and there were these early models on basically thinking about transforming the space. A standard solution when you don't have anything on the constrained space is you ask, is there an unconstrained problem that is isomorphic? It turns out that if you model what are called the logits, basically the log of—you choose some party to be the numeraire, doesn't matter which one. Then you model the log of—let's say Party C is the numeraire party—the log of vote share A divided by the log of vote share B, of C, and a second equation that models the vote share of log of B over the share of C. On the two-dimensional, that's unbounded, and it's a monotonic transformation, so I can convert back to the original—if the votes live in a triangle, so a vertice is one party getting all the votes—so Liberals get all the votes here, Labor gets all the votes here, and Conservatives get all the votes here. I had this epiphany that these people were—this was actually being developed in geology, for the most part actually, geochemistry, because they get these sediments. They do a core sample, and they want to know what fraction is iron, what fraction is some sort of salts. They got very interested in this. This guy Atkinson developed these techniques on these, but he had the same problem, that he didn't want to have this Dirichlet structure; he wanted a more general structure. We basically did that, but there were some problems. For core samples, there's no real zeroes. The problem is you can't divide by zero. But literally you can have a party—the party might not contest a particular constituency for strategic reasons. We had to figure out how to do that, so we figured out how to do that. It's a missing data problem informed by our understanding of politics. He and I wrote this paper that also appeared in the APSR.
It was funny, because fast forward, we're now—I was a postdoc in 1994, 1995, so we're almost 30 years—we actually have a paper that we just presented with a grad student here, Danny Ebanks, that basically goes full-circle. Gary wrote a series of papers with Andrew Gelman in the early 1990s about how to add on the two-party data with this structure. Although they had to cheat, because some things they wanted to do weren't feasible given computational techniques in the early 1990s. We can now do that fully, but also it turns out that we noticed something that no one noticed in their data; their model actually doesn't fit really well. Their model assumed conditional on covariance. Things like what's going on in the district, the vote share in district one is independent from the vote share of district two. Well, that's just not true.
It turns out that when we go to this logit scale, the transform model, and we allow correlations across districts, we get a model that fits way better. What no one looked at back in their day is—their model predicts events that should happen like one in ten thousand times happen like 20 times an election year. It's not a very good model. But we weren't thinking that way. The change with machine learning and thinking about out-of-sample prediction was not really something we did. This was the best we could do with the model. "It looks okay." [laughs] This paper we have is in fact going full circle and uniting all the stuff that Gary and I did for these multiparty elections to his old stuff, and then also the stuff that Neal Beck and I did, which is really playing in the importance of this non-independence. That's a research tie-through that would not have been expected necessarily, but it's only possible because of changes in technology.
ZIERLER: Did you have a good time at Harvard?
KATZ: It was weird. Being a postdoc is weird. Remember, at the time, social scientists really didn't do postdocs. This is now a change; there's lots of postdocs now. But when I was a postdoc, you were a unicorn. The Harvard students didn't know what to make of me, because I was from San Diego, and I knew all the faculty members, and I knew them by first names. I get there the first week, and a very well-known political scientist, Ken Shepsle, another now-member of the National Academy—one day I'm at Harvard, I've known Ken for a long time, I've seen him at meetings—I knock on Ken's door and I say, "Ken, do you want to go grab lunch?" "Sure." So we go off and grab lunch. And literally I was accosted when I got back from lunch by Harvard graduate students—"You went to lunch with Ken Shepsle!" I'm like, "Yeah. He eats lunch most days. Knock on his door!" [laughs]
ZIERLER: "You can do this!"
KATZ: I was friendly on a collegial level with Harvard faculty, and that was something that the Harvard grad students—and I was having this great job market, and I had all these publications, and so we didn't really get along with the grad students, because they didn't—and we were only going to be there for a year, not even a year. I was on the job market and we were stressed about that. Looking back, I had no reason to be stressed, but it's easier said than done. It was a weird year. We were making money but we were still broke. We lived in a crummy, crummy apartment in Somerville, in Davis Square. We made a couple friends who were not in academia. That worked out. But it was sort of a holding year. Intellectually, I had a great time being around Gary and the rest of the people I knew at Harvard. I got projects started. But it wasn't a great year. [laughs]
The Unique Positioning of Social Science at Caltech
ZIERLER: I think as a narrative turning point, this is a great place to think about our next discussion, but last question for today—when you were considering all of these offers, especially since you had so many to choose from, was there something special about Caltech?
KATZ: Yeah. It goes back to the Howard Rosenthal—obviously I knew a lot more about Caltech the second go-around. They made a great offer. It's not true as much anymore, but back in the day, Caltech, especially in the social sciences, made well above market offers. Because Caltech has this different structure; you may or may not know this. Most universities operate on quote-unquote nine-month salaries. For a bench scientist, that doesn't matter very much, because you always have grants, so you always get your 12-month salary. Social scientists, it's very difficult to raise. Caltech pays everyone on a 12-month calendar, so it was a great salary offer. I also knew intellectually—I knew people who had been faculty members here. One of the people at Harvard was Mo Fiorina, who had come from Caltech there, and he was a friend. John Ferejohn, he would come in often to Harvard. Matt and Gary. There was definitely a lot of push to come to Caltech, and it was a great offer. I also knew it was a weird place.
ZIERLER: A good kind of weird?
KATZ: Both, actually. Two dimensions. And I'll leave this to you, because this is part of the reason I left Caltech; it's weird being a social scientist at Caltech. It has gotten a lot better, but literally when I first got here, you definitely felt that you were a second-class citizen. Leading to this was the provost at the time—when Natasha and I were formally trying to solve our two-body problem, I went out on the job market. I had been at Caltech for two years. Again, I get a whole bunch of offers—I'm very fortunate—the most viable one being Chicago. She had a great postdoc offer there. They made me a ridiculous financial offer. At the time, her paternal grandparents were living in the suburbs. Steve Koonin, who was the provost—now, it's a cool thing: I'm an assistant professor at Caltech, and the provost is taking me out, ostensibly for retention. At no other university does the provost take an assistant professor out who is considering an offer somewhere else. And he means this genuinely. He's trying to recruit me to stay at Caltech. But he's a physicist, and he does the physicist thing. He goes, "If physicists did social science, we'd be done in two years."
KATZ: "This is your pitch about why I should stay at Caltech?" [laughs] Besides the fact that my wife doesn't—they did some things for Natasha. She could teach here for three years. But there was really no community for her here. That's why it was weird to be here. The other thing that was weird for me was at Caltech, in the social science group, at the time I was probably one of the least technical people, whereas in any other political science department, I'd be like two standard deviations above anyone else. That was weird. Again, the political science group here, you have to remember, is small. The main group are economic theorists and experimentalists. So, you're a second-class citizen within a second-class citizen. It's a little weird doing what I do here.
ZIERLER: I'll just point out, in historical perspective, this is a good 25 years after the development of HSS, that you would think that some of these issues would have been worked through a little better than they had been at that point. I guess it depends what your starting point is.
KATZ: Yes, it depends what your starting point was. I had been at San Diego, where social sciences were really important, Harvard, where they were super important. MIT, not so much. It was just a weird thing. HSS at any other university is two thirds of the faculty. We're one sixth of faculty. Yes, it has improved a lot. Back in the day, the colleagues here thought, well before I arrived here, literally that, "What do you do? That's not science." I think that is gone. A statement like Steve Koonin's, Dave Tirrell would never utter such a—in part because Dave Tirrell is a really nice guy.
ZIERLER: Also not a physicist.
KATZ: He's also not a physicist. But even the physicists today would probably not say that. So, it has definitely improved. But like I said, how a superstar social scientist is treated—like I'm not treated the same as a superstar—I'm a member of the American Academy. I'm not a member of the National Academy yet. I didn't have a named chair until after I was elected to the American Academy, and I had to raise it. This division had almost no named chairs. In every other division, there's just a bazillion. Literally they have empty ones, because they don't have bodies to fill them.
ZIERLER: In a way, you're really carrying the original mantle of Fiorina and Roger Noll and all of those guys. Still happening.
KATZ: Still happening.
ZIERLER: Let's pick all of this up next time. We'll develop it more.
KATZ: Sounds good.
[End of Recording]
ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Monday, September 12th, 2022. It is great to be back with Professor Jonathan Katz. Jonathan, once again, thanks for having me to your office.
KATZ: My pleasure.
ZIERLER: Today I want to pick up on the theme we left off last time, where you got to Caltech, you joined the faculty, and it was weird, as you said, in both senses of the word—the good weird that we all know and love about Caltech, but also the not-so-good weird, just by virtue of being in HSS and what HSS means within Caltech.
ZIERLER: Maybe as foreshadowing to your decision to move to Chicago only three years later, how much of that was about the initial frustrations you felt about where Caltech fit in and where it didn't circa the mid 1990s?
KATZ: The main driving force was, because of Caltech's limited scope in the social sciences, when my wife and I were trying to solve our joint career issues, there were just not really any good opportunities at Caltech, or particularly in the area. That was the primary thing. But there was a little bit of a push/pull, in the sense that—like I said, Caltech is a great place. The things that I love about Caltech in general—I always like to say that lots of universities—it has been a buzzword since I was in graduate school and continues to be a buzzword—is "interdisciplinary." At most places, it's just lip service. I would say the places that do it best actually are Caltech and Chicago, for different reasons, although for similar reasons on scale; they're smaller places. Caltech does it because we don't have departments, with the exception being that EAS is so large it actually does divide into departments, although broader departments than most places. In HSS, we're weird because we actually run as two separate halves, the humanities half and the social science half. But the social science half, as is the humanities half, is incredibly broad. I have colleagues who are historians, anthropologists, now cognitive neuroscientists, economists, political scientists, and everything in between. We have almost all of the social sciences with the exception of we don't have any sociologists, we don't have any physical anthropology. But physical anthropology is more separate anyway, these days. That's something that's really good about—
ZIERLER: What about statisticians? Where do they fit in?
KATZ: That's funny, because Caltech, like MIT, my undergrad alma mater, has no statistics department. Until recently, everyone who did statistics at the Institute is in social sciences. It's myself, Bob Sherman, Matthew Shum, and then a bunch of other—Yi Xin. As an aside, statistics always has an uncomfortable thing, because statistics is usually grouped with mathematics, because that's where it came out of, but mathematicians, especially at places like Caltech which are very theoretical, they think statistics is boring, because basically the math was basically worked out. The hard math was worked out in the 1800s. It's just not interesting theoretically. Did you ever talk to Gary Lorden?
KATZ: You should, actually. He's still around. He's retired. He was the lone statistician at Caltech in the Math Department. He was also VP for Student Affairs for a very, very long time.
ZIERLER: Oh, wow.
KATZ: He lives in the neighborhood. I haven't seen him in a while. I think he's still okay? He might be someone interesting for you to talk to. He must have retired a decade ago. To the best of my knowledge, he lives in the area. So, that's the other thing—statistics is sort of not something we do here. The plus side, what I loved about Caltech, in the social science group in particular, was there weren't these—as my work sits between these borders, there aren't borders. That's great. It's highly technical. As you probably know in the social sciences, there's this tension between what I would call the science end of the social sciences and the anti-science or "science is impossible"—like, "It's an oxymoron to say social sciences." This is particularly big in anthropology—the legacy of Clifford Geertz and company at Chicago, actually—which is all you can do is thick descriptions; generalizations are just hopeless. What I liked at Caltech was this interdisciplinary nature, technical, science, uniform. And it's a great place to be a faculty member. Jean-Lou Chameau, when he was president, he didn't like me describing Caltech this way, but I actually think it's a very true and honest evaluation—Caltech is not a university. It's a research Institute that has a few students.
KATZ: It's just silly. We have 300 tenure-track faculty, roughly, and we have 950 undergraduates.
ZIERLER: It's much smaller than the local high schools.
KATZ: Yes, exactly! It's crazy. Those things, I really liked. That's the positive side. We all work together. The agreement in the Division is that we teach broadly and that we interact with people. I wrote papers and did things which I would never have done—I had a side project that I loved—it didn't do well—on applying something called ambiguity aversion to problems of political economy. Typically when people think about measuring uncertainty, they think about it probabilistically. But it turns out there's lots of things where that's not enough uncertainty. The most famous example of it is the equity premium problem. The reason you get a high return for stocks is because there's more volatility. You could lose money, so you have to be paid for that over, say, investing them in treasuries, which are basically presumably 100% safe. But the premium is like six or seven percentage points annualized. So you ask, how much risk aversion must there be, and the answer is, people would have to be so risk averse that they would like never leave their house. Like we should all be agoraphobics, if you really think this is the level of—so it turns out ambiguity aversion is the notion that some problems, you can't actually capture by just a simple probability distribution. It's the old Haig line, "the unknown unknowns." I don't even know how to characterize what the unknowns are. They are these black swan events. There are ways of calculating preferences that incorporate this ambiguity aversion, and it turns out that it becomes much easier to explain certain behaviors, because people are very ambiguity averse, even if they're only mildly risk averse. I wrote this really cool paper with my colleague Paolo Ghirardato. We started at Caltech together at the same time. That was really exciting and great. The downside is twofold. It's really, really small. This was even a greater issue in the mid 1990s when I got here. The internet was there. I literally have had an email account since 1984. [laughs]
ZIERLER: That's early adoption right there.
KATZ: But it wasn't this online world. Literally, when you submitted papers to journals, I would photocopy and send three hard copies. There wasn't this notion that there was easy access. There was no Zoom. Being at a small place meant that we were limited both in my colleagues, and we were also limited in how many speakers in my area we'd bring through. That's the downside. Then there's the complicated problem, which is that at any other university, humanities and social sciences would be basically two thirds of the faculty, and at Caltech, we're 55 over 300, so less than [laughs] a sixth, somewhere around a sixth of the faculty. Those are the tensions. So, there was a scale issue. That also comes into this joint hiring issue. Everyone in my generation, and definitely the younger generations now, we all need our spouses, our partners—we often meet in graduate school. We're often two academics, or academic and academic-adjacent researchers. In a small place, it becomes very hard to solve the two-body problem. If you're Ohio State University with thousands upon thousands of slots and lots of other opportunities, it's just easier. The downside to Caltech for this was the size.
ZIERLER: But you knew all of this going in.
KATZ: Oh, sure, sure, sure. You know it all going in. But there's a difference between knowing it, and living it [laughs], one. Then, two, there was this very serious issue that Natasha and I were trying to solve our joint career issues. I went on the job market. I was very successful. I had job offers from Yale, Harvard, Chicago, Minnesota, maybe one or two other places. John Ledyard was the division chair at the time, when I was dealing with this issue. John tried to come up with something for Natasha, but again, we don't do sociology, so they offered her sort of a three-year deal where she would teach some classes and would have an office, but there was no community for her here.
We had actually decided to say no to all the other places, and then Chicago sort of came in at the last minute. Chicago was a pull for us because I think it's a great university—I still think it's a great university—Natasha had family in the area, and she had this amazing postdoc opportunity there. I told you that Chicago is one of these interdisciplinary places. One of the interdisciplinary institutes is something called Wilder House, which houses sociologists, anthropologists, philosophers, computer scientists, who are interested in issues around immigration, comparative political economy broadly construed, and she had this two-year postdoc there. She is an undergraduate alumna of Chicago, so that was also a draw. But we weren't even set to go. I will also say that Caltech is famous in my era for offering above-market salaries, in part just mechanistically. As you know, most universities pay faculty on a nine-month salary. If you're a bench scientist who raises lots of grants, you actually pay your summer support, so it is nine months and then you raise another two to three ninths depending on the university policies. Caltech agreed a long time ago that we didn't want that, and everyone is on a 12-month contract. So, we literally pay 20% more than everyone else because we did this. For social scientists who typically find it very difficult to raise summer salary, that was a big deal.
But Chicago being Chicago, they offered me a ridiculous salary. It was about a 40% pay increase. So, Natasha had a great opportunity, they made this big raise to me. Caltech did some stuff to try and keep us, and then I told you, I think I prefaced this—Steve Koonin was the provost, and he took me to lunch, which again is something that would never happen—I won't say never, because it happened at Chicago too—only at small places would the provost care about the retention of a junior faculty member. So, he took me to lunch, ostensibly to try and convince me that I should stay at Caltech. But I told you at the start of this conversation, Steve, who is a computational physicist and was big in the sort of Santa Fe Institute, he said, "Oh, if physicists did social sciences, we'd be done in two years." I'm like, "That's not really a selling point." So, we ended up going. In part, there was this pull. The grass is always greener on the other side. I got to Chicago, and my main appointment was in Political Science, and I had complementary appointments, zero-time appointments in Economics and Statistics.
Focus on Cross Sectional Modeling
ZIERLER: If we can just pause on the admin side—the research. What were your big topics that you were looking at circa 1997, 1998, the time of this change?
KATZ: I had two major projects at that time. I had the work I was doing with Neal Beck. He was on my committee but not my advisor at UC San Diego. We had written this paper on—I think I had mentioned partly how I came to Caltech and I knew my colleagues—I was interested in what are called time-series cross-sectional models or panel models, so data where you have repeated measures over time. The examples that I was interested in were like macro political economy. The quintessential examples were things like you're interested in knowing, does having left-wing governments lower economic growth rates in advanced industrialized countries? People would go out and get the OECD data on all the OECD countries, which would tell them GDP, other economic measures, and they would go out and code things about their political economy. Do they have left-wing governments? Is it a parliamentary system versus a presidential system? And so forth.
What had happened was—this was my first big paper, and actually still ends up being my biggest paper. Neal and I were interested in this question. We were interested in this idea, which has now become very popular, of cross-validation. In machine learning, cross-validation just means you split out some of your data as a training data that you fit the model, and then an out-of-sample portion which you want to use to test how well it performs out of sample. The problem is data, especially in the 1990s, was really expensive to gather. It wasn't computers gathering data. Literally, I entered, by hand [laughs], data. The question is, how might you think about doing this analysis without losing some of the data? The idea is to do cross-validation, which is that you repeatedly redo your analysis on subsets.
Now, the question with panel data or this time-series cross-section data is how to separate out—because there's a structure, right? You don't want to randomly pull out years, because there's a dynamic component. If Germany has a good year last year, it also probably has a good year this year. You have to maintain this structure. There's also what is called contemporaneous structure, so when Germany has a good year, France likely has a good year, just because they're trading partners, even after you've controlled for everything you can observe. We were playing around with this data. Someone named Geoff Garrett, who was at Duke at the time—he's now dean of the Business School at USC—gave us this data. We were looking at how they did things, and we were looking at the measures of uncertainty. One thing in the frequentist statistics world that people care about is testing, are our results statistically significant? A common thing is that people talk about a 95% confidence interval. A rule of thumb—that means that when I estimate a model, I estimate a parameter, like for a regression, that's something I don't know and I've got to fit, and it has associated statistical uncertainty, and so the way that we talk about that statistical uncertainty, if it's measured by a standard error. If the coefficient is twice as large as the standard error, that means that we're reasonably confident that we know, for example, is this coefficient positive or negative in the range. That ratio is called a t-statistic, so basically you want a t-statistic greater than two, roughly. Anything more than that is better, because that means you're estimating the thing with more precision. It's not really true, but it's close enough for government work.
People were using this model that was developed by this guy Parks in 1968, and we were looking at their standard errors, and they seemed really, really small. Like they had these t-ratios—like I said, two is really good. If you look at published results in social sciences, t-statistics of two, three, four—you don't really see much bigger than four. Social sciences have small effect sizes and lots of noise, and back then not lots of data. [laughs] We were seeing things like 10, 20, 100. Something went on. It turns on that Neal and I noted that there was a problem with this model that Parks developed. In statistics, we often prove results about asymptotic. We figure out the properties of a statistical estimator, with the thought process that we let the data get infinitely large. Why do we do that? Not because we think we have infinite data; it turns out there's a whole bunch of math tricks we can use. In particular, there's two important things called the law of large numbers. The law of large numbers is really easy. If I have independent samples, as they get really, really large—say, the sample mean—that is, if I just take the simple average—that should converge to whatever the population mean is. That's the law of large numbers. And if I want to think about the distribution of that estimator, of that sample mean, in large samples, I get to apply what's called the central limit theorem, which says regardless of what the underlying population is, that estimator has a normal distribution, converges to the true population mean, and with a known have variant, a known statistical uncertainty, which goes to zero as the sample size gets large.
So, because we had these two tools—"I have hammer"—we solve every problem by basically doing asymptotics. The problem is, well, there's 20—at the time, in the 1990s, I had maybe 20 or 30 years of data on 17 countries. That's not infinity. It turns out this estimator, we show that Parks and why everyone liked it was because it was easy for researchers to find statistically significant effects. Because while it had great asymptotics properties, Neal and I were able to show, in realistic samples, that the standard errors were biased too small. Researchers like too small. That means they can say that, "Ahh! My results hold! They're statistically significant." We all figured this out from doing this problem. Then Neal and I actually then went on to develop a standard error estimate for these types of data that was—it actually was biased the other way. So it wasn't correct, but it was biased the way you should think about it in science, which is it's going to be too large, not too small. [laughs]
We wrote this paper, and actually we thought it was a pretty good paper, and we sent it actually to the AJPS. This was actually when I was a Caltech faculty member, or just before. It got rejected. AJPS was sort of like the number two journal. We got some new comments, and we actually then, on a lark, sent it to the APSR, the American Political Science Review, which is the flagship journal. It gets accepted. Actually right now—there was just the centennial of the American Political Science Review, about a decade ago—it's actually one of the ten most cited articles ever published in the American Political Science Review.
ZIERLER: What were the key conclusions?
KATZ: It wasn't a conclusion. What we provided was a tool. We showed that these models that everyone—this type of data is very commonly used, this sort of repeated measure, in social sciences, for good reasons—we provided people with a tool to how to make proper inferences from it. In some sense, it's unfair that we have lots of cites. We have lots of cites because we didn't have a result, a substantive result; we had a tool that other people could use in a variety of questions they were interested in. Something which is now common—we actually wrote code—there was a standard statistical package that people used, that they could implement this relatively easily. Now, in political methodology, my area, this is very common. Whenever you develop a new technique, you always develop software now so Joe Average User doesn't have to figure out the math. I have mixed views about this, because in part why this world went so astray was because people didn't really understand the math, the assumptions of what they were doing, but that's the world we live in. Neal and I then developed other methodological issues for these time-series cross-sectional data. I was working on that at the time.
The other big project which ended up being a book that I wrote with Gary Cox—Gary Cox was one of my advisors. Gary Cox actually is a double Caltech alum. He was an undergraduate and grad student at Caltech. We were interested in looking at American elections. At the time, in the 1990s, it's hard to imagine now, but elections weren't very partisan. They weren't very polarized. In fact, members of Congress were thought to not run on partisan labels but to try to run—and incumbents did very well, and so the question was, what was the incumbency advantage? Why did incumbents seem to do better than non-incumbents? We were working on that question, and we wrote a paper before I got to Caltech.
We were able to adjudicate two theories. It's not really relevant what the two theories are. But it got us to think, "What changed?" This all changes in the mid 1960s, and it got us thinking, "What changed in the 1960s?" It turns out something that had been noted before but sort of dismissed was—redistricting. In 1962, the U.S. Supreme Court passes Baker v. Carr, which holds that one person, one vote rules. That actually only held to state legislative elections, not to Congress. Then in 1964, in Wesberry v. Sanders, by actually slightly different legal reasoning, they not only required that there should be equal size districts but in fact actually the standards are even greater. Literally, state legislatures can have a variance of deviations in population by a few percentage points. Literally, in Congressional elections these days, it has to be absolutely perfect, which is an absurdity, but that's where we're at. Because then what happens is that there's this massive wave of redistricting that happens in the 1966, 1968, 1970. Literally about a third of the states redraw their districts. We wanted to know did that have an impact on congressional elections that people seemed to have missed, including actually someone who was here; Bruce Cain did a lot of work on this. They sort of treated the problem often times, as a natural experiment as scientists did, given the technology at the time, they just looked at redrawn districts and non-redrawn districts, and they didn't see any real effect. Then Bruce would also argue, "Well, yeah, so there's a gerrymandering—there's a Republican gerrymander in California—" It's hard to imagine, but California was a Republican state back in the 1970s. That was undone by a Democratic gerrymander in Illinois or in New York.
What Gary and I did, by thinking a little bit about the politics, basically the modeling, how districts got drawn and what the Supreme Court's entry did, we were able to say, "Well, no, there's actually this heterogeneity. It matters what the configuration was." Was the map being drawn by a uniform Democratic legislature, or was it actually a compromise between Democrats and Republicans, because the states had to divide control, perhaps because there's a Democratic governor and a Democratic governor and a Republican legislature. We showed that there were massive effects of these things. A lot of what I was doing in that early period was Gary and I were working on what ended up being a series of articles and a book. That was probably the majority of the time, and then I had a bunch of side projects. That book came out in 2002, but like all publishing—the research was actually done in the late 1990s, and the writing sort of late 1999, 2000, to eventually coming out in 2002.
ZIERLER: Was the two-body problem such that you were putting out feelers to see what else could happen? Did Chicago reach out to you?
KATZ: I actually applied. I was the one who was reaching out. I was fortunate that I had a great portfolio, and so it was not hard for me to get job offers. I let these places know that I would be interested in moving, because Natasha and I were trying to solve our joint career issues. Like I said, Chicago moved very late, because Chicago has a crazy structure where there are two parallel processes. You get hired in a department. My official title at Chicago was I was an assistant professor of political science, and in the College. The Liberal Arts College actually is a separate administrative entity, and so they have to approve your offers. That means that basically it takes twice as long [laughs] for them to get offers approved, because you basically have these two parallel tracks of administration that have to sign off on everything. They were very late in the game. We had said, "We're going to stay" and maybe Natasha would take this deal, or figure it out.
Long story short, they made me this offer. I was a little frustrated. I'm jumping around a little bit, but the other thing is that—so I do political science, political economy, which is a small part of what the social scientists do. So it's not only that I was a second-class citizen being a social scientist at Caltech, but in the Division, I was not in the mainstay of what the social scientists did. So, this looked like an attractive opportunity. It's a great university. Natasha had a great opportunity. So, we went. It's weird, I went from being—at Caltech at the time, I was probably considered one of the least mathy people in the Division, in the social sciences, to in my main department, and even in Economics, I was anchoring the sort of most technical person [laughs]. I actually wrote proofs of things, which like that was just something they didn't do. It was a weird time at Chicago, but Chicago was the center point of this fight between the social science view of political science and the interpretivist or the non-social science view.
ZIERLER: Why Chicago? What's the history here?
KATZ: Because the first place this "impossibility" of the social sciences really came out of was Chicago Anthropology. Then, because Chicago is a small, integrated place, that had influence in other social science departments, in particular Sociology and Political Science. Economics was always exterior to this. Economics has always been a sort of insular discipline.
ZIERLER: A School, if you will.
KATZ: Yes, a School, exactly! [laughs] Particularly at Chicago, which also ended up being its own School of Economics. I think that's what insulated them. There were these two juggernauts, and they had an uneasy détente. You don't introduce this stuff into the Economics Department, and we're not going to try and colonialize [laughs] your department. In fact, the reason I was hired was in fact to try and—Chicago was famous—they had just lost a whole bunch of really good people, including David Laitin and Jim Fearon, who both are members—Gary Cox is a member of the National Academy of Sciences. David Laitin was. And then when he moved to Stanford, Jim Fearon became elected to the National Academy of Sciences in political science. I think there are 18 political scientists in the National Academy. But they left in part for this, and Stanford was building up the juggernaut on the other side. They were building up the very social science-y, quantitative—so they left. In fact, them bringing me in, even though I was junior, was to try and rebuild what I would call mainstream or quantitative social science, which is a little weird, because I was a junior person. It was particularly weird, because—although I don't know salaries—I was better paid [laughs] than many of my colleagues who had been there a long time.
The university, I loved. My main department was dysfunctional, to say the least. It was like a comedy, if I didn't live it. We would have four-hour faculty meetings. Literally, we'd think about hiring someone, and they're literally going through footnotes of people's books. Like I'm all for close reading, but this is not why you should be deciding whether or not you're going to hire a tenured faculty member, what they said in footnote 32 on page 19 of their book. Anyway, that's where we were. Because all the people like me had left—there was only a handful of us left—I was just constantly outvoted even though the goal was we were supposed to rebuild some of this area. It was very frustrating to be there. The other thing that—well, it wasn't problematic; it was good for me—my first year, I was a normal faculty member, but I actually applied—you've heard of the Olin Foundation?
KATZ: Which is now out of business. They intentionally went out of business. It was always the plan. It had a cap on length of life. I was one of the last—one of the things the Olin Foundation had, for people who do work that I do, was to basically pay for a year of sabbatical. So, I got to Chicago. The first year, I taught my regular course load, but then I won this Olin Fellowship where basically the Olin Foundation paid my entire Chicago salary, and so then I didn't have to teach. That also caused [laughs] a little bit of—
KATZ: —tension [laughs]. I was a young kid. To be fair, I'm pretty strong-willed, but I was probably just downright obnoxious in my late twenties. [laughs]
ZIERLER: You came in with tenure?
KATZ: No, I wasn't. I was hired as a junior faculty member.
ZIERLER: They didn't reset the clock, did they?
KATZ: No. We'll get to the end of the story. So, I was really unhappy at Chicago. My wife also then decided that she didn't want to be an academic, that this sort of sitting in a room, writing papers—she was actually interested in the policy. She does work on immigration, and she hated the methodological discussions. But when you talk to academics about these things, you talk about your research, that's often what they want to talk about. So, she decided that wasn't what she wanted to do. She didn't want to be an academic. I was clinically unhappy in my department.
ZIERLER: Mostly because of this dynamic where you were getting outvoted, and what that represented?
KATZ: Getting outvoted, and what that represented. I was brought in with the hope—I was actually catholic with a little "c." Although one of my closest friends at Chicago—she's still a dear friend—Lisa Wedeen, she studies Arabic politics. Her first book was on comedy in the Middle East and its political ramifications. Like, dyed in the wool interpretivist, super smart, but anyway, I had no problem with that; it just wasn't what I do. I thought I was brought in with the understanding that these would both be allowed to be in the Department, and it was pretty clear, even in my short time there, that this was not going to happen.
ZIERLER: Who won the argument to hire you, and then who lost the argument that this wasn't going to fly?
KATZ: Mark Hansen, who is still there, although he left Chicago for a while and came back. There were a bunch of people still there, and a bunch of them, as they were leaving, voted for my appointment, and that was enough sort of to get it over the hump. The problem that the other side had for not making me an offer is that I had this like phenomenal record. The usual reasons you kill someone—
ZIERLER: They couldn't do anything.
KATZ: At the time I went there, I had a book and three APSR articles. This was not a concern. So, they couldn't get me on the quality issue, and there was enough votes, but then these people all left. I knew they were leaving; there was no deception. Talking about tenure, this is really funny. One of the things that Chicago did to try and keep me is they offered me tenure. To tell you how unhappy I was at Chicago, I left without tenure. When I came back to Caltech, I was tenured. Now, to be fair, for the record, you were surprised—you gave a big face, and most people would have—maybe I was too cocky, but I was never particularly worried about getting tenure, and I'll come back to that, at Caltech.
ZIERLER: How come you couldn't simply leverage the tenure offer at Caltech?
KATZ: Because it was just too quick. Because I wanted out. Here's the issue. When I wanted to leave, I decided to leave like January or February time, which—actually how the political science market works, that's actually late. We actually—because our main meeting is actually this week, the political science job market moves in the fall. Natasha was like, "You've got to leave. You're unhappy." I called up two places. I had said "no" already to Harvard once before. I called up Harvard. They made me an offer. And I called John Ledyard, and Caltech did.
ZIERLER: And you left on good terms? That wasn't a difficult call to make?
KATZ: No, no, I left on good terms, although I did resign. Chicago demanded that I resign, so I legally resigned from Caltech, so they had to hire me back. But because I was only gone for, at the time, sort of a year and eight months, the provost had agreed that they didn't have the do the whole thing. Basically as long as the faculty voted, I could come back. Had I demanded tenure, that would have been a long process.
ZIERLER: Was the line unfilled?
KATZ: Yeah, they didn't hire anyone to replace me. Then, Harvard put together an offer, and Caltech. In fact John Ledyard called me up. Basically his offer returned to me was—he figured out what my pay increase would have been [laughs] for my two years gone, so say I would have gotten 4% or 5% raises; we were still doing those back in those days. But I'm already making like way, way, way more; I literally hung up on him. [laughs] Anyway, they did eventually come to terms equal to my Chicago terms. Harvard came up with a great financial offer, but there actually my concern was tenure. I had a lot of confidence, but the politics about how the tenure system works, particularly back then, was sort of crazy. At Harvard, when they want to tenure someone, typically most universities—like when Caltech wants to tenure someone, we send out a letter to experts in the area, and we say, "Jonathan Katz is up for tenure. Tell us what you think about his work. Is he a leader in his field?" That's not how Harvard works. Harvard, the Dean's Office constructs a list of eight to ten people, and they ask the reviewers to rank everyone on the list. They don't tell you who they're considering. That's internal or external. And, it's a very large department that has very, let's say, tough politics. Let's say I'm being really optimistic, I'd say my tenure probability at Harvard was 30%. Now, given that it was zero for most junior faculty members there, that was a huge thing, but I basically know I'm going to be tenured at Caltech. So, that's what I did. I came back. Although I liked Harvard, and I did a lot of work with Gary, it just didn't make sense. So, I came back to Caltech. Before COVID, we used to always on Friday go to the Rathskeller at the Athenaeum. I remember one time being down there with John, and he was joking around, and he goes, "We'll see if you have tenure." I said, "I will have tenure in the fall. It's your choice if it's at Caltech." [laughs] So, yeah, I wasn't very worried about tenure.
ZIERLER: It worked!
KATZ: [laughs] Yeah, although I would say tenure was very bittersweet. I came in with this cohort of five other, plus or minus a year, social scientists. I'm the only one who got tenure. In fact, one of them, I told you I wrote this paper on ambiguity aversion, and that was with my colleague Paolo Ghirardato, who is a very, very dear friend. We lived in apartment buildings next door to each other on San Pasquale. Natasha, I, and Paolo would go to the movies, go to dinner. We'd cook at each other's houses. We were super close friends. The same day I got tenure, because of how the system works here, the IACC, the division chairs and the president and the provost meet once a month as an appointments committee—the same day that I got tenure, he didn't get tenure.
KATZ: Yeah. Neither did this woman, Caroline Fohlin, who was an economic historian. Normally, an email goes out, and we go out and have—we didn't celebrate, because these two people didn't get tenure. It was a weird thing. And, I found out on a Monday. There's not even a decent restaurant to go to on a Monday! [laughs] It was weird.
From Chicago Back to Caltech
ZIERLER: Did coming back to Caltech change your research purview? Or maybe I should ask, what did Chicago do to you and your exposure to the fissures in the field at that point?
KATZ: I think I wasn't there long enough, and I was sort of on the outside, that it didn't really—other than I had already had respect for Chicago, and I had more respect, although not the Political Science Department. I was fortunate. As I told you, I was a precocious kid. The provost there at the time, Geoff Stone, who was an academic lawyer, a constitutional lawyer, he actually took me to lunch, too, trying to convince me to stay. It was classic Chicago. Chicago is all about money. Not surprising. The economists really did—and the president was an economist, the late Hugo Sonnenschein. So, I go to lunch with the provost, and he's like, "What zeroes do I put on this check to have you stay?" And I'm like, "It's not actually money, and I'm going to tell you what you need to do, but you're not going to do it." I said, "You need to put the Political Science Department in receivership." Because Chicago, like Caltech, like all our peers, is a very faculty-driven place, that I knew that was not an option.
What I think Chicago did is it gave me a little greater appreciation for Caltech, not having those tensions. I've always been a broad reader. Again, I probably have a reputation in the field for being very narrow in what I do, but I actually read quite broadly in it. It didn't really affect my research career. It did make me probably like Caltech a little bit more. [laughs] The grass is always greener. But I still sometimes to this day miss what it would be like to be in a big political science department, where I would have tons of colleagues in my area. Those are always things. I think what has made this better, and as I say the only upside to COVID; now there's even more seminars online. I already interact a lot with my colleagues. I have coauthors all over the world, because communication has gotten easier and things like that. I think the technology changes made places like Caltech more viable, because you're not so reliant on your local community. You can hear talks in other places. You can more easily interact and do research with people in other places. But as we know—we're sitting here together—there's nothing that beats sharing a meal, sharing a room face to face.
ZIERLER: As you were facing a career to spend at Caltech, did that force you into collaborations with Caltech faculty that you might not have pursued, where there were more likeminded people at a bigger institution?
KATZ: I came back here, and the tenure process started immediately, and I got tenure the following—so I periodically have started up projects; none have been successful with my colleagues here, other than Michael Alvarez, who I write a lot with, but that's really later. We always talked to each other about our work. We wrote some early papers together that didn't do very well. They're actually still in the file drawer. Right now, we're working on a book together, and we co-advise a bunch of students together, so now we do much more together. But I think when I came back, I did more of what I was already doing. I finished up my project with Gary. Then I started doing a series of papers with Andrew Gelman, who is a statistician, political scientist at Columbia. So, my collaborations have been this.
In fact, it's really weird; I'm probably the only social scientist at Caltech who has never done an experiment. Caltech is the home of experimental economics. Literally Charlie Plott is one of the primary reasons why it exists as an accepted subfield in economics. It's also a big deal now in political science, with people like Tom Palfrey. Periodically, Tom and I have worked on things that didn't work out. I tell my grad students that just as important as knowing when to start a paper or a project is knowing when to say, "Mm, this is not going anywhere." For example, Tom and I had this idea about, in models of political economy—this is right around when I became involved with the Voting Technology Project, with the 2000 presidential election, Bush v. Gore, this ultra-close election. It turns out there are some problems in game theoretic models of politics which is, why does anyone turn out to vote? Because the probability that you're going to actually influence the election is close to zero, empirically. In fact, Andrew Gelman and I have a series of papers on this. Empirically—let's not even talk about flipping coins—your probability of casting a decisive ballot in the U.S. presidential election is in the order of ten to the minus eight. If you believe those numbers, you should go play the Powerball today, because those are the same odds you have of winning the Powerball. What Tom and I thought was, well, suppose that people have beliefs that there's sort of this region—it's not one vote, but because of litigation and stuff, it's in some range. We thought that might be an idea, that that would be enough of a wedge that you could actually get positive turnout. It turned out not to be true.
I'm a big fan of working on projects, even for a while, and saying, "You know what? This isn't going." There was something I really wanted to do—later, so now we're zooming ahead—Jean Ensminger and I are very dear friends, she's the anthropologist—we were going to have this great project. She has these three decades worth of census in Eastern Kenya. She has been working in this community, so she knows who is related to everyone, who knows everyone, and literally she has done a census every ten years, in networks. We were interested in corruption. There's a way to use a simple game called the corruption game to—you basically have people roll a die, and there's two cups, so you can do it with people who are illiterate, and you do heads and tails. If it comes up heads, you put a dollar coin in the community cup, and if it comes up tails, you put it in your cup. You can do this all on your own, so I don't see. You're going to do it ten times. We've got to do it more than ten times; you have to do it like 20 or 30 times. On average, basically they should get close to 50/50, if the coin is fair, so you can actually see, do they cheat and put a little more in theirs. We were interested in these norms of corruption, and what were the paths. Like was it within families, was it within social networks? We were going to do this, and then of course the political and safety issues in Kenya, before COVID, became disastrous, and then COVID hit, and now she's retiring. We had even lined up some funding. We had lined up some seed funding for it from the Linde Institute to do this. I'm a big one to start projects; sometimes they work, sometimes they don't.
ZIERLER: Did you start taking on grad students when you came back right from the beginning?
KATZ: I tell grad students here they probably shouldn't work too much with junior faculty, because in some sense you're a direct competitor. I didn't really start seriously working with grad students probably until about the time I came back from Chicago and I got tenure. The first student I worked with was Christina Ramirez. She's a biostatistician at UCLA. We had a colleague at the time, Jeff Dubin, who got access for her to one of the largest HIV practices. She was one of the first people who found the waning effectiveness of the antiviral protease inhibitors. But she was interested in the statistical questions. It's a tough population because you have lots of attrition. People stop taking the drugs because they have side effects, because they become homeless, because—right? So, estimating the effects of these things ended up being very hard. That's what she did. That was her distribution. She was my first student.
Then the second student was Maggie Penn, Elizabeth Penn. We had a colleague, a very important colleague here, Richard McKelvey, Dick McKelvey. Unfortunately, although he worked right up until literally two weeks before he died, he died her penultimate year, so I took over as her chair. Richard had done all the hard work; I was more just coaching her through the process. Then my first real student that I would say was firmly mine—mine and Mike—was Betsy Sinclair who is now a tenured faculty member at WashU. In fact, a funny aside—Betsy, who is probably the nicest person I have literally ever met, and she's married to a fellow Caltech grad student, who is an economist—they actually solved their joint career issues right out of the bat. He got a job offer at Northwestern in the Business School, and she got my job at Chicago. I told her not to go. I said, "I understand, you're solving the two-body problem, but it's a snake pit." They did take those jobs, because it was too hard not to. It was good universities, good places, solved the joint issue the first time around, sounded great. Four years in, she was like, "Uncle. I can't take this." I know if Betsy couldn't take it, and she's way nicer than I am [laughs], that it wasn't me! [laughs] They both moved to WashU subsequently.
Then I've had a whole string. Silvia Kim, a Korean woman who is now on faculty at American University. Now, almost all of the students that I train, I jointly train with Mike. There was this period we sort of skipped. I took on a few students early on—Maggie, and Christina—just after I got tenure, but then I became division chair in 2007. There's not enough hours in the day. So I helped Mike do a few students but that was it. I really didn't bring on any new students, full-time, being a main advisor, until I was done being division chair, and that was seven years ago.
ZIERLER: What were the circumstances of you being asked to be division chair?
KATZ: This was crazy. Jean Ensminger was the division chair. She was an odd choice, partly because she's even more fringy of what the social scientists do than me, and she had only been at Caltech a short while. I think she had been at Caltech maybe two or three years? Three years, I think. I knew Jean before she came to Caltech. She became division chair, and just her style—it's a job that's very difficult to do. It's a very large division. People don't really think about this. There are roughly 200 people in the Division, and it's multi-million dollars of budget. It's just as lot to do. As an anthropologist, you're used to always doing your own work, and so she literally did everything herself. And she's a perfectionist, which I love her for it—it's why she's a great researcher—but it's very hard when part of the deal is you've just got to make the trains run on time. By the fourth—it's a five-year term, normally—she was done.
Wasn't a problem. We had an heir apparent, a gentleman by the name of Peter Bossaerts, who was solidly in the wheelhouse of the social scientists. He started out in finance and econometrics, statistics, but had moved into doing behavioral stuff, behavioral neuroscience. Great. The problem is, Peter is from Belgium, originally, as was his wife at the time, and the had two teenage children. The one thing his wife made him promise was that his kids were going to have one year living in Europe before they graduated high school. So, he had arranged to have a visiting professorship in Lausanne, Switzerland. He thought it was going to be the year before he was going to become division chair. But, Jean steps down early, because she just can't deal with it. So, okay, then we do this crazy thing. Peter Bossaerts is appointed division chair in absentia, like he's going to do it—and then Dave Grether, who unfortunately recently passed away during COVID, who had been division chair in the 1980s, was going to be the guy on the ground to sort of sign the checks. And—it's a disaster. Doing the division chair job from in the building is awful; I couldn't imagine doing it from a nine-hour time-shift.
ZIERLER: This is before Zoom?
KATZ: Oh, this is before Zoom, yeah. I at the time had just been appointed executive officer for the social sciences. I had been DGS for a long time, the director of graduate studies for a long time, which we also skipped over. It's okay; not that exciting. By Thanksgiving time, Peter is AWOL. Susan Davis, at the time, was the division administrator, which I don't know how much you know about the structure, but that's like a chief of staff. Susan and I were just making decisions, because decisions needed to be made. Then shortly after I want to say late January, Peter calls the provost and says, "Not only can I not be division chair, I'm not coming back to Caltech." So there's this frantic search.
I was at the time 39 years old. I was not expecting to be a division chair. I'm a Gen X'er. We are a thin population. We're a really thin population at Caltech among academics, because Boomers sort of packed everything in. So, there weren't a lot of people who were viable candidates. Jean and John Ledyard sort of convinced me to be—that, "Well, you're going to have to do it at some point, anyway. You might as well just do it now." It was actually really stupid [laughs] in the sense that I was at my prime, and so basically you had zero time. I did publish papers during the time I was division chair, but basically you have no uninterrupted time. I tried everything. I blocked Fridays. Part of the portfolio was this Division does a lot of stuff with the Huntington. I even got Steve Koblik, who was then president of the—I had an office in the Huntington, so I could go hide somewhere not on campus. I was basically just finishing up projects that I had somehow started or was working on. It was just next to impossible to do any new work. Anyway, I took over, and the first year was disastrous. First of all, a lot of problems festered, including our relationship with the Huntington, that would not have, had there been an adult in the room to deal with them when they were little problems. But little problems have a way of, when not dealt with, becoming big problems. You get no training. I like to say what makes you a good academic is negatively correlated with what makes you a good manager.
KATZ: Because being a good academic is all about, "Me, me, me, me, me, me, me, me." A manager, my job is to make other people successful. Plus you're at a place like Caltech where we still insist that our division chairs be leading scholars, which is a good thing, too, and unusual in American higher education these days. These days, university administrators pretty early on in their career typically decide they're going to go the administrative route, or they're going to become active researchers.
So, I agree to this. The first year was me putting out all these problems, learning on the job. I'm like, "Okay, great." The first year, I got through it. Year two comes; that's '08, the financial meltdown. I had to deal with the financial crisis. One day—I forget the date; I should remember it—in late October, I laid off 17 people, personally. I remember doing one of them. HR offered to do it. I said, "No, that's not right," because I said, "The person who made the decision should tell these people. They have the right to face their executioner." I remember one of the people I let go, he said, "Oh, I know you're just the messenger." I said, "I'm not the messenger. I made the ultimate decision." I said, "I got input, and this is not something I do happily, but I just—I have a budget to close. We agreed on x percent cuts, and almost all our expenses are people."
ZIERLER: What kinds of jobs did you feel were the first to go?
KATZ: I took it to heart, which is, to be fair—Jean-Lou Chameau, who was president—it's not his line; I think it is sometimes ascribed to Winston Churchill, but I think it goes back further than that—"You never let a good crisis go to waste." I used it as a chance to restructure some things and to get rid of some people who, to be honest with you, should have probably been gotten rid of—
ZIERLER: So there was fat to trim?
KATZ: I don't call it fat. There were not the right people in the right jobs. Like the head of the Hixon Writing Center—the Hixon family, a very prominent Pasadena family, had given this money. To be honest with you, they didn't really give us enough. Jean Ledyard, who was the division chair at the time, hired this person who had a stellar CV but had come from a very big program, didn't really know what Caltech needed. To be honest, she really wasn't doing the job was being paid an incredible sum of money. So, for example, I used the crisis to get rid of them.
I had a bunch of lecturers who were partners of faculty. Again, that was done not because—some of them were not doing good jobs, but some of them, it was just fairness. Literally, 30% of the faculty have spouses who could teach in the Division. I operated on a rule, where I said, "Listen, as part of hiring and retention, we will hire faculty spouses for some fixed period, say three years. But the goal is that that's a transition time for them to find something in the area." So I let go some faculty spouses. That was very popular! [laughs] Anyway, that was the second year.
The other thing is I restructured—the Division had changed a lot. Remember, in the early days, the Division was really teeny, and it was mostly economic theorists and Charlie. Well, there's no postdocs, there's a small graduate program. There's no equipment. There's no labs. So, no one knew where the money was. We had an administrative structure from the days when we were this little unit. So, I ended up spending the first two years basically just finding where every dime was, finding out how we were spending it. To be honest with you, we still at the time had lots of secretarial support. Three or four faculty members shared a secretary. The answer is, even by that point, I didn't need a secretary. By the time I would tell a secretary how to book my travel for some trip online, I could have made all the choices faster than I could explain to him or her what I wanted to do. That's what we did. Then unfortunately, Susan Davis, who was a dear friend, she started to show signs of very quick onset dementia, so another thing I had to do in my third year as division chair is I had to let Susan go. That's when we hired Candace. So, I had to tell my friend and my chief of staff that she was not capable of doing the job. I thought Natasha and I did the right thing. I didn't want to do it in the office. We had her to dinner on a Friday night. We had a nice dinner, and I said, "We have to have a conversation. You're showing signs that you just aren't remembering things, and it's impacting your work life, and we're concerned about you." There were tears. It was awful. We tried to negotiate her leaving on good terms. Ultimately, I had to fire her, which was something I never wanted to do. But you had to make hard choices; this was the thing you had to do.
So, my time as division chair was very hard. The other thing was that because we didn't really need money, we spent no time fundraising, so I spent an inordinate amount of my time, probably 50% of my time—Natasha and I would do three to four events a week. I would travel two to three times a month. I raised a ton of money. I think in my time I raised about $30 million. A lot of things don't come in until later, and we raised more. Again, it was because we treated division chair and like a department, and that's not what it is.
ZIERLER: Does HSS have an alumni network to draw on, in the way other divisions do for fundraising?
KATZ: We do, actually. One of the people who is now giving a gift—Paul Young was an undergraduate here. We do, in the broad sense. A lot of the people didn't graduate in HSS, but they do things that are HSS-related. For example, Caltech had no presence in New York, pretty much, but I would go to New York all the time, so I revived what is called the Caltech Wall Street group. Through that, for example, I met Paul Young. Paul Young was one of the youngest partners ever—he's a Caltech undergraduate, PhD from MIT. He rose to be one of the youngest partners ever at Goldman Sachs. When he retired from Goldman Sachs like three years ago, he ran all of quantitative trading for Goldman Sachs. We don't know his net worth, but I know one thing—Paul was there when Goldman went public. If you remember, Goldman Sachs went from being a partnership to being a publicly traded corporation. At the time of that, every partner at Goldman was offered the chance to buy one million shares of the new entity, the new corporation. Most didn't take it, by the way, or they took a very small amount. Paul—because these are such holders, they had to list—we know Paul took his full allocation. This is back ten years ago. Ten years ago, if he had held just his Goldman Sachs shares through, they'd be worth, then, $100 million. [laughs]
ZIERLER: He could write a check for HHS.
KATZ: So, he is writing a check to Caltech now. He's my age; that's the other thing. People don't usually think about serious philanthropy until they're fifties, sixties, seventies. So, that's what I did. I spent all my time fundraising and rehabilitating the infrastructure to make the Division, which was based on the structure of this little—appendage, really—as we talked about in our first interview, HSS started out as one guy who had a house on Hill who had a couple of master's degrees from Cambridge, sharing and having the undergraduates read poetry as a way to tame the engineer masses. That was what I did. It was incredibly time-consuming.
I did my first five years as division chair. They wanted to reappoint me. I said I would not do another five years. John Ledyard, who is a dear friend, who had hired me, he had done it for ten years, and he said that was a mistake. He goes, "There's about seven years' worth to do. In the beginning, you're learning and you're getting things done. By years like seven, eight, nine, you're just—it's annoying. You now know way too much about your colleagues." I told Ed Stolper, who was the provost—our terms completely overlapped—-that if he reappointed me, I would only do it for two or three years. I didn't have it in me to go ten. He was fine with that. It was funny; when they convened a committee of the faculty to do the reappointment, I was told after the fact that only two faculty members spoke up about not wanting me reappointed, at which point I told Ed he should definitely not reappoint me, because if I in five years have only managed to piss off two of my colleagues, then clearly I'm not doing my job right! [laughs]
ZIERLER: [laughs] During this time, obviously you have so little to spend on the research that it forces you to prioritize. What was most important to you?
KATZ: To be honest with you, it was what I could get—I wish I had some grand rationale that somehow I had these priorities, but it was basically—what I could do. It was projects where I had coauthors, either graduate students or former graduate students or collaborators in other places that I could lean on to help. I had published some papers with Gabriel Katz, one of my former students who was then faculty in the U.K. I did some papers with Andrew Gelman. I had a paper with Gary Cox. Basically, it was sort of fumes. What was in the tank is what got done. I did try, especially later on, in my time as division chair, to have more carved-out time, and there was just so much to do. At least for me in order to do the division chair job right.
ZIERLER: In your interactions with the provost and the IACC, as division chair, did you have an ability to shape the conversation and those early frustrations about where HSS was within the broader constellation?
KATZ: Yes. We should come back to this. I do think that has improved in my time at Caltech, in part I think because of strategic things that we've done, in part for hiring. We have hired people who are demonstrably easier for—what I love about Caltech is that most of the faculty here are very, very successful. There's this norm of not tooting one's own horn about accolade-this or accolade-that. But what we did do is we've had more people elected to things like the American Academy of Arts and Sciences. On the social science side, we have the joint hires with Engineering. We have joint hires with Biology.
I think the Institute as a whole has realized a little bit more that a lot of what they care about, things like climate change, that these are not just engineering or scientific problems. I think the first thing where this came to light was the Voting Technology Project. You're old enough to remember election night 2000. We learned about "the hanging chads" in Florida. A couple of weeks later, then-president David Baltimore of Caltech and then-president of MIT Chuck Vest get together. They see this debacle on TV and in courts. We all learned what a chad was, even though people who weren't computer programmers from the 1970s. They were like, "This is just stupid. It's an engineering problem, so we're just going to put together a team." So they put together a team, mostly of engineers at MIT and at Caltech, including Ron Rivest—RSA, right? A Turing-winning computer scientist. Stellar people. And they really thought it was an engineering problem. The project still exists today, although it's pretty much on fumes. The reason it still exists is because it's not just an engineering problem. Elections in the United States are managed by counties, for the most part—well, townships in New England, but basically counties. There are basically 500 electoral jurisdictions, and they incredibly vary. There's L.A. County. It has ten million people. It is the largest electoral unit in the United States. The next biggest one is Cook County, at one third the size. Then we have Alpine County, California, which I think the entire county has—if it has 1,000 people—it's on the border with Nevada. There are large parts that are basically uninhabitable; they're mountains or the desert. They run elections very differently. L.A. County has a very professionalized registrar of voters. Not surprisingly, the person who runs elections in Alpine County also is their livestock person and [laughs] their—anyway. They learned that the problem was not so much the technology, although there were some technological things you could do to improve things, and we did. It was partly things like, "How do you measure performance," even. But it was also just the social sciences. How do we get this process to work?
The other problem we also learned is that basically no one cares about how the sausage gets made unless someone gets food poisoning. In 2000, and the aftermath, the Feds passed a bunch of bills, they gave a bunch of money to states to buy new equipment. That equipment is now 20+ years old. There has been no new cache of money from the federal government or anyone else to update this. Now, it's even worse. Now with the divisions in the states and we've politicized elections—the Trump lie that he didn't lose—we have registrars of voters and people not wanting to volunteer to be poll workers. Because who wants the aggravation? Things like that, were getting them to realize a little more about the social sciences. It was also our professionalization, our getting bigger. But we're still too small.
The Import of Computation
ZIERLER: In the way that things have gotten better for HSS, where is computation in that? It's computation-everything around the Institute, as a basis for divisions collaborating with other divisions.
KATZ: It has been incredibly important in HSS. A lot of the problems that people were interested in, you couldn't solve, because computation just wasn't possible. One thing that Charlie Plott did, and John Ledyard, and Jacob Goeree when he was here, was designing—the FCC was going to sell off Spectrum for the next generation of wireless carrier technologies. The problem is that you have very different players in the market. You have Verizon, AT&T; they want to buy a swath or spectrum across the entire United States. Then you have local carriers, small players, who just want in California. It turns out everything we know about auctions, we know about—simple actions, like what's caused a Vickrey auction—actually, he won the Nobel Prize of it. We know that if I just want to sell this cup of coffee that's in front of me, we know from theory and in practice what the optimal mechanism is; it's called a Vickrey auction. It's also called a second-price auction. We all submit sealed bids, and the high bidder wins, but he pays the price of whoever bid the second-highest price.
Why that works is because if they don't do that—you want to give people an incentive to put out what it's actually worth to them. This way, the top guy is willing to get it, because he knows that he's always going to have excess profit, because he's paying less than he's willing to pay. Other auctions don't have that. There are incentives for people to lie about what their valuation of a thing is, and you can get distortions, and bad things happen. But in a complicated auction where you're selling Spectrum that's sort of a jigsaw, where I'm selling multiple goods, even if they're complementary—they did it by having Caltech students—they ran sample auctions with Caltech students who were really smart, and they came up with a thing—I view it as like how we used to design airplanes. When we used to design airplane wings, aerospace engineers would have a pretty good idea of the basics of fluid flow over a wing, and then they would use those heuristics to design one. Then they'd take it to the wind tunnel over on the other side of campus, and they'd put it in and they would actually take measurements saying, "Mm, no, that didn't work quite right. Maybe if we tweak it, we can get some better lift." That's literally how you designed airplanes. Now, that's not how you design airplanes. Now, we actually know the physics. We don't know closed form solutions for the physics, but we can simulate the system. We don't need the wind tunnel; we just need my super-fast computer or the HPC computer on campus. That has fundamentally changed everything.
It's the same in social science. The things I do, the models I now fit and can do, were not possible in 1995 when I came to Caltech. Even though Caltech was willing to give me huge—for a social scientist, I had a huge startup fund. I bought a $50,000 HP workstation. It was state of the art at the time. Literally, your iPhone has more power than the $50,000 computer I had on my desk when I came to Caltech. [laughs] So, yes, computation has changed everything. That has allowed a set of questions to be answered, or asked even, that we couldn't ask. You probably think about computation as a tool. It turns out the growth in computation and sort of the networks behind it actually end up lots of fundamental theoretical questions you have to answer. We have a collaboration in the Division between the social scientists and the computer scientists. Most computer networks are these decentralized networks. Well, that's actually isomorphic to a market. [laughs] It turns out when you have very busy but these sort of peak load models, it turns out how you design the network and how you price it, for example, to get optimal performance, that's actually a question that you didn't ask in the 1970s, 1980s, 1990, because you couldn't saturate the network. [laughs] So, I think that computation has both made lots of questions easier and also shown lots of connections between areas that we didn't know these problems were really isomorphic. We were speaking a different language. Also, there are some fundamental things that actually—how we understand neural nets, which were originally designed to be how people think, well, it turns out that's not actually a very good model. At least that's not what we think is a very good model for how the brain works. So, computation has had interesting things on both sides.
ZIERLER: What about undergraduates? From your vantage point as division chair, did you have a say in the curriculum, and what kind of exposure students who came here for physics, engineering, computers, how much time they should spend in HSS?
KATZ: Jein is the answer. The core curriculum, the things that we require of all students, that is actually done by the faculty board. Obviously they take input from us, but the overall requirements are decided across the Institute. We definitely have things to say internally about what offerings we make. Caltech is a weird place. On some things, we're centralized. On the core requirements, we're very centralized on the overall requirements. But on the actual content and what we think counts as social sciences, for example, that's a prerogative of the social science faculty, and we have changed. When we're designing a major or designing requirements of what we're offering, we definitely are thinking about what are things an educated person should know, but also always having to modulo that with the fact that we're a really small place.
Part of the solution has always been that we think people are willing to teach relatively broadly, probably more broad than they are in other universities, where you just have more bodies and more specialists and you can do it. We also supplement it. We have a large number of lecturers. That's true in our division; that's also true in EAS for the same reason. There are areas that we think are important. For example, we teach psychology, because we think it's important for students to have the opportunity to learn about modern psychology and neuroscience. Now, our neuroscientists do teach some classes, but none of them are going to teach like Psych 101. [laughs] For that, we typically hire one of the postdocs who is in one of the neuroscience labs to teach that. Similarly, we had one anthropologist. We think anthropology is an interesting and important field, but we're not going to hire another anthropologist. For a long time, we've always had a long-term lecturer in anthropology who would cover, again, typically the more basic program. But we also want to balance that. We want to make sure upper division classes and other things are being taught by tenure-track faculty. That's one of selling points to undergraduates is, "Listen, you're actually going to have and meet and know Caltech faculty members." So, we definitely get to control internally what we have.
The overall numbers—Caltech demands a lot. Caltech still requires basically one course a quarter, 12 courses in Humanities and Social Sciences. That's unheard of. Maybe at Williams or at Amherst, maybe? I doubt it, actually. I joke with them, like Chicago has similar requirements, but they don't have it on the math side. So like Natasha went through all this humanities and social sciences, but she took like one calculus class, and no physics. Whereas our undergrads have to take 12 courses in HSS, and then they have to take now a year of physics, a year of mathematics, two terms of biology, plus some breadth sciences, two other labs. No one else does that. But those determinants are done at the faculty board level, but clearly we do interact. What is it that a Caltech student needs to know?
ZIERLER: You alluded to the poet, with the house on Hill . In the way that Caltech founders saw the humanities as a civilizing influence for the engineers and the scientists, does that remain at all? Or is HSS doing things that are just required for what we expect a Caltech education to be?
KATZ: I think it's more the latter. I think this notion that somehow we were taming or humanizing the engineering students, I think that went by the wayside in the 1940s and 1950s when they decided to make HSS—well, at the time just "H"—a full division, where you'd hire. I'm yet to meet a Caltech colleague outside of HSS who didn't think that HSS was important. You never get that. There's always the question about how much. Everyone gets sort of jealous about their amount in the core. I think that has definitely changed. That changed pretty quickly as Caltech matured relatively quickly in the postwar period into a mainstream Institute. Again, we're still smaller, but that yes, this is something important, and done. It does go to this sort of—we've always had free reign, and I think the HSS faculty have all decided that the way to teach HSS was to actually have them exposed to how cutting-edge researchers in the field do things.
Even in our introductory courses, we tend not to do, "You're going to know almost nothing about an entire field." This is even more clear in how the freshman humanities work. We don't do like, "We're going to tour all of Western literature in ten weeks." What do we do? You take Frosh Hums, where their requirement is that you're going to hang out for ten weeks with an expert in Medieval literature, Jenn Jahner, and she's going to tell you about that. We're not going to try and teach you everything there is to know about English literature. Similarly, Cathy Jurca is going to teach something on film studies. Or Morde Feingold is going to teach something on 20th century history of physics. Because there's always this competing world about how you should teach people about the humanities and social sciences. Do you teach them breadth, or do you teach them what HSS scholars actually do? Caltech has always I think taken the right approach. We're the only place where almost all of our students go on to graduate school. The goal was we teach them like they're young graduate students. Kid gloves, but that's what we do.
ZIERLER: Caltech undergraduates are sufficiently scholarly enough that even if they came here for—
ZIERLER: —even if they came here for robotics, they understand that hanging out and doing medieval literature or learning about politics is good for their overall education? Or is part of what HSS is doing that they have to show them that it's good for their education?
KATZ: It's a heterogeneous group. I think some students come here and thrive, and they're great students, and they do it. There are a lot of Caltech undergraduates who are like, "Okay, I'm going to figure out how I do the minimum number of things to get through my HSS requirements." Over my time at Caltech, that has become a smaller and smaller subset, I think partly just because of the boom in the college-age education has allowed them to become very picky, and so now it's not just good enough for a student to be great in STEM; they also have to show interests outside of that, which often means humanities and social sciences. So, I think that has improved, but I think that's more on the supply side. [laughs]
I definitely, when I teach them, tell them why I think this is important for them to know. A large part of life is about politics. Some of them do come in very naïve, thinking that—something we've unfortunately come to know—that just because you're a scientist and just because this is what the science says doesn't mean people are going to believe you or follow what you have to say! [laughs] The other thing I tell them, and I don't think I'm alone in this—when we do surveys of alums, the one thing—because we are also charged with teaching them communication skills—writing and presentations. They do the most of that in HSS. If you ask our alumni eight, ten years out what they wish they had more of, it's that. I tell them, "You can come up with the most brilliant idea, but if you can't explain it to someone else who's not as smart as you, it's not going anywhere." Again, I think some of them get it. I think some of them think, "Whatever, Professor." [laughs] I think the ages of them coming in and they only want to take physics and they're only going to do physics, that's pretty rare.
ZIERLER: That's a good thing.
KATZ: It's a good thing. Again, I think it's just because it has become so much more difficult to get into Caltech, and our peers. I'm glad I didn't have to apply today.
ZIERLER: Last topic for today, some self-assessment questions about your tenure as division chair. You mentioned earlier that there were a rather large number of people who did not get tenure in your cohort. What did those numbers look like when you were division chair as a sort of metric for the overall health of the Division? Because when someone doesn't get tenure, that's also to some degree a shortcoming on the part of the Division.
KATZ: First of all, we're all dealing with small n, so I can tell trends about overall groups. Tenure rates are always going to be lower in the humanities and social sciences, just because of how we hire. We tend to hire people—although now postdocs are becoming a little more common in social sciences—we hire people when they're finishing up their PhD. They might have some good ideas; you don't know if they're going to be able to make that impact. In some sense, we're always much more uncertain. Compare that to when they hire in biology, literally the average biology grad student has been in graduate school for eight years, and a postdoc for six to ten years, before you hire them as a professor. Which I think is a dysfunctional system. But when you're hiring someone who is eight years past their PhD, you know what you're getting. We're always going to be noisier, just because of this.
I think tenure rates have improved. I don't know if this speaks to health. I'm on the recruiting committee—we actually just had the Social Science Recruiting Committee, and we have a ton of recruiting to do, because we have a wave of retirements, plus some leavings, but mostly a wave of retirement. Over the next three to four years, we're going to probably need to hire ten to twelve social scientists, to just maintain numbers. When Geology makes an offer to someone, the conditional accept rate is probably 85% or 90%. In social sciences, we're a weird place. I would say to a new PhD, we're talking our accept rates average probably around 25% to 30%. So, there's going to be more turnover. I don't think actually tenure rates are the way to think about evaluation. I do think we should worry about is the Institute providing resources, mentorship. All those things are incredibly important. But what I don't want—when you say that you can only hire someone you think you're going to tenure, then you become really risk-averse on who you're going to hire.
ZIERLER: And that doesn't work for 25% conditional accept.
KATZ: Yeah, so I actually think that one out of five is probably way too low, but something around 50% is probably not—and that's probably close to where we're at—is actually probably not a bad number. Because I want to take risks. I want to hire someone who I think they're doing interesting things, but maybe a little out there. Maybe it's going to hit, maybe it's going to work out, and maybe it's not.
ZIERLER: What did you learn during your time as division chair to minimize that risk, when you see talent, it's interesting, and you want to maximize the chances of success?
KATZ: You want two things. Doing successful research. Well, there are three components to research, one which you have no say over, and I don't kid myself—being lucky is always good. There's a significant amount of luck in all of this. You want someone who has out-there ideas, but who is professionalized enough to know that they need to hedge their bets a little bit. They need some—I don't want to say boring research—but you want to see someone who has a portfolio of things. Maybe one or two things are sort of out there, and then more standard stuff that's just pushing the envelope but not necessarily sort of crazy. What I look for is someone who has an intellectual maturity. Again, this ability to talk to new hires, like, when do you call it quits on a project? There's actually an economic term for it called the sunk cost fallacy. Yeah, you spent years on this, and that's great, but if it's not working out, you shouldn't finish just because you spent two years doing it. Those are the things you look for.
You look for professionalization. Can they communicate their idea? It's not enough that they have these great ideas, but can they write? Can they sell? If you ask who makes it, who's successful, it's the people who have these great ideas, who are great salespeople. That's hard, because I think a lot of Caltech faculty, a lot of people, we tend to be a little bit on the spectrum. We tend to be a little bit more introverted. I've met lots of colleagues at other places or students who have great ideas. I have a former student of ours right now—he's brilliant, Yimeng—but he is definitely shy. On his own, he came up with a paper for—we have this second year paper requirement, for grad students—it is published in the top field journal. It's all his idea. I don't feed it to him. It wasn't like, "Here's some simple project." Literally, from start to end, he came up with the problem, he solved it, he wrote a good paper. He needed some help on writing the paper. Mike and I helped him on that. But it got published. But on his own, he can't do that. Even though he's this brilliant guy—
ZIERLER: It's not the complete package.
KATZ: It's not the complete package. He's a postdoc, but he's a postdoc at Florida State. If you were just going by his intellect and his ideas, he should be at Harvard, Princeton, Michigan.
ZIERLER: Last question for today; I'll circle back to the very first question. It's really an intangible question to answer, but—your early impressions when you got here, you knew these things coming in, but it still made things more difficult than they could have been. In your time as division chair, do you think you were able to minimize the likelihood of the next generation of young faculty coming in having those perceptions as well? Is that even doable? Is that something you can go about accomplishing? Or is that just what Caltech and HSS is, and that's the deal?
KATZ: I think to a certain extent, that's just the deal. I think it has improved just because as the Division has gotten bigger, and as our colleagues have become more understanding of what we do here, which has improved, that has gotten better. But it is never going to be 100%. If you're in English or you're in economics and you're at Princeton or Yale, you are at the center of what those places do. I don't kid myself that I am ever going to be at the center.
ZIERLER: That's never going to change here.
KATZ: We didn't talk about this, but I actually liked being an administrator. I actually went out and interviewed for a bunch of jobs, some very good ones, dean of art and science jobs. It's my fault; I could never convince them. Caltech looked so weird to them that I could never convince them that my experience somehow translated to this. I also was this weird person, because I had a very successful academic career, which is threatening to some places. They don't like a dean who has tastes and views. Anyway! So, we're never going to be that. We're never going to be the center of the universe.
ZIERLER: So in a way, the junior faculty who understand that or either deal with it or embrace it, those are successful.
KATZ: On the social science side, it's easier. The ones who embrace it—like, by the way, reaching out to connections to people like in computer science or in math—I think they're the ones who really like it here.
ZIERLER: That rewards a particular personality.
KATZ: Yes, it does. It really does. Or those in areas where we have a large group. For example, experimentalists feel very at home here, because yeah, Caltech might not be the center of what they do, but we have a ton of them. I think it's harder on the Humanities side. They don't have a graduate program. I think the historians of science have it a little bit easier, the internalists who are interested in talking to scientists about how they do what they do. But if you're a cultural lit person—you do race and literature and philosophy—or you're Jennifer Jahner, and you do Medieval literature; you're not going to be the center of what Caltech does. Again, we tried to improve that. That's why I was a big proponent of maintaining and actually expanding our relationships with the Huntington. But you can't kid yourself and ever think that we're going to be the center or even a major bit of what Caltech is about. That said, Caltech has been very good—I do say our research gets—we do Watson lectures. Actually in some sense, some of our stuff is easier to sell, so the Communications Department often comes to us.
ZIERLER: Sure, because it's an oddity, almost.
KATZ: It's an oddity, but it's also easier, right? It's things that they know about. If you have someone coming in and talking about theoretical computer science—okay? Or worse, a mathematician! You're an algebraic geometrist? Even people at Caltech don't know what that is. [laughs]
KATZ: The few mathematics things that maybe connect to some things in their lives, like cryptography, maybe you can sell that, to like why you should care about this. It's the same thing about raising money. The closer you are to saving kids, the easier it is to raise money. That's why it's hard to raise money in mathematics. It's hard to raise money in literature. It's easier in biomedical. It's easy in biology generally. There's this pecking order about where people want to make an impact with their gifts. So, it takes a little bit more to sell places like us, areas that are more fundamental.
ZIERLER: On that note, we'll pick up next time, post division chair. We'll bring the conversation right up to the present, and focus on your academics.
[End of Recording]
ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Friday, November 4th, 2022. It is great to be back once again with Professor Jonathan Katz. Jonathan, it's great to be with you again. Thanks for having me.
KATZ: My pleasure. Nice seeing you again.
ZIERLER: We're going to pick up back in 2014 when you stepped down as division chair. Just to pick up from where we were last time, all of the responsibilities, all of the roles that you played as division chair, was there any scholarship that you needed to put on the back burner that you could revisit once you stepped down?
KATZ: No, it was more the other way around. I continued to do some publishing and stuff while I was division chair, but really it was building down accumulated research. The hard part when you're division chair, especially how I did the job, you have very little uninterrupted time. Things that you've already thought about, like if I just needed to write something or work briefly with a grad student and read something, you can do that. But for me to start a new project, I need thought time. I need time to read and to think, a block of it. I tried in the beginning. I said, "Oh, I'm going to block Fridays, and I'm not going to do anything." The answer is that I did, kind of, but it's really hard. I had a sabbatical after I stepped down, and part of that was just to recharge and figure out new projects. It was almost like starting from zero again, because I had published most of the stuff that was in the pipeline.
ZIERLER: Where did you do your sabbatical?
KATZ: We didn't do it anywhere. We have a very large dog, who doesn't like to travel. We did talk about going someplace, but I just stayed here. I stayed away from campus a little bit, and I would hide up in my office. I wouldn't really see anyone. So I was on sabbatical, but here. Other than with my students—I was just starting to rebuild students, so I didn't even have that many around. I just hung out here and did my work.
ZIERLER: Did the vantage point that you gained of HSS as division chair influence next projects, just knowing at a greater view what was happening in HSS?
KATZ: No, it didn't. A little aside—Caltech is a weird place in higher ed. At most universities, academics choose early on in their career that they are going to be an administrator, and they are very different. At Caltech, it is both good at bad, we decided that in order to be an administrator, a division chair at Caltech, you have to be a distinguished scholar. The expectation is that you're promoted back to the faculty at the end of your tenure.
KATZ: In fact, the only job we look externally for in the academic administration is the president. We wouldn't even contemplate a provost or—I think once, back in the day, I think it was Geology, I think in the 1960s, they hired someone from outside to be division chair, and it ended up being a disaster. Of course, although we're scientists here, we all overgeneralize from an n of 1. Just because you had one bad choice doesn't mean it's a bad idea! When I became division chair, I actually liked it. I liked the strategic aspect. I liked fundraising. So, I threw my hat in. I had some great interviews. I thought I was going to go be a university administrator. My weirdest one, I was interviewed by the Board of Trustees to be president of Reed College, which was a long shot. The only reason I got interviewed was because Steve Koblik, who was president of the Huntington when I was division chair—one of the things I did was coordinate all the relationships with the Huntington. Steve, prior to his being president of the Huntington, he was president of Reed College. He pushed his friends in the trustees to interview me. I also interviewed for the dean of Arts and Sciences at Northwestern, NYU, Johns Hopkins.
ZIERLER: This was all during the division chair years?
KATZ: Yeah. And Duke. The problem was—and I ultimately decided—I could never convince them—Caltech is so weird that I could never convince them that my experience, even though in some sense much broader than anyone they would ever have for a dean of Arts and Sciences—because of how Caltech is structured, the division chairs actually run everything here. But Caltech, we're not a student-run institution. As I said to you before, Caltech is a research institution that has a few students. I could never convince them that I had—
ZIERLER: A transportable skill set.
KATZ: —a transportable skill set. I was also threatening, again because I didn't look like the mode of a normal university administrator, because I was a distinguished scholar. That used to be the norm. If you go back in history—Millikan here, but there were others—most senior university administrators are not particularly distinguished, and I think a lot of the faculty search committees, they're conservative with a little "c." I would have taste, and I'd have standards, and I—right? That's not something they were in the mood to I think deal with. Also, clearly I didn't effectively communicate my skillset to them.
After those interviews, and it became clear to me that nothing was going to happen, I said, "All right, I'll go back to being a faculty member, back to being a researcher." But basically, the wells were empty. I had to basically drill new holes [laughs] for things to work on, and to build up. Again, I didn't take on any students when I was division chair, so I had to get back in the pipeline. I started teaching again. I had the sabbatical, and then I started teaching after a year and a half. We have a required graduate sequence, and so that's when I got to know students. Now I have five students working with me, but until you have that pipeline and get to know the students—I would say SS is a different grad program. In other divisions, you typically apply to a graduate program, but you also kind of know for the most part whose lab you're going to work in. You're kind of admitted both to a graduate program and to someone's lab. That's not how it works here, so if you don't get to know the students—and most of the students we admit, it's kind of the Caltech problem, right? I told you before we were on the record that we don't—almost no one comes to Caltech to major in social sciences. Very few of the students coming into the program—they all want to be economic theorists. Even though the job market is terrible. It's like the mathematics of economics; the job market is terrible. Basically, we have to convince some students to come over to the empirical political science side, that "Listen, our job market is way better." All my students get jobs. I had a student, Lindsey—Lindsey is giving job talks this week and next week at Chicago and Duke. [laughs] Our economists don't place that highly. So there's this sort of salesmanship. Literally, basically for three years I was just rebuilding students, starting up projects again.
ZIERLER: Drilling new holes.
KATZ: Exactly. And like all things, some work out, and some don't.
Measuring Democracy Outcomes
ZIERLER: What was interesting to you at that point? What were your exciting new projects?
KATZ: I was doing a lot of stuff with my late advisor—now-late advisor; he was alive then [laughs]—Matt McCubbins who actually was a PhD from Caltech. We were interested in a lot of stuff related to a new area for me, which was measuring democracy and its outcomes. I'm a big believer that most research happens by happenstance, that it's always better to be lucky than good. We had this idea—you may or may not know this—NGOs like the World Bank and other NGOs literally spend millions of dollars generating annual scores of democracy and rule of law. If you actually know how the sausage is made, they are truly terrible, but they get used—
ZIERLER: Methodologically, you mean?
KATZ: Yeah, methodologically. They're based on experts, and unfortunately, especially the ones that the NGOs produce, it's a lot like U.S. News and World Reports—or "World Distorts"—rankings of colleges. They produce a new ranking, and the numbers have to change every year, because otherwise why would you look at the ranking? Countries' democracies, despite all the concern about the U.S., don't actually change all that much year to year. Our idea was that there's this cool dataset, which is a bunch of academic lawyers put together a project called the World Constitutions Project, where they literally translated and hand-coded basically every constitution they could get their hands on for the last century, on about 800 different characteristics. We said, "All right, let's see if we can forecast these things like Freedom House's polity score just from these constitutions data." We'll fit in some training data, which is past—the constitutions'—all these coded factors.
We did an okay job, but what was interesting is—here's the problem with machine learning models. They're not really interpretable. They're just looking for prediction, not causation. So, a lot of the things that were predicting countries being democratic were mechanical. They were like, "Do you have an age limit for entering Congress?" The legislature. The only one that had any surface plausibility as something that was interesting was something that's called exceptions clauses. Many democratic constitutions have exceptions clauses. The U.S. Constitution has a very narrow exceptions clause: in the time of war, the president can suspend the right of habeas corpus. But other countries have much more expansive exceptions clauses. They can suspend elections. They can suspend the legislature. So, we got this weird finding. The predictions weren't great, but there was this weird thing about these exceptions clauses, and basically all the transitions of democracy to non-democracy, with nine exceptions, were only in democracies whose constitutions had these exceptions clauses.
We looked into it. The first thing we looked at was, what about these nine cases that don't fit the model? Well, eight were easy. Eight were basically foreign powers invading, so like the Soviets invading Afghanistan, the U.S. invading Iraq. Eight of those, you could clearly show, were basically foreign intervention to topple a democratic—an elected regime. Whether it was democratic, we can—[laughs]. The lone exception to this day is actually Myanmar, or Burma, depending on your political views about what you want to refer to the country as. That was the only country that has successfully had a transition from democracy to non-democracy without an exceptions clause in their constitution. So, we went down that rabbit hole. We were working on that. That was actually one of these papers that I actually really like [laughs], and Matt kind of messed us up. Matt gave this talk at this conference, and he therefore felt obliged to submit it to this okay journal that had a special edition for the conference papers. It's a better paper than the journal it's in! In fact, we were pursuing more on this topic, and then his illness got worse, and he passed away last summer. I actually have a potentially new grad student who is going to take the reins of this project with me.
ZIERLER: Oh, wow. So, this is an ongoing effort?
KATZ: Yeah. I still think it's a really—this dataset is really cool. I think this question about how you measure things like democracy and the rule of law are very important. The other thing that is very hard methodologically is—I tell this to students all the time—there has been this move afront, and for the most part, I'm quite supportive of it, a movement that's called the credibility revolution. It's basically about the importance of causal inference in the sciences, but in social science in particular. The gold standard for estimating causal effects is randomized controlled trials, but for lots of questions we're interested in, in the social sciences, that's not possible, either ethically it's not possible or just practically it's not possible. Everything about countries is basically—you can't do a randomized controlled trial. When that doesn't work, there are statistical techniques that really on effectively quasi-randomization, but with countries, you just don't get there. So, I think unfortunately, I do actually think this is a bad trend, that we've now said, "Oh, there's a whole bunch of questions that we all agree are really important, like what drives democracy, or does rule of law lead to growth for the economy, but you can't do them, because there's no causal framework." That's my own little aside. But, I have this student, and we're interested in pursuing this project. Also some of the more statistical issues around how these things are measured.
We also did another paper which didn't do as well, that was actually related, in corporate finance. Corporate finance is the portion of finance that works on basically governance structures within companies. One thing, unsurprisingly, that scholars are interested to know is, for example, does having better governance, that is, having more transparent, more democratic—shareholder democratic—governance lead to better outcomes? There's an economist at MIT's Sloan School, Gompers, who basically got some data together, these sort of 24 characteristics, and basically created his own index. These are all binary indicators, so basically he just sums them up—zero, one. When you do that, when you sum up those things, first of all you think every one of them is the same, so like all companies that get a six or a seven are the same. But there's lots of different ways. There's literally hundreds of thousands of ways to get a six. The other thing that is really weird about them is that they—think about this: some of these factors, the companies have no choice over. It's just a function of where they are incorporated, because it's based on state law. Six of these characteristics, they're called autocratic right? Just where you're incorporated. You may or may not know that 50% of all publicly traded corporations in the United States are incorporated in one state.
KATZ: Delaware. [laughs] Who happens to have all six of these bad things. Now, I guess you could assume they incorporated in Delaware because of these provisions in state law. The answer is probably more complicated than that. There are lots of things that make Delaware an attractive place to incorporate, in part, as we saw with the Twitter fiasco, the chancery courts in Delaware are very efficient about dispatching lawsuits. [laughs] So, some of the stuff is the state where they were. Some of the stuff is in basically the founding documents of a corporation, and they are very difficult to change. Basically you need shareholders' votes to change them. Then some stuff is just like their by-laws that the Board of Trustees accepted. Treating these all the same seems sort of like an odd thing. So, we wrote this whole paper saying this measure that people use, there's other problems. If you thought these things were all measuring the same thing, they should be highly correlated. They're actually not. There's ways of looking at the dimensionality of a set of covariates, a set of variables, in this case these 24 measures. If you thought that these things were all measuring the same latent—call it corporate democracy—than if I looked at something called—do you know what eigenvalues and eigenvectors are?
ZIERLER: Sure, yeah.
KATZ: Right. Basically there should only be one primary eigenvalue, eigenvector, and the other one should just be noise. That turns out not to be the case. [laughs] So, there's all this evidence that this thing that he claims to be this sort of great measure of democracy in corporations is not. Of course, we're not in finance. We did write it with a former grad student of Mat's , who is actually on faculty in the School of Accounting at Indiana, but we still couldn't get it published, because it's going after questions that people want to answer. So, I've been getting a lot into measurement, and some of the stuff into measuring democracy and rule of law and their consequences.
Then some stuff, I went back to. I did a paper with my postdoc advisor—we still often collaborate—Gary King, and a former undergrad of his, Lizzie Rosenblatt—we wrote—the Supreme Court vacillated on whether or not partisan gerrymandering was judicial, right? They basically said in the 1980s, "It could be judicial, but we just don't know how you measure this." In particular, the swing voter at the time, which was the swing voter on most things, was Justice Kennedy. Political scientists have agreed a long time how we measure partisan fairness. It's a pretty simple idea. It's based on symmetry. We'll say electoral rule, or in this case a map, is fair if it treats parties, candidates, equally. That is, if I change their label, I don't get the outcome. So thinking about it in this case, if in Wisconsin, when the average vote share for the Republicans is, say, 54% of the vote, they get 65% of the seats, that's fair as long as in the counterfactual world, were the Democrats to get this average, they too would get 64% of the seats.
What happened is in the aftermath of Kennedy's claims, all these academics, some lawyers, proposed all these various measures of partisan fairness. Kennedy actually explicitly didn't like this counterfactual reasoning, which is weird, because actually a large part of the law is all based on counterfactual reasoning. Like all of tort law is based on counterfactual reasoning. You ask, "If I hadn't harmed you, what would you have gotten?" That's exactly how we figure out the damages under American tort law. So this idea that we're not happy with doing counterfactuals—he didn't like that in order to do these counterfactuals, practically you had to employ someone like me to actually do a statistical model. But that's okay. Again, that's also true in tort law. If you do a big case, you hire—actually, my former colleague who used to be here, Jeff Dubin, was once hired—Michelle Pfeiffer backed out of a movie contract from some independent movie producer late in the game, and they sued her for breach of contract. Basically, you have to ask, "How much would the movie have made had Michelle Pfeiffer been in it?" Versus—
ZIERLER: —whoever else replaced her.
KATZ: [laughs] So, there were all these proposed measures, but they weren't really well-thought-out. They were kind of clever. They had what we deem in the statistics and measurement literature some "surface validity." They kind of looked like they might be plausible measures. But what we actually showed is—there has been actually eight or 12 of these measures—we went through them all systematically. First of all, we laid out, "Here's how you think about fairness, and here's what you're trying to measure." Then in statistics, you ask, "How does my measure relate to this actual concept I'm trying to measure?" We basically showed that while some of these measures get some aspect of the electoral system, none of them really comport with this idea of partisan symmetry or what we actually mean by fairness in an election. So, we did that paper.
Then Gary and I—I have a student, Danny Ebanks, here—Gary King and Andrew Gelman, another one of my often coauthors, they wrote the statistical model that is the workhorse for how we analyze legislative elections. Now, they made some choices, because they developed these models in the late 1980s and early 1990s, and technologically things that we can do today, they couldn't do, so some of it's not their fault. But in particular, for a statistical model, we will call a statistical model good if it's generative; that is, if it can actually generate the data we observe. And they never really tested it. It turns out their model actually fits kind of poorly. Gary, Danny, and I now have this project that is basically to tweak—I think the nice way of putting it is extending their model to actually fit the data better. We've been working on that and validating it with basically 70 years' worth of—70 elections' worth—of Congressional elections. Once that's done—we're in the process of finishing that. The model that we do computationally intensive. I have a very high-powered computer on my desk. It has 24 cores, and 256 gigs of memory. On my machine, our model for all of the elections we analyzed takes about 18 hours.
KATZ: [laughs] Yeah. In the 1990s, it wouldn't have been practical. Some of the techniques we are using to make this possible didn't even exist them. Forget the hardware. We will also have a free software package library that others can use to analyze this. Then, Gary and I paid to generate data for—we basically have every state legislative election in the United States going back to 1968, so it's hundreds of thousands of elections. Once we get this model working and validated, we will start going back and then analyzing the plethora of historical election data. That's another big project that I have going.
ZIERLER: Going back to when you joined the faculty here and the startup package, taking advantage of the software advances, the hardware advances, when you came up for air after being division chair, were there computational advances that you might not have been alive to over those seven years that were available at that point?
KATZ: Yeah. The real key—have you ever heard of the term "Markov chain Monte Carlo"?
KATZ: A little digression. It used to be the fight between Bayesians and frequentists in statistics was more of a theological/philosophical debate. Because basically, I think many statisticians would agree that on straight philosophy of science grounds, the Bayesians have the better position, although the one thing they object to is the idea of priors, where do priors come from, but we'll come back to that. But the real issue is that it didn't matter; even if they were ahead on the philosophy of science front, except for toy problems you couldn't do anything. Because in order to get a posterior, which summarizes all of your model, you need to do integration. You need to multiply the likelihood times the prior to get the posterior.
Now, for trivial problems, where the prior had a very particular form—it's called a conjugate prior—so that we can analytically solve this integration—there was no hope. Well, there was some hope. Part and parcel, not related to statistics, in the 1950s, there were some physicists who developed this set of tools for doing integration by simulation. The idea was these random walks. Essentially, in a smart way, you randomly look over the search space, and you can build up—even though I can't actually analytically solve what the posterior distribution is, I can get samples from that posterior. That's what the Markov chain means. When we say the Markov chain is converged, it means that the draws from the parameters of the model match this posterior distribution.
Now, there's a whole bunch of hard stuff—first of all, for high-dimensional models, that is models with say lots of covariates or otherwise have lots of parameters, this random search is very, very slow. There was also a problem which was that it wasn't actually very clear how you knew that the—the theoretical results are all asymptotic. That is, as my draws of the chain go to infinity—and clearly we do things in finite time. What happened in the late 1990s first was the development of this idea independently called either Hamiltonian Markov chains or hybrid Markov chains. Instead of doing random walks, like when we actually do maximum likelihood estimation, we're maximizing an objective function—the simple way to do that if the function is differentiable is you do some sort of like Gauss-Newton method. You basically use the gradients to find the maximum. The same idea was applied to Markov chains, that instead of looking randomly around the space, the Hamiltonian is actually literally the vector field of the gradients, in a spatial sense, they are actually directional derivatives, you basically look in places of higher likelihood, with higher probability. You do search the entire space. It turns out this does a couple things. First of all, I can basically get estimates, plausible draws from a posterior distribution with many fewer iterations of my sampler. Then that was implemented in a probabilistic programming language called Stan, which started to be developed in the late 1990s and it really comes to be in alpha/beta versions in the 2000s, and now there's a full-fledged—it's a workhorse. We can use it. These models, for example the model estimating these congressional elections, the project with Gary King and with Danny Ebanks, literally our model has thousands of parameters. It has nothing to do with computation. I guess maybe if I had a supercomputer, we could have used old-school random walk Markov chains to estimate this model, but—
Big Data and Machine Learning
ZIERLER: Have you ever been compelled to visit one of the national labs and hop on one of those computers?
KATZ: No. [laughs] I have always found solutions [laughs] short of having to actually use a supercomputer. I'm actually on the steering committee for Caltech's High Performance Computing Center, and I have used that somewhat. But the answer is that these techniques have become so effective that really with an expensive computer—my computer was $10,000—that's all the computational power for this incredibly complicated model. Now, for bigger projects, like some of the stuff that my colleagues do in geo, or in physics, or in neuroscience, where they're dealing with terabytes of data, then you need to go to something like an HPC. But I do mid-sized models, and that's now perfectly feasible to do on a high-end workstation on my desk.
ZIERLER: Given the amount of data you're working with now, does machine learning take on a new level of relevance for you?
KATZ: I would say yes. I teach in IDS, the Information and Data Sciences program. I like to say I've been doing data science since we called it applied statistics.
KATZ: The answer is, kind of. First of all, most machine learning that's really good is actually mostly focused on prediction problems. That's not to say that prediction is not something we often care about, but that's not the main thing that card-carrying social scientists care about. At the end of the day, it's why the credibility revolution took place; we care about understanding the causal mechanism, like what drives elections, what drives coups in countries. We might want to predict them. If I'm at the State Department or the CIA, I might care very much about predicting coups. But most academics care more about like why the coups happen.
So, a couple things. Machine learning of tabular-type data, so the type of data like elections, financial records, the answer is there's actually nothing new there. It's basically just nonparametric regression. I can tell you why this is true. For some problems, if I wanted to parameterize that, there are some uses in machine learning. Machine learning has been incredibly successful is in things like image recognition, or in language models, like translation, but its impact on tabular-type data is pretty mild. In fact, when you actually compare really complicated machine learning models to much simpler models, like just linear regression or slight generalizations thereof, it's rarely the case that you get actually much performance gain from the more complicated models. For tabular data. That's not true— the things they do with images—like what Pietro Perona and his group does with images, that's not possible with off-the-shelf statistical models. But tabular data, machine learning rarely outperforms.
ZIERLER: Because you're not going to miss things?
KATZ: No. It's actually the other way around. The problem for machine learners, because at the core with tabular data they are just nonparametric regression, they have a tendency to overfit the data. There is this well-known issue that, yes, I can fit very well on my training sample, but then when I go out of sample, these very highly parameterized models—which is how to think about it—a nonparametric model is essentially a model that has as many parameters as there are data points, loosely. Those don't generalize very well, because they sort of get every squiggle in my in-sample, but those squiggles might not be present in my out-of-sample.
The best example I know of this—one of the pioneers in relating machine learning to more traditional statistical techniques was a professor of mine at UC San Diego when I was there, Hal White. This was in the 1990s. The Washington Power Company put out a call. Building power plants is really expensive, so you only want to build a new power plant if you really need more capacity. Instead, what power companies do is they go to some industrial customer and say, "Listen, 14 times a year, we can cut your power with 24 hours' notice, basically to reduce demand on the grid." That's one way that operators manage their peak load problem. How do they do it? What that means is the power companies need to generate a one-day-ahead forecast of power demand on the grid. The Washington company had this guy, and his job—he had all this data. He'd get weather data, he had all these reams of data, and he would say, "Tomorrow, I'm going to need to reduce the demand by 127,000 kilowatt-hours." [laughs] So they said, "Oh, it's the age of computing. We'll put out a call. Come do your fancy statistical models to forecast this. And here's the deal. You win—" It was like a million-dollar contract. It was a huge contract at the time. This was the 1990s; a million dollars was—[laughs]. But you had to beat the guy. Not a single one of the fancy machine learning techniques could beat the guy. [laughs] It's not to say that I don't think machine learning is important. But machine learning doesn't encode a theory of the world. That's clearly what this guy had. He had a very good working model, not one he could probably write down, about how all these different variables that he saw—what the temperature was, rain, what day of the year it was—would mean for demand, that these guys just couldn't encode in their fancy models. Because the machine learners are all a-theoretical. They're just like, "We're just looking for patterns." Sometimes they find real patterns, and sometimes they find erroneous patterns.
ZIERLER: What aspects of this statement are timeless, that what you're saying is these are fundamental limitations of machine learning, and what are, machine learning simply isn't there yet, and it might be relevant in the future?
KATZ: I think they're timeless in the sense that it's a fundamental way about this. Like I said, and again my statements are really about machine learning applied to tabular-type data, that they tend to overfit. If you talk to anyone who does statistical forecasting, simpler models just do better. I do think they're getting better. I should say this: Like any technique, I think there are better and worse ones, and good practitioners are good about guarding against this. There are techniques—how much you do out-of-sample and making sure there's no bleedover, so that you're actually doing out-of-sample. Good practitioners who have some domain knowledge can make good machine learning models. What I do think is timeless—you'll never be able to fully automate the type of analysis I do. Now, maybe if we ever have sentient computing, maybe they will. But given that machines are really dumb—currently there's no way to encode this sort of knowledge—it's good for my job security. You kind of need to build these things up with some knowledge.
ZIERLER: As a social scientist, are the systems that you're studying chaotic to some degree? Is that part of it?
KATZ: I'm agnostic about whether or not they're chaotic. They are highly complex. We're dealing with humans, right? Humans don't always follow everything you think they would do. I would put it differently. There's too much heterogeneity. The problem with these machine learners, since they are looking for patterns, they find sometimes non-meaningful patterns, non-meaningful groups. Maybe there could be advances. There are a bunch of companies out there pitching, "You just give us your data, and we will get you insights." I call BS. Without any domain knowledge, I just don't think we're anywhere near the world where you can just plug a bunch of data into a model. Now, companies have tried it. For example, Facebook actually is built on Stan, this language I told you about, this probabilistic Hamiltonian Markov chain. They have a Stan—a library called Prophet, which basically tries to—agnostic time-series forecasting. It does okay. But again, anyone who knows—but it won't do better than someone who knows both time-series econometrics and some knowledge about the process.
The counterpoint, and why you might think I'm wrong—where Google and the machine learners were right was on translation. It used to be that translation software, literally they would hire linguists and they would try to come up basically with encoding rules. That's hard, because language has lots of—in a language like English, it's terrible. We have lots of rules and then a bazillion exceptions to every rule. You and I as native English speakers, you just hear it. They tried to encode more and more complicated rules, generative rules for how you could build sentences, and it just never got anywhere. Google said, "Eh! We're not doing that. We're just going to throw tons of translated text—so the original text and a translation—and we're just going to have our machine learner learn what actually observes in practice." In that sense, more like how a child learns language. That turns out to have been way better. That's why there's almost no translation tools—everyone uses Google Translate.
ZIERLER: Would you call that a brute force approach?
KATZ: That's definitely a brute force approach. But it works in that case because it turns out that the rules of language are too hard to try to encode in any meaningful way, because there are just too many exceptions. And it's worse, because the exceptions change over time—what are acceptable words for things, et cetera. Even if someone asks how you're doing today, and we say, "Good," that's actually not—you should say "Well." You're supposed to respond with an adverb. But, you know, that rule is gone. [laughs] If you had that rule, the computer would get it wrong, but you don't object. You know exactly what someone says when they say, "I'm good."
ZIERLER: As you started taking on graduate students, what were the kinds of things that they were interested in, especially with Trump coming into office in 2016 and Citizens United, those kinds of things?
KATZ: [laughs] There's an old adage—"Political scientists don't really care about politics. Sociologists are not really social." I would say some of them care a little bit about this, but they're more focused on understanding more fundamental things. Not that Trump isn't interesting. It's not something we study much here. I don't study the presidency, not because I don't think it's important; it's just, for me who does statistical analysis, it's an n of 1, right? I can talk about things in general. Again, it's not mine as much, but there are a large number of students who get interested in things like misinformation and the flow of misinformation in networks. I think it has affected those types of things. But studying Trump? No, not so much. And really it's pretty incredible how apolitical [laughs] political science graduate students are.
ZIERLER: What about in the kinds of things that—? Not so much Trump the individual, but the issues that led to the rise of Trumpism.
KATZ: Again, I think those are really interesting questions, but there's no one on the faculty here who really that's what they study. That's really more in the domain of a little bit in political psychology and political communication. There's definitely work out there about the rise of authoritarianism, the rise of fascism. But students self-select, right? The political science group here is pretty small. These days, it's myself, Mike Alvarez. Mike does more political behavior, but again more on the computational, statistical side of things. We have a bunch of people who do formal theory, like game theoretical models of politics. The person in the office next door is Alex Hirsch. He studies things like lobbying. We have Gabriel Moctezuma. He does what are called structural models. So, there's no one here who does the sort of "Let's think about what drives—" There's questions about what drives partisan polarization. Those are valid research questions; there's just no one at Caltech who does them. It's a sort of chicken and an egg problem, so students don't come here—that's not what they want to study.
ZIERLER: Working with Mike Alvarez, what was your point of entry on the campaign finance stuff?
KATZ: It was actually me. As I told you, research is all about happenstance. A guy emailed me out of the blue, partly because of my consulting work. I'm sort of known in some—the guy who sent it to me was actually a Republican election consultant. He said, "Oh, I put together this data, campaign finance data." In principle, it's available from the Federal Election Commission. It's in a terrible format. It's not well cleaned. So, some of the stuff gets cleaned, but there's actually explicit prohibitions on use from the Federal Election Commission, by statute. I as a researcher can analyze donor patterns, but a campaign can't pull down FEC data to look for potential—"Oh, this person gave to Senator McConnell, so I'm a Republican running, and actually I'm going to hit them up. I can pull their address." You're not allowed to do that. His company had spent a long time building a product on the expenditure side of things, and that was actually pitching to companies that wanted to pitch to campaigns. But because the data is all together, he had put together this dataset of the donors, but he said, "I can only do it for researchers. Do you think this might be of value to you?" He basically gave me a copy.
It turns out the data he gave me, there were problems with it, so then I just had one of our joint grad students, Mike and mine, Silvia Kim, who is now on faculty at American University, we had her just start pulling down her own FEC data. In fact, this was a time when the FEC was hiring an outside company to try and make the data more accessible. We ended up being a beta tester for them, because we were pulling down so much data. We basically just cleaned it ourselves. So, I got into campaign finance. I had done some campaign finance very early in my career, but the data was much more aggregated. Now, literally we have 100 million transactions.
The first paper that came out of that project was another quirk. Typically, most donations these days, because of changes in technology, are actually small donations. If you look back in the 1980s and 1990s, campaigns would mostly go for larger donors because credit card transactions, getting in touch with these people was too hard. Enter the modern world with computing, email, and easy to get transactions—"I'm happy to take your $25 for my campaign." In fact actually there have been now, on the Democratic side, ActBlue came into being. You might know ActBlue. You might have given money through ActBlue. ActBlue is an aggregator, but they also act as a pass-through. For example, the Bernie Sanders campaign used ActBlue as basically handling all their financial transactions. Turns out there's a quirk in the regulations about campaign finance. Typically you only get identified if your aggregate giving in a year to a campaign is more than $250. However, if you give through a third party, that third party has to list everyone. So we actually finally got to know something about these third-party donors.
ZIERLER: And ActBlue is one of those third parties?
KATZ: Yeah. Then what happened? ActBlue ended up being so good, the Republicans didn't like that they didn't have an operation like it, so they created something called WinRed. You may or not know WinRed.
ZIERLER: That one, I didn't know. I'm not on those mailing lists, I guess. [laughs]
KATZ: We wrote this paper analyzing these, like what does the censoring do, to how we understand campaign giving. Campaign finance data is great, and it's really important, but it's one of these areas where we have lots of data, but we actually don't have lots of covariants. All we get from the FEC is we get your name, your address, how much you gave. We can link that to if you gave to other campaigns, but that's it. We don't even know your gender. So then we had to build up a dataset to make this more interesting, so we could actually ask social scientific questions. There are some ways to intelligently guess [laughs] gender and race, for example, by your address and name. We got this dataset together and we had to clean it, but then we had to run it through these algorithms to basically impute what their gender and race was. We then also pulled together aggregate data on economics like from their zip code that they lived in, so we could actually say some interesting things about who these donors were.
What got us into it was that we had a student—someone sent me data. I said, "Hey, Silvia, can you look at this data?" We spent a summer—it was actually her first project she did for us. It was the summer after her first year of the program. "Here's a little project for you to whet your [laughs]—to start with. See how it goes, and we'll see if it works out." It ended up working out, but not the way—it never works out the way you think it's going to work out. This idea that we're going to talk about hidden donors, that was again not something we thought we were going to deal with. That's actually part of a larger project we're still working on. All these laws come out of the aftermath of Watergate in the United States. There were a whole bunch of books and research articles written in the 1980s about donors, but the landscape really changed, and we wanted to know, did the things that Michael Melvin said in the 1980s really still hold today? The answer is no. We're still working on that aspect of the project, the bigger aspect of it.
Answering Answerable Questions
ZIERLER: Have you looked at some of the political ramifications of what comes out of this data?
ZIERLER: That's too applied for you?
KATZ: No, it's not too applied. I like to answer questions I can answer. Ultimately one does care about campaign finance and other things because of how it says things about American democracy, and I'm perfectly willing to sit here and be an armchair theorist about how these things do, but really what I think of myself doing is trying to do the science of social science, the science of political science. I try very hard in my published articles to say things—"Here's what the data can tell me."
There's lots of things I would like to know, but those are really hard to do. There are people who are interested in it, and there have been some attempts at studies that, for example, look at like basically buying votes, is what you're sort of getting at. It turns out it's really hard. I will tell you what the general consensus is of those studies. A couple things. The most counterintuitive thing I'm going to say about campaign finance—because everyone thinks there's so much money in campaign finance—actually, among people who study campaign finance in the United States, is how little money is spent on elections. If you think about that Congress controls a trillion-dollar entity, literally we spend maybe $100 million, now maybe several hundred million dollars? That's trivial. In any other field, if you said your return on investment for a trillion dollars, you would invest way more than—you know. So, political scientists are all asking, "Why is there so little money in politics?" Now, it has ratcheted up in the age of things like Citizens United.
The other thing in terms of buying votes—the causation usually goes the other way around. It's not so much donors giving money to get access or to get a particular bill passed. I think that happens somewhat at the local level, at small elections, where you can actually easily buy elections. For congressional elections, it's basically rewarding your friends. The coal industry knows who their buddies are in Congress, and they give money to their buddies. Now, do I think that this person gives money because coal gives them money? The evidence seems that the causation is the other way around, that they just give money to the people that they think are their friends.
ZIERLER: Has Citizens United just advancing the idea that corporations are people changed the data in interesting ways, regardless of what it says sociologically?
KATZ: Yeah! It definitely has. A large part of the data, we don't get to see. It allowed for this large influx of basically dark money, that we know of indirectly because some state laws require—like California has very strict reporting requirements, much stricter than the federal government. What it has done is ratcheted up spending. Still not high enough, if you think what the total that's at stake, but it has definitely ratcheted it up, and it has made it easier for a large donor. It has definitely become much easier for very wealthy individuals or corporations to give huge sums of money to try and influence elections. So, yes, it has changed the data. But it has changed on both sides. The other thing is that technology has made it much easier to aggregate a lot of small donors. I would also point out, until recently, the Dems actually rely on large donors much more than the Republicans. That's something you may or may not know.
KATZ: [laughs] So, yes, the Koch Brothers get written about a lot, but there's Tom Steyer, George Soros. Not to raise conspiracy theories; he actually does spend huge sums of money on elections and elections-related things in the United States. That has gotten easier. It's not just Citizens United. It's other rulings about disclosure and what they can do.
ZIERLER: We're talking campaign finance 2019, 2020. What have you been up to in the past few years?
KATZ: As I said, I have this large project with Danny Ebanks and Gary King on updating these models of elections. The first paper is the proof of concept, actually putting the model out there and proof that it's better than what people have been doing. We're also writing a software package that does these models.
ZIERLER: What have people been doing?
KATZ: The gold standard is an extended version of a regression that Andrew Gelman and Gary King wrote up in the 1990s. That is still what is mostly used. We're applying it to this also larger dataset of legislative elections. I have another grad student, Jacob Morrier, and he and I are working related to this question about term limits and candidate entry. This credibility thing—what we really would like to do is I'd like to randomly assign whether or not an incumbent can run again, and what challenger can go against them. I can't do that. Some people have suggested that we can use campaign term limits as a quasi-instrument. I'm actually very skeptical of that, because while voters might not be very rational, high-quality candidates are very rational.
The problem is that a term limit changes who challenges. If you think about a high-quality challenger who might want to contest the seat, if she is looking and it's an election prior to the incumbent legislator being termed out, she says, "Maybe I shouldn't run this term, because I know for sure there's going to be an open seat next time, and I'd have better shots." This logic screws up a lot of this idea that you can think about term limits as somehow randomly assigning whether or not an incumbent can run. You can't ever test randomization at some level, but we can test the consequences—if there are strategic cascades. We have been working on both some theory and some empirics related to this. I have this campaign finance book with Mike, and some former grad students—Silvia Kim. Then I have this project on measuring democracy.
The other thing I've been doing, which is quasi-related, quasi-research, is Mike Alvarez and I have been charged—we've been analyzing data for Caltech. We just did a study of whether SAT scores predictive of anything at Caltech. We've been looking basically before and after there was this moratorium. During the COVID pandemic, we didn't require, because people didn't want to sit in a room and take tests with other people. The question is, how has that changed? But it's a broader thing. That was our first study. That is why actually in fact Caltech has agreed to continue the moratorium. Now we're doing some other projects being funded and sanctioned by the Office of Student Affairs, looking at admissions data. Can we predict students who might have a tougher time at Caltech? Another question I think we're going to look at is looking at student athletes at Caltech. Caltech is a weird place. We don't recruit athletes, but we actually have a number of people who are very serious athletes. How do they do compared to their peers?
ZIERLER: I'm sure you're following the Supreme Court case with Harvard and UNC.
KATZ: Yeah, I'm following it, and the answer is that we're going to do away with affirmative action.
ZIERLER: What is your perspective on Caltech's different approach to—let's call it what it is, an Asian, Asian-American quota system—and Caltech's unique approach relative to places like Harvard. What are your thoughts on how Caltech has made its admissions decisions in the past 20 years?
KATZ: I would actually put it differently. While we talk about Asian quotas, the real issue at places like Harvard are actually what are so-called dean's admits, which is actually legacies and donors. And athletes. Literally 50% of the admits to a Harvard class are so-called dean's admits, where the dean has put their finger on the thumbs of the scale to get them in. You ask why Harvard has so many fewer Asians? It's because they have this system where they can do this. Where I think Caltech can be really proud of itself is in fact we've never had that. The dean's office, development, has nothing to do with it. I've had friends—"Can you get my—?" I say, "I have literally no influence over who gets into Caltech."
ZIERLER: Is that a platonic ideal for how things should be?
KATZ: No. I think it's mixed. It's good and it's bad. It still means that the population is not diverse in some sense. It has gotten better. Comparatively, for example, I went to MIT, and my class, the class of 1990, was literally 46% women. Caltech didn't hit that benchmark until like three or four years ago. [laughs] I think Caltech has always been late in the game to trying to think about this. I do think Caltech is thinking about it in the right way, which is not trying to put the thumbs on the scale of the admissions decisions, but trying to improve the outreach and getting them to apply. For example, women just don't apply to Caltech. That's true in general at science-based institutions but Caltech is even worse than that. I think Jarrid Whitney and his team have gone out of their way to try and improve the pipeline of applicants we get. I think it's doing pretty good. There's always these tradeoffs about this.
We have been fortunate—and it's true of all elite universities—that there is so much excess demand for slots. If you look at higher education in the United States, the number of positions at the elite universities, however you want to define them, has basically increased a teeny, teeny bit. The only real Ivy League-plus school that has had any sizable increase in enrollment is Chicago. That was an intentional choice by now late President Hugo Sonnenschein. Most other places basically had the same size classes, maybe a little bit increased. But the college-age population is at an all-time high, because the baby boomers have kids. [laughs] We're actually past the peak, but we're just past the peak. That has made life in some sense easier for places, so we get way more demand, and there's lots of students who could come to Caltech, and so now you get to choose on other criteria. What has happened on many dimensions of diversity—ethnic, first gen—also just on skill sets. We actually get more athletes now than we used to get. We get more musicians than we used to. The students are competing by—literally we could fill this class five times over [laughs], so how do you choose the one-fifth you're going to actually take? The answer is it's all these ancillary things.
ZIERLER: But you're saying Caltech is basically on the right track?
KATZ: I think it is. We could always do better.
ZIERLER: What does doing better look like?
KATZ: A little more outreach. I think it's only recently that Caltech has come to really think of itself as having to worry about the diversity issues.
ZIERLER: Outreach on the basis of "We want the student body to look more like Southern California or the United States" kind of thing?
KATZ: Yeah, I think so. But I think always mindful that—I tell students—I've had friends, and friends of friends ask me, "Could Caltech be right for my student?" I was like, "Caltech is a very special place. For the right student, it can be an amazing place. But we are like nowhere else. If you do not want to do STEM and you are not super good at it, and you are not willing to work incredibly hard, this is not the right place for you." We are still unabashedly—I think sometimes a little too much—the core program is still pretty overwhelming. We've reduced it. If I were a dictator, I would reduce it more. I think we still require too much Institute-level requirements. It means that most students, for example, unless you're majoring in math or physics or chemistry, you get almost no exposure to what you might major in, until your second year. [laughs] And, the world has changed. The core program was designed in the post-war period when it was the era of physics, chemistry, and math. Biology is arguably more important today. Computation is definitely more important. And they get none of that. In fact, the only computation that most students get de facto—CS 1 is in the core, because basically almost 90% of the freshmen take CS 1, but it's not actually computation. It's actually not a course I would want them to take. It's about programming and Python. Don't get me wrong; that is an incredibly valuable skill for other things they're going to do.
ZIERLER: What would CS 1 look like if you had your way?
KATZ: It would look more like 6-double-oh-1, which is the equivalent course at MIT, which is more about algorithmic thinking. In fact, it is actually taught—I don't know if it's still taught—when I took it, it was taught in Lisp, which is a totally useless language, except that it is a really good pedagogical way to teach key concepts in algorithms, issues like recursion, which Python just doesn't have. I'd like them to get a course on algorithmic thinking as opposed to just a programming language.
ZIERLER: How would that be responsive? One of the trend-lines at Caltech is it's "computation everything" now. How would that work?
KATZ: It's computation everything, but I don't think our undergraduates, unless you major in CS and go take advanced courses in CS, that the average undergraduate gets much exposure to what I think are key concepts. Instead they get them piecemeal in applied classes without ever a big picture about why these are important concepts and how they are generalized across all the sciences, unless they are a CS major.
ZIERLER: I'm going to ask a question; you can laugh me out of your office. But to go back to this idea that you were interviewing for administrative jobs while you were division chair—you liked these aspects of it. You stayed at Caltech. Is it within the realm of possibility that you or even somebody from HSS would become provost here?
KATZ: Zero. Never.
ZIERLER: Now, why? That's where you can laugh me out of your office, but—
KATZ: No, I'm not going to laugh you out; it's a perfectly reasonable question. The answer is—this goes back to I think our first discussion. While I think it's true that HSS and HSS faculty get more respect than even when I started here, it's still the case that we are a second-class citizen. We are not the basis. There were options. For example, John Ledyard who was actually the division chair who hired me, apparently he was talked about as a possible provost when Steve Koonin became provost, and basically a bunch of trustees said, "We're not hiring an economist to be a provost at Caltech. That's not what we're known for." I think the Board would have a very hard time. It would be an uphill battle. And I think the faculty would have a hard time. Again, they like us, they respect us, to a point, but we look different. Our groups are different. It is true that anybody who has been a division chair knows—I mean, I know a ton about—I even oversee some actual wet labs. We have wet labs in HSS. But it's a teeny fraction. But the majority of all the other divisions are dealing with the issues about lab rehabs, equipment purchases. So I don't see the faculty being very happy about a social scientist, and I don't think the trustees would go for it.
ZIERLER: You gave the perspective outside looking in. When you were division chair and you interacted with other division chairs in the forum with the president and the provost, would you self-censor yourself, or do you think that—?
KATZ: I never self-censor myself. [laughs]
ZIERLER: But from your perspective, despite their concerns and biases, perhaps, could you or somebody from HSS serve perfectly well in the position of a provost?
KATZ: Oh, yeah. There's nothing—in fact, there are many dimensions I think that Caltech would benefit by someone—
ZIERLER: That's kind of what I was getting it, if that's what you were thinking.
KATZ: —would benefit by an economist, someone with economics training. I think often times, Caltech makes stupid decisions because they actually didn't really think through the strategic—like how is someone going to game this. The perfect example I give, which would have been solved had they actually listened to the economists, it pre-dates you here. We spent $25 million of Caltech's own money building a parking structure underneath the athletic fields in front of Broad. Why was that? Well, when I first got to Caltech, we didn't charge for parking. The older faculty, most of them lived walking or biking distance to campus, but they all had a parking spot, a named parking spot. I have a named parking spot. That's really inefficient, because if I'm not here, then that space just gets unused. So, Caltech, they were running out of parking. Particularly they were running out of parking for—I felt badly—for the staff, who basically have a license to hunt, but no dedicated spots. In light of this, they wanted to build a new parking structure, and the only option basically, given other uses on our grand plan, was to build this subterranean parking lot, which is an incredibly expensive thing to do. What the economists wanted them to do was just charge for parking. "No, no, no. We're just going to build a new parking structure." They then charged for parking, because "We need to basically recoup the $25 million that we paid for this parking structure that is now—"
KATZ: But when they started charging—so actually now I pay a lot. For my named spot, I pay—is it $85 a month? Lo and behold, a whole bunch of faculty members who lived close to campus all gave up their parking spots, and people carpooled to campus, because parking was now not free. Lo and behold, that's why no one uses Broad, the parking structure there, because when you actually charged and made people pay a cost for their externality, they use less of it!
ZIERLER: Who would have thought!
KATZ: [laughs] Who would have thought! Another area of this, which is more direct on academic things—Caltech is weird. We're a small place, but we actually run incredibly decentralized. The divisions basically run their own budgets. They run their own spaces. Dabney and Baxter are the divisions. Some divisions are space-rich. Some—like HSS is literally out of space. I actually recommend—and this is done at other universities—that there be a market for—that the provost charge for space, and you can do a transfer system. Right now, if I needed space for some of my neuroscientists—Steve Mayo had zero—he was division chair of Biology, now Biology and Biological Engineering—he had zero incentive to give me this, because he gives up space that he may want at some time in the future, and he gets nothing for it. I said, "We should just have a market for space. We should charge." What other universities do is they do an implied rent. He would either forego that rent, or I could pay—you could get assigned the property rights in different ways. But if you actually had a market, there would be more efficient allocation of space, just like there was with parking. [laughs] Yeah, so I think there would be things, but it will never—well, never say never, but I don't see in my lifetime at Caltech there being a provost who is a social scientist.
ZIERLER: Now that we've worked right up to the present, for the last part of our talk, some retrospective questions about your career, and we'll end going to the future. To stay on the administrative side, HSS of course has a new division chair. What are the challenges and opportunities looking ahead?
KATZ: The challenges remain the same. They are a bit more extreme right now. The real issue, one of the areas where HSS is different, is just hiring and retention of faculty. I can tell stories. We've had a bunch of retirements. We've had a bunch of leaves. So, the social scientists are way down. Within the next two years we will be down at least eight faculty members.
ZIERLER: How will Caltech fare in recruiting the superstars against your Stanfords and your Harvards?
KATZ: We have a very hard time. In certain areas where we're very strong in, it's a lot easier. The conditional probability of ever moving a superstar, if you're being wildly optimistic, is 10%. You should expect on a senior offer of someone you actually want to hire, it's about 10%. In fields that are a growth area or something else, it's way lower than that.
ZIERLER: Was your successful recruitment—are there lessons to be learned there?
KATZ: From mine?
KATZ: Well, no, because it's always easier to hire junior people. I was hired as a junior person.
ZIERLER: No, but that's who I mean, to hire junior people in these slots.
KATZ: I'm generally in favor of hiring junior people, but you also want to fill in, right? We also have a missing—my generation—it's not uncommon—I'm Gen X—we sort of have a missing generation of like people in their forties to mid-fifties. That's a problem because those are the people who actually, like, run things. [laughs] I'm in favor of hiring senior people when you're either going into a new area—and I mean tenured people—going into a new area or rebuilding for the age distribution. The age distribution is just we have a whole bunch of old guys, we have a bunch of young people, and then—. The other problem—I like hiring junior, but the problem at Caltech in the social sciences—I think we talked about this at one of our previous sessions—we have the lowest tenure rate in the Institute. It's not because we have higher standards; it's because of the point at which we hire people. Us and Computer Science are the only people who hire right out of graduate school.
ZIERLER: Where there's no track record.
KATZ: There's no track record! It might work out; it might not. When they hired me, it might work out; it might not. In my cohort, plus or minus a couple years around me, there were eight of us; I'm the only one who got tenure! Hiring junior is great, I do like it, but you've got to think about what your long-term yield is going to be, of people. The other problem we have is often times we hire really good young people, and then we lose them. If they turn out to be superstars, we do often sometimes lose them to the MITs, the Princetons.
ZIERLER: This is a longstanding problem?
KATZ: It's a longstanding problem!
ZIERLER: And it hasn't gone away.
KATZ: It's never going to go away. It's a scale problem. Again, something that I mentioned in a previous session—when I first became division chair, I put up a graph at the IAC retreat—that's the retreat for all the senior administrators—the average size of a top-five department in all the fields that we cover, and the size of the corresponding group at Caltech. In economics, every top-five economics department is over 50 people, and that's actually an undercount at every place but Princeton, because every place but Princeton also has a business school. While you might not think all the economists in the business school are of the same snuff—they have a different objective function—there are many good economists in those business schools, so that's an undercount. What we always fight against is this, "Do I want to be at a place that has lots of people who do what I do, or not?"
It is also students. We're always going to be disadvantaged. It's always why when I stressed as division chair—and this is the same, and Tracy still has this—you have to choose areas where we have some comparative advantage, that we're doing something—I don't want to be going head to head with Harvard, Princeton, Stanford. I'm going to lose more than I'm going to win. I want to do things like what we did early on, where we did economic theory, mathematical economics, when no one was doing it. We did experimental economics when no one was doing it. We did political economy when hardly anyone was doing it. One bet we took, and it's doing well—we did social neuroscience. No one is doing it. Or—they're starting to do it, but—
ZIERLER: Is Caltech a victim of its own success, that it innovated these fields that took off, and now they're flourishing elsewhere?
KATZ: Yes! Yes. We lost good people to Princeton. We lost good people to Stanford. The answer is yes. Same thing—we were doing a lot of stuff in networks. We had Matt Jackson here, who was a superstar here. Stanford lured him away, with a giant center and a bazillion graduate students. [laughs] Leeat Yariv is a great experimental behavioral economist. Princeton made her an amazing offer, and Princeton is arguably one of the best departments in the country.
ZIERLER: Is this basically a parallel story for recruiting graduate students? Can you get the best graduate students that are out there?
KATZ: I'm shaking my head. No. The answer is the same reason. I think rightfully so, students are often motivated by rankings. We don't even get ranked in political science. [laughs] We sometimes get ranked in economics. So, we're not ranked. The good thing is that actually no one is good about admitting grad students. It is actually very hard to know who is going to be a successful graduate student, because here's what happens. What do you have to admit someone? You have their college career. You admit someone who was a good student. That's not what makes them a good researcher. In fact, it's somewhat negatively correlated with what makes them a good researcher. You want them to be bold. Sometimes it has been less of an issue with the graduate students, because I think the higher variance actually often works in our favor. That yeah, we didn't get the kid who was valedictorian out of MIT, but it's not clear that guy is going to become a good economist. [laughs]
ZIERLER: Undergraduate, you're a good synthesizer of information.
KATZ: Exactly. That's not the goal in graduate school. But yeah, we struggle to recruit graduate students. This is a perennial problem.
ZIERLER: But your students do do well on the job market.
KATZ: The ones who work in the areas that I work in, yeah.
ZIERLER: So that seems like—
ZIERLER: —there should be some connection to getting better graduate students if they do so well coming out. What does that say about you as a mentor pushing against the tide?
KATZ: My advisor, Matt, his graduate students—they just did a conference in his honor when he passed, this past June, and all these students came back, and yeah, Matt was always an amazing—that was one of the things he cared about, was always his students, and I share that view. This actually came up last night at the graduate HHS welcome party last night. I said, "I'll always take a graduate student who is hardworking and listens over one who is quote-unquote brilliant and doesn't listen." [laughs] I can turn hardworking and listens into a very good yeoman researcher at the worst. [laughs]
The other problem we have—I have a slight disadvantage, which is the graduate program is designed—the first-year program is designed kind of like how Caltech is designed. It assumes everyone is going to be a math and physics major. Our program assumes that basically they are going to be economic theorists. They basically take effectively eight courses in very high-brow mathematical-based theory. Now, I want my students to know theory, but they don't need eight courses in it. [laughs] They don't get to take any of my stuff until the second year and on. That's just a particular peculiarity of what I cover and what I do. That's not a good match for how the program is. One of the things you always learn as an administrator is you never let a crisis go to waste. Since in fact we've had a bunch of losses, like we just lost Federico Echenique—he's my age, roughly; he's like four years younger than I am—to Berkeley, where he got his PhD. Great place. I'm like, "We've had all these losses in economic theory. Maybe we could think about reworking the first-year program so that it's not—" I said, "Empirically, the program is designed that everyone is going to be economic theorists, but in fact, we may produce one a year." [laughs] Maybe this is not the right model?
ZIERLER: Did you try to address this as division chair and you didn't get anywhere?
KATZ: I had bigger fish to fry than the graduate program. I was much more concerned about the fiscal state of the Division, and so my efforts went into first writing the underlying administrative structure in the Division, and then raising money. One, you have to choose your battles. The other thing is you don't want to be too heavy-handed as a division chair. Bending the graduate program to my views, that's not a very good selling point to your colleagues. [laughs]
ZIERLER: [laughs] Right.
KATZ: Whereas now that I'm a back-bencher, I can be the crotchety old guy who says, "How about we do something a little different?" [laughs]
ZIERLER: But you're interested in having that voice in faculty meetings and things like that?
KATZ: Of course, yeah. I'm very involved. I'm still very involved.
ZIERLER: is there more traction now, do you think?
ZIERLER: Your leadership. You're a back-bencher, as you say.
KATZ: Yeah. More because we've had all these losses. The old guys are retiring. This system was put in place when—so some of the people who support this were people who actually went through this program, like Tom Palfrey. So, yeah, the answer is there's some more traction. The world changes. But a recurring theme in our conversations is, academics are really conservative. It's funny, because we get tenure, and then people become the most conservative-minded.
ZIERLER: It should go the other way.
KATZ: [laughs] It should go the other way, but we are not. Everyone is about maintaining the status quo.
ZIERLER: Retrospectively, looking at your scholarship, it's very eclectic, the kinds of things that you've looked at over the course of your career, really going all the way back to graduate school. What do you see as the connecting threads or the guiding stars, it's all feeding into this one meta-narrative or key interest of yours?
KATZ: I like to use my ADD; I think part of it literally is that. It's just I get bored and I work on different things. Always, the overlying theme, if there is one, is about trying to do science, and how do we do so in an empirical way? How do we answer these questions that we want to answer, and do it as best we can? That could be generating a new statistical tool. That could be doing my own substantive analysis. That's the thread that goes through it. Always wanting to tie together my substantive work, both theory—that is thinking about strategically how people interact—and thinking about how we can use data to test those theories. That's the overriding theme.
ZIERLER: I'll just note you said "science." You didn't say "social science."
KATZ: Yeah. I view what I do as science. It is a social science, but I view myself as a scientist.
ZIERLER: Is that an uphill battle that you're still fighting in terms of how you are perceived at Caltech, even in HSS?
KATZ: Not in HSS. My colleagues might not be as blunt as I am, but we all are on the science end of things. At Caltech, I'm treated like a scientist, I think. To be honest with you, I don't think most of the rest of the Institute thinks much about HSS.
ZIERLER: One way or the other?
KATZ: One way or the other. Except in the usual ways of friendships and things like that. There's interactions with a few subgroups, a little bit of people in computer science, a little bit of people in neuro and biology, but there's just not a lot of interaction. There's not a lot of intellectual interaction. There's a lot of social interaction.
ZIERLER: For all of the talk about how much more interdisciplinary Caltech has gotten, is that overstated, from your perspective?
KATZ: No, I think it's actually honestly true. But it's more almost faint praise. It's more how little it is done at other places. [laughs] Here, there's lots of interesting work in the social sciences, for example, between the social sciences, between economists and political scientists, between economists and neuroscientists. That just doesn't really go on at other places. And there's some connections between economists and computer scientists, but at some level—you know, I like Jackie Barton, we're great colleagues and good friends, but I have nothing to say about inorganic chemistry. [laughs] My work is not related. The oddballs are always the physicists who think they can do everything. [laughs]
KATZ: In some sense, I think Caltech is about as interdisciplinary as a university can get. There is always more. But there are going to be some levels that we're just interested in different things. I also don't want things to be interdisciplinary for interdisciplinary's sake. I want to answer a question. We'll see. My one concern is the whole Resnick giant gift, and the Resnick funds, and there has been a push to try and get some more social scientists involved in climate change. I think climate change is incredibly important; it's just not something I would even know what question I would want to ask and could answer that isn't already known. We know a lot about who is willing to participate, who is willing to give up things for—
ZIERLER: Sure. But you're talking about you personally? Institutionally, HHS needs to be part of the equation.
KATZ: Oh, yeah, yeah, we are.
ZIERLER: Human behavior is—
KATZ: Oh, yes, it's driving all of it. At the end of the day, it's a lot like when we talked about the Caltech-MIT Voting Technology Project, that Caltech's general take on this problem has been technological. We need to come up with better batteries or better solar cells or better—you know. The answer is the problem is not technological. There are some technological things we need to solve, but it's fundamentally how you get them implemented and used and policy changed. Scientists tend to undervalue and underestimate the importance of adoption and regulation and politics. This is all foreseeable about what is going on in Germany and the rest of Europe about Russian gas. The answer was, there just wasn't good cheap alternatives. So yes, there are alternatives; they're just not—and why is solar and wind taking off in California? It's not because—well, some of it is because of tax incentives, but really what's the case is that the per-kilowatt hour cost is now at or below other generation technologies. [laughs] Now, government policy helped get it there. We funded, we rebated to get it there. But it wasn't like there has been some massive change in solar technology. It has gotten a little cheaper to make. It has gotten a little more efficient. But we're actually near the theoretical limits. Not quite in fully commercial. Commercial panes are like probably about 20% to 25% efficient, and I think the upper bounds are something on the order of 32% to 33%.
ZIERLER: Has your consulting work gotten more intensive over the years? Do you have an entrepreneurial aspect where you see a startup in your future at some point?
KATZ: I've been involved in three startups already. Yeah. Potentially. Again, I'm opportunistic. If something came forward or someone came forward—actually my colleague Gary King and I who we work a lot together, he has actually been a very successful entrepreneur. He has actually successfully founded three companies. We've talked about—he goes, "Oh, it's much easier to found a company than it is to write a paper. You should just do that." [laughs]
KATZ: So, yeah. But for me, I'm fortunate. I have enough money [laughs] that I don't do things purely for the money. It would have to be interesting to me, that I'd want to devote that much time. To be successful, it's a huge amount of time and effort. I would also say that things structurally at Caltech—Caltech actually doesn't make it very easy for faculty to be entrepreneurs compared to some of our peer institutions.
ZIERLER: Sure. Old habits die hard.
KATZ: Old habits die hard. That's another downside to this. It's actually why we've lost some faculty members, to be honest with you.
ZIERLER: Because they can pursue this more easily at Stanford, for example?
KATZ: Yeah. Stanford is—yeah. Also, Caltech is draconian about making people step down. Caltech has a rule that you can't be a chief executive officer of a company as a faculty member. Well, in the early days of startup, who are you—?
ZIERLER: What else are you going to be?
KATZ: What else are you going to be? You're not going to pay someone $200,000 a year to come run your company. First of all, you'd have to give them a ton of the company, and you'd have to pay them something, because they've got to live. I think that's some of the reason why Caltech underperforms on its external—we don't make it very easy.
ZIERLER: Finally, last question, looking to the future. I know you're going to refer to your ADD as not having some sort of long grand plan. But for however long you define a research agenda, what's on deck, or at least what are the big areas that you feel like you haven't touched yet that could take up the next five, ten years, however long you plan out these things?
KATZ: At this point, I really don't plan. I like to be open to opportunities. I like to be entrepreneurial in that sense. I like to be open to opportunities. When a new dataset comes up, I see some interesting idea. I think in the long run, they will all be related to doing highbrow quantitative work. For the near future, it's these projects on models for elections and analyzing these large datasets. It's these models for understanding democracy. But I don't literally sit down and plan. I'm fortunate—I have research support. Typically academics who plan a lot, it's because you have this grant cycle where you've constantly got to be selling. I get outside grants but the model is not quite the same. I have to fund some graduate students, but I'm definitely not David Baltimore running a 70-person lab. Literally his only job is like raising the tens of millions of dollars it takes to keep his lab functioning. So, I don't really plan out a future. I work with students, come up with interesting ideas. I think I work like most entrepreneurs, which is I start a lot of things, but a lot of them get killed, because they just end up not working out. That's okay.
ZIERLER: And that's okay with you? That's fine?
KATZ: That's great. Yeah. I tell students that "no" is the most important word. Knowing when to quit a project, knowing when to—as I told you, people say, "Oh, fundraising is hard." I say, "Fundraising is not hard. The problem is when to say ‘no.'" Because it's too easy as an administrator to say, "Oh, someone is dangling this big check in front of you."
ZIERLER: We won't adapt to figure out how to use it wisely, or that's the open question?
KATZ: That's the open question. It's very hard. You're much better off building to strength.
ZIERLER: A follow-up question, how do you tune your antennae to know—when you're being entrepreneurial, information is coming at you all the time, what are the kinds of things that you think might capture your attention, where you would say it's worth a go?
KATZ: I wish I knew how to quantify it. I wish I knew how to teach it. It's just I'm a voracious reader. I'm fortunate that I read very quickly. A lot of it, I just read very broadly, keep on the lookout.
ZIERLER: I'll just note the irony—it's not a very scientific approach.
KATZ: But I don't actually think research breakthroughs come from a scientific approach. You're asking where do ideas come from. How you implement those ideas is very scientific. I'm sure there's some business school guy who will tell you there's some strategy to how to come up with successful idea generation. I don't know of another one without just immersing yourself and thinking about a broad range of things. I'm fortunate that I have access to that. I also do things that put me in positions. I'm a journal editor. I literally read across the entire social sciences. For Social Science Advances I read about 800 papers a year, just for Science Advances.
ZIERLER: You know what's out there.
KATZ: Yeah. You get to know what's out there. I wish we would slow down the publishing. To be honest with you, think there's too much written. I think academics forget that the peer-reviewed model is actually pretty recent, and that we've created really bad incentives on both the production of knowledge and publishing it. Peer review is not up to the task, and there's too many what look like legitimate publishers publishing crappy things with basically little in the sense of curation.
ZIERLER: Because they need the content?
KATZ: They need the content, and it has gotten worse. I will say things that the U.S. and the British governments are doing are making it worse. They're requiring open access. Well, open access—I actually hold up Science Advances for being the right model. We have an acceptance rate of about 10% at Science Advances. Science Advances is published by AAAS. It has got its issues; don't get me wrong. No organization is perfect. But we are not in it to make money. If we wanted to make money—it's a $4,000 APC—the answer is you follow what they're doing at Nature. They literally publish almost every paper they get. They used to tell authors that. "We'll just send it down the chain." Because you only get paid, you only get revenue, for accepted articles. It's a really bad incentive. I know libraries and university budgets were put on strain about the cost of journals. Elsevier, especially, is a really terrible organization, terrible company. But there at least was some incentive to keep quality up, because you have to justify your subscription cost. In the world in which you get paid by authors to publish and you're a profit-maximizing firm, the answer is you publish everything, and you create more and more niche journals to publish everything, to the point where it's going to be a needle in a haystack. Effectively the notion that—
ZIERLER: Just finding the good papers.
KATZ: Just finding the good papers. The whole value of journals as a place of curation to find important and probably correct papers, that's going away really quickly. That's now hitting the social sciences. It has been true in biology forever. NIH and the Wellcome Trust have been pushing this for the last two decades.
ZIERLER: You're fighting the tide, do you think, with the Science Advances?
KATZ: A little bit, but we're not going to win. It's the economic incentive. You created a system. I think NSF and the Office of Science and Technology and the President's Office, they think they're doing the right thing. They want to democratize. They want to make information that taxpayers pay for available. Which, I understand that, but no one thought even the second level down about what incentives this creates.
ZIERLER: And that's not good, ultimately, for the consumers.
KATZ: No! A lot of junk science is going to get published in scientific journals, furthering distrust in science.
ZIERLER: Again I'm thinking people need to listen to the HHS perspective. That's the big takeaway.
KATZ: Yeah. It's funny, this is all a push by lab scientists, without thinking, "What's the economics of this?" [laughs] "What incentives are you creating?"
ZIERLER: I'm glad I got your perspective on this. Jonathan, thank you so much.
KATZ: It has been fun.
- The Meaning of Interdisciplinary
- Election Law and MIT
- The Science of Social Science
- Lower East Side Roots
- From York to MIT
- Research on Legislative Political Economy
- The Unique Positioning of Social Science at Caltech
- Focus on Cross Sectional Modeling
- From Chicago Back to Caltech
- The Import of Computation
- Measuring Democracy Outcomes
- Big Data and Machine Learning
- Answering Answerable Questions