Nathaniel Beck is a Professor at New York University’s Wilf Family Department of Politics, and an Affiliated Faculty member at the Center for Data Science.
What did you study in school? How did you get to what you study now?
I went to the University of Rochester almost half a century ago. I started off as a math major, but I soon realized I would never be a really good mathematician. Around that time I was excited by Isaac Asimov’s science fiction series, Foundation, and I became interested in his psycho-historian characters who could forecast mass movements in society, so I turned to political science. It was sheer luck I was at the University of Rochester, as they had just hired Bill Riker who was one of the key founders of modern political science (both in terms of game theory and top empirical work). Rochester was also the kind of place where an undergraduate could just walk into a Ph.D. course and easily chat with someone like Riker. I loved the research and have never looked back.
How did you start to incorporate data science into your political science studies?
I developed an interest in non-linear models very early on, so moving to data science was not a hard sell. And I had been doing supervised machine learning for a long time before I even knew that term existed
Within the cross-section of political science and data science, what are some of your research interests?
One of my main interests is figuring out the usefulness of machine learning methods for political science, where we want to describe what the so called, “black box” looks like. I am also actively trying to see if there are ways to combine machine learning methods with structural assumptions we have about political systems.
How does machine learning factor into your work?
The use of regression models is just a very simple form of machine learning. 50 years ago regression was state of the art. Now more complicated models (trees) are in that state. We still need to figure out how to get trees to tell us about social phenomena (as opposed to simply classifying things, our task is not to read zip codes) and we need to continue to put them on a firm statistical basis. Fortunately Bayesian models can be very useful here.
What data science methods have you used in the past for your work in political science? Natural language processing seems to be an obvious one, but I’m curious if there are other disciplines in data science you’re using to study politics.
While I’m interested in text analysis, my own interests have always been in non-linear methods. My first presentation in a statistics class at Yale was on an early tree based method (Automatic Interaction Detector), and though at the time it was laughed at, with more modern thinking about cross-validation and such, this has become a rather useful tool (essentially the move from stepwise regression to LASSO).
As a person interested in time series, I have always thought about the possibilities with out of sample validation, and cross validation has proved increasingly useful. I’ve worked with Jonathan Katz on time-series–cross-sectional models, and I also did some work with Simon Jackman on generalized additive models. Gary King, Langche Zeng, and I have looked into neural nets in to study conflict under the hypothesis that one needed high order complex interactions to model conflict.
These days I expect all of our graduate students to become conversant with these methods. There has also been a revolution in computations, with Markov chain Monte Carlo making things possible. We’ve gone from this being cutting edge to a standard tool in a bit over a decade.
What aspects of politics and political methodology have the largest untapped potential for work in the world of data science?
While we are doing a lot, the area is just taking off. There is a lot of potential in new sources of data, such as the work coming from NYU’s Social Media and Political Participation lab. There’s obviously lots of potential in text (where I have some undergraduate honors students studying thinks like rhetoric in the Federal Reserve), geo-spatial analysis (colleagues using satellite imaging to study fires as a means of ethnic cleansing in Kenya), novel data collection (colleagues tracking refuges with cell phones), etc. etc.
Your faculty bio says your “undergraduate teaching is primarily related to a new Public Policy major which should be announced soon.” Can you tell me about this new major, and data science will enter into the equation?
It is easier to fit data science into public policy than political science. We have all these great new applications, whether it be using tax returns to measure inequality, looking at how people are moving around, on a daily or permanent basis, or something like NYU’s Center for Urban Science and Progress’ metered city project. Public policy is also taking advantage of new ideas in research design, whether in the use of experiments or quasi-experiments. So many of the courses stress what we can learn from data, but also, how we learn from data.
I’m teaching the first capstone seminar next year, and expect that most of the projects to be data oriented. NYU’s College of Arts and Science is particularly interested in training students who can live and work and understand things in the 21st Century, and our major fits very nicely into that. The Politics Department is also considering revamping its methods sequence to give undergraduates more skills and appreciation for what modern data science can do (our Ph.D. students are already great at this).
What drew you to the CDS program at NYU?
When CDS started, Yann LeCun reached out to me because he had seen my work on neural nets. I started going to CDS events, and I met people with very different substantive interests, but using similar methodological processes. Who would have guessed that David Hogg and I are using similar methods, even though he is studying distant solar systems?
On a chalk-board at CDS we have the question: “What does it mean to be a data scientist?” Could you answer this question?
I think what distinguishes data science from statistics is a real appreciation for exciting new sources of data and a willingness to deal with the very messy problems of such data. The trick is to not lose all the good done by statisticians (understanding causality and uncertainty) while devising methods to deal with this very messy data.
Interview by Jack Lowery