Networks are ubiquitous in science and have become a focal point for discussion in everyday life. Formal statistical models for the analysis of network data have emerged as a major topic of interest in diverse areas of science, as many scientific inquiries involve collections of measurements on pairs of objects. Probability models on graphs date back to 1959. Along with empirical studies in social psychology and sociology from the 1960s, these early works generated an active network community and a substantial literature in the 1970s. This effort moved into the statistical literature in the late 1970s and 1980s, and the past decade has seen a burgeoning network literature in statistical physics and computer science. The growth of the World Wide Web and the emergence of online networking communities such as Facebook and LinkedIn, and a host of more specialized professional network communities has intensified interest in the study of networks and network data. In this talk, I will review a few ideas that are central to this burgeoning literature. I will emphasize formal model descriptions, and pay special attention to the interpretation of parameters and their estimation. I will conclude by describing open problems and challenges for machine learning and statistics.

The setup is that there are many markets in which buyers and sellers are distinct types of actors, for example, the market for spouses has, until recently, been such a market (although I make no claim as to which side of the market is buying and which is selling). This market, in the form of college applications, was analyzed by Gale and Shapley in a famous 1962 paper in which they proved that there was a solution to this type of matching problem.

Reading a few articles in the "physics of politics" as a political scientist, one has the sense of observing an alternate universe. For example: a paper on the effect of election results on party membership in Germany that has no references to work outside of physics; features many exotic (to me at least) terms like Wegscheider potentials, the Sznajd model, and the Kronecker symbol; and takes a time-series approach to causation that I suspect would be unacceptable to most reviewers in political science and economics these days.

In general, it's clear that physicists doing work on political phenomena (or "sociophysics" more generally) are primarily interested in exploring the individual-level social interactions that might underpin the macro-order we observe in, e.g., regularities in turnout or vote share distributions. As such, political institutions (which are the major preoccupation of political scientists) necessarily disappear from the model and are typically not even mentioned, even when they would seem to be of first-order importance in explaining a particular phenomenon. (Another example of the alternate universe: a paper that argues that party vote shares in Indonesia follow a power law, but which does not describe or mention the electoral system.) These omissions seem foolish on first reading, but it's clear that they reflect a different choice of explanatory variable: physicists seek their explanations in micro-interactions, and we seek them primarily in political institutions. It's probably both of course, but models can only be so complex.

Despite my overall sense of disorientation in reading these papers, there were also somewhat surprising moments of familiarity. Physics heavily influenced economics in an earlier period of colonization, and much of what we read in economics and political science descended from those models. In reading these newer physics papers, there is therefore a sense of distant kinship, the knowledge of a common ancestor several generations back.

I wonder about the scope for collaboration between physicists and social scientists. Based on my admittedly very cursory reading of one area in which physicists have ventured, it's hard to know whether the potential gains from trade are sufficient to overcome the apparent difference in goals. For all I know there already is a lot of productive collaboration going on -- if you know of something interesting, share it in the comments!

We consider the doubly robust estimation of the parameters in a semi-parametric conditional odds ratio model characterizing the effect of an exposure in the presence of many confounders. We develop estimators that are consistent and asymptotically normal in a union model where either a prospective baseline density function or a retrospective baseline density function is correctly specified but not necessarily both. The case of a binary outcome is of particular interest, then our approach yields a doubly robust locally efficient estimator in a semi-parametric logistic regression model For general types of outcomes, we provide a strategy to obtain doubly robust estimators that are nearly locally efficient We illustrate the method in a simulation study and an application in statistical genetics. Finally, we briefly discuss extensions of the proposed method to the semi-parametric estimation of a parameter indexing an interaction between two exposures on the logistic scale, as well as extensions to the setting of a time-varying exposure in the presence of time-varying confounding.

The Applied Statistics workshop meets each Wednesday in room K-354, CGIS-Knafel (1737 Cambridge St). We start at 12 noon with a light lunch, with presentations beginning around 12:15 and we usually wrap up around 1:30 pm. We hope you can make it.

While a copy on your computer is very handy, a desk copy of this book is essential if you are interested in machine learning or data mining. The book is also a sight to behold. You can buy a copy at Amazon or Springer.

Regression discontinuity design (RDD) is a powerful and increasingly popular approach to causal inference that can be applied when treatment is assigned deterministically based on a continuous covariate. In this talk, I will present an application of RDD from French municipalities, where the system of electing the municipal council depends on whether the city's population is above or below 3500. First I show that cities above the population cutoff have fewer uncontested elections and more opposition representation on municipal councils, consistent with expectations. I then trace the effect of these political changes -- which amount to a heightening of the scrutiny imposed on the mayor -- on policy outcomes, providing evidence that more opposition scrutiny leads to more moderate policy.

The Applied Statistics workshop meets each Wednesday in room K-354, CGIS-Knafel (1737 Cambridge St). We start at 12 noon with a light lunch, with presentations beginning around 12:15 and we usually wrap up around 1:30 pm. We hope you can make it.

The Referendum is a phenomenon typical of (but not limited to) midlife, whereby people, increasingly aware of the finiteness of their time in the world, the limitations placed on them by their choices so far, and the narrowing options remaining to them, start judging their peers' differing choices with reactions ranging from envy to contempt. ...Friends who seemed pretty much indistinguishable from you in your 20s make different choices about family or career, and after a decade or two these initial differences yield such radically divergent trajectories that when you get together again you can only regard each other's lives with bemused incomprehension.

Those familiar with casual inference will recognize this as stemming from the Fundamental Problem of Causal Inference: we cannot observe, for one individual, both their response to treatment and control. The article is an elegant look at how we grow to worry about those mysterious missing potential outcomes--the paths we didn't choose--and how we use our friends' lives to impute those missing missing outcomes. Kreider goes on to make this point exactly, with a beautiful quote from a novel:

The problem is, we only get one chance at this, with no do-overs. Life is, in effect, a non-repeatable experiment with no control. In his novel about marriage, "Light Years," James Salter writes: "For whatever we do, even whatever we do not do prevents us from doing its opposite. Acts demolish their alternatives, that is the paradox." Watching our peers' lives is the closest we can come to a glimpse of the parallel universes in which we didn't ruin that relationship years ago, or got that job we applied for, or got on that plane after all. It's tempting to read other people's lives as cautionary fables or repudiations of our own.

Perhaps the only response is that, while so close to us in so many respects, friends may be poor matches for gauging these kinds of effects. In any case, "Acts demolish their alternatives, that is the paradox" is the best description of the problem of causal inference that I have seen.

Despite their popularity, conventional propensity score estimators (PSEs) do not take into account the estimation uncertainties in the propensity score into causal inference. This paper develops Bayesian propensity score estimators (BPSEs) to model the joint likelihood of both the outcome and the propensity score in one step, which naturally incorporate such uncertainties into causal inference. Simulations show that PSEs treating estimated propensity scores as if they were known will overestimate the variation in treatment e_ects and result in overly conservative inference, whereas BPSEs will provide corrected variance estimation and valid inference. Compared to other direct adjustment methods (E.g., Abadie and Imbens 2009), BPSEs are guaranteed to provide positive variance estimation, more reliable in small samples, and more flexible to contain complex propensity score models. To illustrate the proposed methods, BPSEs are applied to evaluating a job training program.

This has received attention from the social science community and others. Even Paul Krugman, mentioned in Coburn's press release as an example of (wasteful? political?) NSF funding, has something to say about it. There's no need to rehash the arguments here, which ever-so-nicely point out that Senator Coburn doesn't really know what he's talking about nor do his arguments make a whole lot of sense.

Regardless of the arguments, I just wanted to put a graph up to put all of this in perspective. In the 111th Congress, Coburn has had very little success with his amendments:

Seven of the rejections are instances when Coburn's amendment was tabled without discussion. Most of the rejections have been of proposed budget cuts or banning funds from certain projects And this is just in this year. Out of all the roll call votes on Coburn-sponsored amendments in the Senate over his tenure, only 8 out of 68 have actually passed.

I understand trying to tackle his critiques, as they track with an internal debate already in the discipline. But I think it may be a tad knee-jerk to start letter-writing campaigns to our Senators. Tom Coburn knows that putting out no-win amendments is a great way to take positions in the Senate without committing to anything. Minority amendments are a costless signal of the blandest kind--even a political scientist can see that.

They find that 10 year olds that ate candy daily were much more likely to be convicted of a violent crime at age 34 than those who did not eat candy daily. They cite this as evidence that childhood diet has an effect on adult behavior. One of their hypothesized mechanisms is that using candy as a reward for children (e.g. for behavior modification) inhibits the child's ability to delay gratification. And there is evidence that children that posses problems with delayed gratification tend to score lower on a host of measures, including the SATs (see also: the marshmallow studies).

The longitudinal data gives them leverage. For instance, the authors are able to control for parenting style at age 5 along with other variables, such as various scales of behavior problems or mental abilities at age 5 (some of these were discarded in the final analysis because of their variable selection rules). These ease my main concern that "problem children" might lead to a certain type of parenting and also indicate a propensity for violent adult behavior. Their controls help to eliminate this possibility (though, I will say that I am not familiar with this literature and they use fairly complicated scales to measure these concepts).

Strangely, at least to me, they do not seem to control for parental income or socio-economic class. I have a few ideas as to why this might matter. First, candy is relatively cheap compared to a good diet, thus poorer families might be forced to choose the cheaper option when feeding their children. Second, financial pressures lead to time pressures, which could force parents to take shortcuts--feeding their children junk food because it is quick or using it to induce behavior because it is easy. Thus, parental income may matter greatly for candy intake and it also may increase propensity to commit violent crimes. I am not certain this is true, but it seems plausible and unmentioned in the paper. Even if the finding is not causal, however, it is still interesting.

We review recent developments in the estimation of an optimal treatment strategy or regime from longitudinal data collected in an observational study. We also propose novel methods for using the data obtained from an observational database in one health-care system to determine the optimal treatment regime for biologically similar subjects in a second health-care system when, for cultural, logistical, or financial reasons, the two health-care systems differ (and will continue to differ) in the frequency of, and reasons for, both laboratory tests and physician visits. Finally, we propose a novel method for estimating the optimal timing of expensive and/or painful diagnostic or prognostic tests. Diagnostic or prognostic tests are only useful in so far as they help a physician to determine the optimal dosing strategy, by providing information on both the current health state and the prognosis of a patient because, in contrast to drug therapies, these tests have no direct causal effect on disease progression. Our new method explicitly incorporates this no direct effect restriction.

A copy of the paper is also available.

The Applied Statistics workshop meets each Wednesday in room K-354, CGIS-Knafel (1737 Cambridge St). We start at 12 noon with a light lunch, with presentations beginning around 12:15 and we usually wrap up around 1:30 pm. We hope you can make it.