22 October 2012
We hope you can join us this Wednesday, October 24, 2012 for the Applied Statistics Workshop. Chad Hazlett, a Ph.D. student from the Department of Political Science at MIT, will give a presentation entitled "Kernel Regularized Least Squares: Moving Beyond Linearity and Additivity Without Sacrificing Interpretability" (this is joint work with Jens Hainmueller from MIT). A light lunch will be served at 12 pm and the talk will begin at 12.15.
"Kernel Regularized Least Squares: Moving Beyond Linearity and Additivity Without Sacrificing Interpretability"
Chad Hazlett and Jens Hainmueller
Department of Political Science, MIT
CGIS K354 (1737 Cambridge St.)
Wednesday, October 24th, 2012 12.00 pm
We propose the use of Kernel Regularized Least Squares (KRLS) for social science modeling and inference problems. KRLS borrows from machine learning methods designed to solve regression and classification problems without relying on linearity or additivity assumptions. The method constructs a flexible hypothesis space that uses kernels as radial basis functions and finds the best fitting surface in this space by minimizing a complexity-penalized least squares problem. We provide an accessible explanation of the method and argue that it is well suited for social science inquiry because it avoids strong parametric assumptions and still allows for simple interpretation in ways analogous to OLS or other members of the GLM family. We also extend the method in several directions to make it more effective for social inquiry. In particular, we (1) derive new estimators for the pointwise marginal effects and their variances, (2) establish unbiasedness, consistency, and asymptotic normality of the KRLS estimator under fairly general conditions, (3) develop an automated approach to chose smoothing parameters, and (4) provide companion software. We illustrate the use of the methods through several simulations and a real-data example.
15 October 2012
We hope you can join us this Wednesday, October 17, 2012 for the Applied Statistics Workshop. James Scanlan, an Attorney-at-Law, will give a presentation entitled "The Mismeasure of Group Differences in the Law and the Social and Medical Sciences". A light lunch will be served at 12 pm and the talk will begin at 12.15.
"The Mismeasure of Group Differences in the Law and the Social and Medical Sciences"
CGIS K354 (1737 Cambridge St.)
Wednesday, October 17th, 2012 12.00 pm
This paper addresses the problematic nature of efforts in the law and the social and medical sciences to appraise the comparative circumstances of advantaged and disadvantaged groups on the basis of standard measures of differences in outcome rates, given that such measures tend to be systematically affected by the prevalence of an outcome. The rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it. Thus, for example, as mortality declines relative differences in mortality of advantaged and disadvantaged groups tend to increase while relative differences in survival tend to decrease; as procedures like immunization and cancer screening become more common, relative differences in rates of receipt of those procedures tend to decrease while relative differences in rates of failing to receive them tend to increase; relaxing mortgage lending criteria tends to increase relative differences in mortgage rejection rates while reducing relative differences in mortgage approval rates. Similarly, among subpopulations where adverse outcomes are comparatively rare (e.g., persons with high education or high income, British civil servants), relative differences in adverse outcomes tend to be larger, while relative differences in favorable outcomes tend to be smaller, than among subpopulations where adverse outcome are more common. Absolute differences between outcome rates and differences measured by odds ratios are unaffected by whether one examines the favorable or the adverse outcome. But such measures tend also to be affected by the overall prevalence of an outcome, though in a more complicated way than the two relative differences. Broadly, as uncommon outcomes become more common absolute differences tend to increase; as already common outcomes become even more common, absolute differences tend to decrease. Differences measured by odds ratios tend to change in the opposite direction of absolute differences as the prevalence of an outcome changes. The paper will explain these patterns and the misinterpretations of data on group differences arising from the failure to understand them. It will also describe a method for appraising the size of the difference in circumstances reflected by outcome rates of advantaged and disadvantaged groups that is theoretically unaffected by the prevalence of the outcome.
8 October 2012
We hope you can join us this Wednesday, October 10, 2012 for the Applied Statistics Workshop. Thales Teixeira, Assistant Professor of Business Administration at the Harvard Business School, will give a presentation entitled "Viral Video Advertising". A light lunch will be served at 12 pm and the talk will begin at 12.15.
"Viral Video Advertising"
Harvard Business School
CGIS K354 (1737 Cambridge St.)
Wednesday, October 10th, 2012 12.00 pm
To become viral, online video ads need to be viewed and then shared. Yet, what works for one decision may not work for the other. In this research we propose a novel consumer-centric model of viral advertising consisting of viewing and sharing decisions. We apply the model to assess the role of humor, present in 91% of viral ads, by teasing out the differential impact that type of humor (pure or shocking) has on each decision. In the lab, we record the facial expressions of consumers as they watch online ads containing either pure (i.e., smile, laughter) or shocking humor (e.g., shock from profanity), and examine its impact on their decisions. The video data is processed using face tracking software and used to calibrate a dynamic sequential model that accounts for both within and cross-decision dynamics. We find that shocking humor increases viewing but reduces sharing compared to no humor at all. Yet, content isn't the only factor of viral ad success; individual traits also matter. We also find that highly extraverted and self-directed consumers share humor ads more often and to a broader group of people each time. The magnitude of the effects of these two novel findings is then measured in a viral field study in which we selectively sent ads to participants and tracked views derived from sharing. We find that extraverted people garnered 300% more total views by sharing non-shocking humor ads than introverted people sharing ads low in humor.