Matching for Adjustment and Causal Inference

Matching for Adjustment and Causal Inference

Description

This class is an introduction to statistical adjustment using matching and propensity scores in the style pioneered by Rubin and Rosenbaum and currently in rapid development by many other methodologists across the social science and statistical disciplines. An important motivation for matching is to approximate an experimental design. And, since such a motivation arises from a desire to make transparent and defensible statements about causal relations, we will introduce the counterfactual conception of causal inference and the potential outcome formalization of these ideas. We will also spend some time on statistical inference (hypothesis testing, confidence interval creation) after the creation of a matched design. Finally, we will grapple with some of the questions that are current research topics in this area: When and how one can claim to have adjusted “enough”? How can we engage with concerns about unobserved confounds even if we have adjusted for what we observe?

Since methods of matching are rapidly developing in the methodology literature, we will here focus on the simplest and oldest form: post-stratification. The general concepts and work-flow should be transportable to more sophisticated methods of matched adjustment.

Required background

I assume some previous engagement with high school mathematics, probability and statistical computing in the R statistical computing environment. If you have not used R, you are welcome to take the class, but I encourage you to get a little experience with R before the first class session. Feel free to email me to ask for advice about how to practice with R before the class begins.

Readings

Required  
Rosenbaum, P. R. (2010). Design of Observational Studies. Springer

Recommended 
Becker, H. S. (1986). Writing for Social Scientists: How to Start and Finish Your Thesis, Book, or Article. University of Chicago Press
Berk, R. (2004). Regression Analysis: A Constructive Critique. Sage
Gelman, A. and Hill, J. (2007). Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press (particularly chapters 9,10 and 23 see http://www.stat.columbia.edu/~gelman/arm/).
Morgan, S. L. and Winship, C. (2007). Counterfactuals and Causal Inference: Methods and Principles for Social Research (Analytical Methods for Social Research). Cambridge University Press. See http://www.wjh.harvard.edu/~cwinship/cfa.html for some links and background reading)
Rosenbaum, P. R. (2002b). Observational Studies. Springer-Verlag, second edition (see http://www-stat. wharton.upenn.edu/~rosenbap/index.html for lots of papers and presentations).
Rubin, D. B. (2006). Matched sampling for causal effects. Cambridge University Press, Cambridge; New York

See the course outline

Instructor

Jake Bowers
Associate professor
Department of Political Science
Department  of Statistics
University of Illinois @ Urbana-Champaign
jwbowers@illinois.edu
http://jakebowers.org






Jake Bowers, es Profesor Asociado en los departamentos de Ciencia Política y Estadística de la Universidad de Illinois en Urbana-Champaign, Estados Unidos. Jake es un muy destacado académico en métodos de investigación para las ciencias sociales. Su trabajo en metodología se centra en diseños de investigación, inferencia estadística e inferencia causal. Ha investigado en experimentos con redes, experimentos de campo, así como ajustes estadísticos para la inferencia causal en base a técnicas de “matching”. Su trabajo sustantivo investiga contextos, catalizadores, e inhibidores de la acción política de los personas. Sus publicaciones aparecen en Sociological Methodology, Political Analysis, The Journal of Politics, Journal of the American Statistical Association, y Political Psychology. Jake ha sido además editor de The Political Methodologist y es miembro del consejo de redacción de Political Analysis y Observational Studies.