In this paper, we more fully develop the properties of the distributionfree test for nonnested model discrimination introduced by Clarke (2003). We prove that the test is both consistent and unbiased. We demonstrate that the test is asymptotically more efficient for highly leptokurtic distributions than the well-known Vuong test. Using a Monte Carlo experiment, we then establish that the distribution of individual log-likelihood ratios (the data to which both tests are applied) is highly leptokurtic. Finally, we use the same Monte Carlo to measure the performance of the distribution-free test and the Vuong test. The Monte Carlo advances previous efforts in that it allows for two misspecified models that vary in distance from a true, but “unknown,” data generating process. The results indicate that the power of the new test is as great as or, for many alternatives, significantly greater than the power of the Vuong test.

Quantitative political science is awash in control variables. The justification for these bloated specifications is usually the fear of omitted variable bias. A key underlying assumption is that the danger posed by omitted variable bias can be ameliorated by the inclusion of relevant control variables. Unfortunately, as this article demonstrates, there is nothing in the mathematics of regression analysis that supports this conclusion. The inclusion of additional control variables may increase or decrease the bias, and we cannot know for sure which is the case in any particular situation. A brief discussion of alternative strategies for achieving experimental control follows the main result. Keywords omitted variable bias, specification, control variables, research design Quantitative political science is awash in control variables. It is not uncommon to see statistical models with 20 or more independent variables. An article in the August 2004 issue of the American Political Science Review, for example, reports a model with 22 independent variables (Duch & Palmer, 2004). 1 The situation is no different if we consider

In many political science applications the linear regression model is an inappropriate tool for answering substantive questions. This course serves as an introduction to a multitude of probability models that are appropriate when the linear model is inadequate. Many of these models can be estimated using the method of maximum likelihood. This course serves as an introduction to maximum likelihood estimation, and as a survey of models that are broadly applicable in the social sciences. The purpose of the course is three-fold: (1) to prepare students to conduct applied research using appropriate statistical models; (2) provide a foundation in maximum likelihood estimation so students can investigate and implement other models from the statistics literature; and (3) provide students with the tools necessary to develop their own statistical models of political phenomena. In the weekly class meeting the course will be conducted as lecture-based workshop. Throughout the semester the I will lecture on the key material from the readings, and answer student questions. We will also work through examples collaboratively. Our pace will be unpredictable. The bulk of learning in the course will take place outside of the classroom by reading,