We discuss a method for improving causal inferences called “Coarsened Exact Matching” (CEM), and the new “Monotonic Imbalance Bounding” (MIB) class of matching methods from which CEM is derived. We summarize what is known about CEM and MIB, derive and illustrate several new desirable statistical properties of CEM, and then propose a variety of useful extensions. We show that CEM possesses a wide range of statistical properties not available in most other matching methods but is at the same time exceptionally easy to comprehend and use. We focus on the connection between theoretical properties and practical applications. We also make available easy-to-use open source software for R, Stata, and SPSS that implement all our suggestions.

This paper proposes entropy balancing, a data preprocessing method to achieve covariate balance in observational studies with binary treatments. Entropy balancing relies on a maximum entropy reweighting scheme that calibrates unit weights so that the reweighted treatment and control group satisfy a potentially large set of prespecified balance conditions that incorporate information about known sample moments. Entropy balancing thereby exactly adjusts inequalities in representation with respect to the first, second, and possibly higher moments of the covariate distributions. These balance improvements can reduce model dependence for the subsequent estimation of treatment effects. The method assures that balance improves on all covariate moments included in the reweighting. It also obviates the need for continual balance checking and iterative searching over propensity score models that may stochastically balance the covariate moments. We demonstrate the use of entropy balancing with Monte Carlo simulations and empirical applications.

The validity of empirical research often relies upon the accuracy of self-reported behavior and beliefs. Yet eliciting truthful answers in surveys is challenging, especially when studying sensitive issues such as racial prejudice, corruption, and support for militant groups. List experiments have attracted much attention recently as a potential solution to this measurement problem. Many researchers, however, have used a simple difference-in-means estimator, which prevents the efficient examination of multivariate relationships between respondents' characteristics and their responses to sensitive items. Moreover, no systematic means exists to investigate the role of underlying assumptions. We fill these gaps by developing a set of new statistical methods for list experiments. We identify the commonly invoked assumptions, propose new multivariate regression estimators, and develop methods to detect and adjust for potential violations of key assumptions. For empirical illustration, we analyze list experiments concerning racial prejudice. Open-source software is made available to implement the proposed methodology.

The Comparative Manifesto Project (CMP) provides the only time series of estimated party policy positions in political science and has been extensively used in a wide variety of applications. Recent work (e.g., Benoit, Laver, and Mikhaylov 2009; Klingemann et al. 2006) focuses on nonsystematic sources of error in these estimates that arise from the text generation process. Our concern here, by contrast, is with error that arises during the text coding process since nearly all manifestos are coded only once by a single coder. First, we discuss reliability and misclassification in the context of hand-coded content analysis methods. Second, we report results of a coding experiment that used trained human coders to code sample manifestos provided by the CMP, allowing us to estimate the reliability of both coders and coding categories. Third, we compare our test codings to the published CMP “gold standard” codings of the test documents to assess accuracy and produce empirical estimates of a misclassification matrix for each coding category. Finally, we demonstrate the effect of coding misclassification on the CMP's most widely used index, its left-right scale. Our findings indicate that misclassification is a serious and systemic problem with the current CMP data set and coding process, suggesting the CMP scheme should be significantly simplified to address reliability issues.

Social scientists spend considerable energy constructing typologies and discussing their roles in measurement. Less discussed is the role of typologies in evaluating and revising theoretical arguments. We argue that unsupervised machine learning tools can be profitably applied to the development and testing of theory-based typologies. We review recent advances in mixture models as applied to cluster analysis and argue that these tools are particularly important in the social sciences where it is common to claim that high-dimensional objects group together in meaningful clusters. Model-based clustering (MBC) grounds analysis in probability theory, permitting the evaluation of uncertainty and application of information-based model selection tools. We show that the MBC approach forces analysts to consider dimensionality problems that more traditional clustering tools obscure. We apply MBC to the “varieties of capitalism,” a typology receiving significant attention in political science and economic sociology. We find weak and conflicting evidence for the theory's expected grouping. We therefore caution against the current practice of including typology-derived dummy variables in regression and case-comparison research designs.

The Cox proportional hazards model is ubiquitous in time-to-event studies of political processes. Plausible deviations from correct specification and operationalization caused by problems such as measurement error or omitted variables can produce substantial bias when the Cox model is estimated by conventional partial likelihood maximization (PLM). One alternative is an iteratively reweighted robust (IRR) estimator, which can reduce this bias. However, the utility of IRR is limited by the fact that there is currently no method for determining whether PLM or IRR is more appropriate for a particular sample of data. Here, we develop and evaluate a novel test for selecting between the two estimators. Then, we apply the test to political science data. We demonstrate that PLM and IRR can each be optimal, that our test is effective in choosing between them, and that substantive conclusions can depend on which one is used.