Uses two new estimators (see below) that are implemented in my R package, FAiR, to test traditional political economy theories of preferences for redistribution and other welfare-state programs against alternative theories that stem from sociology, psychology, and lab experiments. In short, the alternative theories perform better, and the new methods produce stronger evidence for this conclusion than do standard methodological techniques.

This paper reanalyzes the empirical evidence presented in Lisa Blaydes, 2004, ???*Rewarding Impatience: A Bargaining and Enforcement Model of OPEC*,??? International Organization 58(2):213???237. The IO published Lisa Blaydes??? response to this critique in the same issue, which is also available on her website.

The "stable" version (0.4-x) of my R package to estimate structural equation models with latent variables is available here. The "development" version (0.6-0) is available here. It is mostly functional but some things may not work (correctly) at any particular point in time, and they may or may not correspond to the (lack of) documentation. Some of the working papers depend on functionality in FA$i$R 0.6-x.

This paper uses FA$i$R (0.4-x) to evaluate a new Risk Context Scale that evaluates recidivism risk of male parolees.

This paper develops graphical tools to help researchers understand the (often poorly understood) properties of a distribution of covariance or correlation matrices. For example, when choosing a prior, researchers often use something from the Wishart family. An alternative is to specify a jointly uniform prior over a correlation matrix and some marginal distribution for the standard deviations to form a prior over a covariance matrix. See also my working paper “Generating Correlation Matrices via Canonical Partial Correlations” below.

This paper is a companion to the previous one that focuses on how best to make finite-sample inferences about the number of latent variables that generated the observed variables. The previous paper proves that the right answer can be found as the sample size goes to infinity, so this paper compares the finite-sample performance of several new and old ways to make this inference.

This paper builds on recent work by Lewandowski, Kurowicka, and Joe (2009), which expresses a correlation matrix as a function of partial correlations. For the “canonical” parameterization, we provide expressions for the Cholesky decomposition of the correlation matrix as a relatively simple function of the partial correlations, which allows for much faster generation of random correlation matrices. Also, the extension to generating random correlation matrices of a given rank is straightforward.