Jackknife methods for improved cluster–robust inference

Inferential problems have long been known to exist in finite samples when using the conventional cluster–robust variance estimator for ordinary least squares. Many improvements to inference have been suggested, including bootstrap and jackknife methods, in addition to alternative standard errors and degrees of freedom. This presentation will discuss how to use jackknife methods in Stata for improved inference. We detail the new Stata ado-command, summclust, which offers both improved inferences and diagnostic tools for assessing when conventional errors can be problematic. We also discuss jackknife methods for two different situations: linear models with multiway clustering and nonlinear models with one-way clustering. These alternative methods considerably improve upon the finite-sample overrejection problems.