A New Estimator for Standard Errors with Few Unbalanced Clusters

In linear regression analysis, the estimator of variance coefficients should take into account clustered nature data, if present, since using standard textbook formula will in that case lead to a severe downward bias errors. This idea cluster-robust (CRVE) generalizes clusters classical heteroskedasticity-robust estimator. Its justification is asymptotic number clusters. Although an improvement, considerable could remain when low, more so regressors are correlated within cluster. order address these issues, two improved methods were proposed; one method, which we call CR2VE, was based on biased reduced linearization, while other, CR3VE, can be seen as jackknife The latter unbiased under very strict conditions, particular equal cluster size. To relax this condition, introduce paper CR3VE-?, generalization CR3VE where size allowed vary freely between We illustrate performance CR3VE-? through simulations and show that, especially sizes widely, it outperform other commonly used estimators.