Small sample validity of latent variable models for correlated binary data

Abstract

Recently developed models for correlated binary responses assume that the binary outcomes are manifestations of latent normal variables. In these models, the covariance structure is based on the tetrachoric correlations and is more flexible than similar models which use Pearson correlations. The parameters are estimated by the approach of generalized estimating equations(GEE). This simulation study investigation the impact of sample size and incorrectly assuming an independence model on parameter estimates in terms of bias, efficiency and coverage probability is close to the nomial level unless the sample is small (≤20), or an incorrect independence model is assumed for correlated unbalanced data. Methods for reducing bias are proposed using a weighed jackknife technique