Logistic Regression Analysis of CPS Overlap Survey Split Panel Data

Robin Fisher


The complex nature of survey data changes the distributions of statistics associated with logistic regression models. The variances in the traditional logistic regression models are generally too small, leading to overly liberal tests. Roberts, Rao, Kumar (1987) provide a method for constructing approximate tests when a separate estimator for the covariance matrix is available. I apply this method with some modifications and the variance estimator described in Fisher et al (1993) to the CPS Survey split panel data.