This paper examines the finite-sample properties of the random effects
probit estimator in comparison to the standard probit estimator and t
he standard probit estimator with a corrected asymptotic covariance ma
trix. The Monte Carlo experiment considers data-generating processes c
onsistent with longitudinal data and also data from sample surveys. Th
e probit estimator with corrected asymptotic covariance matrix works s
urprisingly well over a wide range of parametric configurations and is
recommended as long as an estimate of the error correlation is not of
high importance.