RANDOM EFFECT MODEL AND GENERALIZED ESTIMATING EQUATIONS FOR BINARY PANEL RESPONSE

Panel data models are widely used in empirical analysis because they allow researchers to control for unobserved individual time-invariant characteristics. However, these models pose important technical challenges. In particular, if individual heterogeneity is left completely unrestricted, and then estimates of model parameters in nonlinear and/or dynamic models suffer from the incidental parameters problem. This problem arises because the unobserved individual characteristics are replaced by inconsistent sample estimates, which, in turn, biases estimates of model parameters. Logit model or probit model on panel data with using univariate approximation (neglect correlation) result consistent estimator but not efficient. In many cases, data are multivariate or correlated (e.g., due to repeated observations on a study subject or for subjects within centers) and it is appealing to have a model that maintains a marginal logistic regression interpretation for the individual outcomes. In this paper, we studied modeling binary panel response using Random Effects Model (REM). Using Monte Carlo Simulation, we research correlations effects to maximum likelihood estimator (MLE) of random effects model. We also compare MLE of REM to Generalized Estimating Equations (GEE) of logit model. Data were generated by using software R.2.8.1 as well as the estimation on the parameters. Based on the result, it can be concluded that (a) In some value of individual effect, random effects model is more better GEE. (b) REM can be accommodating individual effects and closer to parameter than the other. (c) REM is appropriate method to estimate covarians of utility at individual effect having value about one