Dynamic Panel of Count Data with Initial Conditions and Correlated Random Effects : Application for Health Data

One of the well known model is the ‘dynamic fixed effects panel model by Blundell, Grifffiths, and windmeijer (1995, 2002)’, which is called linear feedback model. This model uses first difference transformations and GMM for estimations. It assumes that the lagged dependent variable has liner relationship with current dependent variables, and the other covariates are related by nonlinear ways, exponential function. Therefore, we cannot use prepackaged computer program like STATA and SAS for the estimations. Also, the model employs fixed effects model, but if we use fixed effects model, we lose information of time-invariant variables like gender, race, and so on, which is regarded as important variables in microeconometrics. Finally, the initial values are important in the case of short panel. For example, if the number of doctor visits of this quarter is 5, then the number of doctor visits for next several quarters may highly related to the number of this quarter. However, the previous studies have not explicitly included the effects of initial values into their models. So, what I want to do in this paper is to make model which has low estimation cost, explicitly includes initial values as one of its covariates, and use adjusted random effects model where individual specific effects are related to covariates.