The new experimental TCOUNTREG procedure is an transitional version of the COUNTREG procedure. It includes all features of the COUNTREG procedure. In addition to features implemented in the COUNTREG procedure, PROC TCOUNTREG provides the following new features:
Two new variable selection methods are provided. The greedy search method can be used either with the forward or backward selection. In each step, the AIC or BIC criterion is evaluated, and the selection continues until the selection criterion is met. The second method uses the penalized likelihood approach to select significant variables. This method is not path-dependent as in the case of greedy search—it falls into the family of LASSO estimators. Using the penalized likelihood method, PROC TCOUNTREG fits a model to the set of all candidate variables and evaluates it simultaneously to find a subset of best-fitting variables.
Several conditional (fixed- and random-effect) count panel data models have been added to the TCOUNTREG procedure. The unconditional panel fixed-effect models can be easily estimated in the TCOUNTREG procedure by using the CLASS statement and the dummy variable approach. This technique is relatively simple but is suitable only for a model with small number of cross sections. If the number of cross sections is large, a conditional model is typically preferred to overcome the incidental parameters problem. The TCOUNTREG procedure enables you to estimate the following types of models:
Poisson regression model with fixed and random effects
negative binomial regression model with fixed and random effects