Estimation of Count Data Models with Multiplicative Fixed Effects and Spatial Dependence

This note looks at estimation of spatial autoregressive models for non-negative and count outcomes with multiplicative fixed effects. We show that in presence of significant proportion of zeros in the count variable, control function and Instrumental Variable estimation to model such spatial de5pendence yields biased and inconsistent estimates, which can be primarily attributed to the assumed form of proxy for the spatial autoregressive term. This may be remedied by using a different form of proxy. We also show that consistent estimates can be derived with Conditional Poisson Maximum Likelihood based on modified ‘reduced’ form conditional mean specification. The properties of these estimators are subsequently investigated in Monte Carlo trials