In this article, we identify and illustrate some shortcomings of the poisson command in Stata. Specifically, we point out that the command fails to check for the existence of the estimates, and we show that it is very sensitive to numerical problems. While these are serious problems that may prevent users from obtaining estimates or may even produce spurious and misleading results, we show that the informed user often has simple workarounds available for addressing these problems.
AlbertA., and AndersonJ. A.1984. On the existence of maximum likelihood estimates in logistic models. Biometrika71: 1–10.
2.
CameronA. C., and TrivediP. K.1998. Regression Analysis of Count Data.Cambridge: Cambridge University Press.
3.
GourierouxC., MonfortA., and TrognonA.1984. Pseudo maximum likelihood methods: Applications to Poisson models. Econometrica52: 701–720.
4.
HabermanS. J.1973. Log-linear models for frequency data: Sufficient statistics and likelihood equations. Annals of Statistics1: 617–632.
5.
ManningW. G., and MullahyJ.2001. Estimating log models: to transform or not to transform?Journal of Health Economics20: 461–494.
6.
SantnerT. J., and DuffyD. E.1986. A note on A. Albert and J. A. Anderson's conditions for the existence of maximum likelihood estimates in logistic regression models. Biometrika73: 755–758.
7.
Santos SilvaJ. M. C., and TenreyroS.2006. The log of gravity. Review of Economics and Statistics88: 641–658.
8.
Santos SilvaJ. M. C.2010. On the existence of the maximum likelihood estimates in Poisson regression. Economics Letters107: 310–312.
9.
TobinJ.1958. Estimation of relationships for limited dependent variables. Econometrica26: 24–36.
10.
WinkelmannR.2008. Econometric Analysis of Count Data. 5th ed. Berlin: Springer.
