DID- OLS vs PPML

[Niyaj Mohammed]

I'm running a DID regression where my outcome is the number of crimes committed by group g for each day. I run OLS and poison regression and I get different results. I wanted to know which result should I rely?

[Jared Greathouse]

Presumably poisson.

[Joao Santos Silva]

Dear Niyaj Mohammed,

When do you use OLS, is the dependent variable in levels or in logs? If it is in logs, then Poisson is indeed preferable. If it is in levels, note that the two estimators are based on different assumptions: OLS assumes parallel trends while Poisson assumes proportional trends. Only one of them can be valid.

Best wishes,

Joao

[Niyaj Mohammed]

The dependent variable is in levels.