Interpretation of PPML and QMLE results

[Sara Signorelli]

I am estimating a difference in differences specification to test the effect of a reform on the hiring of foreign workers in firms. My main outcome variables are the number of foreign workers hired by a firm during a specific year (entryF), and the share of hires done by a firm in a specific year made of foreign workers (sh_entryF). The data I have is administrative and it records all labor movements experienced by each large firm per year for a panel of 12 years.

Since both outcome variables include a large number of zeros (many firms do not hire any foreigner), I decided to use a ppml model for entryF (since it has a count structure) and the QML estimator proposed by Papke and Wooldrige (1996) for fractional data for sh_entryF: http://onlinelibrary.wiley.com/doi/1...O;2-1/abstract . My main coefficient of interest is the one associated with the interaction between a dummy for treated group and a dummy for post-reform period, it therefore only takes 0 and 1 values.

ppml entryF treat treat_post i.year, cluster(firm)

glm sh_entryF treat treat_post i.year, family(binomial) link(logit) cluster(firm)

My question is, how can I interpret the ppml and the QML stata output? Can I interpret the coefficient on "treat_post" as a partial elasticity? (the reform increases entryF / sh_entryF by \beta %), or should I run a postestimation command to get marginal effects that are more readily interpretable?

Thank you!

[Ashani Abayasekara]

Hi Sara,

I have a similar question about a Poisson model I want to run in a DiD setting where the dependent variable is real income which has many 0 values. Just wondering if you were able to figure out the best way to interpret coefficients, and whether the PPML and QMLE is the better model?

Many thanks,
Ashani