4

I am trying to run a fixed-effects Poisson Quasi Maximum Likelihood estimator on 3-dimensional(year, country, industry) Panel data.

The dependent variable is the number of patents(non-negative and non-integer) and the main independent variable is the deregulation(a dummy variable which equals 0 before the year deregulation was implemented in a country and 1 starting from the implementation year). I also have some independent variables as controls(size and share) and want to control for the year, country, and industry fixed effects.

I tried several packages and functions on R and mostly ended up on glm and feglm, which I expected to give the same results, but they are different.

model1 <- glm(pt ~ dereg + size + share + 
                factor(year) + factor(country) + factor(industry), 
              data=pdata, family=quasipoisson)

library(fixest)
model2 <- feglm(pt ~ dereg + size + share|year+country+industry, data=pdata,
                cluster=c("country","industry", "year"), 
                family=quasipoisson)

What is the difference between the two methods?

Which one is more reliable in this case, or is there a more suitable package that will allow to include all fixed effects and run quasi-Poisson estimator on the panel data?

Laura
  • 51
  • 3
  • Welcome. What differences did you observe in your summary output? – Thomas Bilach Feb 26 '21 at 19:54
  • Thanks. I made a correction in the model2, and now the difference is only in std. errors (which I think is because of the clustering), and unlike model2 there is an intercept in model1. – Laura Feb 26 '21 at 21:06

0 Answers0