Expand data with estimated fixed effects

[pp2382]

Thanks so much for your work Laurent!

I am estimating a Poisson regression with several fixed effects and could not find a method to extract each fixed effects and merge them with their observation.
The function fixef extracts the FE but I could not find a way to merge them back to the initial dataset.

Am I missing something? Is there a simple way to assign the correct FE to each observation? It does not seem easy with the current format of the output of fixef.

Thanks very much

[lrberge]

Hi and thanks for your kind words!

Indeed, this is not possible currently. That's because the use case is fairly limited so I didn't implement it.

But maybe you've found a solution by now?

[pp2382]

Yes, I did manage to find a way. Here is the code I use in case it can be useful for someone else (even though I am still learning my way around R)

##I first create my fixed effects (method doesn't work with combination of fe using ^)

dt[,var1_x_var2 := .GRP , by=c("var1","var2")]
dt[,var3_x_var4 := .GRP , by=c("var3","var4")]

##Then I run my regression

pois_reg <- fepois(Y ~ X1 |  var1_x_var2 + var3_x_var4 , data=dt)

##I use fixef to extract the fe and convert them in data.table
fe <- fixef(pois_reg)
fe1<-data.table(names = names(fe$var1_x_var2 ), fe1=fe$var1_x_var2 )
fe1[,names:=as.numeric(names)]

fe2<-data.table(names = names(fe$var3_x_var4), alpha=fe$var3_x_var4)
fe2[,names:=as.numeric(names)]

##Finally, I merged the two tables of fixed effects with the initial one

dt<-merge(dt,fe1,by.x=c("var1_x_var2"),by.y=c("names"),all.x=T)
dt<-merge(dt,fe2,by.x=c("var3_x_var4"),by.y=c("names"),all.x=T) 

It seemed to work for me, but I am not sure how general this solution is.

Thanks for following up and thanks again for a great package.