Dear Stataforum
First, I want to thank all Stataforum contributors for your yeoman service. I have benefited enormously over the years from the forum.
I have a dataset of retail workers and would like to study the effect of store renovation on their exit, controlling for worker fixed effects. (From interviews, I understand that the renovation affects workers differently.) I tried a duration model (stcox) with worker fixed effects (frailty) but it would not converge, and anyhow, the plot shows that the data do not satisfy proportional hazards.
Next, I tried clogit with group(worker) but the frequency of exits (about 2.7 percent of the data) is so low that the routine does not converge.
So then, I tried firthlogit. It worked as follows:
. firthlogit exit conv age_ln ot_ih i.cashier i.year if $sample & cashier == 1 & sgpr == 1 & new == 1
where exit = exit, conv = renovated store (time varying), age_ln = ln(worker age), ot_ih = overtime, and $sample, cashier, sgpr, new are restrictions on the sample.
Not sure how to incorporate the worker fixed effects, I tried,
. firthlogit exit conv age_ln ot_ih r i.cashier i.year if $sample & cashier == 1 & sgpr == 1 & new == 1
But firthlogit complained and returned
"note: 1.cashier omitted because of collinearity
initial vector: matrix must be dimension 10
r(503);"
Grateful for your advice.
Best
IP
0 Response to firthlogit: initial vector: matrix must be dimension 10
Post a Comment