Hi,
I have an unbalanced panel data of stocks with N=20,000 and T=50 years. I have 50 industries. I want to estimate the effect of X on Y where Y is a dummy variable. I want to control for "industry*Year" fixed effects where I have around 2,500 pairs (50*50). I have two questions:
1) If I use clogit y x, group(ind_year) or "xtlogit y x, fe by setting xtset ind_year", I get the same results and I believe I can only interpret the sensitivity of the odd ratio to the changes in X. Is that correct for these two commands?
2) If I use logit with "industry*Year" dummy variables, I can get coefficients for each pair although very hard computationally. However, the number of observations for a given "industry*Year* pair ranges between 1 and 950 but most of them are greater than 25. I know that we need a high number of observations for a given fixed effect (This is different than T). In this case, can I just use predict command to estimate the probabilities? I might drop "industry*Year" observations with fewer than 25 obs if necessary.
Thanks a lot.
Best,
Ulas
I have an unbalanced panel data of stocks with N=20,000 and T=50 years. I have 50 industries. I want to estimate the effect of X on Y where Y is a dummy variable. I want to control for "industry*Year" fixed effects where I have around 2,500 pairs (50*50). I have two questions:
1) If I use clogit y x, group(ind_year) or "xtlogit y x, fe by setting xtset ind_year", I get the same results and I believe I can only interpret the sensitivity of the odd ratio to the changes in X. Is that correct for these two commands?
2) If I use logit with "industry*Year" dummy variables, I can get coefficients for each pair although very hard computationally. However, the number of observations for a given "industry*Year* pair ranges between 1 and 950 but most of them are greater than 25. I know that we need a high number of observations for a given fixed effect (This is different than T). In this case, can I just use predict command to estimate the probabilities? I might drop "industry*Year" observations with fewer than 25 obs if necessary.
Thanks a lot.
Best,
Ulas
Comment