Dear forum,
I am running a correlated random effects logit model (cre) for a binary dependent variable (labor fore participation) and want to estimate marginal effects for different subpopulations of my data.
After running my model:
I used the margins command with subpop(if) option:
.
This gives me the marginal effects for the specified subpopulation.
I´ve become sceptical about the results, as the p values are very similar for each subgroup (always lies between 0.01 and 0.015). First I have doubts that the effect is really significant for all subgroups and second and more important some subgroups are very small. The smallest subgroup contains of 1,039 observations and 592 individuals and only 6(!) observations have a change in the independent variable (indepvar)
I would expect very large SE in this subpopulation because of the view changes in the dependent variable and thus the effect not be significant. Even if the effect is really significant in this group, I think its odd that the p values are so similar for all of my (30) subgroups. Besides, the size of the effects vary more (between -0.02 & -0.045).
So either the significance of the efffects are really similar distributed in the population or something is wrong with the calculation.
Can anyone give me some details how stata calculates the SE and p values for the margins command with subpop(if) option and/or any suggestions why the p values are so similar?
I thank you in advance and if you need any further information please let me know.
Best,
Claudio
I am running a correlated random effects logit model (cre) for a binary dependent variable (labor fore participation) and want to estimate marginal effects for different subpopulations of my data.
After running my model:
Code:
xtlogit depvar indepvar $controls indepvar_bar $controls_bar, re cluster(clusterid)
Code:
margins, dydx(indepvar) subpop(if subpop==1)
This gives me the marginal effects for the specified subpopulation.
I´ve become sceptical about the results, as the p values are very similar for each subgroup (always lies between 0.01 and 0.015). First I have doubts that the effect is really significant for all subgroups and second and more important some subgroups are very small. The smallest subgroup contains of 1,039 observations and 592 individuals and only 6(!) observations have a change in the independent variable (indepvar)
I would expect very large SE in this subpopulation because of the view changes in the dependent variable and thus the effect not be significant. Even if the effect is really significant in this group, I think its odd that the p values are so similar for all of my (30) subgroups. Besides, the size of the effects vary more (between -0.02 & -0.045).
So either the significance of the efffects are really similar distributed in the population or something is wrong with the calculation.
Can anyone give me some details how stata calculates the SE and p values for the margins command with subpop(if) option and/or any suggestions why the p values are so similar?
I thank you in advance and if you need any further information please let me know.
Best,
Claudio
