Dear all,
I'm a Swedish research who is struggling with two issues related to a probit panel model (data has approx 30,000 observations and 3 years) specification:
1. Marginal effects:
After running the probit model with random effects ("xtprobit y x"), I want to run an additional command in order to get marginal effects (and hence interpret more than simply the sign and significance). However, if I run the command:
margins, dydx(*)
The marginal effects are identical to the coefficients from the xtprobit-regression (?!). I did some research on alternative commands and found that the following two commands "worked better":
a) mfx compute, predict(pu0)
and
b) margins, predict(pu0) dydx(*)
However, I'm not sure if I understand WHY these give me different numbers and if which one, a) or b), that I should go with. Is anyone familiar with these two command?
2. Log Likelihood test:
I would also like to run a LR-test in order to compare the xtprobit regression with the probit regression (Random vs. Pooled for panel data). What would be the best war to do this? Simply compare the LR-test automatically generated by STATA in the regression output for the xtprobit regression? Or is there an easier way?
Sorry about these, probably, silly questions but I really hope someone can help me out. Any advice would very much appreciated!!
Thanks in advance.
Best,
Karolina
I'm a Swedish research who is struggling with two issues related to a probit panel model (data has approx 30,000 observations and 3 years) specification:
1. Marginal effects:
After running the probit model with random effects ("xtprobit y x"), I want to run an additional command in order to get marginal effects (and hence interpret more than simply the sign and significance). However, if I run the command:
margins, dydx(*)
The marginal effects are identical to the coefficients from the xtprobit-regression (?!). I did some research on alternative commands and found that the following two commands "worked better":
a) mfx compute, predict(pu0)
and
b) margins, predict(pu0) dydx(*)
However, I'm not sure if I understand WHY these give me different numbers and if which one, a) or b), that I should go with. Is anyone familiar with these two command?
2. Log Likelihood test:
I would also like to run a LR-test in order to compare the xtprobit regression with the probit regression (Random vs. Pooled for panel data). What would be the best war to do this? Simply compare the LR-test automatically generated by STATA in the regression output for the xtprobit regression? Or is there an easier way?
Sorry about these, probably, silly questions but I really hope someone can help me out. Any advice would very much appreciated!!
Thanks in advance.
Best,
Karolina
Comment