[John Mullahy]

Sa Sa: After you estimate your logit model, can you generate predicted probabilities

predict pc

Then can you show us the results of

summarize pc cesarean physicians, d

[sa sa]

Yes Clyde, i have the same error message with log_physicians as the predictor and margins, eydx(log_physicians)

. Here the results John
summarize physicians, d

physicians
-------------------------------------------------------------
Percentiles Smallest
1% .5154454 .4125226
5% .60027 .4125226
10% .6411128 .4125226 Obs 99728
25% .7966681 .4125226 Sum of Wgt. 99728

50% 1.13773 Mean 1.101283
Largest Std. Dev. .3380353
75% 1.324555 2.038605
90% 1.4614 2.038605 Variance .1142679
95% 1.505399 2.038605 Skewness .4005778
99% 2.038605 2.038605 Kurtosis 3.235274

[sa sa]

. summarize pc cesarean physicians, d

Linear prediction
-------------------------------------------------------------
Percentiles Smallest
1% -2.388353 -2.388353
5% -2.371035 -2.388353
10% -2.336569 -2.388353 Obs 99728
25% -2.264516 -2.388353 Sum of Wgt. 99728

50% -2.215366 Mean -2.22541
Largest Std. Dev. .0733672
75% -2.174733 -2.080387
90% -2.123587 -2.080387 Variance .0053828
95% -2.116638 -2.080387 Skewness -.382055
99% -2.105139 -2.080387 Kurtosis 2.608612

cesarean
-------------------------------------------------------------
Percentiles Smallest
1% 0 0
5% 0 0
10% 0 0 Obs 99728
25% 0 0 Sum of Wgt. 99728

50% 0 Mean .1027094
Largest Std. Dev. .3035804
75% 0 1
90% 1 1 Variance .0921611
95% 1 1 Skewness 2.61738
99% 1 1 Kurtosis 7.850676

physicians
-------------------------------------------------------------
Percentiles Smallest
1% .5154454 .4125226
5% .60027 .4125226
10% .6411128 .4125226 Obs 99728
25% .7966681 .4125226 Sum of Wgt. 99728

50% 1.13773 Mean 1.101283
Largest Std. Dev. .3380353
75% 1.324555 2.038605
90% 1.4614 2.038605 Variance .1142679
95% 1.505399 2.038605 Skewness .4005778
99% 2.038605 2.038605 Kurtosis 3.235274

[John Mullahy]

Sa Sa: What estimation method are you using? If you are using logit, then the predicted probabilities should be between 0 and 1, but the values you show here are all negative.

If you used

logit cesarean physicians

then did you use

predict pc

or

predict pc, xb

Your output shows a header of -Linear prediction- so my guess is that you did the later but I was hoping to see the former.

[sa sa]

I use xtlogit cesarean physicians, re vce (robust)
and then predict pc

[John Mullahy]

Thanks Sa Sa. I'm running out of ideas. One more thought: If you estimate the model using -logit- instead of -xtlogit- does -margins- give you the same error with -eyex- ? If -margins- works with -logit- works not with -xtlogit- then my guess is that the culprit is related to the random effect. But this is just a guess.

[sa sa]

when i use predict pc, i get the following message : (option xb assumed; linear prediction)
When i use logit, i can estimate margins, eyex(pysician) and i get the same result as the coef of physicians
However, since i have panel data, i have to work with multilevel logit model....

[John Mullahy]

Yes, I can understand why you want to use -xtlogit-.

I do find it strange why, with -logit-, your -margins, eyex(physicians)- gives the same result as the parameter estimate for physicians.

I'm sorry, but I've run out of ideas. Hopefully others who understand -xtlogit- better than I do will provide commentary.

[sa sa]

Thanks John...

[Nick Cox]

sa sa: Please re-register with given name and family name (or accept my apologies if those are really your given and family names).

More at https://www.statalist.org/forums/help#realnames