Using Predict after suest in logit regression

[Ujjawal Sawarn]

Hi,

I have used 3 logit equation and used "suest" command to combine parameter estimate and covariance matrices. Now, I want to use "Predict" command to estimate predicted probability.
When I code:

predict probab if e(sample) == 1

The values I am getting are greater than one. Hence, they are not predicted probability. I need the predicted probability to calculate concordance Index.
Please help me to calculate the predicted probability or concordance index directly.

Very Thanks in advance.

[Ujjawal Sawarn]

Sorry, I forgot to mention. The three equations have different depvar but same indepvar as:
logit Y1 X1 X2 X3
logit Y2 X1 X2 X3
logit Y3 X1 X2 X3
suest Y1 Y2 Y3, vce(clustercode)

Please share the solution if you have to calculate the predicted probability or concordance index from here.

[Andrew Musau]

What you are getting with suest are the linear predictions. The -pr- option of predict is not supported. Depending on whether it makes sense to do what you want (I can only vouch in the case of one estimated model as predictions will relate to that model), the logit and probit transformations are straightforward and only need the use of the functions -invlogit()- and -normal()- respectively.

webuse lbw
qui logit  low age i.race smoke
estimates store logit
predict lowhat1, pr
suest logit
predict xb1
gen lowhat2= invlogit(xb1)
qui probit  low age i.race smoke
estimates store probit
predict lowhat3, pr
suest probit
predict xb2
gen lowhat4= normal(xb2)
*ASSERTIONS ADJUSTING FOR ROUNDING
assert round(lowhat1, 4)==round(lowhat2, 4)
assert round(lowhat3, 4)==round(lowhat4, 4)

Res.:

. *ASSERTIONS ADJUSTING FOR ROUNDING

.
. assert round(lowhat1, 4)==round(lowhat2, 4)

.
. assert round(lowhat3, 4)==round(lowhat4, 4)

.

[Ujjawal Sawarn]

Thank you so much for responding. I have removed "suest" and instead used simple logit in all 3 regression with vce (cluster). So, I ran 3 different regression with vce and then used predict command for estimated probability.
logit Y1 X1 X2 X3, vce(clustercode)
predict probability
logit Y2 X1 X2 X3, vce(clustercode)
predict proability2
logit Y3 X1 X2 X3, vce(clustercode)
predict proability3
I think this also will give correct result.

[Andrew Musau]

Yes, prediction for each model is fine. You typically do not want to use the robust variance estimator in logit unless either you believe that your model is misspecified or you have clustered data. See https://www.stata.com/support/faqs/s...nce-estimator/

[Ujjawal Sawarn]

Okay. Thank you.