# [Limdep Nlogit List] Fitted values in the sample selection model

elantkoj at ae.krakow.pl elantkoj at ae.krakow.pl
Thu Sep 13 02:13:12 EST 2007

```Hello!
I am analyzing the regression model with sample selection, so I have
commands like:
PROBIT; Lhs=SELEKCJA; Rhs=XSEL; Hold; Quiet\$
SELECT; Lhs=ZAL; Rhs=X; MLE; Keep=CondMMLE; Res=ResMLE \$
Values for CondMMLE are calculated as conditional mean separately for
SELEKCJA=1 and for SELEKCJA=0.
But when I add to my program  ;Mar  specification like:
PROBIT; Lhs=SELEKCJA; Rhs=XSEL; Hold; Quiet\$
SELECT; Lhs=ZAL; Rhs=X; MLE; Keep=CondMMar; Res=ResMef; Mar\$
fitted values CondMMLE and CondMMef are completely different.
And residuals ResMLE and ResMef are also different.
Does anybody know why?
How does LimDep calculate CondMMef?
Which results (which fitted values) are correct? Should I use CondMMLE or
CondMMef?
There is nothing about it in Manual.
In fact R2Decomp for MLE and for MLE with ;Mar are also VERY different!
For example 0.11 and 0.62.

Next problem is, that in the first specification:
PROBIT; Lhs=SELEKCJA; Rhs=XSEL; Hold; Quiet\$
SELECT; Lhs=ZAL; Rhs=X; MLE; Keep=CondMMLE; Res=ResMLE \$
LimDep gives parameters for Alfa, Beta, Sigma and Rho (in B).
But in the second one:
PROBIT; Lhs=SELEKCJA; Rhs=XSEL; Hold; Quiet\$
SELECT; Lhs=ZAL; Rhs=X; MLE; Keep=CondMMar; Res=ResMef; Mar\$
LimDep gives parameters for Alfa, Beta, Sigma, Rho and one more parameter
– but what is it? (also in B).

Thanks for all suggestions!