heckman selection on multinomial probit first stage

[togarra]

Hi All, I am looking for a way to run an OLS with a Heckman correction on data in which the first stage involves 3 choices - I realise there is no specific command in Stata for this, but wonder if someone could advise on an simple way to go about this ? This would be so helpful! I mostly want to check whether my data has sample selection issues - I don't think it does, but I want to be sure. Any advice on how to go about this, especially for someone who is not very fluent with econometrics? Thanks! Tanya

[Mengqian Chen]

Hi togarra,
May I ask do you find the answer as I have the same difficulty? Many thanks.
Best,
Mengqian

[FernandoRios]

You may want to look into the community contributed command
-cmp-
it should be able to estimate a model like the one you describe

[Mengqian Chen]

Hi FernandoRioa,
Thanks a lot for your comment. I find -cmp- helpful. However, when I use -cmp- to conduct Heckman with a first stage of -mprobit-, STATA reports some errors.

My first-stage Heckman regression is mprobit because I have three categories in the dependent variable, selectvar. I could have used mlogit, but IIA is violated. Therefore, I guess mprobit is a better choice.

My code is below:

. cmp (outcome = $VarList2 i.fyear) (selectvar= $VarList1, iia) if Obs==1 & Universal=="Yes" , ///
ind(selectvar*$cmp_cont $cmp_mprobit) nolr qui

STATA results:

. cmp (outcome = $VarList2 i.fyear) (selectvar= $VarList1, iia)
> if Obs==1 & Universal=="Yes" , ind(selectvar*$cmp_cont
> $cmp_mprobit) nolr qui

Fitting individual models as starting point for full model fit.

Fitting full model.
Likelihoods for 336 observations involve cumulative normal distributions above dimension 2
> .
Using ghk2() to simulate them. Settings:
    Sequence type = halton
    Number of draws per observation = 37
    Include antithetic draws = no
    Scramble = no
    Prime bases = 2 3
Each observation gets different draws, so changing the order of observations in the data s
> et would change the results.

               cmp_lf1():  3301  subscript invalid
    mopt__calluser_lf2():     -  function returned error
      opt__eval_nr_lf1():     -  function returned error
             opt__eval():     -  function returned error
opt__looputil_iter0_common():     -  function returned error
opt__looputil_iter0_nr():     -  function returned error
          opt__loop_nr():     -  function returned error
             opt__loop():     -  function returned error
            _moptimize():     -  function returned error
           Mopt_maxmin():     -  function returned error
                 <istmt>:     -  function returned error
Mata run-time error
Mata run-time error
r(3301);

Do you have any suggestions to improve the code? Many thanks.

Best,
Mengqian

[Marry Lee]

I wonder if any one would confirm the validity of Mengqian's code.

cmp (outcome = $VarList2 i.fyear) (selectvar= $VarList1, iia) if Obs==1 & Universal=="Yes" , /// ind(selectvar*$cmp_cont $cmp_mprobit) nolr qui

Is it ok to proceed this way when we want to run a selection equation (for Heckman procedure) that has more than 2 options (i.e. a multinomial selection variable)?

Particularly, would the generation of the IMRs through this code be correct?

predict xb, eq(selectvar) xb
predict e, eq(selectvar) e(0 .)
gen cmp_mills = e - xb

Thank you
Many thanks.

[Marry Lee]

Sorry about this.

[Stephen Jenkins]

Marry Lee -search svyselmlog- for an SSC implementation. Also the original -selmlog- available here. Reading: see also Bourguignon et al. JOURNAL OF ECONOMIC SURVEYS Vol. 21, No. 1, doi 0950-0804/07/01 0174–32

[Marry Lee]

Thank you Stephen Jenkins. I will look at the references you mentioned.

[Mudabira Fayaz]

Marry Lee -search svyselmlog- for an SSC implementation. Also the original -selmlog- available here. Reading: see also Bourguignon et al. JOURNAL OF ECONOMIC SURVEYS Vol. 21, No. 1, doi 0950-0804/07/01 0174–32

Hi Stephan ,

I am trying to install selmlog command in stata but the search function doesn't show any results for installlation of the command.
I also checked the URL you supplied but I can't open the zipped file. Is there any other way to install this command?

Thanks in advance.

Regards,
Mudabira

[Stephen Jenkins]

. ssc d svyselmlog

----------------------------------------------------------------------------------------------------------------------------------------------------------
package svyselmlog from http://fmwww.bc.edu/repec/bocode/s
----------------------------------------------------------------------------------------------------------------------------------------------------------

TITLE
      'SVYSELMLOG': module to compute selectivity adjustment based on the multinomial logit for survey design

DESCRIPTION/AUTHOR(S)
      
       -svyselmlog- estimates the parameters of the main equation
      correcting  for selectivity using the conditional probabilities
      from a multinomial logit.  -svyselmlog- allows for three
      parametric selectivity-correction methods like the  ones
      developed in Lee (1984), Durbin and McFadden (1984) and
      Bourguignon, Fournier  and Gurgand (2004). Additionally,
      selectivity can be adjusted following Dahl's (2002)
      semi-parametric method.
      
      KW: multinomial logit
      KW: selectivity
      KW: conditional probabilities
      
      Requires: Stata version 8
      
      Distribution-Date: 20050921
      
      Author: R. E. De Hoyos, University of Cambridge
      Support: email [email protected]
      

INSTALLATION FILES                           (type net install svyselmlog)
      svyselmlog.ado
      svyselmlog.hlp
----------------------------------------------------------------------------------------------------------------------------------------------------------
(type ssc install svyselmlog to install)

I have also verified that the -selmlog- code in the zip-file is downloadable, and can be extracted by me. (You might have to override a browser's blocking the opening of the file -- the probable reason is that the help file has "hlp" extension, not modern-day "sthlp")