Heckman model with more observations than reg

Marta Oliveira

Join Date: Nov 2015

Posts: 82
#1

Heckman model with more observations than reg

01 Oct 2019, 03:57

Dear Statalist,
I am running heckman models and compare the results with reg (OLS). For example
heckman lghwage wexpfull, select(age)
reg lghwage wexpfull
But Stata tells me that the number of observations in the heckman model is much higher than with the OLS. Is this normal or does it indicate a problem in my data?
Thank you!
Tags: None

Carlo Lazzaro

Join Date: Apr 2014
Posts: 17704

01 Oct 2019, 04:13

Marta:
ii is always difficult (for me, at least) to give a positive reply without seeing what Stata gave you back.
That said, what you experience may be due to missing values, as you can see from the following toy-example:

Code:

use http://www.stata-press.com/data/r15/womenwk
. reg wage i.educ

      Source |       SS           df       MS      Number of obs   =     1,343
-------------+----------------------------------   F(3, 1339)      =    126.78
       Model |  11803.0955         3  3934.36516   Prob > F        =    0.0000
    Residual |  41551.7992     1,339  31.0319635   R-squared       =    0.2212
-------------+----------------------------------   Adj R-squared   =    0.2195
       Total |  53354.8946     1,342  39.7577456   Root MSE        =    5.5706

------------------------------------------------------------------------------
        wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   education |
         12  |   1.853433   .4027279     4.60   0.000     1.063387    2.643479
         16  |   5.422342   .3910016    13.87   0.000       4.6553    6.189385
         20  |   10.36177   .6331415    16.37   0.000     9.119714    11.60383
             |
       _cons |   20.44185   .2956586    69.14   0.000     19.86184    21.02185
------------------------------------------------------------------------------

. heckman wage i.educ,select(age)

Iteration 0:   log likelihood = -5400.8769  
Iteration 1:   log likelihood = -5383.0138  
Iteration 2:   log likelihood = -5382.4071  
Iteration 3:   log likelihood = -5382.4068  

Heckman selection model                         Number of obs     =      2,000
(regression model with sample selection)              Selected    =      1,343
                                                      Nonselected =        657

                                                Wald chi2(3)      =     344.40
Log likelihood = -5382.407                      Prob > chi2       =     0.0000

------------------------------------------------------------------------------
        wage |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
wage         |
   education |
         12  |   1.357748   .3987013     3.41   0.001     .5763079    2.139188
         16  |   4.920785   .3878262    12.69   0.000      4.16066     5.68091
         20  |   9.764491   .6296831    15.51   0.000     8.530334    10.99865
             |
       _cons |   22.92084   .3929268    58.33   0.000     22.15072    23.69096
-------------+----------------------------------------------------------------
select       |
         age |   .0468164   .0034605    13.53   0.000     .0400339    .0535989
       _cons |  -1.227563   .1249251    -9.83   0.000    -1.472412   -.9827143
-------------+----------------------------------------------------------------
     /athrho |   -.811626   .0976252    -8.31   0.000    -1.002968   -.6202842
    /lnsigma |   1.832919   .0310983    58.94   0.000     1.771968    1.893871
-------------+----------------------------------------------------------------
         rho |  -.6704863   .0537376                     -.7628377   -.5513259
       sigma |   6.252111   .1944302                      5.882416     6.64504
      lambda |  -4.191954   .4454765                     -5.065072   -3.318837
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0):   chi2(1) =    47.86   Prob > chi2 = 0.0000

. sum wage education age

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
        wage |      1,343    23.69217    6.305374    5.88497   45.80979
   education |      2,000      13.084    3.045912         10         20
         age |      2,000      36.208     8.28656         20         59

.

Kind regards,
Carlo
(Stata 19.0)

Comment

Marta Oliveira

Join Date: Nov 2015

Posts: 82
#3

01 Oct 2019, 05:06

I apologize for not showing imediately what Stata gave me back. I attached.
Your example was still able to help me. So Stata is only considering in his calculations only the selected observations (uncensored observations)? And the nonselected (censored) observations are observations where there are missings in the dependent and independent variables?
Attached Files
Comment
Carlo Lazzaro

Join Date: Apr 2014

Posts: 17704
#4

01 Oct 2019, 06:31

Marta:
you may be interested in https://stats.stackexchange.com/ques...he-second-step

As an aside, which version of Stata are you using?

Last edited by Carlo Lazzaro; 01 Oct 2019, 06:34.

Kind regards,
Carlo
(Stata 19.0)
Comment
Marta Oliveira

Join Date: Nov 2015

Posts: 82
#5

01 Oct 2019, 07:30

Thank you, Carlo Lazzaro for all help. Stata estimates for the uncensored observations. I am using Stata 14.
Comment

Carlo Lazzaro

Join Date: Apr 2014
Posts: 17704

01 Oct 2019, 07:38

Marta:
now things are clearer to me (by the way: that's why posters are requested to s'pecify which version of Stata they're using, unless it is not the last released).

when I run the same code in my old Stata 14 version, I got censored/uncensored...

Code:

. use http://www.stata-press.com/data/r15/womenwk

. heckman wage i.educ,select(age)

Iteration 0:   log likelihood = -5400.8769 
Iteration 1:   log likelihood = -5383.0138 
Iteration 2:   log likelihood = -5382.4071 
Iteration 3:   log likelihood = -5382.4068 

Heckman selection model                         Number of obs     =      2,000
(regression model with sample selection)        Censored obs      =        657
                                                Uncensored obs    =      1,343

                                                Wald chi2(3)      =     344.40
Log likelihood = -5382.407                      Prob > chi2       =     0.0000

------------------------------------------------------------------------------
        wage |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
wage         |
   education |
         12  |   1.357748   .3987013     3.41   0.001     .5763079    2.139188
         16  |   4.920785   .3878262    12.69   0.000      4.16066     5.68091
         20  |   9.764491   .6296831    15.51   0.000     8.530334    10.99865
             |
       _cons |   22.92084   .3929268    58.33   0.000     22.15072    23.69096
-------------+----------------------------------------------------------------
select       |
         age |   .0468164   .0034605    13.53   0.000     .0400339    .0535989
       _cons |  -1.227563   .1249251    -9.83   0.000    -1.472412   -.9827143
-------------+----------------------------------------------------------------
     /athrho |   -.811626   .0976252    -8.31   0.000    -1.002968   -.6202842
    /lnsigma |   1.832919   .0310983    58.94   0.000     1.771968    1.893871
-------------+----------------------------------------------------------------
         rho |  -.6704863   .0537376                     -.7628377   -.5513259
       sigma |   6.252111   .1944302                      5.882416     6.64504
      lambda |  -4.191954   .4454765                     -5.065072   -3.318837
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0):   chi2(1) =    47.86   Prob > chi2 = 0.0000

.

...instead of selected/non selected (as reported in Stata 15):

Code:

. use http://www.stata-press.com/data/r15/womenwk

. heckman wage i.educ,select(age)

Iteration 0:   log likelihood = -5400.8769 
Iteration 1:   log likelihood = -5383.0138 
Iteration 2:   log likelihood = -5382.4071 
Iteration 3:   log likelihood = -5382.4068 

Heckman selection model                         Number of obs     =      2,000
(regression model with sample selection)              Selected    =      1,343
                                                      Nonselected =        657

                                                Wald chi2(3)      =     344.40
Log likelihood = -5382.407                      Prob > chi2       =     0.0000

------------------------------------------------------------------------------
        wage |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
wage         |
   education |
         12  |   1.357748   .3987013     3.41   0.001     .5763079    2.139188
         16  |   4.920785   .3878262    12.69   0.000      4.16066     5.68091
         20  |   9.764491   .6296831    15.51   0.000     8.530334    10.99865
             |
       _cons |   22.92084   .3929268    58.33   0.000     22.15072    23.69096
-------------+----------------------------------------------------------------
select       |
         age |   .0468164   .0034605    13.53   0.000     .0400339    .0535989
       _cons |  -1.227563   .1249251    -9.83   0.000    -1.472412   -.9827143
-------------+----------------------------------------------------------------
     /athrho |   -.811626   .0976252    -8.31   0.000    -1.002968   -.6202842
    /lnsigma |   1.832919   .0310983    58.94   0.000     1.771968    1.893871
-------------+----------------------------------------------------------------
         rho |  -.6704863   .0537376                     -.7628377   -.5513259
       sigma |   6.252111   .1944302                      5.882416     6.64504
      lambda |  -4.191954   .4454765                     -5.065072   -3.318837
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0):   chi2(1) =    47.86   Prob > chi2 = 0.0000

.

Kind regards,
Carlo
(Stata 19.0)

Announcement