Conditional Mixed Process (CMP) and Heckprobit have different results

[Rio Nguyen]

Hi everyone,

I am using Heckprobit and CMP (https://www.stata-journal.com/articl...article=st0224) for a probit model with sample selection bias but they gave different results. Could anyone help me have a look at my codes if there is something wrong with them? Codes and results are reported below:

Thank you very much for your help.
Thao.

heckprobit DB reject i.rela_no hardlend finobsnew, select (neednew= reject expgrowth)

and

cmp (DB= reject i.rela_no hardlend finobsnew) (neednew= reject expgrowth), ind(neednew*$cmp_probit $cmp_probit) qui

Probit model with sample selection              Number of obs     =     30,106
                                                Censored obs      =     24,729
                                                Uncensored obs    =      5,377

                                                Wald chi2(5)      =      84.06
Log likelihood = -17489.53                      Prob > chi2       =     0.0000

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
DB           |
      reject |    .077246    .058208     1.33   0.184    -.0368395    .1913315
             |
     rela_no |
          2  |  -.0458923   .0802493    -0.57   0.567    -.2031779    .1113934
          3  |  -.4600466    .271659    -1.69   0.090    -.9924885    .0723952
             |
    hardlend |  -.1949219   .0382624    -5.09   0.000    -.2699148   -.1199291
   finobsnew |   .6397209   .0789153     8.11   0.000     .4850496    .7943921
       _cons |   1.195207   .1331406     8.98   0.000     .9342563    1.456158
-------------+----------------------------------------------------------------
neednew      |
      reject |   .2790752   .0224358    12.44   0.000     .2351018    .3230486
   expgrowth |  -.1708988   .0168584   -10.14   0.000    -.2039406    -.137857
       _cons |  -.8663599   .0133933   -64.69   0.000    -.8926102   -.8401096
-------------+----------------------------------------------------------------
     /athrho |  -.9685172   .2305167    -4.20   0.000    -1.420322   -.5167129
-------------+----------------------------------------------------------------
         rho |   -.748052   .1015237                      -.889666    -.475159
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0):   chi2(1) =    16.76   Prob > chi2 = 0.0000

Mixed-process regression                        Number of obs     =     30,403
                                                LR chi2(7)        =     612.05
Log likelihood = -17686.097                     Prob > chi2       =     0.0000

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
DB           |
      reject |   .2992867    .046445     6.44   0.000     .2082562    .3903172
             |
     rela_no |
          2  |  -.0458934   .1053684    -0.44   0.663    -.2524117     .160625
          3  |  -.6566487    .362371    -1.81   0.070    -1.366883    .0535854
             |
    hardlend |  -.2756387   .0400279    -6.89   0.000     -.354092   -.1971854
   finobsnew |   .8335987   .0613801    13.58   0.000     .7132958    .9539015
       _cons |    .076203   .0789435     0.97   0.334    -.0785234    .2309294
-------------+----------------------------------------------------------------
neednew      |
      reject |   .2795577   .0224325    12.46   0.000     .2355908    .3235246
   expgrowth |  -.1672135   .0172992    -9.67   0.000    -.2011193   -.1333076
       _cons |  -.8685423   .0135838   -63.94   0.000     -.895166   -.8419186
-------------+----------------------------------------------------------------
/atanhrho_12 |   .0345465   .0532142     0.65   0.516    -.0697515    .1388445
-------------+----------------------------------------------------------------
      rho_12 |   .0345328   .0531508                     -.0696386    .1379591
------------------------------------------------------------------------------

[Phil Bromiley]

You didn't get a quick response. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

For some reason, the two estimations have slightly different sample sizes. The only thing that looks really different is the parameter on reject. The calculations of rho are probably reporting different functions of rho. You'll need to fix the sample size, and then look closely at the specifics of the calculations in each estimator. You might ask the author of cmp about this.