ivprobit reports first stage results even w/o `first` option

[Michael Rose]

I'm using ivprobit with Stata 14 and I think I discovered a bug. There is the option "first" to display the first stage regression results, and the documentation says "The default is not to show these parameter estimates".

I think in the ML case (the default), it shows the first stages results:

use https://stats.idre.ucla.edu/stat/stata/dae/binary.dta, clear
ivprobit admit (gre=gpa) i.rank 
esttab

produces

. esttab

----------------------------
                      (1)   
                    admit   
----------------------------
admit                       
gre               0.00501***
                   (4.16)   

1.rank                  0   
                      (.)   

2.rank             -0.363   
                  (-1.87)   

3.rank             -0.617** 
                  (-2.82)   

4.rank             -0.763** 
                  (-2.97)   

_cons              -2.967***
                  (-3.91)   
----------------------------
gre                         
1.rank                  0   
                      (.)   

2.rank             -5.055   
                  (-0.31)   

3.rank             -34.56*  
                  (-2.08)   

4.rank             -25.86   
                  (-1.38)   

gpa                 117.2***
                   (8.37)   

_cons               207.0***
                   (4.12)   
----------------------------
/                           
athrho2_1          -0.418*  
                  (-2.47)   

lnsigma2            4.659***
                 (131.78)   
----------------------------
N                     400   
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

whereas the 2-step case with

estimates clear
ivprobit admit (gre=gpa) i.rank, twostep
esttab

reports only the second stage, as promised:

. esttab

----------------------------
                      (1)   
                    admit   
----------------------------
gre               0.00545***
                   (3.32)   

1.rank                  0   
                      (.)   

2.rank             -0.395   
                  (-1.91)   

3.rank             -0.671** 
                  (-2.99)   

4.rank             -0.831** 
                  (-3.16)   

_cons              -3.230** 
                  (-3.17)   
----------------------------
N                     400   
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

Or am I doing something wrong? What would be a workaround?

[George Ford]

 
 ivprobit admit i.rank (gre=gpa)

[Michael Rose]

?

[Andrew Musau]

I am not sure what you are getting at. The manual refers to the Stata output and you are showing us the result of esttab (a user written command that is part of estout from SSC) to refute what the manual is stating. No first-stage result below, as far as I see.

. use https://stats.idre.ucla.edu/stat/stata/dae/binary.dta, clear

. 
. ivprobit admit (gre=gpa) i.rank

Fitting exogenous probit model

Iteration 0:   log likelihood = -249.98826  
Iteration 1:   log likelihood = -229.29667  
Iteration 2:   log likelihood = -229.20659  
Iteration 3:   log likelihood = -229.20658  

Fitting full model

Iteration 0:   log likelihood = -2660.4766  
Iteration 1:   log likelihood = -2660.4609  
Iteration 2:   log likelihood = -2660.4609  

Probit model with endogenous regressors         Number of obs     =        400
                                                Wald chi2(4)      =      49.08
Log likelihood = -2660.4609                     Prob > chi2       =     0.0000

-------------------------------------------------------------------------------------
                    |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
                gre |   .0050071   .0012041     4.16   0.000     .0026471    .0073671
                    |
               rank |
                 2  |   -.362674    .193766    -1.87   0.061    -.7424485    .0171004
                 3  |    -.61669   .2190426    -2.82   0.005    -1.046006   -.1873743
                 4  |  -.7629496   .2570431    -2.97   0.003    -1.266745   -.2591544
                    |
              _cons |  -2.967458   .7588781    -3.91   0.000    -4.454831   -1.480084
--------------------+----------------------------------------------------------------
 corr(e.gre,e.admit)|  -.3951277   .1426047                     -.6345953    -.086435
           sd(e.gre)|   105.5513   3.731802                      98.48477    113.1249
-------------------------------------------------------------------------------------
Instrumented:  gre
Instruments:   2.rank 3.rank 4.rank gpa
-------------------------------------------------------------------------------------
Wald test of exogeneity (corr = 0): chi2(1) = 6.11        Prob > chi2 = 0.0134

[George Ford]

#2 is the correct statement. you need the IV part at the end.

[Michael Rose]

But if it's not in the Stata output, then why is it in esttab?

[George Ford]

I see. The issue is that the coefficients from both models are stored and esttab just grabs it all. you can -estab, unstack- and it reports the results in separate columns.

This works:

esttab , drop(admit:*)

[Andrew Musau]

But if it's not in the Stata output, then why is it in esttab?

The output refers to what you see on the results window. But this may or may not coincide with what is stored in e() or r().

[Zara Khan]

Wald test of exogeneity (corr = 0): chi2(1) = 0.12 Prob > chi2 = 0.7311 can anyone please tell me how to intrepret this line generated at the end of the ivprobit?