Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Wald and LR test after probit

    Hi,

    I noticed that stata sometimes reports Wald test, sometimes it reports likelihood-ratio test after running probit. I use the same command but different subsamples (see below). How can I force stata to rePort Wald instead of LR?

    N 4821 2067 1722 306 737 527 725
    pseudo R-sq 0.136 0.159 0.188 0.247 0.262 0.337 0.142
    AIC 363.892 215.735 144.143 64.480 71.449 74.992 123.830
    BIC 493.507 311.511 220.460 101.716 117.474 117.664 183.450
    chi2 99.511 141.603 67.843 14.583 34.736 54.798 68.610
    chi2type Wald Wald Wald LR Wald Wald Wald
    p 0.000 0.000 0.000 0.103 0.000 0.000 0.000
    p_corr 99.357 99.081 99.303 98.039 99.186 98.482 98.483


  • #2
    Your output is very hard to read. I can't tell what commands you gave or what it is you are showing. Code tags might help; see pt 12 in the FAQ.

    But one way to get a Wald test is to use test or testparm after running the estimation command, e.g.

    Code:
    webuse nhanes2f, clear
    probit diabetes i.female age
    testparm i.female age
    -------------------------------------------
    Richard Williams, Notre Dame Dept of Sociology
    Stata Version: 17.0 MP (2 processor)

    EMAIL: [email protected]
    WWW: https://www3.nd.edu/~rwilliam

    Comment


    • #3
      Unfortunately, you didn't show the commands for the models.

      That said, with the option vce(robust) Stata gives Wald, whereas without vce(robust) it gives LR.

      Code:
      . sysuse auto
      (1978 Automobile Data)
      
      .  probit foreign weight mpg, vce(robust) nolog vsquish
      
      Probit regression                               Number of obs     =         74
                                                      Wald chi2(2)      =      30.26
                                                      Prob > chi2       =     0.0000
      Log pseudolikelihood = -26.844189               Pseudo R2         =     0.4039
      
      ------------------------------------------------------------------------------
                   |               Robust
           foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
      -------------+----------------------------------------------------------------
            weight |  -.0023355   .0004934    -4.73   0.000    -.0033025   -.0013686
               mpg |  -.1039503   .0593548    -1.75   0.080    -.2202836    .0123829
             _cons |   8.275464   2.539177     3.26   0.001     3.298769    13.25216
      ------------------------------------------------------------------------------
      
      .  probit foreign weight mpg, nolog vsquish
      
      Probit regression                               Number of obs     =         74
                                                      LR chi2(2)        =      36.38
                                                      Prob > chi2       =     0.0000
      Log likelihood = -26.844189                     Pseudo R2         =     0.4039
      
      ------------------------------------------------------------------------------
           foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
      -------------+----------------------------------------------------------------
            weight |  -.0023355   .0005661    -4.13   0.000     -.003445   -.0012261
               mpg |  -.1039503   .0515689    -2.02   0.044    -.2050235   -.0028772
             _cons |   8.275464   2.554142     3.24   0.001     3.269437    13.28149
      ------------------------------------------------------------------------------
      
      .
      Best regards,

      Marcos

      Comment


      • #4
        Marcos is right, but I'll note the Wald Chi-Square with his approach is different than the Wald Chi-Square with my approach. His Wald is 30.26.

        Code:
        sysuse auto, clear
        probit foreign weight mpg, nolog vsquish
        testparm weight mpg
        
        . testparm weight mpg
        
         ( 1)  [foreign]weight = 0
         ( 2)  [foreign]mpg = 0
        
                   chi2(  2) =   20.75
                 Prob > chi2 =    0.0000
        So if you think vce(robust) is appropriate/necessary do it like Marcos says but if not then use my approach,
        -------------------------------------------
        Richard Williams, Notre Dame Dept of Sociology
        Stata Version: 17.0 MP (2 processor)

        EMAIL: [email protected]
        WWW: https://www3.nd.edu/~rwilliam

        Comment


        • #5
          After reading Richard's insightful advice and looking more carefully at the output in #1, it seems the chi-square differ much among the models, no matter it was Wald's or LR. In short, they must be quite different models.

          A second aspect I wish to comment before ending this post: Stata only presents one chi2type as the ereturn. There seems to be no way to select one of them. So, I gather we cannot get, say, Wald, having just LR as the ereturn, and vice versa. Therefore, I fear the matter relates mostly with the vce, and I speculate it has something to do with the assumptions (IEE, for example): more restrictive under LR (and Stata goes for it in the standard model). However, Stata "decides" automatically to estimate under Wald whenever we select vce(robust) or include a cluster variable..
          Best regards,

          Marcos

          Comment


          • #6
            Thank you very much. vce(robust) is the reason...

            Comment


            • #7
              Thank you for informing the original command. I was really curious about it. That helps to clarify the point in #4 and in #5: vce(robust) and use of clustering will prompt Stata to provide the Wald chi2type.
              Best regards,

              Marcos

              Comment

              Working...
              X