Announcement

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

  • Gologit2 output presentation and interpretation

    Hi,

    I've been seeing different ways to present the gologit2 output such as default output coeff, mfx2 results, predicted probabilities, and use of gammas based on the R. Williams, Long and Freese, etc.. However, may I ask the following question: when and why to use (or choose) one type of output over the others.

    Also, I'm not sure whether I correctly understood the interpretation of the default gologit2 output. Is it, "even the output says it's a "coefficient", should it be interpreted like the output of a logit? Hence, getting the odds ratio? and read it as the log of the odds of being in the 1st category versus the others? and so on?"

    Would be grateful for your response to the questions.

  • #2
    This article from the 2016 Journal of Mathematical Sociology addresses most or all of your Qs.

    http://www3.nd.edu/~rwilliam/gologit...ologit2016.pdf

    Also possibly useful is a handout entitled "Adjusted Predictions & Marginal Effects for Multiple Outcome Models & Commands (including ologit, mlogit, oglm, & gologit2) " You can find it at

    http://www3.nd.edu/~rwilliam/stats3/Margins05.pdf

    If still unclear you can email me or post followup queries.
    -------------------------------------------
    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
      Thank you, Richard. The articles have been very helpful. The second article particularly made the interpretations clear for me.

      However, i'm having a bit of difficulty getting margin all the way to the end (I'm using stata 12). I'm attaching a sample of what shows up and the sumstat for the particular variable. Please note that all my explanatory variables are continuous. I would suppose interpretations would have been more specific and direct had my explanatory been categorical like the ones in your example. Nevertheless, I'm excited to see the margins (direction and magnitude) across different cumulative logit models.

      Btw, for your reference, my dependent variable is poverty decile (1st decile being the least poor, 10th decile the poorest). RAI_local, one of the predictors, is an accessibility index, the proportion of the population falling within 2km of a local road.

      Click image for larger version

Name:	Screen Shot 2017-05-01 at 8.47.37 AM.png
Views:	1
Size:	46.2 KB
ID:	1386253


      Thanks.

      Comment


      • #4
        I think I found the margins for continuous variables.

        It's also by you: http://www3.nd.edu/~rwilliam/stats3/Margins03.pdf

        However, this shows up upon trying the mcp.

        Click image for larger version

Name:	Screen Shot 2017-05-01 at 4.52.16 PM.png
Views:	1
Size:	13.9 KB
ID:	1386285


        i'm using gologit2, is mcp also applicable to gologit2?

        Thanks.

        Comment


        • #5
          It is good to show your commands, including those that precede margins. mcp isn't as useful for multiple-outcome commands like ologit or gologit2. I would suggest paying attention to the margins05 handout I linked to above. It'll be especially good if you are condemned to using Stata 12, which lacks some of the very nice features that were added by 14.

          You may want to look through all the handouts on adjusted predictions and marginal effects that are at

          http://www3.nd.edu/~rwilliam/xsoc73994/index.html
          -------------------------------------------
          Richard Williams, Notre Dame Dept of Sociology
          Stata Version: 17.0 MP (2 processor)

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

          Comment


          • #6
            I finally managed to have access to one with a version 14. Thought of using quantiles instead of the integer form of the independent variables as the error message earlier indicated.

            ollowing the handout, i'm getting this:

            Click image for larger version

Name:	Screen Shot 2017-05-01 at 8.49.19 PM.png
Views:	1
Size:	27.9 KB
ID:	1386405



            Thanks.

            Comment


            • #7
              If you didn't get that error you would get a different one. Only factor variables are allowed to the left of the comma in margins, e.g. you would have needed to specify i.rai_local_quart. If it only has 4 values that might be fine.

              You may want to do something like margins, dydx(*). Or compute adjusted predictions where you plug in different values for variables. Again, those are described in the margins05 handout.

              Make sure you have the most current versions of everything. Stata is now at version 14.2. gologit2 is on version 3.1.1:

              Code:
              . which gologit2
              c:\ado\plus\g\gologit2.ado
              *! version 3.1.1 18oct2016  Richard Williams, [email protected]
              -------------------------------------------
              Richard Williams, Notre Dame Dept of Sociology
              Stata Version: 17.0 MP (2 processor)

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

              Comment


              • #8
                Here is an example:

                Code:
                webuse nhanes2f, clear
                gologit2 health weight height, auto
                margins, dydx(*)
                The last part of the output is

                Code:
                . margins, dydx(*)
                
                Average marginal effects                        Number of obs     =     10,335
                Model VCE    : OIM
                
                dy/dx w.r.t. : weight height
                1._predict   : Pr(health==1), predict(pr outcome(1))
                2._predict   : Pr(health==2), predict(pr outcome(2))
                3._predict   : Pr(health==3), predict(pr outcome(3))
                4._predict   : Pr(health==4), predict(pr outcome(4))
                5._predict   : Pr(health==5), predict(pr outcome(5))
                
                ------------------------------------------------------------------------------
                             |            Delta-method
                             |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
                -------------+----------------------------------------------------------------
                weight       |
                    _predict |
                          1  |    .001162   .0000938    12.39   0.000     .0009782    .0013459
                          2  |   .0019459   .0001438    13.53   0.000     .0016641    .0022278
                          3  |   .0012075    .000095    12.71   0.000     .0010213    .0013937
                          4  |  -.0011963   .0000909   -13.17   0.000    -.0013744   -.0010182
                          5  |  -.0031192   .0002309   -13.51   0.000    -.0035717   -.0026667
                -------------+----------------------------------------------------------------
                height       |
                    _predict |
                          1  |  -.0018732   .0002727    -6.87   0.000    -.0024077   -.0013387
                          2  |  -.0051043   .0003859   -13.23   0.000    -.0058606   -.0043479
                          3  |  -.0036269   .0004502    -8.06   0.000    -.0045093   -.0027445
                          4  |   .0027937   .0004332     6.45   0.000     .0019446    .0036428
                          5  |   .0078107   .0004583    17.04   0.000     .0069124    .0087089
                ------------------------------------------------------------------------------
                If you have installed spost13, somewhat easier to read is

                Code:
                . mtable, dydx(*) dec(5)
                
                Expression: Marginal effect of Pr(health), predict(outcome())
                
                           |     poor      fair   average      good  excellent
                 ----------+--------------------------------------------------
                    weight |  0.00116   0.00195   0.00121  -0.00120   -0.00312
                    height | -0.00187  -0.00510  -0.00363   0.00279    0.00781
                If you can't reproduce these results your software may not be up to date.
                -------------------------------------------
                Richard Williams, Notre Dame Dept of Sociology
                Stata Version: 17.0 MP (2 processor)

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

                Comment

                Working...
                X