Announcement

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

  • Help with interpreting Probit estimates with categorical variables

    Hi,

    So firstly, I just registered right now and I need immediate help. So, I will be really grateful if someone could guide me,

    Now, I'm doing a small university project and my research is basically seeing if child marriage affects higher education etc.

    I was first using a LPM model with the independent variable (education) and the main explanatory variable (early marriage) as binary. Now, since it's a survey data there is a wealth index provided which is a categorical variable and a wealth score. I'm for now using the wealth score as another explanatory variable but I also want to see the effect of the wealth index as a categorical variable.

    Since, LPM is very basic, I decided to do a Probit estimation as well. Now, when I run the probit regression. To get the marginal effects I know we can use the "mfx" command. But when I incorporate the wealth index which is a categorical variable, the "mfx" command no longer works. I'm just controlling for the categorical variable and my main point of interest is the early marriage binary variable. How do I estimate and interpret the coefficient of the binary variable now since the mfx command does not work?

    the commands I run:
    probit educ i.early_mar mar child i.windex5, robust
    mfx

    And I get this error: default predict() is unsuitable for marginal-effect calculation

    In my model: educ =1 if a specific number of years of education have been completed or higher, early_mar=1 if married below 18 years of age, mar=1 if married, child=1 if children born, windex5 is the wealth index quintile from lowest to highest

    I'll be really looking forward to a bit of help in this.

    Thank you!

    Regards,
    Saad


  • #2
    Could you just hard code dummies for different groups of windex5?

    Comment


    • #3
      ie
      Code:
      tabulate windex5, generate(groupname)

      Comment


      • #4
        mfx is obsolete. You should be using margins in combination with factor variable notation (help fvvarlist). I would not recommend hard coding.

        Comment


        • #5
          Originally posted by Felix Holub View Post
          Could you just hard code dummies for different groups of windex5?
          I needed to add this that I've got very very basic Stata knowledge since this is our introductory econometrics course. We haven't been even taught logit or probit models but I just thought it would be good to add them.

          So, you're saying that I should just create dummies for each of the group and then run the model? Like ( lowest = 1 if windex5 is lowest and 0 for the rest and so on?) I didn't think about that at all so I will try that! Thank you!

          Also, I used the margins, dydx(*) command as well. I wanted to know if it's going to give me the same estimates as the mfx command.

          Comment


          • #6
            Sorry for double posting.

            Originally posted by Stephen Jenkins View Post
            mfx is obsolete. You should be using margins in combination with factor variable notation (help fvvarlist). I would not recommend hard coding.
            So, the margins, dydx(*) should give me the estimates I want right?

            Comment


            • #7
              Unless you are using an ancient version of Stata, do not use mfx. Use margins instead, e.g.

              margins, dydx(*)

              For more, see

              https://www3.nd.edu/~rwilliam/stats/Margins01.pdf
              -------------------------------------------
              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
                Originally posted by Saad Khan View Post

                I needed to add this that I've got very very basic Stata knowledge since this is our introductory econometrics course. We haven't been even taught logit or probit models but I just thought it would be good to add them.

                So, you're saying that I should just create dummies for each of the group and then run the model? Like ( lowest = 1 if windex5 is lowest and 0 for the rest and so on?) I didn't think about that at all so I will try that! Thank you!

                Also, I used the margins, dydx(*) command as well. I wanted to know if it's going to give me the same estimates as the mfx command.
                Exactly! This is actually what the (recommended) "i." notation does automatically for you.

                Comment


                • #9
                  Hi,
                  I am using Probit estimation for my model, where my dep var is polychotomous and indep variables include docio-demographics, likert data and factor variables (like Income). However, as soon as I include the 'Income' var with the prefix i. , Stata generates endless iterations , stating "No concavity". I have tried generating dummies as well, using the 'Tabulate' command but to no avail. Can someone kindly guide me on this matter ?
                  My second problem concerns the interpretation of coefficients. For variables other than dummies, I use the formula Pr(y=i)=(e^coeff)/(1+e^coeff) and for dummies I use the 'margins' command right after the Probit estimation. Is this approach correct?

                  Comment


                  • #10
                    Originally posted by Ayesha Wajih View Post
                    Hi,
                    I am using Probit estimation for my model, where my dep var is polychotomous and indep variables include docio-demographics, likert data and factor variables (like Income). However, as soon as I include the 'Income' var with the prefix i. , Stata generates endless iterations , stating "No concavity". I have tried generating dummies as well, using the 'Tabulate' command but to no avail. Can someone kindly guide me on this matter ?
                    My second problem concerns the interpretation of coefficients. For variables other than dummies, I use the formula Pr(y=i)=(e^coeff)/(1+e^coeff) and for dummies I use the 'margins' command right after the Probit estimation. Is this approach correct?
                    Hey Ayesha,

                    If I am correct, margins is good enough for variables other than dummies as well. Although, I think for continuous variables you can use the margins coefficients for small changes. I could be wrong in this, but that has been my approach.

                    Moreover, I recently faced the iterations getting backed up issue as well. Maybe it has to do with some variables having very little variation which was what happened in my case. What I did was run a simple OLS regression instead of probit and checked the standard errors. Removing the variables with very high standard errors fixed the issue for me. Maybe you could try that and see what variables might be causing the issue. Alternatively, you can also use the iterate(#) option but just so you know the estimates are not correct.

                    Hope this might help you!

                    Comment

                    Working...
                    X