Announcement

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

  • Hausman-Taylor estimator for binary dependent variable

    Hi all,

    I am a business student who has had some introductory econometirc courses and has some basic knowledge of Stata.

    In one of our lectures we briefly talked about the Hausman-Taylor estimator (i.e. the xthtaylor command) as the model to be used when one wants to get a parameter estimate for a time-inavriant variable (e.g. an individual's sex) which would not be possible with fixed effects.

    Unfortunately, our lecturer could not tell me how to implement the Hausman-Taylor estimator when the dependent variable is binary (i.e. when one would use xtlogit, xtprobit or xtcloglog, unless one was interested in getting an estimate for a time-invariant variable like sex). Is there a specific command in Stata for that case or do I have to write it on my own?

    Thank you very much in advance,
    Nina

  • #2
    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.

    I don't know about HT, but xthybrid does almost the same thing and does handle binary dv's.

    Comment


    • #3
      Originally posted by Phil Bromiley View Post
      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.
      I don't have a specific dataset/problem. My question is more of a general nature.

      Comment


      • #4
        In case you need a specific example I try to explain my question/problem using one of the examples given in the Stata documentation:
        Code:
        use http://www.stata-press.com/data/r15/psidextract
        Let's assume that I was interested in the role that sex (the dummy variable "fem" takes a value of 1 if the individual is female) plays in explaining whether an individual is a member of a union or not. As my dependent variable is binary and I further assume that there are some unobserved individual-specific time-invariant effects, I usually would use a regression like this:
        Code:
        xtlogit union occ south smsa ind exp exp2 wks ms lwage fem blk ed, fe
        However, as individuals in the dataset do not change their sex over time, no paremeter estimate for "fem" is obtained. Since the key interest of my analysis is to determine the role of an individual's sex, however, I've learned that the Hausman-Taylor estimator is the one to be used. It could be specified the following way:
        Code:
        xthtaylor union occ south smsa ind exp exp2 wks ms lwage fem blk ed,endog(exp exp2 wks ms lwage ed)
        As far as I understood from our lecture (and as one can see when performing the following commands):
        Code:
        predict unionhat, xb
        sum unionhat
        The xthtaylor-command predicts negative values for some observations, i.e. neglects the binary structure of my dependent variable.

        Does anyone know how I can "solve" this issue by using something like a Hausman-Taylor estimate for Logit/Probit-models?

        Many thanks for your help.
        Last edited by Nina Bauer; 19 Apr 2019, 11:51. Reason: typo

        Comment


        • #5
          Hi Nina,

          I am also working on exactly same problem. Have you already gotten any clues or solutions? Many thanks.


          Yanjun

          Comment


          • #6
            Originally posted by Yanjun Ren View Post
            Hi Nina,

            I am also working on exactly same problem. Have you already gotten any clues or solutions? Many thanks.


            Yanjun
            Unfortunately, I have not gotten any clues nor solutions yet.
            My hope was that someone in this forum can help me or maybe has some idea/literature at least.

            Yanjun Ren do you have found any clues or maybe literature on it?

            Comment


            • #7
              Is there really nobody in this forum who has at least some hint/idea how to tackle this problem?

              Comment

              Working...
              X