Announcement

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

  • Fixed-effects with a bounded dependent variable

    Hello all,

    We are looking for a command that (1) can handle a dependent variable whose value is bounded between 0 and 1 and (2) includes firm fixed-effects.

    We are using Stata 13 for Windows.

    Commands such as “xtivreg2” have the option to include firm fixed-effects. Since our dependent variable is bounded between 0 and 1, however, OLS is inappropriate.

    The command “glm” can handle dependent variables whose value is bounded between 0 and 1. We have used “glm” using the following syntax:

    glm `dv’ `iv’, link(logit) family(binomial) robust nolog

    `dv’ denotes the dependent variable, and `iv’ denotes the independent variables.

    We tried manually creating firm fixed-effects by de-meaning each observation, where the mean is based only a firm’s own observations

    But this approach does not work with “glm.” For our dependent variable, suppose a firm has two observations, and its values are 0.80 and 0.90, so that its mean is 0.85: The values for the fixed-effects observations would be 0.80 – 0.85 = -0.05 and 0.90 – 0.85 = 0.05, respectively.

    “glm” works only if all of the values for the dependent variable lie between 0 and 1. Hence, when we use GLM with our manually created fixed-effects, we get an error message saying that the “dependent variable . . . has negative values.”

    In principle, with “glm,” we could include an indicator variable for each firm, but we have ~2,000 firms. Hence, we would rather use another option.

    If it matters, some of the observations have a value of 1, but none have a value of 0.

    Any help would be very much appreciated.

    Best regards,

    Jeff




  • #2
    Have you looked at the following article, which is the main reference on panel data methods for fractional data that I am aware of?

    "Panel data methods for fractional response variables with an application to test pass rates", by Leslie E. Papke, Jeffrey M. Wooldridge, in Journal of Econometrics 145 (2008). 121–133.

    See their discussion on p. 123 (bottom of right hand column) about why they do not use a fixed-effects approach. On Leslie Papke's home page at http://www.econ.msu.edu/faculty/papke/, there is link to "Papke and Wooldridge (2008) Stata Files"

    Comment


    • #3
      Originally posted by Stephen Jenkins View Post
      Have you looked at the following article, which is the main reference on panel data methods for fractional data that I am aware of?

      "Panel data methods for fractional response variables with an application to test pass rates", by Leslie E. Papke, Jeffrey M. Wooldridge, in Journal of Econometrics 145 (2008). 121–133.

      See their discussion on p. 123 (bottom of right hand column) about why they do not use a fixed-effects approach. On Leslie Papke's home page at http://www.econ.msu.edu/faculty/papke/, there is link to "Papke and Wooldridge (2008) Stata Files"

      Thank you very much for the response.

      I have not seen those references before. I'll look into them to see what options we have.

      Best regards,

      Jeff

      Comment

      Working...
      X