Announcement

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

  • Tobit with missing values as censored observations?

    Dear Statalisters

    I am conducting analysis on the age at which a person has their first child. For anyone who does not have any children, this variable is coded as missing.

    Does this warrant a Tobit model? On the basis that the variable is censored for individuals who do not have any children.

    This might involve replacing the missing data with zero and running the following model:

    Code:
    replace ageat1st=0 if mi(ageat1st)
    tobit ageat1st X Y Z, ll(0)
    I have a feeling this isn't appropriate - if so, any ideas what is a viable alternative model?

  • #2
    Julia:
    welcome to this forum.
    I would go -regress- with an -if- qualifier, restricting the sample to those persone who have at least one children.
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment


    • #3
      Thank you Carlo. I thought that might be the case!

      Comment


      • #4
        While I suspect Carlo's suggestion makes most sense, there is a separate issue on selection.

        That is, you could think of this as duration data and some observations (those who do not have a child) do not get an event in the observed times. The survival analysis literature worries about this a lot. It depends a bit on the point of the research whether restricting the sample to those who have a child makes sense (i.e., you generalize to the population that have at least one child) or if you're trying to generalize to the population as a whole which includes folks who do not have children.

        Comment


        • #5
          I do share Phil's point.
          This kind of research can be made pretty feasible or terribly difficult conditional on enrollment criteria (that is, restricting the sample to those women who had at least one child or not).
          I still think that an -if- qualifier can be the right tool in Julia's case.
          That said, things might be different if those women who do not had children tried to get pregnant unsuccessfully.
          Kind regards,
          Carlo
          (Stata 19.0)

          Comment


          • #6
            Julia hasn't said anything about the study design. For a cross-sectional design (at least), I'd suggest a current status analysis via stintreg or stpm (SSC).
            Last edited by Steve Samuels; 27 Jul 2018, 12:09.
            Steve Samuels
            Statistical Consulting
            [email protected]

            Stata 14.2

            Comment

            Working...
            X