Announcement

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

  • doubleb for willingness-to-accept

    Hello,

    I am trying to use the doubleb command to estimate a respondent's willingness to accept a certain intervention.

    However, I am receiving the following error code:

    There is an inconsistency in at least one of your observations.
    Check for situations where the response to the first question is yes but the second bid is lo
    > wer than the first
    or for situations where the response to the first question is no but the second bid is higher
    > than the first.
    After solving this issue try the command again.
    r(498);

    The first question (answer1) asks how much the respondent thinks OTHER PEOPLE are WTA for the intervention while the second question (answer2) askes how much the RESPONDENT is WTA for the intervention. The amount in the second question increases or decreases based on the yes/no response to the first question. So it is possible and not an error for the response to the first question to be no and the second bid to be higher than the first (for example).

    How can I use a double-bounded model given this structure of the two survey questions?

    Is there another STATA package that is more appropriate?

  • #2
    -doubleb- assumes that your latent dependent variable is univariate. From your description, I don’t think you are placing two bounds on a univariate latent WTA: you are working with a system of two WTA equations. My guess is that you should apply some sort of bivariate or censored probit model rather than a double bounded model. Could you elaborate on how exactly the first question relates to the second question in your experiment?

    Comment


    • #3
      Hi Hong,

      Thank you for your response. If the answer to the first question is "Yes" (e.g., the survey respondent thinks that other people would be WTA $XX for the intervention), then the amount given in the second question (e.g., how much the respondent themselves would be WTA for the intervention) is increased by an increment of lets say $5. If the answer to the first question is "No", then bid2 = [bid1 - $5].

      I think that your recommendation to use a bivariate probit model may be correct. I have been able to get a WTA estimate for a double-bounded model with the following code (which currently excludes covariates):

      biprobit (answer1=bid1) (answer2=bid2), robust
      wtpcikr bid2, reps(100000) eq(answer2)

      However, any suggestions for how to strengthen this analysis is appreciated.

      Comment


      • #4
        Thanks. I think you may find it useful to look at a classic paper by Cameron and Quiggin [1994]: https://doi.org/10.1006/jeem.1994.1035. I think it makes sense for you to apply the bivariate probit approach, for similar reasons as what they discuss in the context of their application. Indeed, you may even have a stronger case, considering that your first and second stage questions are inherently referring to two different latent variables (one for the decision maker's guess about other people's WTA and the other for the decision maker's own WTA).

        Comment


        • #5
          Agreed. Is there a publication that shows how, after running biprobit, the two individual WTA estimates from the two questions can be combined into a single WTA estimate with 95% CrIs?

          What is your recommendation for obtaining a single WTA estimate from the two questions? This is a benefit of the doublb command.

          Comment


          • #6
            I don't think the double-bounded model is applicable to your data, unless there's a theory which says that the decision maker's guess about someone else's WTA (wta1) and the decision maker's own WTA (wta2) are one and the same thing (wta1 = wta2). Your setup is quite different from double-bounded CVs, where one is asking information about the same type of WTA twice: For example, by including two questions pertaining to wta1, or alternatively by including two questions pertaining to wta2. What you are working with is a system of two singled-bounded models, and that's quite a different thing from a double bounded model.

            Comment

            Working...
            X