Announcement

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

  • Interpretation of -xtlogit- regression

    Hi everybody, My regression is as follows:

    optimismt = Beta0 + Beta1 * optimismt-1 + Beta2 * Size + Beta3 * Tobin's Q...

    optimismt and optimismt-1 are both dummy variables being 1 if the person is optimistic in this period or was optimistic in the period before, respectively.
    Regressing without fixed effects I get Beta1= 0.9
    Regressing with fixed effects I get Beta1= 0.7

    How can I interpret these results? And how can I interpret the coefficients for Size and Tobin's Q?

    Thank you in advance.

    Best regards,
    Jens

  • #2
    Jens:
    as per FAQ, the chance of getting helpful replies would increase if you post exactly what you typed and what Stata gave you back (including output tables, if appropriate).
    Just click on the #-icon in the Advanced editor (A-icon) and put Stata codes and results in between delimiters.

    Kind regards,
    Carlo
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      I am always kind of to suspicious of these types of models. Forget for the moment that this is a non-linear model, and write the basic model as

      \[
      y = \gamma*y_{t-1} + \beta*X + \epsilon
      \]

      Now the assumption for unbiased estimation of the parameters of the model require \epsilon to be uncorrelated with the predictors, right?. However, as per model definition, \epsilon is correlated with \y - in each and every time-period. Given this \epsilon is - also per model definition - correlated with \y_{t-1}. Therefore the estimates are biased and thus interpretation gets pretty shaky.

      Best
      Daniel

      Comment


      • #4
        Thank you a lot for your answer.
        But if I want to interpret coefficient Beta1, how would I do this in my logistic regression?

        Best regards,
        Jens
        β

        Comment


        • #5
          Short answer: you may interpret your estimate of beta_1 as biased! Note well Daniel's points. You are trying to fit, it appears, a dynamic binary dependent variable. You don't appear to have dealt with the well-known "initial conditions" problem. Do a websearch on "dynamic random effects probit" model, and read up the extensive literature about these sorts of models. Other relevant issues will include the length of your panel data series, and the number of individuals followed over time.

          Comment


          • #6
            I understand. Thank you.
            Could you please help me with how my specific model would be changed and which stata commands I have to apply in order to implement the dynamic random effects probit model?

            Thank you.

            Best regards,
            Yannick

            Comment

            Working...
            X