Announcement

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

  • #16
    There is no contradiction between these results. The arithmetic mean and geometric mean are different things and they respond differently to the same changes in the data. Both are correct in terms of what they mean. In most contexts, when we talk about means, we usually refer to the arithmetic mean. Geometric means are not all that commonly used--although they can be valuable ways of looking at data that is very skew to the right because outliers don't pull the geometric mean up as much as they pull up the arithmetic mean. Since income distributions in some populations are very skew, this might be useful. It's up to you to decide which way of looking at central tendency is more appropriate to your data and work with that. Just be clear with yourself, and with anyone you show your results to, what you are doing.

    Comment


    • #17
      If confused about interpretation, you can always check with the margins command (if not use it for such calculations).
      Code:
      margins, at(x = (0.8 0.9))
      Set the x at whatever you like.

      When your DV is log, you can add
      Code:
      margins, at(x = (0.8 0.9)) expression(exp(predict(xb))*exp((`e(rmse)'^2)/2))
      to get the exponentiated prediction

      but with logDV and a X on the unit interval, the effect is exp(B*d)-1, where d is the percentage point change of interest.
      a citation:
      HTML Code:
      https://www.researchgate.net/publication/226044301_Interpreting_semilogarithmic_regression_coefficients_in_labor_research


      Comment


      • #18
        Dear Clyde and George,
        Thank you very much for your help! It is now very much clear to me.

        Regards,
        Gunjan

        Comment


        • #19
          I have a similar question. I am running an OLS model exploring the relationship between the reported likelihood of a student going to university (independent variable ranging from 0%-100%) and the log expected earnings at age 30 (dependent variable). The coefficient for the reported likelihood of a student going to university is 0.002 and i have interpreted as follows: a percentage point increase in the reported likelihood of a student to pursue tertiary education is associated with a 0.002% increase in earnings expectations. Is this correct or the coefficient should by multiplied by 100, thus becoming 0.2%. Many thanks in advance for any contribution.

          Comment


          • #20
            A unit change in the variable representing likelihood of going to a university (i.e. a 1 percentage point difference) corresponds to an increase of 0.002 in log expected earnings. If log expected earnings increases by 0.002, that corresponds to expected earnings being multiplied by exp(0.002) = 1.002 to more than 3 decimal places. So this corresponds to a 0.2% increase in expected earnings..

            Comment


            • #21
              Thank you very much Clyde!

              Comment

              Working...
              X