Announcement

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

  • Standardized coefficients with clustered data

    Hello,

    Im running an OLS regression, where I regress years of education on the sex ratio in china, using the following model:
    Where Y_ipt represents the outcome variable, completed years of education of respondent i, from cohort t, in province p and county c. Male is a dummy = 1 for male, of individual i. Sex_Ratio_tp is the sex ratio of cohort t in province p. Male_i * Sex_Ratio_tp is the effect of the sex ratio on males in from cohort t in province p. X_itpc represents our set of demographic and socio-economic characteristics at the individual, cohort, province and county level. Further, I add a control for province fixed effects. Standard errors are clustered at the province and age level to control for serial correlations within the groups.

    Anyways, in order to interpret the economic significance of my model, i want to standardize my beta coefficients. When i run the beta command in Stata it reports "options vce(cluster clustvar) and beta may not be combined".

    Is there any way to standardize my coefficients when clustering my standard errors?
    ​​​​​​
    Best,
    Anna

  • #2
    Perhaps one option could be to do the standardization on your own.
    1) define your sample
    2) create new variables
    3) run the model with new variables and clustered standard errors
    The only question there is if, theoretically, Standard errors should be estimated using clusters, or if it is fine to use simple Standard errors. Perhaps, for that, you can see what other people (papers) have done in those cases.
    F

    Comment


    • #3
      I see, thank you for your reply! How is it possible to compute the standardized variables by myself? Dont i need to divide the beta coefficient with the std deviation in order to to that?

      Comment


      • #4
        Nop.
        Beta coefficients are nothing else than the estimation of coefficients after the variables in the model have been transformed.
        see this example
        Code:
        sysuse auto, clear
        foreach i in price mpg trunk {
            qui:sum `i'
            gen double sd_`i'=(`i'-r(mean))/r(sd)
        }
        
        reg price mpg trunk, beta
        reg sd_price sd_mpg sd_trunk
        HTH

        Comment


        • #5
          Hi again, that works - thank you! One final question, when i run the regression, stata will not control for fixed effects: "sd_provcd_birth: factor variables may not contain noninteger values". Do you know why that is the case?
          Best, Anna

          Comment

          Working...
          X