How inverse probability weighting （IPW）is applied into regression analysis to deal with attrition?

[Jianhong Xu]

Dear all,

I am just wondering that after the weight is calculated by Stata however the weight will be applied to the observations? Will the dependent and independent variables just multiply the weight to generate new value and stata use these new values to run regression?

Thanks!

[Andrew Musau]

The description of analytical weights is provided at https://www.stata.com/support/faqs/s...ar-regression/.

[Jeff Wooldridge]

Typically one uses the pweight option, which appropriately forces calculation of robust stand errors. For the OLS obective function, the sum of squared residuals is weighted by 1/phat, where phat is the estimated probability of remaining in the sample. Mechanically, you can divide all variables, including the intercept, by sqrt(phat). I don’t know one would want to. You can apply the pweight option to lots of estimators.

[Jianhong Xu]

Typically one uses the pweight option, which appropriately forces calculation of robust stand errors. For the OLS obective function, the sum of squared residuals is weighted by 1/phat, where phat is the estimated probability of remaining in the sample. Mechanically, you can divide all variables, including the intercept, by sqrt(phat). I don’t know one would want to. You can apply the pweight option to lots of estimators.

Hi Prof. Wooldridge

Thanks for your reply! I am just wondering that why we need to divide by sqrt (phat) rather than phat directly?

Best,
Jianhong

[Jianhong Xu]

The description of analytical weights is provided at https://www.stata.com/support/faqs/s...ar-regression/.

Thank you for providing this link!