My coauthor is using a panel survey to estimate the effect of obtaining a second degree. Individuals who start a second degree are in the treated group. The control group contains individuals who did not start a degree, but these individuals are duplicated for every year that they are in the survey i.e. person 1 is in the survey from 2003-2006, they did not start a second degree, so they appear in the control group 4 times. As such, the sample contains duplicate records of individuals.
My coauthor uses Python to run machine learning commands, and then bootstraps to obtain Individual Treatment Effects. I'd like to note here that the bootstrapping is done on the whole sample, meaning that individuals who have been duplicated can be selected more than once into the bootstrap sample. As such, we have not accounted for clustering during this step.
How is this relevant for Statalist? Well, after receiving the extract of person id's and Individual Treatment Effects, I import these into Stata so that I can compute the mean (Average Treatment Effect), standard errors, confidence intervals, create graphs etc. Since we have not adjusted for clustering/duplicated individuals at any point before this, is it possible to adjust for clustering AFTER bootstrapping?
I have seen clustering accounted for DURING bootstrapping in Stata by applying the cluster option to the bootstrap command. But since the bootstrapping has occurred outside of Stata, is it possible to weight the Individual Treatment Effects by a duplication factor before calculating the standard errors?
Thank you,
Tessa.
My coauthor uses Python to run machine learning commands, and then bootstraps to obtain Individual Treatment Effects. I'd like to note here that the bootstrapping is done on the whole sample, meaning that individuals who have been duplicated can be selected more than once into the bootstrap sample. As such, we have not accounted for clustering during this step.
How is this relevant for Statalist? Well, after receiving the extract of person id's and Individual Treatment Effects, I import these into Stata so that I can compute the mean (Average Treatment Effect), standard errors, confidence intervals, create graphs etc. Since we have not adjusted for clustering/duplicated individuals at any point before this, is it possible to adjust for clustering AFTER bootstrapping?
I have seen clustering accounted for DURING bootstrapping in Stata by applying the cluster option to the bootstrap command. But since the bootstrapping has occurred outside of Stata, is it possible to weight the Individual Treatment Effects by a duplication factor before calculating the standard errors?
Thank you,
Tessa.
