I have been working through some steps to prepare for a GEE analysis with inverse probability of treatment weight (iptwt)) and have been using some of Garrido's resources as a guide to generate propensity scores:
https://onlinelibrary.wiley.com/doi/...475-6773.12182
http://npcrc.org/files/NPCRC.Observa...p.10-20-14.pdf
I first ran the --pscore-- command [pscore treatment covariate1 covariate2 … covariate#, pscore(pc_pscore) blockid(pc_block) detail] to obtain estimated propensity scores ('mypscore'), estimate balancing properties, etc.
I then ran the command --qui dr [varlist], genvars-- to obtain 'iptwt'.
In looking at the additional variables that were generated by the --dr-- command, specifically the 'ptreat' variable, I noticed that the estimated propensity score values in 'mypscore' and 'ptreat' are slightly different. I'm guessing that each command estimates propensity scores slightly differently, hence the slightly different values, but I'm now faced with needing to choose between 'mypscore' (and generating ipt weights with my own syntax) and 'ptreat' (using the stata-generated 'iptwt').
Does anyone have insight as to how/why the propensity score values are different (i.e., how is each command calculating the propensity scores)? Additionally, any guidance on which variable might be the better choice (i.e., 'mypscore' or 'ptreat') would be appreciated.
Thanks
https://onlinelibrary.wiley.com/doi/...475-6773.12182
http://npcrc.org/files/NPCRC.Observa...p.10-20-14.pdf
I first ran the --pscore-- command [pscore treatment covariate1 covariate2 … covariate#, pscore(pc_pscore) blockid(pc_block) detail] to obtain estimated propensity scores ('mypscore'), estimate balancing properties, etc.
I then ran the command --qui dr [varlist], genvars-- to obtain 'iptwt'.
In looking at the additional variables that were generated by the --dr-- command, specifically the 'ptreat' variable, I noticed that the estimated propensity score values in 'mypscore' and 'ptreat' are slightly different. I'm guessing that each command estimates propensity scores slightly differently, hence the slightly different values, but I'm now faced with needing to choose between 'mypscore' (and generating ipt weights with my own syntax) and 'ptreat' (using the stata-generated 'iptwt').
Does anyone have insight as to how/why the propensity score values are different (i.e., how is each command calculating the propensity scores)? Additionally, any guidance on which variable might be the better choice (i.e., 'mypscore' or 'ptreat') would be appreciated.
Thanks
