I would like to estimate the impact of medium and heavy flooding (compared to noflood) on post-flood household income using a linear regression framework. Noflood, medium and heavy are categorical variables. Since pre-flood characteristics x1 and x2 are not balanced between the flood groups, I would like to use x1 x2 as matching variables and then weight observations inversely proportional to their propensity score, pscore1. So the code will be:
Code:
reg income medium heavy x1 x2 [pweight=pscore1]
1) Can someone please tell me how I can calculate the weight pscore1? My understanding is that i first estimate the propensity scores using a multinomial logistic regression. But it is unclear to me how one should proceed from there.
2) Is the above code equivalent to
Code:
teffects ipwra (income x1 x2) (flood x1 x2)
where flood=1 if noflood==1, flood=2 if medium==1 and flood=3 if heavy==1? If so, how can I retrieve pscore1?
Many thanks!
0 Response to Propensity scores for multi-valued treatment
Post a Comment