I'm trying to do a PSM, Full matching, in complex survey data. I first estimated the propensity score matching using the MatchIt package of R and diagnosed the covariate balance.
But, a problem occurred when trying to use svyglm in survey package to estimate the treatment effect. In order to use svyglm, the complex survey must be designed through svydesign in survey package. I wonder what weight is used at this time.
After full matching, weight exists in two ways, Survey Weight and Matching Weight.
In the figure above, the wt_itvex is the survey weight, and the weights is the matching weight generated after matching.
design_org <- svydesign(id=~psu,strata=~kstrata,weights=~NULL,data=full_att1_data)
fit_boot <- svyglm(formula, data = md_boot,
family=quasibinomial(link="logit"),
design=design_org)
How do I specify the weight of svydesign to apply both weights? (part of "weight=~NULL")
I first thought of the product of the two weights as the code below.
design_org <- svydesign(id=~psu,strata=~kstrata,weights=~wt_itvex * weights,data=full_att1_data)