How to deal with extreme propensity scores after IPTW

Epi_Stats — Sun, 11 Feb 2024 21:13:58 GMT

Hi,

I'm trying to run a propensity score analysis using proc psmatch and IPTW stabilized weights.

When I do this, there are a handful of observations with weighted propensity scores >1.

I can't share my full dataset, but here is a brief example of the matched IPTW data I have with 2 key confounders, the propensity scores (_PS_) and weighted propensity scores (PS_Weight)

data Matched_IPTW;
input ID AGE TREAT CFD1 CFD2 _PS_ PS_WEIGHT;
DATALINES;
1 83 1 1 2 0.95419 0.4235
2 69 1 1 2 0.94767 0.4264
3 71 1 1 0 0.06560 6.1600
4 83 1 1 1 0.05468 7.3904
5 71 1 1 1 0.04666 8.6598
6 76 0 1 1 0.93772 9.5675
7 70 1 1 0 0.04194 9.6356
8 69 0 1 2 0.95491 13.2150
9 76 1 0 1 0.02980 13.5610
10 68 1 0 0 0.01467 27.5379
;
RUN;

CFD1 is a binary confounder (1,0; yes/no), CFD2 is a categorical confounder (0,1,2; none/mild/moderate), _PS_ is the propensity score and PS_Weight is the stabilized PS generated from IPTW using SAS Proc psmatch and requesting WEIGHT=ATEWGT(STABILIZE=YES) in the assess statement.

In the SAS psmatch documentation, I can't find how to address extreme PS values that have been weighted - I have read that you can trim or truncate such values, but I don't know how to request this; or what method I should use to address this?

Any help would be really appreciated, please let me know if anything above is unclear etc.

Thank you in advance

topic How to deal with extreme propensity scores after IPTW in SAS Procedures

How to deal with extreme propensity scores after IPTW