topic Re: PS 1 to 5 Optimal Matching in Statistical Procedures

PS 1 to 5 Optimal Matching

cici5723 — Sat, 09 Mar 2024 21:28:20 GMT

ERROR: A feasible optimal fixed ratio matching that has the specified parameters does not exist.

I meet this problem when I do the 1-5 optimal matching, here is the code:

/* Perform the 1-to-5 optimal matching */
proc psmatch data=DRUGS region=allobs;
    class Treatment Gender;
    psmodel Treatment(Treated='1') = Gender Age BMI;
    match method=optimal(k=5) exact=Gender distance=ps caliper=0.5; 
    output out(obs=match)=matched2 matchid=_MatchID;
run;

If I change the k=1, it could run, while it showed error when I put k=5.

Re: PS 1 to 5 Optimal Matching

sbxkoenk — Sun, 10 Mar 2024 13:27:41 GMT

I have moved this topic from "SAS Studio" - board to "Statistical Procedures" - board.

@paulT can help you with this.

(... yet if paulT is the Paul I think he is.)

Koen

Re: PS 1 to 5 Optimal Matching

MichaelL_SAS — Mon, 11 Mar 2024 14:52:00 GMT

Assuming the data that you are using are the same as the Drugs data set used in the PROC PSMATCH documentation, an optimal fixed ratio matching with 5 control units for every treated unit will not be feasible. There are only 373 control units and 113 treated units, so there aren't enough controls for that to satisfy that fixed ratio.

It might be that the prompt is requesting an optimal variable ratio matching, with 1-5 control units matched to each treated unit. You would request such a matching my using the METHOD=VARRATIO(KMAX=5) specification in the MATCH statement. Using the DRUGS data set from the PROC PSMATCH documentation, that matching is feasible.