1-to-(n) Propensity score matching without replacement

[sam101]

Hi,

I was hoping if someone can help me with this. I have a data set with about 100 cases and 6000 controls. I want to create a propensity score matched cohort of 1 case:3 controls (propensity score generated based on a set of baseline variables like age, gender, kidney function etc.). The -psmatch2- command does not let me create 1-to-many matching without replacement when using the n() option

"psmatch2 treatment_variable , pscore(logit1) caliper (.2) noreplacement n(3)"- returns error message

"psmatch2 treatment_variable , pscore(logit1) caliper (.2) n(3)"- does propensity matching with replacement (not what I am looking for)

Can anyone please suggest on how to do this/share code to overcome this? I am using Stata 15 version. I can't use the -teffects- command as I need the id for matched controls to do survival analysis on the final matched cohort.

Thank you so much in advance.

[Mike Lacy]

Take a look at -ssc calipmatch-, which will do greedy matching and create a new variable to identify matched n-tuples. I think it's a convenient command to use.

[sam101]

Thank you so much Mike. This suggestion was really helpful !!
-calipmatch- is exactly what I was looking for.

[Alexander Rodriguez]

Hi sam101 gmatch is another you may wish to consider.

Does anybody know if there is a succinct compendium of 'treatment effects in observational data' commands?

My impression is that broadly there are two sets of commends that exist: (1) commands to test for treatment effects (for example teffects, psmatch2) and (2) commands like gmatch, ccmatch or calipmatch that are capable of performing matching from your large data sets.

Many thanks,
Alexander

[David Radwin]

Does anybody know if there is a succinct compendium of 'treatment effects in observational data' commands?

For an overview of built-in Stata commands (e.g., teffects), see https://www.stata.com/features/treatment-effects/

I don't know of a summary of all the user-written commands (e.g., psmatch2, pscore), though.