Closed jwbowers closed 3 years ago
If I'm not mistaken, this describes behavior of the formula method, e.g. match_on(z ~x1 + x2)
. Perhaps also the function method.
I think the glm method, e.g. match_on(my_propensity_mod)
, is different, and will try to impute the NAs. @josherrickson would know best.
For now I propose to move this blurb over to the docs segment about the formula method, i.e. the block lower down that begins with "@details \bold{First argument (\code{x}): \code{glm}.}
". This may not be ideal from user perspective, but it seems an improvement.
Thanks for speaking up about this, Jake, and for the pull request.
@benthestatistician You're correct, glm (and bigglm) attempt imputation via scores
. The others do not.
@josherrickson I'm passing this on to you. I'd request your attn to:
match_on()
If so then my guess it that it's ready for merging in to master.
I discovered how match_on was handling missing data on covariates by some sleuthing and figured that users might want to know this especially if, by chance, they give a formula to match_on where some of the covariates have missing data for some observations and then those observations are kicked out of subsequant matches.