Sensitivity Analysis with a MatchIt Object - Suggestion

[Maxi54321]

Hello,
since it is not completely obvious, how to conduct a Sensitivity Analysis following a Propensity Score Matching with MatchIt, I would like to ask for an opinion on this approach by Matt Bogard.

He conducts this:
`trt <- df1%>%
filter(treat==1)%>%
arrange(distance)

ctrl <- df1%>%
filter(treat==0)%>%
arrange(distance)

trt <- as.vector(trt$re78)
ctrl <- as.vector(ctrl$re78)

psens(trt, ctrl,Gamma = 2, GammaInc = 0.1)`

In my case, for "distance" I planned to use the distances which are the output from e.g. "m.out2$distance" (let "m.out2" be a matchit-object). As preparation for the code above, I would run this: df1$distance = m.out2$distance, so that the code can arrange by distance, as stated above.

Is this reasonable (my preparation and/or Mr. Bogard's suggestion)? I would be very thankful for a statement. Thanks also for your recent fast and comprehensible answers.

[ngreifer]

I would not recommend this approach. psens() is not designed to be used with matchit outputs. Also, the test it runs is quite obscure even though Rosenbaum presents it as a standard. It relies on an obscure test statistic that is quite different from the usual t-statistic used to evaluate the treatment effect. It also is unable to take into account covariate adjustment using regression after matching. In my opinion, software for sensitivity analysis is not mature enough to use out of the box with the results of a matching analysis. You might consider using the e-value, which is easy to compute from the treatment effect estimate and more interpretable than the gamma statistic, or use the tools in the tipr or OVtool packages.