Has anyone found a way to use rbounds and mhbounds in Stata with teffects?
In particular, with nearest neighbor matching. What would one use for the weight and support variable?
Is there a function to use for a matched pair design that has matching with replacement?
Looks like the answer is a resounding "No." I tried three work arounds:
1) Use psmatch2 and set the specifications as close to the model in teffects. The problem is that rbounds and mhbounds assume matching without replacement.
2) Replicate in R using Matching() followed by rbounds().
3) I also tried to manually write a script to calculate it, but am pretty sure that was a dead end. I understand from Rosenbaum (2002) that the original sensitivity analysis is based on a 2 by 2 contingency table, but what I really want is something based on matching with replacement, weighted, and unmatched data. Rosenbaum has written some packages in R, but they are not in Stata.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics