A caliper which means the maximum tolerated difference between matched subjects in a "non-perfect" matching intention is frequently set at 0.2 standard deviation as the default such as used in the PS Matching SPSS R-extension utilitiy. In practice, how large a difference (caliper) can still be tolerated or is caliper selection covariate-dependent?
A good way to do this is to check whether the covariate distributions are balanced after matching. Our Gauss (bin_match, available on this website for download), Stata (in the Stata archieve) programmes maybe also be useful for doing this. You may also find the considerations in our paper (Huber, Lechner, Steinmayr, online at Empirical Economics) helpful as it discusses some of those practical issues of matching estimation.
A good way to do this is to check whether the covariate distributions are balanced after matching. Our Gauss (bin_match, available on this website for download), Stata (in the Stata archieve) programmes maybe also be useful for doing this. You may also find the considerations in our paper (Huber, Lechner, Steinmayr, online at Empirical Economics) helpful as it discusses some of those practical issues of matching estimation.
when the pre_matching ratio of subjects was 1:2:7, the
MSE was only slightly lower than that of the highest MSE when
we used caliper widths of 0.2 of the pooled standard deviation of
the logit of the propensity score. When the pre_matching ratio of
subjects was 1:2:3 or 2:3:5, using a caliper width 0.2 of the pooled
standard deviation of the logit of the propensity score resulted in
the lowest MSE or close to the lowest MSE. This shows that using
this caliper results in greater precision than achieved with other
calipers. Regardless of the pre_matching ratio of subjects, using a
caliper width of 0.1 of the pooled standard deviation of the logit of
the propensity score resulted in the lowest bias, followed by the
caliper width of 0.2 of the pooled standard deviation of the logit of
the propensity. Depending on the pre_matching ratios of subjects,
our findings suggest that a width of 0.2 of the standard deviation of
the logit of the propensity score is the optimal caliper
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS