0

We plan to conduct a research for exploring the risk factors for a cardiovascular disease by a case-control study. The covariates are expected to be balanced between case and control groups by the Propensity Score Matching. We will compare the difference of covariates between case and control groups before and after matched by PSM. We could compare them by paired samples t-test.

However, if we matched the samples between case and control groups as the ratio 1:2 or 1:3 or 1:4，how to compared the difference of covariates before and after matched by PSM ? If the covariates are categorical variable, we might analyze them by Conditional Logistic Regression Analysis.

But if the covariates are continuous variable, which method could we conduct to analyze them between case and control groups（that is , the ratio is 1:2 or 1:3 or 1:4 between case and control groups）？

Thank you!

skywalker21th
  • 59

1 Answers1

1

If the goal is to assess balance, you should not use hypothesis tests to do so. Instead, you should use sample-based measures of balance that do not invlve the sample size. These include the standardized mean difference and the Kolmogorov-Smirnov statistic. After k:1 matching, you generate matching weights which can be used in weighted versions of these statistics. The weights are usually equal to one divided by the number of matches for each unit (e.g., 1/3 for 3:1 matching). All of this is implemented in the R package cobalt. The cobalt vignette describes perspectives on using hypothesis tests to assess balance.

Noah
  • 33,180
  • 3
  • 47
  • 105
  • Dear Noah, thank you very much for so professional comment. We have ignored the SMD. We will learn the describtion in the R package cobalt. – skywalker21th Aug 03 '22 at 01:53