2
$\begingroup$

My original dataset with 48 subjects has a considerable number of missing items. Majority of the data is categorical (dichotomous) and some of it is ranked (ordinal). I performed a multiple imputation using MICE in R. I am interested in running the following tests: Fisher exact test for relationship between two variables, mcnemars test for paired proportions, Wilcoxon signed rank test and the sign test. My problem is that after running these tests I am unable to pool the results from each of the five imputed datasets to come up with a single report-able result. I mean I cannot get a pooled P-value, pooled OR, pooled median. I also cannot get a pooled proportion estimate of any of the dichotomous variables. Corresponding 95% CI for each of the above estimates cannot be computed as well.

Is there a way to combine these results using Rubin's rules in R?

Improve this question
$\endgroup$
3
  • $\begingroup$ Are you asking a statistical question - is it technically possible - or a programming question - how do I do it in R? If the latter you would be better off on and R specific site like R-help or perhaps a programming site like StackOverflow $\endgroup$ Commented Feb 25, 2017 at 13:39
  • $\begingroup$ I need answers for both. Statistically because I have not come across any solid literature that supports pooling of p values, effect and effect sizes from multiply imputed datasets. So is it technically possible? If possible please share any relevant literature. Thanks. $\endgroup$ Commented Feb 26, 2017 at 18:46
  • 1
    $\begingroup$ I have it figured out. The pool.scalar function pools results for proportions and numerical estimates which solves part of my problem. Rubin's rules direct that a simple average of the estimates (effects and p-values) is enough to summarize results of any tests from the five imputed datasets. In addition the variance of the estimates should be computed to give a picture of the variability of the results due to missingness of data. I will do this manually. $\endgroup$ Commented Mar 25, 2017 at 16:51

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.