I need help in knowing which test to perform with this analysis.

I have two groups (lets say group A and B). A study was performed with initial rates of heart failures in group A and B (i.e, baseline rates) were measured. After 6 months of follow up the heart failure rates in group A and B (i.e, re measurement rates) were measured again. By simple mathematics, I know that the improvement rate is better in group A as compared to group B.

  • How do I test whether the difference in rates between the groups is statistically significant?

I am using t-test for this but not sure how to go about it.

  • 1
    $\begingroup$ It seems like you have a binary (failure/no failure) dependant variable? If so, you would rather need logistic regression to get the most out of your data. $\endgroup$ Jul 25, 2011 at 5:40

1 Answer 1


T-tests are for normal or close-to-normal distributions. You need a nonparametric test. Since you have the same two groups being assessed twice, I don't think you can use the McNemar test for dependent proportions. However, maybe you could do a chi-square test on the increases within each group as fractions of the sample sizes. E.g., if group A had 5/100 develop the illness and group B had 12/100, you would run the chi-square test using the numbers 5, 95, 12, and 88.

  • $\begingroup$ Thanks for your reply. Can I not do a paired t-test as the baseline and re measurement was performed at the same institution? $\endgroup$
    – Nupur
    Jul 25, 2011 at 1:19
  • $\begingroup$ To be more explicit. I have the baseline and remeasurement rates for both groups (each group has 10 hospitals in it). I calculated relative improvement rate, RIR (which is [baseline - remeasurement] / baseline) for each hospital in both groups. With t-test, I analysed if the mean RIR for both groups are statistically significant? Is this a right way of doing it? for confirming the results, I also categorized each hospital as improvement/noimprovement based on RIR value and then did a chi-sq. $\endgroup$
    – Nupur
    Jul 25, 2011 at 20:10
  • 2
    $\begingroup$ It is seldom an optimum way to analyze ratios. If relative changes are relevant and you've verified the Bland-Altman conditions, then analysis of log ratios is more appropriate. But even more frequently, analysis of covariance of the final measurement, adjusted for the initial measurement as a covariate, works better. One example of this is when the baseline is measured with much error or is poorly correlated with the follow-up measurement; in that case you lose power by analyzing change instead of using analysis of covariance. $\endgroup$ Aug 24, 2011 at 2:40
  • $\begingroup$ I don't see how ancova can apply when the data consist of group-level aggregate rates, pre and post, or else of binary variables describing each individual hospital, again pre and post. $\endgroup$
    – rolando2
    Sep 26, 2011 at 18:05

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.