Five different groups

Question

We are running a study which investigates platelet aggregation changes during three trimesters of pregnancy and postnatal period.

Then we compare each of these phases to control non-pregnant women.

The sample size is 46 including 10 per each trimester, 10 postnatal and 6 as control.

The cases we recruited during these phases are not the same women (i.e first trimester women are different from those of second trimester and so on). We used nine aggregation agonists for each group.

We used the Kruskal-Wallis test to compare these five groups to each other as they are completely different then if we find p-value is significant, we performed a post-hoc Mann-Whitney test to find which pairs of groups are different.

The questions here are:

1. Is this the right way to analyse the data in this study using the above tests and are we justified?

2. What are the limitations of using these tests?

3. Is there any better way to use in this research?

Answer 1

Post hoc pair-wise tests following rejection of Kruskal-Wallis should be performed with Dunn's test, not a run of the mill rank sum test, because (1) the Kruskal-Wallis test decision was based on a separate set of rankings, and (2) Dunn's test uses a pooled estimate of the variance implied by the null hypothesis. See also: Post-hoc tests after Kruskal-Wallis: Dunn's test or Bonferroni corrected Mann-Whitney tests?.