There is a formula in the R documentation, but it's really strangely written and when I put numbers into it, values seem unrealistic: rdocumentation.org/packages/mStats/versions/3.2.2/topics/mhr‌​r

Hi @Ute - excellent point! I think I can do a Mann-Whitney U Test, taking all articles from all journals together and comparing the rank of the T type (retracted articles) the to non-retracted ones; if I'm understanding correctly, this would avoid the need to normalise and it would already rank the pooled data..

Hi @utobi- there are only two types of article in these journals for our purposes - type T (retracted articles) and all others. Our preliminary analysis suggests this former type in our field generate much more public interest prior to their subsequent retraction than others, and we're looking for a means to compare this fairly. In each journal, publications per year range from 500 - 18000, so there's high individual counts but retracted articles are relatively rare (1-5 per journal year). I like the rank idea - public engagement metrics skew extremely, might help.

@ChristianHennig Sure, but then they report the rate data, not the count data in their other results and main text. They even combine rate data at some points. I dug in to get the raw data and I suspect you're right, but they flit between both in the text.

Thank you, this is really useful! Is there a condition that the two samples should have broadly similar shapes for the RS test or is does it simply mean that it's not a straightforward median difference issue?