Using Mann-Whitney test when all values in one data set are zero (SPSS vs Graphpad)

[sunilaup]

I would be very grateful for advice on the following problem:

I am examining the frequency in which political issues are reported in the national media in the six days before and six days after a particular date. An example of the data set for one issue in the six days beforehand is 8,7,9,5,6,7 i.e. on day one the issue is reported in 8 newspapers.

I am comparing the before/after data sets using the Mann-Whitney U test to see if there is a significant difference in reporting, as I cannot assume normal distribution.

The problem is that there are some issues not reported at all before the date, but reported heavily in the six days after (e.g. 0,0,0,0,0,0 vs 8,7,8,4,6,8). Graphpad will not let me use the Mann-Whitney test when all of the values in one data set are zero, and insists on the Wilcoxon Rank Sum test instead. However, SPSS has no problems with the Mann-Whitney test.

Could I please have some advice as to which test is the most appropriate

Many thanks

[hlsmith]

I though these were the same test. Are your results differing? Ranking seems appropriate in this situations.

[sunilaup]

Many thanks - that's very helpful. The results are not differing in that as you would expect there is a highly significant difference between the two data sets whatever test I use!

[victorxstc]

May I add Wilcoxon is the paired alternative for Mann-Whitney, and considering the given information (that you are dealing with before/after cases [then paired data]) maybe using a Wilcoxon is the correct way. Unless your before and after examinations are not done on the same group.

[hlsmith]

Agreed! The description is a little ambiguous. It seems that you are comparing count data for the six days before and after a day. The underpinnings of this investigation may be much more complicated than it seems, since there may be some covariance or relationships between all observations (days).

[sunilaup]

Many thanks. I agree that the underpinnings are a bit more complicated, and that it is possible to pair the days as there are equal numbers of each, although strictly speaking they are not paired as day 1 pre does not necessarily correspond to day 1 post

Much appreciated
Sunil

[Karabiner]

These are dependent/repeated measures, because each political issue consitutes
a single "case", which is measured several times. So you'll have to perform an
analysis for dependent measures, e.g. Wilcoxon's rank sum test with the pooled
number of reports before the date versus the pooled number of reprorts after that
date.

Kind regards

K.