# Mann-whitney U test? math n00b?

1. Jun 25, 2011

### bcpsleeperkid

mann-whitney U test? plz help math n00b?

I am trying to test the difference in motor performance (measured by the amount of time each rat is able to keep running on a moving beam) between two groups of rats (those with early exposure to a common antidepressant and those with early exposure to Saline as control) in SPSS. There are three rats in each group and over a hundred entries for each of the six rats. I feel that the Mann-Whitney U test is most appropriate to determine the significance of my data. The problem is that I do not think that this takes into account the measures for each particular rat. One of these rats is an ENORMOUS outlier, big enough that it may make the entire group look significantly different if the program does not differentiate between rats within a group and only looks at all the times for each group as a whole. I understand that non-parametric tests are suited to deal with huge variation within a group but my concern is whether or not the mann-whitney U test is suited to deal with consistently large variation coming from one or two samples with hundreds of entries.

2. Jun 26, 2011

### Stephen Tashi

Re: mann-whitney U test? plz help math n00b?

I could interpret your remarks to mean that you are considering using the result of each of the hundreds of trials as your sample, so the statistic would be computed on the basis of ranking them all. That doesn't sound right, since the trials aren't independent if you believe that the ones that come from the outlier rat tend to be different that the trials from the other rats. It's more convincing to use the mean result of each rat, but that only gives you 6 things to rank, which might not be enough of a sample to impress anybody.

If each rat did the test hundreds of times, did they get better with practice?

In scientific investigations, the statistics that are to be used should be decided upon before the experiments are conducted. Otherwise, the investigator is liable to "cherry pick" the statistical tests that are used in order to make the analysis swing one way or the other.

Last edited: Jun 26, 2011