Non-parametric tests, data not meeting assumptions


Im at the stage where im banging my head against the wall!!!!:eek:

Im testing the time of travel (in hours) taken for fish to cross a lake under 2 different conditions. There are 39 and 40 times in each set respectively.

I tested the data for normality using the Anderson Darling test which showed my data wasn't normal, so i decided on a non-parametric test.

However, the assumptions stated on Minitab for the mann whitney, kruskal wallis and moods median test is that the variences are equal. I used the Levenes test and it showed the variances were not equal (P=0.034).

So which test do i use? the data is non-parametric but all the non-parametric tests assume the data has equal variances!

Also I tried to transform my data, to get normal distribution. I achieved this in one set but not the other. Do I use the un-transformed data and a non-parametric test and ignor the assumptions about variances?

I really look foward to help on my puzzle so I can make my dealine for work!!!



I think you can do non parametric if your data is not normal and its not necessary that the variances be equal.
but before it,please you use cox-box transformation on your data maybe your data will have been normal.
if they dont normal not at all then you should do mann whitney test.
Last edited by a moderator:
Thanks for your reply,

I tried box cox and got no success so used Johnson transformation which worked. Now I've used 2-sample t-test.

Maybe this is right?:confused:


TS Contributor
Do your comparison two ways - first, assume the data is normal, and run a t-test. Then, assume your data is not normal, and run the Mann-Whitney test.

Do you get the same conclusion? If so, then go with it. If not, then post back - we'll try to help you sort it out.

Don't sweat the equal variances thing - .034 isn't very significant - it's more "borderline."