Statistics advice please! - Mann-Whitney vs ranked data factorial ANOVA (May/28/2007 )
Hello
Big statistics crisis on my hands and could do with some help.
Basically, I have been looking at iron levels in patient and control blood samples. There is a gender bias in the sample with more men in the disease group.
This means I cannot directly compare the disease vs control groups as gender will influence the results.
So there are two categorical, independent variables (gender and disease) plus a non-parametric (no matter how I try to transform it!) dependent variable (iron level).
My question is should I:
do separate Mann-Whitney tests for the different genders?
or
rank the data and do a factorial ANOVA?
or
something else entirely (like run off to a hot country, where it isn't raining and forget all about SPSS)?
Any help much appreciated, it has taken me all weekend to get this far!
Not easy What did you try for performing the iron levels? Normally, a ln-transformation resolves all the problems... To control for gender differences, a general linear model (ie. a two way ANOVA) would be appropriate, bit that also needs a normal distribution of data. How does your data look - you may judge by eye if it is dereived from a normally distriuted population...
Hello
I have done logarithmic, square root and reciprocal transformations on my data. Although they improved the data there are still outliers and I have no reason to delete these data points.
Can I do linear regression models if there are outliers?
Some outliers should be no worry. I would just try if I were you and look at the goodness of fit an R² values.
Gender you can perhaps use as covariate, i.e. a effect modifier or confounding variable.