3 variables, non normal distributed, which test?

[Kimmie Christensen]

Hello everybody.

I have tried for weeks now to solve my problem.
I have 3 independent variables: time, celltype and stimuli.
My data are not normally distributed, and I have tried testing for transformations but none works (with boxcox and ladder command)

Which test should I use to test for significans? I have been looking at Moods median test and Kruskal-Wallis test, but as i can see, they only work with one independent variable?

I really hope you can help me.

Kimmie

[Nick Cox]

Not very clear.

It sounds as if you have a dependent variable, but you don't say what it is.

Please read FAQ Advice all the way through, but especially #12. A sample of your data prepared using dataex (SSC) would help.

Why do you think it's important that your distributions are normal? It's not an assumption of regression, for example, that any marginal distribution is normal.

What is the role of time? Is this an experiment fit for say repeated measures anova?

[Kimmie Christensen]

Sorry, it is my first time writing here I am a master student and I have never used Stata before :-(.

I have a dependent variable, it is my western blot results for which I have made ImageJ calculations on.

I would like my distribution to be normal so i can do a three-way ANOVA, but as I cant get my dependent variable to be normally distributed, i dont know which test i should use?

I hope you understand my question, and again sorry if do not explain it very well.

[Kimmie Christensen]

This is an example of my data

Mucin MucinN Ligand LigandN Time Cell_type Cell_typeN DateN Plate WB1
MUC1---1------- PBS------ 0-------- 0-- A549 sham---- 1---------- 1------ 1-- 1.7618233
MUC1---1------- PBS------ 0-------- 0-- A549 FIBCD1-- 2--------- 1------ 1--- .3052422
MUC1-- 1------- PBS------ 0-------- 4-- A549 sham----- 1--------- 1------- 2-- 1.1394731
MUC1-- 1 -------AIF -------1--------- 4-- A549 sham----- 1--------- 1------ 5-- .89081251

[daniel klein]

Nick gave excellent advice on posting.

I have not completely figured out what you are trying to do either, but as Nick implies note that, contrary to urban-myths, ANOVA does not require the variables to be normal, but assumes the residuals to be. Given a reasonable sample size (as a rule of thumb say 50 or more observations) the model is rather robust if this latter assumption is violated. Put otherwise, non-normal variables should likely not be of any concern.

Best
Daniel