zscoring multimodal variable according to different means

#1
Hello everyone, I am a researcher and wrestling with a trimodally distributed data that I want to enter into a repeated measures ANOVA (whose residuals are normal after some outlier removals). I have the suspicion that the multimodal means are masking some effects by adding variability. By the way a different variable can group the participants according to their different means. My inclination is to z-score ((x-μ )/ σ) using the the SD of the whole group, but subtract the mean according to the different variable. I've never read about this being done, so my question is, is that a legitimate approach? Or totally bogus?
 
Last edited:

obh

Active Member
#2
Hi Ellie,

You need to be careful with the outlirets. Do you know the reason for each outlier?
What are the IVs and what is the DV?

What do you try to do?
Why don't you use multiple regression (it is the same model as the ANOVA, but maybe you can get what you want this way)
 
#3
Hi, thanks for responding. I don't think I have a big enough N (=28) for multiple regression. The dependent variable is tapping rate (intervals between taps), so I don't think the outliers mean very much, it has to do with (theoretically arbitrary) metrical levels that participants tapped. The independent variables are to do with the timing conditions of the stimuli to which participants tapped.

But, my original question is whether it is legitimate to subtract different means from different cases/participants when z-scoring, while using the SD from the whole group. The purpose of this being to centralize multimodal distribution. Do you have any feedback about this z-scoring?
 

obh

Active Member
#4
Since ANOVA and multiple regression are the same models if you don't have enough data for regression you don't...

I don't understand the second question.
 
#5
"But, my original question is whether it is legitimate to subtract different means from different cases/participants when z-scoring, while using the SD from the whole group. The purpose of this being to centralize multimodal distribution. Do you have any feedback about this z-scoring?"

I'm just looking to centralize the mean of multimodally distributed data...if this sounds bizarre/not understandable then I should try a different approach, thanks anyway