0
$\begingroup$

I'm fairly new to statistics (I'm doing it for my bachelor thesis). I want to do a factorial mixed ANOVA in Jamovi, but not all of my data is homogenous (15 out of 126 measurements are not homogenous). I am able to transform the data with python, but I couldn't find any library that provides a factorial mixed ANOVA.

Experiment design: Subjects dodge virtual objects in virtual reality. I have two groups each with 15 participants. The virtual rotation for group A is amplified (e.g. they turn their head 10° in real space but 15° in virtual space). Group B has no rotation amplification. The groups are the between-subject factor. I measure their greatest turn angle in virtual space (dependent variable), while they try to dodge an incoming object. The 4 independent within-subject factors are form (3 levels), size (2 levels), speed (3 levels) and incoming_angle(7 levels).

I want to find out, if there is a difference in virtual turn angle for each group. I also want to find out, if the greater levels of each factor are causing greater turn angles.

I think because of the experiment design, I really can only go with factorial mixed ANOVA. But none of the assumptions for that method are met (normality of residuals, sphericity, homogeneity of variances).

I tried various transformations (absolute values, raising to higher powers, reciprocal, logarithmic, Box-Cox), but they all fail the assumptions.

My data looks like this:

histogram

qq-plot

Improve this question
$\endgroup$
5
  • $\begingroup$ Maybe you could tell us what your data represents, and what is your ultimate goal with the analysis? $\endgroup$
    – kjetil b halvorsen ♦
    Feb 19 at 20:10
  • $\begingroup$ I'm having trouble determining what your question is. ... If you can run your analysis in Jamovi, that should be a fine software choice. ... The one residual plot you present doesn't look too bad. Do you have a similar plot of residuals vs. predicted values ? ... What do you mean when you say there are 126 tests (for homogeneity ?) ? $\endgroup$
    – Sal Mangiafico
    Feb 19 at 20:19
  • $\begingroup$ @SalMangiafico Yes, the 4 within-subjects factors have 2x3x3x7 levels which make up for 126 measurements per subject. I have 30 subjects, 15 in each group. I think I need to test the homogeneity of variance for all of the 126 measurements, but please correct me if Im wrong. $\endgroup$
    – Snessub
    Feb 20 at 21:20
  • 1
    $\begingroup$ Look at a plot of residuals vs. predicted values. (And please add to your post.) ... In general, running hypothesis tests (for normality or homogeneity) to determine if a model is appropriate isn't very helpful. ... In this case, I'm not sure how you are running these tests. $\endgroup$
    – Sal Mangiafico
    Feb 20 at 21:45
  • $\begingroup$ @kjetilbhalvorsen Thanks for your reply! I updated the description. $\endgroup$
    – Snessub
    Feb 20 at 21:46

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.