[Bug]: RM-ANOVA with Two Within-Subjects Factors: Inconsistent Interaction DF Under Sphericity Violation

[edouard-dendauw]

JASP Version

0.19.0

Commit ID

No response

JASP Module

ANOVA

What analysis are you seeing the problem on?

repeated Measures ANOVA

What OS are you seeing the problem on?

Windows 11, macOS Silicon

Bug Description

I am conducting a repeated measures ANOVA with two within-subjects factors: coh (coherence) and bias.

Theoretically, if bias has df = 3, coh has df = 2, and there are 24 subjects, then the degrees of freedom for the interaction bias_x_coh residuals should be ( 2 \times 3 \times 23 = 138 ). The prompt is correct in this case [see the first RM-ANOVA in the attached image].

However, when I reordered the Model Terms in the Model menu (swapping coh and bias), the df of the interaction residuals dropped from 138 to 46, which suggests that the df for coh was not accounted for in the second model [see the second RM-ANOVA in the image].

Additionally, in the second model, the ratio ( \frac{MS_{AXB}}{MS_{err}} ) does not match the reported F value.

I tried using various data, but the sphericity violation does not seem to be the origin of this issue. However, the order of the Repeated Measures Factors in the initialization impacts the issue.

Expected Behaviour

The order of the factors in the Model Terms box should not affect the df. Is that correct?

Steps to Reproduce

1. Do an RM-ANOVA with two within-subjects factors stats.zip
2. Switch the factor order in the Model Terms box
3. See the df of the interaction residual changing

Log (if any)

No response

More Debug Information

No response

Final Checklist

• I have included a screenshot showcasing the issue, if possible.
• I have included a JASP file (zipped) or data file that causes the crash/bug, if applicable.
• I have accurately described the bug, and steps to reproduce it.

[tomtomme]

@edouard-dendauw
thx for the report and sorry for the delay. I can confirm your results. This seems to be a real bug. Since all F-values are the same in the two models, as your screenshot and my replication confirm. But with different dfs this cannot be true, can it? So this seems to indicate that the reported df musty be wrong int the SECOND model, since if it would be correct there, we would need to see a different F-value....

@JohnnyDoorn: Can you help me out here. Can this be?

[github-actions]

This issue will be automatically closed in 42 days due to inactivity. Feel free to leave a comment if you believe this is still relevant.