[Bug]: Incorrect calculation of partial omega squared in three-way within-subjects (repeated measure) ANOVA

[brianpkeane]

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?

macOS Intel

Bug Description

In a 2 x 2 x 2 within-subjects repeated measure ANOVA, the partial omega squared for a two-way interaction is being reported as .000. However, when I calculated the value by hand--using the formula described by Kroes & Finley, 2023 in the Appendix and using the Mean Squares and Sums-of-Squares reported by JASP--the value ends up matching the “omega squared” value given by JASP. The effect is statistically significant, so I doubt that JASP's partial omega squared value is correct.

Expected Behaviour

If the formulae of Kroes & Finley are indeed being implemented, the value should be above zero. The denominator term of partial omega squared is smaller than that of omega squared, so it is strange that the partial omega squared value is smaller than the partial omega squared value. Therefore, I would re-check the formulae (two-way interactions within a three-way design).

Steps to Reproduce

1. 2. 3. ...

Log (if any)

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]

@brianpkeane To help us reproducing and understanding this error: Could you share your .jasp-file by renaming to .zip and drag and dropping it here? thx

[JohnnyDoorn]

Hi @brianpkeane,

Thanks for taking the time to check the calculations. From your output, I cannot recompute the effect sizes (need the between subjects table as well for the full error information) - could you please share your .jasp file and/or calculations so that I can check them for myself? In JASP, we use the effectsize package - specifically the effectsize::omega_squared function with partial set to TRUE or FALSE. I did verify the results with the Kroes and Finley paper, but only with their examples and not the three-way interaction as you show case. However, I do believe it's possible for the partial effect size to be smaller or greater than the non-partial versions so this is not necessarily an error.

Kind regards Johnny