jasp-stats / jasp-issues

This repository is solely meant for reporting of bugs, feature requests and other issues in JASP.
59 stars 29 forks source link

Bayesian rmANOVA terminates unexpectedly #1346

Closed SteffiPeykarjou closed 9 months ago

SteffiPeykarjou commented 3 years ago

Hi guys,

I am trying to run a Bayesian Repeated Measures ANOVA, a regular analysis that I am using a lot. For some reason, I keep getting an error running the analysis with the attached dataset (never had that before). I tried Jasp versions 0.13.1 and 0.14 on Windows 10 PCs.

I am trying to use two between-subjects factors (cond, median-split (Anzahl Trials Fam), and structure the input using two within-subjects factors (roi + ele: o - Iz, O1, O2, Oz, po: P3, P4, PO9, PO10). The analysis runs very long, > 1 hr, and then terminates suddenly with the error message you find below.

I'd appreciate any help a lot! Thank you, Steffi

grafik

SteffiPeykarjou commented 3 years ago

Hip_FPVS+Hip_Fam_mediansplit_NumTrialsFam.txt

vandenman commented 3 years ago

Hi @SteffiPeykarjou, thanks for the report!

Could you maybe describe the structure of the Repeated Measures Factors a bit more? In particular I'm unsure what I should enter here to replicate the problem: image

You can also upload the .jasp file here directly, if you zip it first.

One thing you should consider is that JASP by default adds all interactions between Between Subject Factors and Repeated Measures Factors. If you open the tab Model you'll see this: image

Here, the highlighted lines contain interaction effects. By default all interactions are considered, which includes the 4-way interaction between ele, roi, cond, and median_split.

This is probably the reason why it takes so long to compute. What I would do is:

  1. Enter the Between Subject Factors but not the Repeated Measures Cells.
  2. Select all Model Terms (Ctrl + A).
  3. Click on the leftward arrow: image
  4. Now there are no models specified.
  5. Manually enter the models you want to consider.
    1. You can enter main effects by selecting individual components and dragging these from Components to Model Terms.
    2. You can enter interaction and main effects by selecting multiple components (hold Ctrl and left-click on multiple components) and then drag these from Components to Model Terms.
  6. If you have no idea what models to consider, I'd recommend entering all main effects, and the interaction effect between cond and median_split.

This way, there are fewer models considered and the analysis should finish faster.

Let me know if this helped!

SteffiPeykarjou commented 3 years ago

Hi @vandenman, thanks a lot for your help, I really appreciate it! I tried with the reduced model as you suggested, which gives me a new error. I think I made an error with my columns, but can't figure out what it was. I am attaching the original analysis and the adapted one, that's probably the easiest way. Thanks for any further suggestions! Best Steffi Hip_FPVS+Hip_Fam_mediansplit_NumTrialsFam_newerror.zip Hip_FPVS+Hip_Fam_mediansplit_NumTrialsFam.zip

tomtomme commented 9 months ago

@SteffiPeykarjou Both files run the analysis just fine with jasp 0.18.3 For the file named "newerror" however, I first had to remove the empty rows and columns. This is afaik a known issue with RM (B)ANOVA. It can't handle missing data very well yet. Advanced Missing value handling is on the horizon for JASP 0.19 however. So lets cross fingers for that to materialize :D

Also note: The computation of that model may need a day or more! I did let it run until I could see the progress bar moving a tiny bit and that was after about 1-2 hours O_o. Also the smaller module was only marginally further in progress

If this is not solved on your side, feel free to reopen.