[Bug]: Bayesian Paired-Samples t-test (Wilcoxon) - Won't run with ordinal data, but will with scale data.

[JayD05]

JASP Version

0.16.4

Commit ID

No response

JASP Module

T-Tests

What analysis are you seeing the problem on?

Bayesian Paired-Samples t-test / Wilcoxon

What OS are you seeing the problem on?

Windows 10

Bug Description

In attempting to run a Bayesian t-test / Wilcoxon I noticed that the analysis would not run if data were ordinal, but would run if the same variables were relabelled as scale data. Further, Frequentist t-test/Wilcoxon will run with ordinal, but Bayesian equivalents will not (even though the same variables are used). See example below:

When data are coded as ordinal:

Frequentist Wilcoxon runs fine:

However, Bayesian returns the following error ("Number of observations < 2 in [Var1 & Var2]"), which doesn't make any sense as there are > 2 observations (as shown by the Frequentist analysis running fine):

Note. The same issue exists for both paired-samples t-test and Wilcoxon.

Changing the same variables from ordinal to scale:

Rerunning the Bayesian analysis shows that it now works fine:

Data:

jasp data.zip

Expected Behaviour

The t-test / Wilcoxon should run regardless of whether data are ordinal or scale.

Steps to Reproduce

1. Run analysis with ordinal variables
2. Change data type on variables from ordinal to scale
3. Rerun same analysis with scale variables
4. JASP will run when data are scale but not ordinal
   ...

Log (if any)

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]

@JayD05 @vandenman
tested with jasp 0.18.3 and your data. I had to switch to scale and then ordinal again - then everything worked. But on that switch the labels were lost. I re-entered labels and the bug was there again.

So this bug is isolated to ordinal data with labels
here the file that shows this directly
ddaf.jasp.zip

Since this may be solved by Joris work for 0.19 I will mark it for 0.19 to remember that I need to test this one after release.

[tomtomme]

@JohnnyDoorn
@vandenman
the original bug is now solved, but if the Var is ordinal, bayesian paired t-Test now does not show any diagrams!
Hence I up this to critical for 0.19.

the error message is wrong. Switching to scale instead of ordinal produces graphs

[JohnnyDoorn]

Somehow the Bayesian t-test (any) looks at the labels instead of the values.. when I use facFive from the debug data set it's fine, but when i adjust the labels to a-e instead of 1-5, it breaks. Descriptives, freq t-test and other analyses still work fine though. @vandenman do you have an idea how to solve this in a clean way? I see that the Bayesian t-tests have their own data reading function.