Model Comparison: How to constrain a model to positive mu-differences?

Questions
4

I think I have to clarify some more why I turned to pymc3 and I’m not just doing a Bayesian ANOVA in JASP or jamovi.
I’m especially interested in predictions from 2 hypotheses I haven’t mentioned before.
1: low variability for orthogonal in all blocks, high variability for parallel projection in block 1 & 3, but not in block 2.

sigma
^
| .       . parallel
|  \     /
|   \   /
|    \./
| .___.___. orthogonal
|------------> block
  1   2   3

2: same as before except that orthogonal variability rises in block 2.

sigma
^
|.       . parallel
| \     /
|  \   /
|   \./
|   /*\
| ./   \.  orthogonal
|------------> block
  1  2  3

The ANOVA on the Synergy Index (a measure of the variability ratio) can’t discriminate these 2 predictions, only that there is some difference in block 2 vs 1 & 3, which doesn’t help much. Therefore, I hope to model these 2 predictions in pymc3, amongst other ones.
That’s why I think counting the samples where parallel > orthogonal would not be able to give me the answer I seek, because it would give me the same result for these 2 examples?

1 Like