issue with Test for Sphericity ('Singularity error')

[ElenaG]

I am contacting you regarding Jamovi software version 2.3.28.0 for Mac OS.

I carried out a simple 3-way repeated-measures Anova ( 2 x 3 x 6 anova for 8 participants) and noticed that the Test for Sphericity does not seem to work as it should when the number of repeats/ participants is 8, or more precisely less than 16. The test for sphericity table returns only ’NaN’’ values and the error ‘ Singularity error. Sphericity tests are not available”. I am not certain that I quite understand what ’singularity error’ means in this context. As the Test for Sphericity does not work properly, the relevant Bonferroni corrections for the p values and degrees of freedom (dfs) are not applied when the number of participants/ repeats is <16. The issue does not seem to occur for larger than 16 repeats.

I am wondering if you are aware of this ‘bug' with the Jamovi algorithm (when the number of participants is less than 16) and whether there is any fix for this?

[jonathon]

Hi Elena,

i think this issue occasionally shows up in the RM ANOVA when the sample sizes are very small.

jamovi uses an R package called “car” to calculate these values, and there’s a post here from john fox, the author of the car package, explaining that it isn’t possible:

https://stat.ethz.ch/pipermail/r-help/2 ... 05880.html

this is from 2012, but it looks as though john hasn’t got to fixing it – I could imagine it being a low priority because of how rarely it pops up.

unfortunately, there’s not much we can do here.

sorry I don’t have better news

jonathon

[reason180]

This repeated-measures ANOVA (2 by 3 by 6, with eight participants) does work, and includes sphericity tests (jamovi 2.5.6 for Windows).

[jonathon]

yeah, i'm not saying it doesn't work for all 2x3x6 with 8 participants ... only that it sometimes comes about when sample sizes are small.

it's true, it would be worth eye-balling the data though.

jonathon

[reason180]

Is the problem simply that, for at least one pair of the y columns, the difference scores are all the same, so the variance is zero, and ANOVAs don't work when zero becomes a denominator of an equation? See below.

temp4376664625Untitled.png (78.99 KiB) Viewed 70794 times

[ElenaG]

The test for sphericity for a 3 way repeated measure anova with <16 partcipants simply does not work. If you look in the test for sphericity table all values are 'NaN' and 'a singularity error'. Hence the dfs and p values are not corrected. This issue occurs with both Windows and Mac versions.
The same data ran in SPSS, Matlab, GraphPad, etc. produce consistent and correct outcomes. The issue is with Jamovi's algorithm which needs to fixed.

[reason180]

I think you'll need to provide a data set that does not work in jamovi, in order for the problem to be fully addressed.