Repeated Measures ANOVA: missing data(?)
I have a dataset consisting performance (D) for participants that each did 3 frequencies on 3 image types. There are 5 total possible frequencies, so there is always 2 frequencies a participant did not see (missing data). However, all of them saw the baseline F0, which means I could potentially subtract the F0 baseline from the other performances to get change in performance.
My dataset has the following structure:
Frequency: 5 levels, ordinal - could potentially be used as a continuous predictor?
Imagetype: 3 levels
ID: within-subject design, random factor
I am trying to perform a Repeated Measures ANOVA, but this is indicating 'Error while loading data, please verify your repeated measures observations'.
Is this because of the missing data or something else? Is it even possible to run a repeated measures ANOVA here? For example, is there a way to deal with the missing data?
Alternatively, what other way would you recommend to analyse this data using Bayesian statistics?
Thank you for any help - I have been reading up on statistics & bayesian statistics in the last weeks and would love to hear your insights
Best wishes,
Emma!
Comments
Hi Emma,
Some general remarks:
E.J.
Hi Emma,
To add to EJ, your data would be suited for either a RM ANOVA or mixed model. For the RM ANOVA, you would take imageType and frequency as the repeated factors. Your data would need to be in the wide format, where each row in your data set lists the observations of one participant, with different columns for the observations. So for instance, row one would be participant 1, with a column for imageType "1" and frequency "1", the second column for imageType "1" and frequency "2", etc. You will have missing data, but that should be fine. See the screenshot below for an xample of RM ANOVA with missing data, in a similar setup as you describe. As you can see, you don't need to manually include the "ID" column, JASP does this automatically.
You can also try sharing a jasp file with me, and I can take a closer look at where things are going wrong. You can send it to j <dot> b <dot> vandoorn <at> uva <dot> nl.
Kind regards,
Johnny
Hi EJ and Johnny, thank you for your responses.
Johnny, thank you for that screenshot, it is very useful to see to confirm that I was correctly structuring my data and analysis.
However, I am getting this error message, which I initially thought indicated my data was wrong/unsuitable.
However, a colleague found this reported bug on the JASP github: https://github.com/jasp-stats/jasp-issues/issues/1114
I have been trying different combinations of column names and factor names in my .csv and JASP respectively, adding and removing any number values, but have not yet found a working combination.
So I think my issue stems from a bug with the interpreter rather than the data. I might instead have to use R. From reading the documentation, I think the RM-ANOVA utilizes the anovaBF function from the BayesFactor library.
Emma
Hi all,
I have investigated this in more detail and discovered it is actually unlikely to be the same bug I described above. The maximum amount of factors I seem to be able to include is 3 x 3.
So the analysis runs for Imagetype Ob, Su, Pr, Frequency level A, B, C (see below - this data has gaps and is indeed working as expected)
But when I try to add a fourth frequency level I get the error (and I have 5 frequency levels in total)
I was able to run the anovaBF function in R with 3 image and 5 frequency levels.
models <- anovaBF(formula = D ~ Freq * Imagetype + ID, data = data.frame(data), whichRandom = "ID")
This gives me decisive evidence towards the alternative hypotheses
The reason I wanted to use JASP was twofold, for the assurance that I am doing the analysis correctly, and for the post-hoc/single model interference analyses.
So my big question is: Is there a max amount of levels for each of the factors in a two-way repeated measures ANOVA? Is this a consequence of how the statistics work or is this a bug? Can I use the anovaBF formula as an alternative?
My data consists of 33 participants, in case that is relevant for the maximum amount of levels each factor can have.
Any help would be very appreciated :)