Here's the study design:
Participants are asked to select up to three (of 23) potential colors for an emotion. For each color they pick, they indicate on a scale 1-10 about how much they think about that color when they think of that emotion. I would like to use a rmANOVA on color intensities, which essentially means that if they did not choose a color, I am coding it as intensity for the color = 0.
Do you see a problem with turning unchosen colors into values of 0 so that I can use a rmANOVA to compare whether color intensities are statistically different across the emotions?
If you see this as an issue, please let me know whether there is a better way (pecentages?) to allow for a repeated design. I don't want to loose this design factor, if possible. Thanks, friends and colleagues!
I think that this is a two stage process. The first question is whether a colour is chosen, and the second question is whether, when chosen, it is chosen at a given intensity.
Your proposed coding confuses these two questions into one question that relies on strong assumptions about the place of 'not chosen' on a scale of intensity.
I would start with modelling whether or not the colour is chosen, then use a hurdle model to examine intensity subject to the colour being chosen.
Ronan is correct. Coding missing values as 0 will result in a nonsense outcome.
There is a slight complication because you probably want to treat this as a multilevel hurdle model which may mean more sophisticated software. I'm pretty sure one could do this in the (free) MCMCglmm package for R (e.g., see link).
https://cran.r-project.org/web/packages/MCMCglmm/vignettes/CourseNotes.pdf
- Badges
- Science topic