I'm examining whether listeners can track strength from the voice, and whether the capacity to track strength from the voice changes depending on vocaliser sex. My variables are therefore actual strength (predictor, x) and attributed strength (outcome, y), and vocaliser sex (predictor, z).
There is little overlap between sexes in the distribution of actual strength values, though there is a fair degree of overlap in the distribution of attributed strength values.
There are other predictors in my model but they're not relevant to the question other than to say that I'm performing Linear Mixed Model analysis, with predictors including actual strength and vocaliser sex and attributed strength as the outcome variable. In this analysis, can I include sex as a factor in the model and interpret interaction effects (i.e. does the relationship between actual and attributed strength differ in males and females) even though there is little overlap in x, or should I split my data by sex and report separate analyses for males and females?
Thanks,
Jordan
Jordan,
It is appropriate to interpret the interaction between factors without having a significant main effect. Considering that you have an interaction, it's actually the main effects that become difficult to interpret.
Great to hear from you Michel! This isn't actually the issue with my data - both main effects are significant, as is the interaction. The issue is rather that the interaction (i.e. whether strength is tracked more reliably from male or female voices) graph shows that there is little overlap between sexes in the distribution of scores for actual strength. Imagining that strength is z-scores, female scores run from e.g. -2 to +0.2, and male scores run from -0.1 to +2.
So, I'm running the interaction between strength*vocaliser sex, despite the fact that the sexes are not being assessed on the same range of strength values. I don't want to change my z-scores to be standardised within sexes, so I'm wondering whether its okay to run the interaction as described, or whether I really should be splitting my analyses by sex.
Cheers, and hope you're enjoying Maastricht!
Interpretation is difficult, because you cannot distinguish clearly between two straight lines of different slope or a non-linear relation between y and the continuous variable, and the sex helps to correct for this non-linearity.
Whether the distinction between the two models matters depends on your subject.
- Badges
- Science topic
- Similar topics
- Psychology