Specifying a multilevel model: an experimental psychologists nightmare Thank you greatly for your help in answering this question.
I am having some trouble figuring out how to define my multilevel model. Firstly, a description of what I have completed. This is the prelude to a wider experimental piece of work. I had participants rate 25 images for their attractiveness (outcome). The images varied in waist to hip ratio (factor: 0.6, 0.7, 0.8, 0.9, 1.0) and BMI (factor: emaciated, underweight, average, overweight, obese). I also measured the individuals sociosexual orientation (continuous: 1-100).
So, I have the following model:
Outcome: attractiveness (ordinal data on a Likert scale 1-7)
Predictors:

*

*WHR (categorical)

*BMI (categorical)

*Sociosexual orientation (continuous)

What I am looking to see is whether the attractiveness scores were influenced by BMI, WHR and their interaction, and whether there was an interaction between sociosexual orientation and WHR, sociosexual orientation and BMI, and sociosexual orientation, WHR and BMI.
Basically, I want to see whether WHR and BMI of the person has an effect on attractiveness scores and whether the sociosexual orientation of the participants influences those judgements. I believe I want WHR and BMI as fixed effects, but sociosexual orientation as a random effect?
I hope this makes sense. I find multilevel models very difficult to conceptualise.
 A: I would start out first by plotting the data.
From the description, you have crossed random effects for subjects and items (all participants rated all 25 images, and each image was rated by each participant). So random intercepts for subject and items would be appropriate.

I believe I want WHR and BMI as fixed effects, but sociosexual orientation as a random effect?

I don' know why you thnk that sociosexual orientation should be a random effect, but since its "effect" is part of your research question, this would not make sense. A model such as:
rating ~ BMI * WHR * SSO + (1 | item) + (1 | subject)

would be a good place to start.
I would also avoid categorising your data - using emaciated, underweight, average, overweight and obese for BMI and 0.6, 0.7, 0.8, 0.9, 1.0 for WHR does not make sense - it loses information, reduces statistical power and makes the model harder to interpret. A 3-way interaction, and 2 factors with 5 levels each is going to be somewhat cunbersome. Do you really want to treat someone with BMI 24.99 and one with BMI 25.00 as completely different ? I would retain the vaiables as numeric and centre them before fitting the model
