I have an experiment where 10 groups of 3 judges are asked to evaluate a list of pictures in 2 evaluation conditions.
The pictures were made in 2 between-subjects conditions (A,B) and each group of subjects created the images in 2 conditions (a,b). Therefore we can think these groups as A(a,b) and B(a,b).
The judges evaluating the images are not the same people that made the images.
Each group of 3 judges evaluates a portion of the whole set of images but, in the end, each image gets 3 votes. Each judge evaluates the same images in 2 different conditions. In the end, I have a list of pictures with a value for each image (the sum of the votes) for each of the evaluation condition (eval_1, eval_2). see image example attached.
What I am trying to find are the effects of (A, B), (a,b) and (eval_1, eval_2) conditions.
Although some of the images across groups were made by the same person, each image belongs to one group only. Can I use a repeated measure mixed ANOVA with eval_1 and eval_2 as dependent variables and A,B and a,b as independent?
What other options do I have?
It isn't entirely clear, but if you have some participants creating a subset images and some judges rating a subset of images it seems likely you have a cross-classified design with three fixed factors AB, ab and eval and two or more random factors (judge, creator1, creator 2, image). I'd probably try and run this as a multilevel or linear mixed model with multiple random factors rather than a repeated measures ANOVA (which would treat image or judge as random effects but not both).
Its possible some of the random effects have too small n to be estimated well and just using dummy variables for those effects might be OK. Its also possible that some of the random effects (or even fixed effects) might be aliased with others and you can't simultaneously estimate all of them. The aliasing of fixed effects could arise because its a fractional factorial design. The aliasing of random effects could arise because the random factors are confounded (i.e., the creator combinations and image id are redundant).
Hi @Thom_Baguley. I will look into linear mixed models then. thanks for your reply.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA