Repeated Measures ANOVA as Linear Mixed Model

Hi there,

I have a study in which one group of subjects are completing a face recognition task with an eye tracker. First, they learn a set of faces, one at a time. In the testing phase they are tested on their memory for those faces, one at a time, with some faces being targets and others distractors. During the entire experiment, we are recording their eye movements to see if they look more to the upper or lower half of the face for each type of trial (learning, target, distractor).

I would like to run a repeated measures ANOVA with log10(duration of looking time) as the DV and two IV factors: trial type (learning v target v distractor) and face half (upper v lower). More specifically, I do not want to run a "traditional" repeated measures ANOVA using the GLM. Rather, I would like to run a linear mixed model so that I can include subject as a random effect.

What I would like to know is: 1) what are the assumptions of this analysis that could potentially be violated, 2) how do I test to ensure that those assumptions are not violated in SPSS? 3) If any of the assumptions are violated, how do I correct for them?

I am unable to find any useful guides online that speak to issues of assumptions in a linear mixed model. There are so many useful guides out there for the GLM, but it seems there is nothing yet for the LMM. If I am wrong, please point me in the right direction!



Less is more. Stay pure. Stay poor.
So for vagueness, but I believe most of the assumptions parallel those of linear reg, though multilevel models (mixed) are preferred to repeated measures ANOVA because it doesn't perform listwise deletion in the presence of missing data. I am looking forward to see what you find out in regards to this thread.
Yes, and it doesn't use up as many degrees of freedom. I too am looking forward to see what I can find as the information out there on the WWW seems to be a bit mixed (no pun intended). I found one source that said that if your assumptions for a GLM ANOVA are violated, then you can run the mixed model. But surely it is not the case that the mixed model has NO assumptions!