Hello,

I have a dataset of 15 subjects. Each subject has 12 scores that come from performing 7 tasks, so I have a total of 180 scores.

The study design is observational and the subjects do not have the same mix of tasks (e.g., some perform task #1 three times then a combination of the other tasks, while another subject may never performed task #1 but performs a combination of the other tasks). Each task has a continous score associated with it and all task scores are on the same scale. Scores are actually how much radiation they were exposed to during the task and there is no learning effect or cumulative dose, each task performed is assumed independent.

I wanted to look at predictors of scores (my continuous dependent variable). Subjects have traits that may affect scores and certain tasks have risk for higher scores. Plus I have a bunch of other variables I can examine as well. I first analyzed these data with multiple linear regression with subject (as a category) and task (as a category) both significantly predictiing scores (plus a significant interaction term). So, certain tasks and subjects have a higher risk for scores and certain subjects performing certain tasks have a higher risk for scores.

My hesitation is that I am not directly controlling for random effects of subjects or tasks in a hierarchical model. If I wanted to do this would I add one or two levels and could I still look at the interaction between them? As of yet, there are no other significant prediction variables in the model. So the model would have two random effects and no predictors, so how do I evaluate the levels' influence on scores? I invite an open dialogue to try to see what may be the best approach.