I'm trying to learn ANOVA and see how it applies to a certain experiment. In the experiment we are trying to predict if a certain brain signal is predictive of whether a mouse would do the correct or wrong action. For simplicitly, let us assume that the signal is a scalar real number.
So, there are two classes of interest: correct and mistake. There are also several nuisance parameters:
- Experiment is performed on multiple mice. There is variance among mice
- Each mouse performs a task, say, 100 times each day for 10 days. For each mouse there may be variances among days.
- Mice actually perform one of two different tasks, call them Task1 and Task2. The tasks are very similar, as for both there is a correct and wrong solution. While it is an interesting parameter, for this particular question the task type is a nuisance parameter
So, the goal is to check if the signal is predictive of the classes (Correct/Mistake), excluding the effects of categories Mouse, Day and Task.
After reading on ANOVA, I learned that in case there is one nuisance category, one can perform blocking over that category, and exclude it by means of procedure called Two-Way ANOVA.
- If I have multiple nuisance categories, what is a good way to exclude their effect? Should I perform blocking over every combination of the categories?
- The number of days and number of tasks per day varies across mice. Is this sort of variability a problem for ANOVA?