I have a sample of participants tested on a two-day memory task where they learn the task on day 1, then are tested on day 2. They are Parkinson's disease patients and are tested on or off their medication on either day (4 conditions: on-on, on-off, off-on, off-off), and patients are tested in all 4 conditions in a randomised order. I also have healthy age matched controls tested once. I am looking to see if medication affects either learning (i.e. day 1 effect) or testing (day 2) on a memory test given on day 2, and have previously used a 3-way repeated measures ANOVA to test this (choiceday 1 medication stateday 2 medication state). Here choice is a measure from the test which isn't too important here, but there are two scores from the test - % times card A was chosen, and % times card B was chosen so choice is a factor too. The ANOVA was fine, but I discovered that for this test you are meant to filter the data so that you only include the conditions where patients scored above 50% on one of the measures on the test (as <50% on this measure suggests they haven't learned the task properly). Doing this removes about 20% of my data in a fairly random manner, and means that when I go to do a repeated ANOVA in SPSS the listwise deletion SPSS uses reduces my sample size by half (as half of the participants have at least one condition's data missing).
How can I do a 3-way repeated measures ANOVA not using listwise deletion? (I'd prefer SPSS but matlab or R would be OK too).
I can't just impute the missing data as they were excluded based on performance so the imputed values would be just as bad as the true ones that were deleted.
Any help would be greatly appreciated. thanks