01-16-2015 11:49 AM
I've been asked to analyze the results of a 'longitudinal' survey that examines the subjects' knowledge about certain safety practices at three time-points: before an intervention, a few days after the intervention and several months after the intervention. I put the term longitudinal in quotes because, while the same people are taking the survey at each point, the survey designers don't individual identifiers and therefore cannot track each individual's responses over time.
My first inclination was to simply run categorical comparisons (chi-square/fisher's), but those comparisons assume that the three 'groups' of answers (before, short-term after, long-term after) are independent of each-other. Clearly this is not the case, but I'm unsure of the best way to correct for this, or if such a correction is even possible.
Any suggestions on how to proceed? Am I over thinking this?