I recently ran an experiment where participants had to rate how different pairs of melodies sounded using a 5 point likert scale. Participants in three age groups rated four different kinds of trials, including no change trials. The problem is that the youngest group of participants had higher ratings in the no change condition. We need to control for this somehow. As we have it now, we are dividing scores in the other three conditions by the no change score. Is this statistically kosher? Our reviewers questioned this, asking why we chose to do it this way (rather than simply using difference scores or using z-scores).
Any advice would be much appreciated. Especially any solid responses we can make to the reviewers! Thanks!
Advertise on Talk Stats