With self-reported health scores (as a dependent variable) - such as CES-D for depression which typically ranges from 0 to 60, I was wondering whether log-transforming these scores would allow me to interpret their coefficient as a % change.
My point is that although these scores provide valuable information, it is often unclear what "12 CES-D points" of change would practically mean. In addition, since they are not interval variables, often scores tend to be concentrated within particular range.
Conversely, I thought that log-transforming these dependent variables would allow me to partially flatten their distribution and - more importantly - interpret their change more intuitively. For instance, while it is unclear what 7 point CES-D decrease would mean, 10% decrease in CES-D score can perhaps be construed as similar decrease in their depression symptom.
I believe there are few articles tackling with these transformation issues, but it seems that there no apparent reason not to perform analysis this way.
Any ideas?!
