I asked this question on Reddit a few days ago but didn't get any responses.
Working on some spatial epidemiology research a while back, that has resurfaced, I was told that I should smooth the incidence rates to account for areas with small population/high variance causing outliers (low population with 1 case may have a huge rate). I am unable to contact the professor who gave me the advice at the moment.
What I don't quite understand is:
- Even if the rate is inflated by the low population, isn't it still valid? The raw rate is what is present in the "real world," so why shouldn't it be used?
- Should I use the smoothed rates in calculations further down the road? For example, if I want to calculate spatial auto-correlation (Moran's I or somesuch), should I use the raw or smoothed rates or is rate smoothing primarily for visualization or for analytical purposes? Could rate smoothing introduce an additional spatial component to the data that may have not already been present?
