Why the number of response categories in rating scales should be large