predicting with interval data with a limited number of levels.

noetsi

No cake for spunky
#1
"To visualize how receipt of key services was associated with employment outcomes, we calculated the rate of employment at VR exit for each group based on a count of key services received (zero to four services)." So they have a predictor that ranges from 0-4. This is not interval data, it is bounded on both sides, but it is not really likert scale data either. I am not sure about the validity of using a predictor this way (essentially its a count). I don't think it makes a lot of sense to convert this to a set of dummies.

Can you use a 4 level count this way in linear regression (they actually used it for logistic regression). I know there are no distributional assumptions about predictors, but I have heard it argued that it can create various problems to treat this as interval data. This is an issue I have wondered about for years.