# Regression on non-normal DV + Heteroskedasticity

I have a DV scored from 1 to 3 from a questionnaire: respondents were given an integer score according to their responses. I need to compare the mean scores between two groups using a regression (restricted to regression since I am using Stata with survey data). The mean scores are, of course, not integers (not sure if this makes a difference). The DV is also not normally distributed and shows heteroskedasticity. This, I assume, means I cannot use a simple linear regression considering the assumptions have been violated. I do have independence between groups.

Any suggestions for which type of regression would best fit would be super helpful. I have considered a logit but the DV is neither ordinal nor nominal.

• The question seems contradictory. If you have original integer scores, then that sounds either nominal or ordinal to me depending on the rules, and in the absence of information I'd guess ordinal. But then working out means is a matter of convention at best. It seems to me that ordinal logit is your first port of all and that allows comparison between groups. – Nick Cox Dec 14 '16 at 19:53
• Also, you say more than once "the DV" as if there really is just one. But you also seem to imply that you are averaging (e.g. across answers to different questions). So, this really isn't clear to me. Concrete data examples would surely help. – Nick Cox Dec 14 '16 at 19:55
• Normality of response is not a requirement for regression, but I think most practitioners would consider it a real stretch to apply it to a response coded 1, 2, 3. – Nick Cox Dec 14 '16 at 20:06
• Apologies, the scores would be ordinal. There are multiple DV's but for the purpose of this particular analysis there is one DV (score) for two populations, those with a particular disorder and those without. I must compare those populations in terms of their mean scores. I have considered an ordinal logit - would that be your suggestion? Thanks for the help – JakeP Dec 14 '16 at 20:50
• Yes, on this information. – Nick Cox Dec 14 '16 at 21:06