0
$\begingroup$

The point of the last analysis in my paper was to check on the basis of which predictor variables the answers to moral dilemmas can be explained. Predictor variables are continuous: dark personality triad (Machiavellianism, narcissism and psychopathy), empathy (cognitive and affective) and thinking style (rational and experiential) The criterion variable is the answers to moral dilemmas that can be utilitarian or deontological. It is also continuous, but has only 6 degrees. The tendency towards lower numbers means utilitarianism, and towards higher ones deontologism.

My original idea was to perform a hierarchical regression in two blocks. However, I got quite pronounced heteroskedasticity. However, I got 6 lines, which somehow justifies such a plot given the characteristics of the dependent variable. The direction of regression and distribution seem normal.

I would ask for help if someone can justify to me what would be best to do. Is it to do White residual correction, ordinal logistic regression or is there a justification to proceed with the hierarchical regardless of heteroskedasticity, given the characteristics of the criterion variable and how then to best justify it?

| cite | improve this question | |
$\endgroup$
  • $\begingroup$ Heteroscedasticity can be assumed irrelevant when the only objective is prediction, pure and simple. $\endgroup$ – user332577 Nov 19 at 13:23
  • $\begingroup$ Hey, thank you for your answer. Can you elaborate that a bit more? $\endgroup$ – LillyAnn Nov 19 at 16:55
  • $\begingroup$ Leo Breiman's paper on the Two Cultures of prediction vs understanding is a good place to start. The classic approach to modeling, even before any data is gathered as would be done wrt many PhD dissertations, assumes a careful process of hypothesis generation leading to insight and understanding based on strict observance of all the assumptions underlying modeling. Prediction, on the other hand, relaxes most of those classic modeling assumptions. Call it data mining but automatic, algorithmic methods break the assumptions and rules with the goal of increasingly accurate predictions. $\endgroup$ – user332577 Nov 19 at 17:40
  • $\begingroup$ Okey, I will definitely check that out. Thank you! So in essence you are actually proposing that I don't do any transformation of residuals or what not but just continue with linear regression? $\endgroup$ – LillyAnn Nov 20 at 9:37
  • $\begingroup$ What is your goal? $\endgroup$ – kjetil b halvorsen♦ Nov 20 at 16:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.