I have come across Kruskal-Wallis, chisquare (would like to avoid discretising my continuous data if possible though) and logistic regression. However, to avoid excessive time consumption, I would like to compare all independents to the dependent simultaneously and not to each other. Is this possible?
Why exactly go for non-parametric? What about box-cox power transformations, polynomials, etc...?
Hi, Sam.
If I understand correctly, you have a continuous (non-normal) DV and multiple IV expressed as dichotomous. In my opinion, no matter if your DV is far from normality. You can develop a multiple linear regression, checking the assumptions:
- Linearity (if you have problems here, try box-cox transformations)
- Non heteroskedasticity
- (very important) Normality of standardized residuals (close to values of mean=0, SD=1)
- Durbin-Watson between 1.5 and 2.5
- Non multicollinearity
If your regression model pass these proofs, it will be appropriate for your research.
Regards.
- Badges
- Science topic