I am analysing variables that are not normally distributed and I use PROCESS to predict DV through the mediator variable. Are there any options with non-parametric tests or is it advisable that I still use PROCESS?
I assume you're talking about Andrew Hayes' macro. If you're using bootstrapped standard error estimates, you should be okay in most circumstances, but I'd want to hear more about what your variables are and how they are distributed.
Thanks for your answer. Right, I mean Hayes' macro. I have a scale predictor and a scale mediator without curtosis or skewness. My DV is negatively skewed. Considering the subscales of DV, they are also negatively skewed.
I'm not familiar with the details of the syntax of the PROCESS macro, but if you use bootstrapped standard errors, you'll be fine. Also, recognize that what you care about in a regression context is a normal distribution of the residuals, not the variables. And even that isn't very important as long as the errors are identically distributed and your interest is primarily on the slope rather than the intercept of the line.
Shahrzad, normality of data is not the assumption of regression. Do the parametric easily.
- Badges
- Science topic