1

I am currently analysing some data including the Autism Quotient, which is a continuous measure. I plan to analyse via hierarchical regression. I have assessed normality statistically and both skew and kurtosis are within acceptable ranges. However, looking at the histogram, the distribution appears bimodal.

My question is: what can I do with this to make it permissible to analyses via multiple regression?

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
Caitlin
  • 11
  • 4
  • 2
    It already is permissible - there are no assumptions in regression about the distribution of independent variables (predictors). See [Should quantitative predictors be transformed to be normally distributed?](http://stats.stackexchange.com/q/12715/17230) & [What is a complete list of the usual assumptions for linear regression?](http://stats.stackexchange.com/q/16381/17230). – Scortchi - Reinstate Monica Oct 05 '15 at 10:46
  • 1
    Echoing @Scortchi, note that every indicator or dummy variable with two distinct values could be considered bimodal, and there is no problem about using such variables: it's standard technique. (Indicator variables are more generally a counterexample to the extraordinary myth that predictors should be normally distributed.) – Nick Cox Oct 05 '15 at 10:54
  • While the above comments are correct, a bimodal IV might have issues with linearity of the relationship with the DV. – Peter Flom Oct 05 '15 at 10:57
  • 1
    @PeterFlom: Well, so might a unimodal IV. I suppose the trouble with a bivariate IV is you might not have many observations to check a linear, or any other, relationship in between the two modes. – Scortchi - Reinstate Monica Oct 05 '15 at 11:35

0 Answers0