2
$\begingroup$

Possible Duplicate:
In linear regression, when is it appropriate to use the log of an independent variable instead of the actual values?

I am running a series of multiple mediation models. Each model includes one IV, two mediators, and a DV. I’m using a macro created for SPSS (provided by Preacher and Hayes) that uses bootstrapping to examine the indirect effects of the mediators. My DV’s are depression and social anxiety (each run in a separate model) and are both positively skewed. Normally I would perform a log transformation on these variables, however, bootstrapping is a nonparametric resampling procedure that does not hold the assumption of normality of the sampling distribution. Therefore, my question is: Is it necessary to transform positively skewed DV’s even though I am using bootstrapping?

Note: I have run my models with the log transformed means vs. untransformed means for the DV’s and the pattern of results are essentially the same. However, I would prefer to use the untransformed means as the unstandardized regression coefficients more strongly support my predictions.

$\endgroup$
2
$\begingroup$

In general when you are using a nonparametric bootstrap normality assumptions are not made so you don't have to worry about the distribution of residuals. If they are iid with finite variance that is sufficient. No transformations of the data is required.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.