Hi there!
I'm wondering if the assumption of normality needs to be met in order to carry out a moderation analysis in PROCESS in SPSS?
Many thanks,
Kind regards Iris
Not necessarily depending on the sample size or total number of observations to be considered in the moderation analysis.
A moderated regression model is an ordinary regression model with a product term for each moderator. For this model to be reasonable the residuals reassured to be sampled from a normal distribution of independent errors with constant variance (variance that is not a function of any predictors).
So yes the residuals need to be approximately normal for inferences in the model to be useful. As sample sizes increase this matters less and inferences will be reasonable accurate (assuming the central limit theorem applies), but it can require very large samples for this to be true if there is severe skew or kurtosis for instance.
thank you both! And what about the bootstrapping that appears to happen when carrying out the PROCESS moderation analysis? Doesn't this correct for nonnormality if that appears to be the case?
I'm not that familiar with PROCESS, but generally bootstrapping isn't necessary for moderation. It tends to be used for mediation tests in PROCESS or other models.
The issue I believe is that the sampling distribution of the indirect effect is very non-normal in mediation and bootstrapping is one way to handle this.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS