What kind of data you have? Continuous, categorical? Your dependent variable? Regression analyses requires residuals to be normal, not the data itself.
My independent are age and sex(continuous and discrete respectively), and my dependent is empathy (continuous I reckon).
Hi Cady,
Yes, more details about the type of data would help.
But assuming uou have continous predictors and outcome, you can consider
- Data transformation if.your outcome is positively or negatively skew
- Mild violation of normality assumption might still enable you to use.OLS regression if your have large sample size
- You can always consider non-parametric tests but the downside is that you cannot control for other vaiables as MLR enables you to.
Thanks, I see. I have 169 as a sample size. I wonder if that would be enough for OLS regression?
If the residuals are skewed or heavy-tailed, one option is to model the errors in the regression using some (non-Gaussian) distribution that captures the features in your data (such as skewness) and then fit the model using maximum likelihood.
If you use OLS, you implicitly assume normality of the errors in the regression. OLS may of course still give reasonable results even if the errors are not Gaussian, but one cannot say that it is OK to use in general just because the dataset is of a certain size.
HI Cady, just in case you need a citation, here's the section you can cite: 4.3.1 in the book below, which is widely cited in social science.
https://scholar.google.com/citations?view_op=view_citation&hl=en&user=SkMe3SEAAAAJ&citation_for_view=SkMe3SEAAAAJ:_B80troHkn4C
basically it's what I said in my second point in my earlier post. Thanks!
- Badges
- Science topic
- Similar topics
- Psychology