Hey all,
I'm using the Herek ATLG. Attitudes toward homosexuality does not distributed normally (positive skew). I have two independent variables with 2 levels each and I want to know the differences between them, including an interaction effect and also 3 covariates variables. I can't use a 2 way anova.
I'm a bit lost because a log transformation didn't help either.
Thank you very much.
How did you test for normality? For each combination of the IVs separately? Which method did you use?
Oz Hamzani,
Rainer Duesing brings up good points. Your response variable is on a 5-point or 7-point Likert scale, I assume?
I typically suggest fitting the parametric model and plotting and, or testing the model residuals for normality (this helps address issues Rainer identified). General linear models (e.g., ANOVA) are pretty robust to deviations from Normality, so you may still get reliable results with slight to moderate violations. Do remember that different assumptions will have greater or lesser influence on the results and interpretation. The way it was explained to me once: independence > linearity ≥ homogeneity > Normal distribution.
However, if you decide that the assumptions are too greatly violated you have a few options. (1) Transformation; (2) You might be able to find a suitable error distribution (e.g., Poisson) and fit a generalized linear model; (3) Randomization/Permutation tests; (4) Non-parametric approaches (e.g., Scheirer-Ray-Hare, Friedman's).
Keep us updated.
Caleb
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS