Due to problems with non-normal residuals, I have run a rank-based ANOVA on my outcome variable (continuous). My indepedent variable is binary (Yes/No). I have 1 continous covariate in the model.
When I look at median values for each group - they are quite different (10 vs. 97) - this is a score outcome. IQR is quite large as well.
When I run my rank based ANOVA - the p value for the group is highly insignificant- both in the crude and adjusted models. How is this possible? The medians are so different.