The goal is to run non-parametric general linear model with no restrictions on the design (e.g. not limited to one-way ANOVA as in Kruskal-Wallis test; quantitative factors should be admissible, etc) and to be able to estimate contrast p-values. Each observation will be converted into rank with no ties.
I saw a few warnings about converting a continuous dependent variable into ordinal response, so I need to clarify: is the problem only that a few distinct original observations can fall into the same category and information is lost? If yes, then do I get it right that everything is fine in my case because I will have only one observation per category?
Also, for the people who tried it in practice, am I likely to get very different results between parametric ANOVA and ordinal regression in case when ANOVA assumptions hold?
