Watch this video for step-by-step procedure to perform Non-parametric (Quade’s) ANCOVA,

https://youtu.be/jozx59H5IHw

Quade's test assumes a randomized complete block design. The test itself is based on computing ranks for range of the data in each block. The implicit assumption there is that only the relative magnitudes of the ranges within each block are important, and not the absolute magnitudes. That is, it only matters that block A has a larger range than block B, but it doesn't matter to what degree it is larger.

That said, I probably wouldn't use Quade's test. It isn't a very common procedure, and I suspect it isn't particularly powerful (like many rank-based testing procedures are, including non-parametric ANOVA/ANCOVA). You don't give enough information in your post to really understand the data or the hypothesis of interest, so it's difficult to recommend an alternative. What is your data? What groups are you comparing, and what comparison are you trying to make?

Thanks for the answer! My independent variable is reading enjoyment, measured on 5-point scale and I'm trying to compare it by region and control the effect of gender. The data is heavily skewed and ordinal so I don't think I can justify using a parametric test. What alternative would you recommend?

Anne,

I've used Quade's before. I agree with Ryan about the lack of power but it wasn't a problem for me. You could try transforming your reading enjoyment variable but if that doesn't work then you have to use a non parametric test.  It is probably out of date and there may be better alternatives within R but the great thing about Quade's is that it is fairly simple to implement using SPSS. It is also fairly intuitive, another reason I like it.

Thanks Mario! I don't think that the lack of power will be an issue with my data so I will go with Quade for now as it is indeed fairly simple to implement with SPSS. Just to clarify my previous answer, I also have variables measured on 4-point scale, the region variable has 3 levels and I have fairly large sample of 6356 participants.

I'm not sure what is implemented in SPSS (just Quade test or the entire ANCOVA procedure with post-hoc)

but Quade's ANCOVA is exactly ANCOVA on ranks (ANOVA on residuals got from regression on ranked DV vs ranked IVs). Personally, I don't use anything using classic ranks, I prefer the Aligned Rank Transform (ART), which is a well recognized method for smaller samples. But there is much better method for handling raw Likert, that is to say, ordinal response (rather than a combined score) which is acclaimed and widely used, at least in pharma - the ordinal logistic regression. This gives also the advantage of employment of random effects, if necessary - repeated measures (longitudinal trials), clustering.

Watch this video for step-by-step procedure to perform Non-parametric (Quade’s) ANCOVA,

https://youtu.be/jozx59H5IHw