I’d like to do some rank correlations on some data. Unfortunately, the data is such that every single point will end up in its own rank. An arbitrary “significant digit” cutoff is rather–arbitrary. Likewise, arbitrary “high, medium, low” categories are–arbitrary. Is it valid to first run the data through a multiple range test and use the categories as “ranks” (A=1, B=2, etc.)?
If so, what would be a valid way to generate “ranks” for samples that might be in categories like “AB”, “BC”, “CDE”, etc? Would the harmonic mean of A=1, B=2, etc. be acceptable in such a case?