Posted By:
GlennRich,
My understanding is this. You are correct that chi-square analyses are usually for non-parametric data, but the test should be robust to violations of this standard for such data as we're analyzing here. I am treating the data as ordinal level only, thus I opted for chi-square (which technically would be more appropriate for nominal data, but I don't believe this presents a problem -- let me know if you have reason to believe otherwise). Regression and/or Pearson's correlations are problematic because the predictor (series) is not a continuous variable. You could get more Power from a Student's t-test, using series as your grouping variable, but that requires your criterion data (grade) to be interval level (which is a bit of a stretch) or ratio level (which is more than a bit of a stretch).
I believe the Mann-Whitney test could be used, and that may be a very good suggestion. Naturally we could use a larger sample of T206s (like every one that SGC has ever graded).
It may take me a while to respond, but email me if you have any other stats questions.
Glenn