Can Chi-Square Tests Help Pick a Contest Winner?
Though “balance” equal sample sizes for groups (or “treatments” in medical research and “factors” and “levels” in experiments) is a desirable property, it isn’t a necessary one. It’s apropos to use the 331 press releases for candidate A and the 47 for candidate B. See below for additional reasoning on why this is so. Using Spearman’s rank is an interesting choice for the correlation examination. Perhaps using the raw numbers as ordered pairs would have significant non-normality and the typical correlation coefficient (the Pearson Product Moment) wouldn’t be appropriate. So using Spearman’s rank to check correlation is probably fine. It would indicate a correlation, but not “prove” whether the candidates agreed. The Chi-Squared test on the rankings seems a bit strange though. Why use the rankings and not on the raw counts and proportions? The two typical ways of handling it are: • Perform an overall chi-square test on the proportions for all topics considered together. Significance indi
