Nominal/ordinal Variable Measures of Association I have an ordinal variable (employee performance ratings on 5-point scale) and I want to do a crosstab with a nominal variable (4 levels of race/ethnicity).  What is the best measure of association for these two variables in SPSS?
 A: To expand on RSK's second point, another option would be to apply a non-parametric test which would account for the ordinal nature of the performance variable -- this would be a Kruskal-Wallis test (see Wikipedia article)
, which is a non-parametric analog to a one-way ANOVA.
(You can do this in SPSS 19 under "Non-parametric tests"/"Independent samples". (performance as the independent variable, so chosen under the "Test Fields" list in SPSS 19, and the race/ethnicity variable in the "Groups" list. -- in older versions of SPSS you select the "K independent samples" option under non-parametric tests.)
It is important to note that:
1) As per most non-parametric tests, you are only looking at the hypothesis testing angle (rather than parameter estimation) and so you don't get succinct summaries like odds ratios;
2) As per one-way ANOVA with more than two groups, because your grouping variable has four levels, the p-value from the test is an omnibus result -- that is, a significant result indicates that at least one of the groups is different from at least one of the other groups. To determine which groups are different, you would then need to run some post-hoc tests (e.g. compare all pairs of groups with a Mann-Whitney-Wilcoxon test, which is a non-parametric analog to an independent samples t-test.) If you do this, you may want to consider adjusting the p-values for multiple comparisons (e.g. with a Bonferroni correction) to avoid type I error inflation.
Another option would be to apply ordinal logistic regression to the data analysis: this would give you a very similar (if not identical) p-value about the prinicipal hypothesis, as well as summaries of individual group estimates in the form of odds ratios.
A: I'm not sure about SPSS, but any test on contingency tables (e.g., Fisher's) can help. Alternatively, you can assume that the ordinal variable is numerical and use nonparametric alternatives to ANOVA maybe ...
