Hi everyone, I rarely create threads but I know some people on here are quite gifted in statistical knowledge and was hoping someone could help me out. Anyways, the statistics I'm doing involves examining the relationship between two ordinal variables. Now, most of the time, many would suggest just using a chi-square test (or logistic regression) to determine a relationship between the two. However, the variables of interest are greater than a 2x2 (think 6x6) and I want to do some type of post-hoc pair-wise comparisons between the categories. Forgetting about experiment-wise (or family-wise) error for the time being, does anyone know how to conduct such post-hoc tests on two ordinal variables either in SAS or SPSS? Also, the standardized residuals provided by the chi-square test in the overall analysis isn't quite as useful (unless I'm reading it wrong) because they only provide significance of a particular category in the ordinal variable instead of a pair-wise comparison (which is what I need). Any ideas guys? Thanks!
i have 2 responses to this thread, but i'll put them in one post and pass the savings on to you: 1. paging all baseball fans who've never played an inning of baseball; 2. you doing that all 220?
sigh, something like three months ago I could have really helped you for this, but after many beers I have forgotten most of everything...so sorry if this is completely out of left field. Have you thought of using the Spearman method?
Hey Northside Storm, thanks for the reply. The Sperman method or Spearman's rho is much like the Pearson r or Kendau's tau. They all provide some form of a correlation coefficient with corrections. Unfortunately, I don't simply need to assess the relationship between the two variables, but also examine the levels with each variable against another (hence pair-wise comparison). Thanks for the try though...
Normally you would want to do a one-way ANOVA to find out if there is a difference in your categories. However, since you just want to do the post-hoc multiple comparisons you can perform one of the many tests available. I can't remember the particulars but Fisher's LSD, Tukey's HSD, Student-Newman-Keuls are just some of the pairwise comparison tests out there. I know both SAS and SPSS can perform those comparisons. I think in SPSS click Analyze --> Compare Means --> One Way ANOVA. Then there should be something about Post-Hoc. That should take you to the different pairwise comparison tests.
I'm pretty sure Jet Blast is right. Tukey's test, from what I remember, is regarded as the best post-hoc for pairwise comparisons. In addition to the ones he mentioned, I think you can use Scheffe and Bonferroni post-hoc multiple comparisons in SPSS as well. Just follow Jet Blast's directions and it should give you all the tests that are available for comparisons in SPSS.
Thanks for the input guys. Unfortunately, One-Way ANOVAs are used to compare group means between one independent variable with greater than two levels using numerical data. Furthermore, it's one continuous DV (hence numerical data) and a categorical IV (like race). My situation is that I have the DV as a categorical variable and the IV as a categorical variable (both ordinal). While using any of the corrections for pair-wise comparisons is fine (e.g., Bonferroni, Tukey's, etc.), I have no idea how to do so with two categorical variables. Running a significance test b/w two categorical variables has never been the problem - it's the post-hoc tests that is.
I thought Jet Blast was on the right track with his post, but this may be a little over my head. I don't have much experience comparing a DV and IV that are both categorical/ordinal level data. I forgot you mentioned that the variables are greater than 2x2. The only other thing I can think of is doing a Kruskal-Wallis test. I think Kruskal-Wallis is used for ordinal variables with ranks of 3 or greater. It might work, but I'm not sure. In SPSS you would click Analyze > Non-Parametric Tests > K Independent. Then just check Kruskal-Wallis and insert the variables into the test variables category.
