Timeline for Non-parametric for two-way ANOVA (3x3)
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 13, 2015 at 23:21 | comment | added | gung - Reinstate Monica |
@toto_tico, I just answered you on the other thread. You need to use clmm() in the ordinal package.
|
|
| Nov 13, 2015 at 23:18 | comment | added | toto_tico | @gung, what about the within design part of the problem? Is it possible with an ordinal logistic regression? | |
| Oct 29, 2015 at 15:58 | comment | added | gung - Reinstate Monica | @RyanSimmons, you needn't necessarily classify anything. The test of the OLR model is a test if some groups are associated with typically larger Y values than other groups. You are simply using only the ordinal component of the Y values. Tests like Kruskal-Wallis, Mann-Whitney, etc. are special cases of OLR. | |
| Oct 29, 2015 at 15:40 | comment | added | Ryan Simmons | Essentially, I think the questions "Are these groups different according to X?" and "How can X be used to classify individuals among these groups?" will only be the same in very specific circumstances, and that you can't necessarily substitute ANOVA for logistic regression and get comparable results. | |
| Oct 29, 2015 at 15:37 | comment | added | Ryan Simmons | Isn't this approach answering a different question, though? In the ANOVA setting, the continuous variable is the outcome, and we are testing the sums of squares within/between factors. In an ordinal logistic regression, you are trying to predict factor membership according to the values of that continuous variable. While there are certainly situations where these will give you similar inference, it isn't clear to me that these are equivalent procedures in GENERAL. In the same way that the regression of Y=BX is not always going to be the same as X=BY. | |
| Jul 25, 2013 at 4:36 | history | answered | gung - Reinstate Monica | CC BY-SA 3.0 |