I have a Chi-Sq test, which is 2 x 4. I would like to try to fit another variable. Is there any such thing as Chi-cubed? I would like 2 x 4 x 3. My 2 is the dependent variable, do i have to group the others to make 2 x 12? The groups are categorical. Should I collapse the groups or is there another suitable test?
-
What are your variables? What are your null and alternative hypotheses? – Dave Apr 16 '20 at 15:37
-
2There is no chi-cubed. You can get a chi-square test for a three-way table. At some point with a binary dependent variable logit regression becomes easier and more helpful. – Nick Cox Apr 16 '20 at 15:42
-
Just remembered this: https://stats.stackexchange.com/a/148174/247274 – Dave Apr 17 '20 at 23:49
1 Answers
The chi-square test can be used in any situation when you want to see if the observed numbers of cases among a set of categories differ significantly from the numbers expected, often from the numbers expected based on some null hypothesis. So you certainly could apply a chi-square test to your 2 X 3 X 4 contingency table. You will have 24 possible categories overall; the shape in which you organize them doesn't matter except perhaps as an aid for calculating the numbers of cases expected in each category under the null hypothesis.
The chi-square test in that case, however, would not take into account the different roles that you have assigned to your categorical variables: the 2-level variable is an outcome while the others are considered independent variables whose values might be predictors of that outcome. In that case a logistic regression would be much more in line with what you are trying to accomplish. That would specify the 2-level variable as the outcome while incorporating the other 2 variables (and, if you wish, an interaction between them) explicitly as predictors.

- 57,766
- 7
- 66
- 187