I have a DV which is an ordinal variable with 3 categories (Low, Moderate, High) and 2 IV which are categorical. I.e. Group: CON vs EXP and Time: Pre vs Post.
At first, I thought to perform a Wilcoxon rank test to asses the pre-post differences and a Mann-Whitney U to asses the between-group differences but, as expected, there are a lot of tied values.
Then I thought to perform McNemar test for each group separately, it would be correct?
Other option that I thought is to change my dataset to long format and perform an Ordinal Logistic regresión or a Generalised Linear Model selecting a Multinomianl categorical dependent variables option. It would be possible?
Is there any tutorial to follow?
Any comments or suggestions are welcome. Thanks in advance.
Analyse an ordinal DV with 3 categories
-
- Posts: 16
- Joined: Wed May 29, 2019 11:39 pm
- mcfanda@gmail.com
- Posts: 460
- Joined: Thu Mar 23, 2017 9:24 pm
Re: Analyse an ordinal DV with 3 categories
you may take a look at this, https://gamlj.github.io/gzlm_example1.html, it might help
-
- Posts: 16
- Joined: Wed May 29, 2019 11:39 pm
Re: Analyse an ordinal DV with 3 categories
Thanks for the response, I have read the example and performed a Generalised Linea Model with my own data but I'm not sure to interpreting it correctly. Please, see the attached image.
The loglikelihood ratio tests indicated that the probabilities of belonging to a risk category are substantially different across the time points of the intervention but not in Group factor nor in Group * Time interaction. However, at the parameter estimates table, it is not a significant value.
How is it possible? It means that the probability of High to Low is not related with any factor? The same with the probability of Moderate to Low
The plots seem to express another thing.
Thanks in advance.
The loglikelihood ratio tests indicated that the probabilities of belonging to a risk category are substantially different across the time points of the intervention but not in Group factor nor in Group * Time interaction. However, at the parameter estimates table, it is not a significant value.
How is it possible? It means that the probability of High to Low is not related with any factor? The same with the probability of Moderate to Low
The plots seem to express another thing.
Thanks in advance.