Summary: I was wondering what statistical test is most appropriate for my data: for each of my four experimental conditions seperate participants rated my stimuli. I have about 250 stimuli rated on multiple scales (ordinal, 10 p. likert scale) for each of these four experimental conditions.
Question: I hope to find out if the experimental conditions explain differences in prediction on these scales. can anybody recommend what test to conscider? It would be very nice to acquire some advice and possibly some resources on the fundamentals of these models and applications using python.
More details/my consciderations: At first I made the error to assume that I could just use likert scale as a continous variable, but realized this is not the case when I was plotting distribution and qq-plot and when I read more into this 'discussion'. Now I am conscidering using a Chi-square or likelihood model since one of my resources recommends this for categorical variables (A, Field 3rd edition). However, when I was reading up on possible models, I bumped into ordinal regression (something not in my books).
This all depends on the DV NOT AN IV. See the attached for full details. Best wishes David Booth
If you are considering your dependent variable to be ordinal, and you are able to use ordinal regression, then ordinal regression is probably the best approach.
David Eugene Booth Thank you for your suggestion, are you implying that based on my ordinal DV I should choose a fitting regression (e.g. Ordinal regression?).
Sal Mangiafico Also thanks for the imput, could you shortly elaborate on why the ordinal regression is the best approach in that situation?
In general any good resources on Ordinal regression? Looking for some more background, assumptions to check and if possible code examples in python.
I don't know about Python, but R has very flexible and easy ordinal regression with the ordinal package.
Yes, ordinal regression does have some assumptions.
It wasn't clear to me. But if you just have four groups, like you could use a one-way analysis of variance (anova), then a Kruskal-Wallis test will work (almost?) just as well as ordinal regression.
One reason to not use something like chi-square test of independence or regression that treats the dependent variable as nominal categorical is that these analyses don't know that e.g. 3 > 2 > 1. They just know that e.g. 3 is different from 2. But there are situations where this is what you want to know.
That's exactly what I meant with an ordinal DV
I would use ordinal logistic regression. The attached notes have a great example of ordinal logistic regression in R. Best wishes David Booth
Thank you both for your answers, you made me more certain that I found the right test for my data.
Whether you should perform a χ² test or ordinal logistic regression depends on your research question. For instance, a χ² test could check whether it is irrational to think that a person's monthly income tier is independent of their ethnicity. On the other hand, ordinal logistic regression is used to predict the monthly income tier of a person with some given characteristics (age, gender, ethnicity, education, etc.).
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python