Dear Colleagues,
Can anybody suggest how to include ordinal multilevel variable (e.g. Likert scale in psychometric studies) as an explanatory variable in Cox proportional hazards model?
1) Should I change the variable to a few dummy variables? This idea is most appropriate, but it multiplies the number of explanatory variables and may result in model overfit.
2) Or should I rather keep one ordinal variable with multiple levels? This solution holds assumption that the distances between levels of ordinal variable are constant, which is actually not true for ordinal characteristic of a variable.
What to do then?
Thank you for your help!
Hi Michal,
How many categories are there in the likert scale? If it is not too much (for example: 4 or even 5 categories), I don't think you have a problem of overfit. Of course, it is in relation with the number of observation you have.
Dear,
Use a single-order variable with multiple levels. Each level will be a separate group. This will allow you to compare the differences in the lifespan of each group. Also use the Kaplan-Mayer test.
Good luck
M.Kaneva
Dear Mariana Kaneva,
This is a 5 categories Likert scale. Do you mean to compare level 1 with 2, 2 with 3, ... , 4 with 5 in Cox regression and/or log-rank test to see if there are similarities in RR between the levels?
Thank you,
Michal
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression