I'm studying the effect of an ordinal variable (eating healthy food [Rarely, Sometimes, Mostly, Always] as an example) on a binary outcome (Smoking [Yes, No] for example)
My question is, which category should I choose as a reference [Rarely or sometimes ...]?
Because the odds ratios will differ accordingly, so which one should be the starting point?
Thank you!
It only matters for interpretation. If you choose "Sometimes" as reference, one of the the ORs will be that for "Rarely vs. Sometimes", call it x. If you choose "Rarely" as reference, one of the ORs will be "Sometimes vs. rarely", call it y. Now, y = 1/x, or x = 1/y. So it's easy to convert from one to the other.
Generally, I would always take one of the most extreme categories as reference, and here I would use the lowest (i.e.: "Rarely"), what for my head is easiest to understand. If the outcome would be negatively correlated to the amount of healthy food consumption, the ORs would by 1 (the logORs would be positive).
Now if you had choosen "Sometimes" as reference, the negative correlation would give an OR > 1 for "Rare vs Sometimes" but
Jochen Wilhelm
You're the hero of ResearchGate, I really look up to you!
And your answers are a huge source of inspiration.
Thanks a lot!
As Jochen Wilhelm said, the most extreme categories as reference does the interpretation easier for the reader.
I agree with Jochen Wilhelm but note that it can be helpful for ease of interpretation to choose the middle category if there is one. The model is then pseudo-centred around a typical response.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis