0

I need to investigate whether age has a bearing on the disease. I have 4 different measures of the disease which I have categorized by levels and in the same way I have categorized the ages. What regression model can I use?

kjetil b halvorsen
  • 77,844
hans
  • 23
  • Welcome to the site! Your question however is not very specific. Since your response is categorical use any categorical regression: multinomial logit, trees (bagging/boosting), LDA/QDA etc. The fact that your regressors are also categorical doesn't matter, you can one-hot encode them. Googling models for categorical response should give you required answers. – PaulG Mar 05 '21 at 10:50
  • See https://stats.stackexchange.com/questions/68834/what-is-the-benefit-of-breaking-up-a-continuous-predictor-variable – kjetil b halvorsen Mar 05 '21 at 17:49
  • I agree that you should question why you are breaking a continuous variable like age into categories (see the question in the question linked by @kjetil above).

    As far as your disease measure: Is there a reason to keep the 4 measures separate? Maybe factor analyze and create a scale or just combine based on theoretical issues (you should check and report the internal consistency reliability and probably also check for unidimensionality).

     – Steve Scher Mar 05 '21 at 19:37

1 Answers1

0

The first question you should ask yourself is whether the outcome variable is nominal (no natural ordering of the categories as in "red","blue","green") or ordinal (with natural ordering as in "low","moderate","high"). If the outcome variable is nominal, a commonly used model is the multinomial logistic; if ordinal - there are several different models you could use, one of them being the proportional odds model. For the proportional odds model, you'll want to think about whether the proportional odds assumption is reasonable; if it is not, you'll want to look for another model type that would be more aligned with the problem you're studying.

ColorStatistics
  • 4,724