0

When estimating an ordered logit using MASS::polr is it then possible or a good idea to interpret the coefficient from the summary output or is it necessary to determine the exp() and use the intercepts to determine the probability also? If so, how can the coefficients then be interpreted? Can you give examples of when it's negative, positive, significant and not significant?

AdamO
  • 62,637
rr19
  • 65
  • 1
    Are you already familiar with these concepts in logistic regression? – AdamO Nov 18 '22 at 14:55
  • @AdamO not very good but I know that when odds ratio is less than 1 then there is a negative relation, if larger than 1 there is a positive relation if equal to 1 there is no relation – rr19 Nov 18 '22 at 21:50
  • Or this is at least how I understooth it – rr19 Nov 18 '22 at 21:51
  • 1
    Yes. And so it is with the ordered logit. The only difference with ordered logit is that the response in this case is membership in a higher polytomous category. That should answer your question. – AdamO Nov 18 '22 at 21:53
  • @AdamO Can you give an example based on your last comment? Another question: for instance would you say that if odds ratio is 1.005 (i.e. almost =1) then there is no relationship or would you interpret it as since it's larger than 1 there is a positive relationship? – rr19 Nov 22 '22 at 22:54
  • 1
    Again no need to reinvent the wheel. Small effects matter for prevalent and recurring exposures. – AdamO Nov 22 '22 at 23:08

0 Answers0