8
$\begingroup$

I have a 3-category ordered outcome (food consumption: 1=no food, 2=less food, 3=more food) and a 3-category ordered predictor (food exposure: 3=no time, 2= less time, 1= more time- whereby 3=no time is taken as reference category in the ordinal regression model). I want to explore hypothesis that more food exposure is associated to more food consumption.

I want to know how I can interpret odds ratios less than 1. For example, I have OR= 0.62 for predictor category 2= less time. I have calculated OR as exp(coeff) in Excel, whereby OR of reference category no time is exp(0)=1.

$\endgroup$
2
  • $\begingroup$ So you ran an ordinal logistic regression & exp(coeff) is <1? What software did you use to get these? Can you provide the model output? Usually, you would get a coefficient for your predictor + a set of threshold values. Which coef is this? $\endgroup$ Dec 2, 2013 at 16:15
  • 1
    $\begingroup$ I used SPSS 19. I can't add a photo or file as it seems, but the coefficients I calculate OR for are those for the predictor categories (the one I gave here is for predictor category 2 (less time spent), which is -0.474, so OR=exp(-0.474)=0.622). The coeff for category 1 (more time) is 0.111 (so OR = 1.117), and that for category 3 (no time=reference category) is coeff=0, so OR=1. This model is adjusted for gender, ethnicity, age. Outcome (food consumption) is from 1= lowest to 3= highest consumption, so that I can interpret it as moving to higher levels of consumption $\endgroup$
    – Andreea
    Dec 2, 2013 at 16:39

1 Answer 1

-1
$\begingroup$

Given the coeffcient is significant, it means that the cummulative odds for being in a higher food category are .62 times as high for people with less time than for people with no time. Put differently, having more time than no time decreases the odds (and also the probability) for consuming more food (across all categories of your dependent variable). I do not know whether this makes any sense theoretically.

Given that the coeffcient for category 1 is positive (OR>1), this suggests a nonlinear relationship across the categories of the iV. That is, the interpretation for catgory 1 is opposite of that for category 2, given the coefficient is significant.

$\endgroup$
4
  • $\begingroup$ can I also say that the odds of consuming more food is 37.8% less likely than that of consuming less food? $\endgroup$
    – Andreea
    Dec 3, 2013 at 14:02
  • $\begingroup$ It means that, regardless of whether it is significant or not. $\endgroup$
    – Peter Flom
    Jun 6, 2014 at 10:48
  • 1
    $\begingroup$ The odds are .62 as high (or .38 times lower), not .62 times lower. $\endgroup$ Apr 19, 2015 at 18:54
  • $\begingroup$ @KarlOveHufthammer yes that's what I meant, thanks. I edited the answer. Imao it deserved the opportunist badge; essentially correct, but still negative rating. $\endgroup$
    – tomka
    Mar 31, 2016 at 17:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.