1
$\begingroup$

I'm running a binary logistic regression and when put in relation with one specific continuous variable, I get a log-odds (Beta) of 0.001 and a odd ratio (Exp(B)) of 1.001.

No matter what other variables I add or remove to the model (ordinal, nominal, continuous), I get the exact same results 0.001 and Stand. Err. of (0.000) always p<0.001 or 0.05

So I am wondering if there is a glitch with the data. What means such a low value. A odd ratio of "1 time more than..." seems odd to me :P

Thanks for helping me clarifying it.

Cheers

$\endgroup$
2
  • 1
    $\begingroup$ It would be good if you can include all the model outputs in your question. It seems that changes in the covariate in question are associated with very small changes in the response. $\endgroup$ Commented Jun 17, 2021 at 21:36
  • $\begingroup$ "1 times more than" sounds like equality to me. $\endgroup$
    – Dave
    Commented Jun 17, 2021 at 21:44

1 Answer 1

1
$\begingroup$

I'll guess that the independent variable here has a reasonably large range and so the change in odds per unit is very close to 1. As an example, if the odds of something increases by an order of magnitude with a ten year difference in age (i.e. OR = 10/decade), this would be an OR of 1.26/year, 1.0006/day, 1.00003/hour, etc.

Sometimes it makes much more sense to use a larger (and sometimes a smaller) denominator when communicating results (e.g., a GP talking about the risk of smoking would be more likely to talk about the effect of a decade of smoking, not the effect of smoking an individual cigarette or one day of smoking). There is a reason we talk about speed using kilometers (or miles) per hour and not meters per hour or kilometers per second. If this is the issue, you'll need to think about a sensible denominator to use.

While you can calculate the OR for this from the coefficients you already have (by multiplying the log-odds or raising the OR to the appropriate power, for both the point estimate and the confidence interval limits), you could also create a new variable to represent 5 or 10 (or 100 or whatever other value makes sense to you) units.

This won't change the p-value, it's just the units that you're using for the effect and its confidence interval and (subject to the precision from the number of decimal places) the reader can always convert this to their preferred denominator if they don't agree with yours.

$\endgroup$
3
  • $\begingroup$ Thanks for your clarification. For more context, the study examine work relations. The mentioned continuous variable corresponds to the number of workers which ranges from 5 to 5000. I am not sure what denominator could be best here (e.g. per 10,000 inhabitants). But I don't have this information. $\endgroup$ Commented Jun 18, 2021 at 0:52
  • $\begingroup$ You can, if this makes sense to you, simply scale the variable to tens or hundreds of workers. This is the same as changing from millimeters to centimeters or from centimeters to meters respectively. Using "per hundred workers", 5 becomes 0.05, 5000 becomes 50.This will change an OR of exactly 1.01 per worker to 1.01^100=2.70 per 100 workers, for example. $\endgroup$
    – user215517
    Commented Jun 18, 2021 at 1:46
  • $\begingroup$ Now it works. Thanks for your help. $\endgroup$ Commented Jun 18, 2021 at 2:30

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.