2
$\begingroup$

I'm fitting a logistic regression model in which my predictor of interest is a ratio of measurements in millimeters. Possible values for this ratio range from 0 to ~2.0, with typical values around 0.9-1.2. I want to measure the association between this variable and a binary outcome. I fit a model and obtained the following results:

Coefficients:
              Estimate  Std.err  Wald Pr(>|W|)    
(Intercept)   -6.52673  1.09162 35.75  2.2e-09 ***
x              0.91680  0.36187  6.42    0.011 *  # OR = 2.5

I know that the generic way to interpret these results would be: "For a one unit increase in X, the odds of Y=1 increase by a factor of 2.5..." I keep getting hung up on the interpretation of this result, as a 1 unit change in x (e.g., from 0.6 to 1.6) would be a very extreme/physically impossible change for this particular ratio (which represents an index). Is there a way I should transform this variable or the results so that I can describe changes of 0.10 rather than 1.00?

Thanks in advance.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
3
$\begingroup$

If a one-unit increase in $X$ results in an increase in log-odds of $2.5$, then a $0.1$-unit increase in $X$ results in an increase in log-odds of $0.25$.

If you drive a tenth as long, you go a tenth as far. That $2.5$ is analogous to your speed.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Great, thank you. So I'm wondering how I would report the OR, given that OR=0.25 would represent a reduction in odds. The log-odds I was looking at was 0.91680 [exp(0.91680)=2.5]. When I move the decimal point over by 1, [exp(0.091680)], I get an OR of 1.1. Or would I report a log odds of 1.25 (corresponding to a 25% increase)? Would it be incorrect to report that OR? Apologies if I'm missing something obvious or overthinking this. $\endgroup$
    – mkpcr
    Commented Mar 22, 2020 at 22:08
  • $\begingroup$ Unless I’ve forgotten how logarithm math works, it’s not an odds ratio. $\endgroup$
    – Dave
    Commented Mar 25, 2020 at 0:56
  • $\begingroup$ I was trying to make sure the relationship was the same after exponentiating the log-odds*0.10, i.e., converting it to odds. $\endgroup$
    – mkpcr
    Commented Apr 4, 2020 at 0:36
0
$\begingroup$

The coefficient of X is 0.91680, so OR for X is calculated as exp(0.91680 )=2.501273. The interpretation of the OR is that for every increase of 1-unit in X, the estimated odds of the event are multiplied by 2.501273. However, If your data is not increased by 1-unit, for instance, falls between 0 and 2, it may make more sense to say that for every 0.1 unit increase in x, the estimated odds of the event are multiplied by exp(0.91680×0.1)=1.096014.

You can find more explanations in PennState Eberly College of Science, STAT 463, Lesson 12: Logistic, Poisson & Nonlinear Regression.

Improve this answer
$\endgroup$
2
  • $\begingroup$ please summarize the content of your link, and provide a reference if possible, in case it dies in the future $\endgroup$
    – Antoine
    Commented Oct 1, 2020 at 9:18
  • $\begingroup$ Thanks. I have edited the hyperlink in my post. $\endgroup$
    – Shenglong Lin
    Commented Oct 1, 2020 at 12:14

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.