0
$\begingroup$

I had a group of 120 patients underwent a surgical operation. Of those, I compared patients who experienced a surgical site infection and patients did not experience with anova. I found two statistical significance. Then, I made a logistic regression analysis using the surgical site infection as dipendent variable and the variables resulted with p<0.1 at anova as Indipendent variables. I did not found any significance, not even those with a p<0.05 at anova. What does it mean? Thanks for your time

Improve this question
$\endgroup$
4
  • $\begingroup$ Hi, what was your dependent variable in ANOVA? $\endgroup$
    – Sointu
    Commented Apr 17 at 9:11
  • $\begingroup$ Hi, with anova I Compared continuous variables (eg. Bmi of patients who had a surgical site infection and Bmi of those who hadn't). $\endgroup$
    – Luca ferraro
    Commented Apr 17 at 9:16
  • $\begingroup$ Can you be more precise? Did you run several separate ANOVAs for several response variables? How many? What were the p-values? How many variables entered into the logistic regression? Where is the idea from to use "the variables resulted with p<0.1 at anova" in the logistic regression? That's quite certainly not a good idea. Why would you do any variable selection before running the logistic regression? $\endgroup$
    – Christian Hennig
    Commented Apr 17 at 9:47
  • 1
    $\begingroup$ Note in particular that data dependent selection invalidates p-values, i.e., if you use ANOVA to select variables for the logistic regression on the same data, all p-values from the logistic regression are invalid. $\endgroup$
    – Christian Hennig
    Commented Apr 17 at 9:48

1 Answer 1

Reset to default
2
$\begingroup$

So, it sounds like you tested the mean differences between infected vs. non-infected patients in your continuous variables separately for each continuous variable, but then ran the logistic regression with several continuous predictors. It can happen that different predictors suppress each others' effects on the dependent variable, which may render their effects non-significant.

EDIT. More generally, ANOVA and logistic regression are essentially different analyses. ANOVA estimates whether there are group differences in means, whereas logistic regression tests whether predictors predict the binary group membership while controlling for other predictors, if any.

But the above is not the main issue, as Christian Hennig says. You should not choose predictors this way, and as commented, any p-values from your logistic regression are meaningless.

Improve this answer
$\endgroup$
3
  • 2
    $\begingroup$ Probably both of these are issues explaining the results. Maybe it's worthwhile adding explicitly that the logistic regression tests do something essentially different from the ANOVAs by testing the contribution of any variable assuming all other variables are still in the model (with the implication that you mention). $\endgroup$
    – Christian Hennig
    Commented Apr 17 at 10:13
  • $\begingroup$ Thanks, I will! $\endgroup$
    – Sointu
    Commented Apr 17 at 10:15
  • $\begingroup$ Thank you very much. $\endgroup$
    – Luca ferraro
    Commented Apr 17 at 11:18

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.