0
$\begingroup$

I have a response variable and 10 predictor variables (all ordinal). I wanted to see if there was any evidence of a relationship between the response and predictors. I used a two proportion z-test to see if there was a significant difference in my predictor at the two different levels of my response. I did this for each predictor variable. At the end, my results were significant for each comparison.

I also used Binary logistic regression where my response is dependent and I put in all my predictor variables under "covariates" (because that's the only option in SPSS). Based on the results, only 6 of my 10 variables were significant.

I know that logistic regression is for modeling and prediction, but do these results contradict each other? Is one method "stronger" than the other? Or maybe I'm not looking at this from a correct perspective.

Any input is appreciated, thank you.

$\endgroup$
3
$\begingroup$

do these results contradict each other?

Your results do not necessarily contradict each other. Consider this [bad] example, I am trying to guess a person's weight from their height. I know their exact height (in whatever measurement system you like) and I have a rough estimate of their height (i.e. short, medium, tall). While I might be able to make a good guess from the rough estimate, if I have the precise measurement then I wouldn't expect the rough categorization to add anything new (no new information). This isn't a great example, but I wanted it to be more obvious.

In your case, it is likely that some of the variables do not account for additional variation in the logistic regression model (i.e. no new information). So it is perfectly acceptable for variables to look associated in a bivariate setting, but often not associated when in a multivariate setting.

$\endgroup$
  • $\begingroup$ Thank you, so would it be okay for me to ignore the other four variables? Considering they don't add anything? $\endgroup$ – Stat01 Aug 19 '13 at 14:25
  • $\begingroup$ That depends much more on what you are trying to do. In general it is good to remove "noise" from your model, however you may be trying to show that those variables don't add anything. Do a google search on manual backwards variable selection. $\endgroup$ – Ellis Valentiner Aug 19 '13 at 14:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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