2
$\begingroup$

Are only the p-values relevant when testing the regression coefficients of a logistic regression? Does the z-value of a coefficient give any further information about the significance of the coefficient?

I know how the z-values are to be estimated, but I have some difficulties with interpreting it for each coefficient, so I am not sure if it gives any information about its significance.

$\endgroup$
  • 2
    $\begingroup$ The $z$-value is directly linked to the $p$-value. The two-sided $p$-value for each regression coefficient is calculated as $2\cdot \Phi(-|z|)$ where $\Phi$ denotes the CDF of the standard normal distribution. $\endgroup$ – COOLSerdash Feb 27 '15 at 21:44
  • $\begingroup$ COOLSerdash's comment covers the central issue; I'd only add that the two tailed $p$-value is monotonic in $|z|$ and so the only additional information you can get from $z$ that you can't see in $p$ would be its sign (which doesn't relate to significance in the two-tailed case). $\endgroup$ – Glen_b -Reinstate Monica Feb 27 '15 at 22:49
1
$\begingroup$

In a logistic regression, are you sure you aren't using chi-square values instead of z-values? I've always seen Wald (if asymptotics are appropriate) and likelihood ratio chi-square values used to measure statistical significance in logistic regression.

Either way, the p-value corresponding to a chi-square value for a logistic regression coefficient has the same interpretation as the p-value corresponding to an F-value for a linear regression coefficient.

In logistic regression, the p-value corresponding to the calculated chi-square value is the probability of seeing a chi-square value at least as high as the calculated chi-square value for the equivalent model without that coefficient.

$\endgroup$
  • $\begingroup$ You can get z-values as asymptotically normal test statistics ($\hat\beta/\hat{\sigma}_\beta$) for individual coefficients. $\endgroup$ – Glen_b -Reinstate Monica Feb 27 '15 at 22:51
  • $\begingroup$ @Glen_b You're right, sorry. I was thinking of the Wald chi-square statistic instead of the Wald normal statistic. $\endgroup$ – Brandon Sherman Feb 27 '15 at 23:35

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.