== What is the Wald statistic? ==

The wald statistic, outputted in logistic and Cox model regressions, is simply the square of the regression coefficient, B, divided by its variance. This is also calculated as the t-statistic squared.

If the regression coefficient is zero the wald statistic should, for large samples, be asymptotically equal to a chi-square statistics on one df.

Univariate Wald = $$ \frac{\mbox{ B }}{\mbox{variance (B)}}$$

[[FAQ/infmles|There are problems interpreting Wald statistics in logistic regressions when there is sparse data.]]

There are multivariate Wald statistics which may be used to compare more than one regression coefficient to zero simultaneously which are approximately chi-squared on p-1 df when testing if p regression estimates are equal to zero.

This uses the Covariance matrix of ther regression estimates (COV). The univariate Wald is a special case.

Multivariate Wald = $$B^T$$ COV B