P‐Values for Two‐Sided Tests

The problem of defining P‐values for two‐sided tests is considered when the null density f 0 of the associated test statistic is strictly unimodal. It is argued that the usual practice of defining P‐values for two‐sided tests as twice the one‐tail P‐value is only justifable when f 0 is symmetric. To accomodate testing problems with asymmetric null distributions a more general definition of two‐sided P‐value is given.