Correcting for multiplicity in the 'emmeans' package

Published at January 21, 2026 ·  7 min read

In my recent book (see below), on page 166 and earlier, I made the point that, with pairwise comparisons and, more generally, whenever simultaneous statistical tests are performed, it is necessary to provide P-values that account for the familywise error rate, i.e. the probability of committing at least one incorrect rejection within the whole family of simultaneous tests (i.e. adjusted P-values). In this respect, it may be useful to recall that, for a single non-significant test, the comparison-wise error rate \(E_c\) is the probability of a wrong rejection for that single test (based on a non-adjusted P-value), whereas the probability of at least one wrong rejection within a family of \(k\) comparisons is much higher.

...