The Art of Statistics: Learning from Data

by David Spiegelhalter

In this case, the null hypothesis would be that there is truly no association whatsoever between arm-crossing and gender, in which case we would expect the observed difference in proportions between genders to be 0%.