If a selection mechanism is orthogonal to a property, it seems weird to argue that the selected subset is distributed differently along that axis than the broader population.