Two-act transitions, partitioned contingency tables, and the ‘significant cells’ problem