His first population based result - "If each <truly awful crime> had one perpetrator we’d be talking about 3.3% of the population" - uses the correct population. After that he's using 208.