For me this was far better explained in the following text (also posted on HN not so long ago): http://www.michaelnielsen.org/ddi/if-correlation-doesnt-impl...