Patrick Murray said at 8:10 pm on October 8th, 2010:
The problem with the so-called math error is that you are using unweighted sample sizes (which pollsters use to estimate error) to back into the horse race numbers, rather than using the weighted adjustments to the subgroups upon which our final numbers are based. You can find those weighted adjustment in our methodology statement (which Art originally did, but Bill didn’t). The difference between what unweighted and weighted “n sizes” is something that every pollster DOES learn in polling school. And since we adhere to principles of full disclosure, we give both the unweighted and weighted information. Unfortunately, it is lack of public knowledge about polling principles that lead many other pollsters to withhold disclosure, so they can avoid uninformed critiques. We at Monmouth believe that all our data should be out there. Art’s original criticism of our voter turnout model is a fair one, since no one knows what Nov. 2’s electorate will look like and I responded to that by re-running our numbers using his assumptions and providing that data. This criticism, however, is just plain wrong.
Editors note: I’m relieved to hear a credible explanation from Patrick. While I will lose no sleep derailing the careers of Frank Pallone, Rush Holt and John D’Amico, MMM means Patrick Murray no harm.
I sat on the math discrepancy story most of the day waiting for Patrick to call. I sought out expert advise that was not available. Next time I’ll wait longer. ~ Art