Sunday, May 2, 2010

How to avoid revising

Pick holes in the statistical arguments of the mainstream media instead. I've been taking issue with the Independent's John Rentoul (if you're not familiar with him, check out his series, Questions to Which the Answer is No) over his treatment of opinion polls.

Essentially, John is taking a number of opinion polls (an estimate of the voting intentions of the population) and taking their average, and hoping that this is somehow a better estimate of the intentions of the population than any individual poll. Indeed, if the polls follow an identical methodology, then this approach is mathematically valid - however, it is still effectively a poll, with a random value. The variance will have decreased slightly, but in real terms, not that much. If the methodologies are different, then we are averaging apples and oranges.

No comments:

Post a Comment