Last week I heard interviews of top pollsters by a public radio host who could not accept that their results could vary so much – from a margin of 14 percent for Obama to only a 1 percent edge over his opponent McCain. Clearly these predictions differed significantly. Given the power of statistics, how could that be?
The main reason from what I gathered was the variation in pollster’s models on who will actually vote. For example, as I reported in my blog of 12/31/05 (“Surveys produce precisely inaccurate findings”), about 60 percent of self-stated voters did not cast their ballot in the previous election. This attrition rate historically varies by party (Democrats tending to slack off more than Republicans, perhaps) and demographics. Furthermore, people are more and more resistant to being polled – two out of three now refuse according to an article by Rick Montgomery of McClatchy Newspapers (11/2/08). Furthermore, the demise of landline phones in favor of mobiles makes it ever harder to even contact prospective voters. Who wants to burn up precious cell time on a poll?
Amazingly enough, despite all these difficulties in coming up with accurate predictions based on pre-election surveys, the political snapshots proved sharp according to Ken Dilanian of USA Today (11/6/08). Despite dire warnings by analysts such as political scientist Steven Schier of Carlton College (Northfield, Minnesota), I suppose that since all of the polls correctly forecast a win by Obama (his margin was 6 percent – in the middle of the pre-election range of predictions), the statisticians will get by with the usual excuse of “random variation – the old bell curve,” as Scott Rasmussen of Rasmussen Reports put it. Hopefully they learned some things that will lead to better models next time on who will actually vote, how unreachable cell phone users vary from readily-accessible ‘land-liners,’ and the behavior of the silent majority who refuse to answer any questions. Good luck and please take me off your call list.
PS. One of the more interesting statistics I heard about voting is that for every 1 inch of rain in any voting precinct the Republicans gain a 2.5 percent edge! Maybe McCain’s supporters should have invested in fire trucks to hose down voters waiting in line to vote.