Several good recent diaries have discussed the W&M/TargetSmart poll of Florida MSN’s Last Word featured the other night ( www.scribd.com/… ) and other indications coming out of what amounts to exit polling of early voters. (Here are two; there were more. www.dailykos.com/… , www.dailykos.com/...) Skepticism about good news is healthy, but in this case I think it’s missing an important reason for optimism.
The main theme of skeptical comments in those diaries is that TargetSmart relied partly on opt-in on-line surveys, and live-interviewed by telephone only 188 early voters. But here’s the thing. Those 188 telephone interviewees were randomly selected from actual early voters, using TargetSmart’s excellent voter files as updated to reflect who has actually voted. Now, 188 voters isn’t a huge sample, but it’s actually worth a lot. Using SurveyMonkey’s statistical calculator, www.surveymonkey.com/…, that large a sample from an early-voting Florida electorate of 5 million (conservatively using an overly large number) would give you a statistical margin of error below 7.5%. I think it’s a fair assumption (here I’m relying on W&M/TargetSmart institutional credibility, but I think that’s reasonable on this point) that their live-interview results were pretty close to the combined interview-online result that they report. So let’s suppose the 188 live interviewees gave Hillary a demographically-weighted lead of 15% (not 17% as was reported for the combined respondents), and further suppose that by luck of the statistical draw, the actual representative lead is 7.5% below that. EVEN SO, BY THIS MEASURE HILLARY CAN BE EXPECTED (WITH >95% CONFIDENCE) TO WIN FLORIDA EARLY VOTERS BY 7.5% AND THE PRESIDENTIAL ELECTION IS EFFECTIVELY OVER.