Polling Idiocy
According to a local New Mexico news station, a congressman trails his opponent, 51%-45%:
Eyewitness News 4’s exclusive Survey USA poll on the congressional race in New Mexico’s first district is a real shocker. It shows Republican challenger Jon Barela with a lead over incumbent Democratic congressman Martin Heinrich, 51% to 45% – with only 4% of the voters undecided.
Survey USA polled 559 likely voters from July 22-25, with a margin of error of 4.2%. Barela’s lead exceeds the margin of error and he is also over the “magic” 50% line that pollsters like to talk about.
There are two major problems with how this got reported. First of all, a margin of error is “plus or minus” 4.2%, and is that way for both variables in the survey. In other words, Heinrich could be anywhere between 41% and 49%, and his opponent can be anywhere between 47% and 55%. The congressman could be up 2 points or down 14. Of course, with a spread like that, it’s more likely that the congressman is down than up, but you can’t say with 95% confidence that he is in fact down.
Reporters misstate outcomes of polls and when the polls don’t match the election result like a scorecard on a videogame, people assume polls aren’t trustworthy. They are trustworthy – but just not to the extent people have grown to assume.
The second issue I have with this is the reported fascination and over-importance on hitting a majority support in a poll. There are several reasons for the origin of that conventional wisdom – if an incumbent (not a challenger!) is above 50%, he’s probably safe, but if he’s not, the CW is that undecides will break against him and he’ll lose (because they haven’t backed the incumbent they know by now, so why would they later). The key thinking here is that with Heinrich at 45%, and you can assume he won’t get much better than that. The 50% rule has nothing to do with Barela’s standing. He could be at 41%, and the CW is that he’d still win enough undecideds to overcome a hypothetical 4-point lead.
This CW is flawed for a number of reasons, but I won’t go into it just yet. My real complaint here is with two other data points in the poll: 1) The undecideds are awfully low for a low-profile race with over three months before the election; and 2) supposedly Hispanics in New Mexico are likely to give the GOP 40% of the vote, which seems unlikely in light of the Arizona law fiasco – the tilt of Hispanics to the Dems in 2008 was why New Mexico voted for Obama and Heinrich in an unexpected landslide. I’m not saying they’ll vote in the same numbers in 2010, but I don’t expect the GOP to do better with Hispanics this year than two years ago. Both of these data points point toward a toss-up situation more so than a Barela lead.