Joel's Two Cents
What a Survey Really Says
This is the season when the polls swarm. They blitz your inbox, they infiltrate your home page, they creep up on you when you're not looking and goose you in the hindquarters. They lurk in the dark in prime toe-stubbing locations. In the morning you'll see a dead mouse on your doorstep, and you'll think: Dang polls.
The profusion of polls is the inevitable consequence of this year's historic disassociation of campaigning and voting. Nowadays candidates mostly campaign as a way to become sufficiently plausible to raise a lot of money. But in the past, campaigning was inextricably and intimately connected to voting. There were RESULTS. The data wasn't (weren't?--even words have a margin of error) hypothetical, weighted, projected or extrapolated. Reporters could say, this is what happened. Bottom line: Voting is clarifying. Polling, not so much.
Can you trust the polls? I dunno, ask President Kerry.
What's tricky when you're a journalist is resisting the temptation to read too much into a narrow lead that's within a poll's margin of error. Like, if Obama is up by three points over Clinton in Iowa, which is within the margin of error, do you say he's in the lead, or that it's a dead heat? If I had to write that story, I'd say, "Sen. Barack Obama (D.-Ill.) is sorta winning in Iowa, kinda," or language to that effect. Journalists are always so reluctant to use excellent words like "might" and "maybe" and "kinda" and "sorta" and "shoulda" and "woulda."
One thing that worries me about the margin of error of a poll is that I suspect that even the margin of error has a margin of error. You know, they'll say there's a margin of error of 4 percent, but do they mean EXACTLY 4 percent? Or 4 percent with a margin of error of .15 percent?
Turning to Wikipedia - speaking of things with an innate margin of error - this is what we read:
"A poll with a random sample of 1,000 people has margin of sampling error of 3% for the estimated percentage of the whole population. A 3% margin of error means that 95% of the time the procedure used would give an estimate within 3% of the percentage to be estimated."
There are different kinds of biases in a poll, including sampling bias, response bias and non-response bias. For example, Wikipedia tells us:
"Survey results may be affected by response bias, where the answers given by respondents do not reflect their true beliefs."
Gettin' kinda murky.
Of course I do my own polling when I'm on the campaign trail, in the sense that I accost strangers and ask their opinion. People wearing hats are the best. You rarely meet a person in a hat who doesn't have interesting thoughts. Also nuns are great. So hats and habits, those are the flares that say "Interview me." Also anyone wearing a sandwich board sign, and punks on skateboards. So perhaps when I do my surveys there's an eccentricity bias.
Compounding the confusion for political prognosticators in Iowa is that it isn't a normal election, but rather a "caucus" process in which people can, over the course of an evening, CHANGE THEIR MINDS. This introduces the little-understood effect known as "squishiness bias." People can switch from one candidate to another, kind of like those football voters who at the last second elevated Louisiana State into the title game.
Listen to what my friend David Von Drehle writes at Time:
"Huckabee has never lost sight of the core fact of the Iowa caucuses: turnout is minuscule. Because of inconvenient scheduling (on a school night and opposite the Orange Bowl this year) and arcane rules for voting, candidates can look like giant killers here with about as many votes as it takes to be elected to the Fresno school board."
Today some New York-based newspaper had a piece about cellphones causing pollsters headaches, because you can't poll people who only have cellphones and don't have landlines. A friend tells me, "The great thing about this trend is that it could at least theoretically put an end to polling as we know it. If the pollsters can't reach you they can't poll you. And even if they can reach your cell phone, it isn't tied to any specific location, so how will they do an "Iowa" poll if half the people in Iowa have 917 or 646 or 202 area-code cell phones?"
Oh, they'll do one, you can bet on it.
Posted at 10:37 AM ET on Dec 7, 2007
Joel's Two Cents
Share This: Technorati | Tag in Del.icio.us | Digg This
Previous: For Some, Negative Turn is a Turn-Off | Next: Reid Saves the Day After All
Please email us to report offensive comments.
Posted by: PulSamsara | December 9, 2007 12:20 PM
Posted by: vote4mikehuckabee | December 8, 2007 10:04 PM
Posted by: teague | December 8, 2007 11:22 AM
Posted by: tabita | December 7, 2007 11:46 PM
Posted by: Nissl | December 7, 2007 11:09 PM
Posted by: avraamjack | December 7, 2007 4:03 PM
Posted by: ShriekingDenizen | December 7, 2007 3:19 PM
Posted by: daiwanlan123 | December 7, 2007 12:29 PM
Posted by: nquotes | December 7, 2007 11:25 AM
The comments to this entry are closed.