Centerfield

Pollsters, Defend Thy Craft!

So everyone is all het up about the impending NH primary. And what's causing the excitement? At least in part it's the polls, and the numbers changing daily, and showing trends.

So the Boston Globe reports their globe/wbz poll showing Kerry with a 20 point lead, 37-17. And Zogby reports Kerry over Dean by 28-25, a statistical dead heat. Common practice is to state a margin of error of about +/– 5%.(I'm a little short on time so I skipped the links to the poll I referred to, they are easily found.)

So what's to account for such a discrepancy? Well, a margin of error is a statistical construct based on probability, and on statistical methodology that makes certain assumptions. So a margin of error primarily reflects only some generic basic assumption about accuracy based on the number of people polled. And guess what, the basic assumptions are based at least in part on the fact that no one would pay for a poll if its margin of error were any greater that what's reported these days. If the polls are less accurate than the stated accuracy based on numbers polled, the pollsters don't want to know.

And note that +/– 5% is a 10 percentage point range. So 28-25 Kerry with a +/– 5% margin of error means that the pollster is willing to state (with only 95 percent confidence, which is another story) that Kerry has between 23 and 33 percent of the vote and Dean has between 20 and 30 percent of the vote. Read that again. That doesn't tell you a whole lot, does it?

You know what might be nice, though? If a place like Zogby (or whoever, I'm not picking on any one group) might be willing to do the same 400 person poll 5 or 6 times ON THE SAME DAY. So they could find out whether their claim that the margin of error is really +/– 5 % is on the mark, or whether it's really much greater. Because of one outfit is getting 37 for Kerry and another is getting 28, this suggests that the polls are not especially reliable.

And you know what might be nicer? How about if in 2008 the people of Iowa and NH refuse to answer pollsters' questions, and demand instead that politicans tell the people what they think the people genuinely need to be told. I mean, can you imagine if the politicans had no polls to rely on, and had to craft their message based on what they thought was right for the county, instead of basing friday's message on the thursday poll that said support among females aged 18-29 was lagging. They'd have to rely either on guessing, or just speaking from their heart and their gut.

Seriously worth a try.