The Blog

Richelieu: Media Pollster, Heal Thyself

1:57 AM, Apr 28, 2008 • By RICHELIEU
Widget tooltip
Single Page Print Larger Text Smaller Text Alerts

Lots of babble in the press and among the cable TV gasbags about the network exit polling being off in Pennsylvania due to race. This is worth some discussion over the entire pagan's Cathedral of media polling. First, the media's obsession with polls is a big weakness in modern political coverage. For some reason many in the media elite forget polls are never static and make silly predictions based on early polls that later prove to be very premature. These polls fuel the worst in mindless herd journalism. I have clear memories of this from more than a year ago when I was predicting Obama would win the nomination despite Hillary's "invincible" lead in the polls. More than one poll waving expert journalist laughed at me.

The problem is that all too many people who loudly opine about polls and campaigns have little idea of how polls actually work. All polling, including Pennsylvania exit polling, has a certain built-in error factor. Statistical science demands a truly random sample of the exact population one is trying to measure and in today's America achieving that is pretty much impossible. Some people don't answer the phone, some people lie on the phone to interviewers, some people don't have landline phones at all. So all phone polling today is pretty much a highly educated (and usually quite accurate) guess.

Exit polls are especially heavy on the guesswork; how do you get a perfect random sample of people who actually turned out and voted in a given election? You can try, as the networks do, with a mixture of phone polling to self-proclaimed absentee voters, real live precinct exit sampling, and computer modeling based on some actual early results to try to project specific turnout. The big problem here -- beyond the sampling stuff -- is that all exit polling is initially modeled to history as regards turnout. But some elections have different turnout from the historical average and it is very hard to adjust that to perfection on the fly. One should always keep in mind that exit polling was designed by political scientists as a tool for analyzing election results AFTER they happened, not to predict the results with pinpoint accuracy before the voting is complete. But one day the TV network news carnival showed up in Polling Town and the result was the game show version of exit polling that people are clamoring about today. When all the error factors exit pollsters face are considered, it is actually pretty impressive that most exit polling results are as close to the final results as they are. To hold exit polling to a perfect election night prediction, however, is silly.

So, back to the original point. Do people lie on polls about race? Yes they do. But not always and not that much. It is probably not the biggest error factor in the PA data.

For your Cardinal at least, the whole thing points to the fundamental journalistic conflict inherent in media horse-race polling. It is one of the only situations I know of where the media will create a story, by purchasing their own poll, and then breathlessly report on it: "Pendergast Support Collapses in New Poll." Since most publishers are cheap and most -- though not all -- media polling is technically sloppy, there is a real chance of distortion here. Even worse, many media outlets show a firm prejudice against reporting on any poll but their own, based on the foolish assumption that only a poll purchased via low bid by their tightwad proprietor is technically excellent enough to warrant publication. My view is that the media should poll issues and leave the horse-race stuff pretty much alone. The current kerfuffle over exit polling is simply an extension of this age old conflict. The electronic media perverts the original purpose of exit polling to create shaky election night predictions, and now damns the whole thing as misleading and potentially a dangerous sign of electoral racism. Media pollster, heal thyself.