Tuesday, May 25, 2010

Pole-axed by Predictions.

Specterectomy - noun - procedure by which a foreign body is removed from the Democratic Party.

(Nothing to do with this article, but my last snark about this race - I promise.)

1 - Dewey Beats Truman.
2 - Sestak tied with Specter.

These were the conventional wisdoms leading up to respective elections.

We all know there was no such thing as a President Dewey. A long time ago studies were done as to how the media got it wrong with the 1948 race. Researchers realized that polling techniques needed to be improved if they were going to be reliable. For the last half of the 20th century the technique was to use the White pages. Pick people randomly, call them and ask what they thought. It worked great, because pollsters could get a decent sample of as a RANDOM cross-section of society.

Polling accuracy improved and the element of surprise is removed. Campaign gurus could sit in the back office with good data to figure out when to pull the trigger to go negative. In campaigns, like comedy, timing is everything. Pull the negative trigger too soon the opponent will get a chance to explain themselves and recover (Specter went negative on Sestak's military career and Sestak was able to muster a big enough group of vets to salute him). Pulling too late just looks pathetic. Good information gives you the sweet spot. If you want to complain about the negative nature of politics today - this is part of the root system.

The week before the Sestak-Specter race the polls looked like sine waves attacking each other (so did the campaigns). To listen to the polls everybody should have been sweating bullets. The dust cleared on Tuesday and Sestak had an impressive 8 point victory.

Why were these polls so far off?

Today, there is no method that gets to a random cross-section of society. There is no way for a pollster to get to everyone with the Do Not Call List and unlisted cell phone numbers. Voters do not need to give a phone number to register to vote.

Looking at trends in society, how many of us are killing our land lines? How many are "simply cell"? How many are active enough to vote without a land line? Those populations aren't being polled because they can't be reached randomly. If the only real "random" universe available to pollsters is a cut of the white pages, then the numbers will skew towards people with land lines.

That is why we had a false-positive hair-on-fire weekend before concluding the Sestak-Specter race. Bad data made the race appear tighter than it really was. Until polling techniques catch up to a tech savvy society, all polls are now suspect from the aspect of non-randomness.

Hopefully, there is a bonus here. We all claim to hate negative campaigns. Without good polling data to tell the campaign gurus when to go negative, maybe they will start to use prudence and not go negative? The campaigns then have to resort to actually telling the public why their guy is a good choice instead of yammering the other guy is evil.

I may do a poll on this - can I have your unrandomized phone number?

No comments: