An email from my dad who is a retired political scientist who studied voting patterns. He is trying hard to quit working, but is clearly having trouble doing so. Nearing 80 now, he has gone to France in order to avoid the election season in order to save his health (because he is an amazingly engaged liberal, he cannot sit still, but his family and physician have suggested he do so). So, from an apartment in Paris (where he is trying to relax) he wrote the following note to his kids, which I am copying in its entirety.
A note on the 2012 polls.
There are at least four continuing problems with the election polls - sampling partisans, sampling phones, states versus nation, and public experience with pollsters. The most difficult of these, sampling partisans, arises in the context of partisan identification of likely voters. If the standard national election studies question is used to determine partisan identification ("When it comes to politics do you generally consider yourself a Democrat (Republican), Republican (Democrat), Independent or What?") subsequent answers to questions about the current election have already been influenced. If the specific questions are followed by the ID question then the ID responses are biased. There probably is no way around this dilemma which has been recognized for a long time. Typical strategies involve different order of presentation of questions for the same overall protocol but it remains a problem. In a very close election, where partisan identification is used as part of the basis of estimating likelihood of voting, it becomes an acute problem. Note the partisan structure of the likely voters in Rasmussen or WeAskAmerica compared with other polling organizations. These two polls typically over represent Republicans and their estimates of Presidential voting are similarly biased.
Access to the public by pollsters has changed radically in the past decade and in the past four years. Perhaps half the phones in the U.S. are now cell phones and the voting habits of each class of owners are not known although some demographics are available and the polls have tried to study this. However, again in a close election, this sampling feature is very important. The pollsters have different approaches to this reality. Partisan ID plus phone ownership thus compound the uncertainty in the estimates. Probably averaging across polls smooths some of this error but the difficulties are real.
The poll changes after the first debate highlight the difference between state polls and national polls. Nationally, all pollsters have moved to a predicted Romney victory although without substantial margin. State polls, where some of the above difficulties are mitigated, show different patterns. Romney has apparently moved ahead in Florida and Virginia, and has reduced Obama's advantage in most swing states. However, the Senate races, although closer, have not changed very much in favorites. It appears that even Arizona, Montana, and North Dakota (where there are no new polls) remain competitive for the parties. It looks like Connecticut and Massachusetts show a Democratic margin and the Independent in Maine surely will win. In Pennsylvania, Ohio, Michigan, Wisconsin, and maybe even Indiana and Missouri the Democratic candidates will win. Nevada still consistently favors the Republican candidate but Obama leads Romney in that state. Colorado has moved by a point or so to Romney but the Senate contest is uncertain.
Finally, the public has decades of experience now with pollsters. What response biases this may introduce are not easy to determine.
All these sources of uncertainty as well as the inherent uncertainty in sampling make riddling the poll reports a challenge. it would not matter if Romney were clearly ahead by 6 or 8 points (or the reverse). In a close election like the current one they heighten the uncertainty.
The sudden reverse in the Pew results showing Romney way ahead of Obama clearly arises from a change in the way in which partisanship is estimated among likely voters (they admitted this). They changed their procedures and from one month to the next the partisan ID fractions changed by 21% in their sample - an extremely unlikely swing. But that change emphasizes how important the estimated partisan ID parameters among likely voters are in determining the predictions of a polling house.
And now a partisan pitch: Today's (Saturday October 13) Huffington Post has an entry by Matt Taibbi on whether or not Biden's scorn at Ryan's evasion of tax cut specifics was appropriate. It was, and more. Taibbi takes the press to task in no uncertain terms, appropriately.
Sent from my iPod - dad
be kind be generous