Dear Editor,
There are eight daily nationwide tracking polls in the US Presidential Election (polls which interview every day and average their results over 3 to 5 days), 3 further daily national "snapshot" polls, and about 15 further State-wide polls on a typical day, all of which produce widely varying results. Yet your correspondents persist in writing "stories" based on one poll, frequently commenting on statistically insignificant movements within the margin of error of that poll.
The only "story" here is that there is a lot of what statisticians call random noise exhibited by polls taken in isolation. If you want to write a truly analytical piece you have to combine the findings of a lot of polls and look for trends as is done on "poll aggregation sites" such as Pollster.com, Fivethirtyeight.com, and Real Clear Politics.com, and even here you will find a lot of disagreements because some polls are commissioned by Democratic or Republican leaning pollsters working for clients with an interest in their outcome. For the record, those statistically combined polls currently have Obama leading by a remarkably stable 6-9% over the past two weeks.
However even this is not the full story, because, as Al Gore found out, the US President is not elected by the popular vote directly, but by the outcome of 50 separate state-wide elections each of which tend to have a "winner takes all" approach to the allocation of their electoral votes in the electoral college made up of a total of 538 electors. (The exceptions are New Hampshire and Nebraska). Thus winning a state by a 20% margin benefits you no more than winning it by 1%, and thus all the action centres on the larger states with more electoral votes and which are close to 50:50 in their Democratic/Republican voting tendencies - typically Ohio, Florida, Missouri and Pennsylvania.
It is Obama's lead in these states which is absolutely crucial to the outcome, and he would currently have a 375 to 163 "blow-out" or landslide victory in the Electoral College if the average of these polls are correct and the vote were held today. 33 states do indeed allow allow early voting, and about one third of all votes are expected to be cast before election day. So many of those who tell pollsters they support a particular candidate have in fact already done so, and Obama leads even more strongly amongst those who say they have voted early.
There are many additional factors which can make any individual poll inaccurate. Each pollster employs a different model to achieve a representative random sample of the population and adjusts his results by different assumptions of what constitutes a "likely" voter. Many pollsters do not call cell-phone only voters at all (resulting in a 2-3% anti Obama bias according to some research findings) and some use non-representative internet panels or computerised "robo-calls" rather than human interviewers.
And finally there is the much debated and disputed "Bradley Effect" which postulates that some voters say they will vote for Obama, but actually will not for reasons of racial or other prejudice which they do not want to reveal to the pollster. All the recent evidence in state polls indicates that there isn't a current "Bradley" effect, but this will be the first time this theory will be extensively tested in a reasonably close Presidential election where it might actually make a difference.
So please, please, please, Editor, if your correspondents must write about opinion polls, don't let them take your readers for fools and write about individual polls as if a 2% swing has any significance whatsoever. Even sophisticated statistical models combining the results of hundreds of polls can be wrong by much more than that. There is only one poll which matters and even on a quiet news day there is much of real significance which can be reported on.
