Monday, November 03, 2008

Opinion Polls in the US elections

An interesting article on US opinion polls in today's Times of London.
Bottom line (literally) is that all polls indicate an Obama win.

An old newspaper photograph haunts the dreams of every US pollster. A grinning Harry Truman, having won the 1948 presidential election despite every prediction, is holding up a copy of the Chicago Tribune. It reads: “Dewey defeats Truman”.

Could it happen again? Every pollster is predicting a victory for Barack Obama. Might a grinning John McCain be pictured on Wednesday triumphantly holding a pile of incorrect polling data?

There are two things that say that he might.

The first is that American pollsters have not yet experienced what happened here in 1992 – when the polls pointed to a Labour victory but John Major won. The conventional wisdom is that 1992 was great for the Tories but terrible for the pollsters. In the long run, the opposite turned out to be true. Victory in 1992 turned to ashes for the Conservatives, whereas the pollsters used the debacle to get themselves sorted out.

Now British polls are properly and carefully weighted, taking account of what is known as the spiral of silence – the tendency of voters for the less fashionable party to keep their intentions to themselves. British pollsters weight their results to allow for these shy voters. US pollsters do not.

It isn’t unreasonable to believe that there could be a Republican spiral of silence. And that US pollsters are all missing it.

There is some evidence of mistakes among US pollsters. Every poll has a margin of error, to take into account the fact that a limited sample has been consulted. But the website has shown that during the primaries there was on average a 2.3 per cent pollster-introduced error, caused by poor methodology. This is not the case in Britain.

The second, widely canvassed, reason why the polls could be wrong is known as the Bradley Effect. In 1982 exit polls showed the African American Tom Bradley to be on course for victory as Governor of California. He lost. It is argued that voters had refused to support him because of his race but didn’t want to tell a pollster. Could this happen to Obama?

The Bradley Effect is talked about as if it were incontrovertible but it is only a theory. One of Bradley’s campaign team pointed out recently that the same exit polls that predicted victory for Bradley also projected that the white Democrat Jerry Brown would be elected US Senator.

And he lost too. These two question marks over the polls are ones that McCain can cling to as the campaign comes to a conclusion. They are not, however, the only reason to doubt the pollsters.

The other ones suggest that the pollsters may be underestimating, not overestimating, Obama.

In an election where only 60 per cent may vote, all pollsters have to weigh their findings to reflect how likely respondents are to cast their ballot. The difficulty is deciding how. Usually pollsters use previous elections to help them to decide who is going to vote. But what if, in this election, different sorts of voters are going to turn out?

There is reason to believe that young people and African Americans will turn out for Obama as never before. Some pollsters are adjusting for this, others are not (hence some of the variability in the polls). The result will depend to an extent upon who is right about this.

A second unknown is the use of mobile phones. A segment of the electorate – on the whole younger, poorer people – no longer have land lines. Yet pollsters use random digit dialling of landlines to build their samples.

Some say that this undercounts Obama support by 2 or 3 per cent.

Lost in all this detail? Then cling on to this. The polls may vary, the methods differ, the lead goes up and down. But every poll by every pollster still agrees that Obama will win.

No comments:

Post a Comment