Having had my rant against the "echo chamber punditry" following yesterday's primary election, let me offer some kind words for one columnist who brought a bit more perspective to his analysis. The columnist is John Nichols, who writes for The Nation and contributed a nice post hoc analysis based on "The Bradley Effect." For those unfamiliar with the terminology, he provided the following background:
The Bradley Effect refers to an electoral phenomenon first identified in the 1982 California gubernatorial election.
Tom Bradley, the popular mayor of Los Angeles, was the Democratic nominee for governor. Polls showed the African-American Democrat running well ahead of white Republican candidate George Deukmejian. Yet, when the votes were counted, Bradley lost by more than 50,000 votes.
The result made no sense. The gubernatorial election was one of the few Democratic losses in what was a good year for the party. Bradley was an able politician with a smooth style and a sound record. Analysts took a new look at the polls, which seemed sound.
It was then that they hit on the notion that white voters, not wanting to be thought to be prejudiced against an African-American candidate, had told pollsters they were for Bradley when they had always planned to vote for Deukmejian.
The phenomenon came to be referred to as "The Bradley Effect."
Nichols was also ready with an explanation for why there was not Bradley Effect in Iowa:
But there was no Bradley effect in last week's Iowa caucuses. Obama led in the polls and he led on election night. But, in Iowa, it was a public caucus where neighbors saw who neighbors backed.
There is a lot to be said for this analysis, particularly since it generalizes beyond political decision making.
For many years I followed a survey called "Eyes on the Internet," which basically tried to determine how much time the general American public spent surfing the Web and where that surfing actually took them. Each year they would approach these questions with a different method, all of which were respectable techniques in the world of survey-taking; but each year there was also a major discrepancy in the results. Anyone who tracked these studies would have concluded that there was no pornography on the Internet (or, if there was, no one way paying much attention to it)! The Bradley Effect is basically about the discrepancy between what people say and what they actually do (particularly when they assume that they are doing it privately, hence the explanation for the Iowa results). Whether one uses methods based on observations or interviews, individual behavior is influenced by the very fact that one knows that one is being surveyed. This is essentially the Heisenberg Uncertainty Principle of social science: The very process of measuring affects the outcome of the measurement. We just cannot quantify the effect of the measuring process the way Heisenberg could for physical experiments.
Needless to say, this phenomenon was around long before 1982; and it even generalized beyond surveying methods. The Hawthorne Effect (named for a series of experiments on factory workers conducted between 1924 and 1932) is basically the same phenomenon applied to the problem of modifying working conditions to increase productivity. In this case the process involved not just observation but also modification, which interfered even more with the validity of the conclusions being drawn from the experiments. The point is that, whether it involves working on an assembly line, surfing the Internet, or casting a vote, the only thing we know for sure is that people do what people do. If we are lucky, we may be able to collect enough data to prepare a convincing argument that explains what they did in a particular situation; but turning an analysis of the past into a projection into the future is an extremely risky proposition, which, as I previously suggested, borders on the arrogance of telling people what reality is. Better we should appreciate the sense of reality that people have and take actions that honor it!