Politics: Whither the "Bradley Effect"?

TORONTO, ONTARIO - As the world learned in the year 2000, the popular vote does not determine the United States Presidential election. Instead, the electoral college, in which each state gets a number of electors equal to its Congressional delegation (and the District of Columbia gets 3 electors), decides the race. In essence, there are 51 separate elections, instead of one big one. With 538 electors available, a candidate must earn 270 to win.

For today, I will avoid any comment on the merit of the electoral college system, or its quirks like some states not being winner-take-all and the fact that electors are not actually legally bound to vote for a given candidate. Let us assume that the system works and will indicate an uncontroversial winner on November 4th.

For the first time in this race, the polls indicate today that John McCain would win the electoral college. The post-convention bounces for McCain in Florida, Nevada, North Dakota, and Virginia by most accounts have shifted those states from favoring Obama to favoring McCain, for a total of 270 electoral votes. Most consider Colorado, with 9 electoral votes, to be slightly favoring McCain as well.

In general, post-convention bounces tend to disappear, and nearly two months remain before the election, so this shift does not necessarily represent a major problem for Obama's candidacy. However, there is a deeper problem--all of the polls may be wrong, and in fact over-estimating the vote Obama will receive on election day.

The reason is the so-called "Bradley effect," or the tendency of polls to over-estimate votes for African-American (and perhaps other minority) candidates. In 1982, Los Angeles mayor Tom Bradley, who happened to be African-American, had been leading in the polls in the California governor's race over George Deukmejian, who happened to be white. Exit polls showed a small but clear Bradley victory. Yet, when the votes were counted, it was Deukmejian that actually had more votes.

There have been numerous similar occurances over the years, not all of them involving a candidate losing an election he or she was predicted to win, but often with a much smaller margin of victory, the most notable probably being the New York mayor's and the Virginia governor's elections in 1989.

What is it all about? Some claim the "Bradley effect" occurs because white voters tell pollsters that they will vote for a minority candidate but really do not actually do so. Others claim that pollsters simply underestimate conservative votes, since most minority candidates that appear to have suffered from the effect have been liberal. The bottom line is that nobody knows for certain.

How big is the "Bradley effect"? In some contests, it may have been as high as 10%, so that is probably the upper limit, and far higher than is likely to be observed in a presidential race. A University of Washington study of the 2008 primaries implied that polls may have over-estimated Obama in states with an African-American population below 8%, correctly gauged his support in states with 10-20% African-American population, and under-estimated him in states with a larger African-American population. The applicability of that seemingly-relevant study may be limited, though, since the pool of voters in a Democratic primary and a general election is not the same.

Let us just postulate for a moment that Obama's support is over-estimated by 4%. Right now, besides the states already cited as recently falling into the McCain column, that would mean that Michigan, Pennsylvania, New Hampshire, New Mexico and even (by some polls) Minnesota, Oregon, Washington, and Wisconsin would also go for McCain. Even ignoring those last four (which strike me as absurd), that still gets McCain to 325. That would be daunting to overcome.

It is important to state that all this does not mean that Obama cannot win. Voters go to the polls and cast votes on election day, and there is nothing stopping Obama's campaign from winning by a landslide in November. He won the nomination of the Democratic Party. It simply means that one must be cautious in evaluating polling data, which is a good idea in general.

Some think the "Bradley effect" may not exist anymore. Indian-American candidate for Louisiana governor Bobby Jindal may have suffered from such an effect in 2003, but not in 2007, when his election was by nearly an identical number as predicted by polls. Some pollsters claim they have ways to account for the "Bradley effect." The bottom line is that nobody knows for certain. So, short of the race shifting toward a McCain or Obama landslide, don't expect to know the results of the US presidential race until the morning of November 5th at the earliest--exit polls could have a "Bradley effect" as well.