Gallup’s daily presidential tracking poll has had Barack Obama and John McCain in a statistical dead heat for the last two weeks.  Now it has them in an actual dead heat.  The difference appears to be rising support for McCain, although the changes are quite narrow:

The latest Gallup Poll Daily tracking update on the presidential election finds John McCain and Barack Obama exactly tied at 45% among registered voters nationwide.

Voter preferences had been fairly evenly divided for the past week, with Obama generally holding a slight advantage of two or three percentage points. This is the first time since Gallup’s May 31-June 4 rolling average that Obama does not have at least a slim advantage over McCain. Obama’s largest lead to date has been seven points.

Even better news: the sample is significantly larger than any used by the news media in recent polling (2600 respondents).  It also uses registered voters, which normally would favor Democrats.

Yesterday, Stacy McCain (no relation) explained the polling vagaries:

Given the fact that huge numbers of eligible voters don’t vote, a pollster — if his poll results are to be useful or credible — must try to screen for “likely voters.” This is a doggone difficult thing to do, but it must be attempted, because voters and non-voters differ significantly in their preferences. Non-voters are more likely to support liberal policies and Democratic candidates (a source of endless frustration to liberal Democrats). So a poll that doesn’t properly screen for “likely voters” will always skew leftward (as was true of the Newsweek poll that surveyed “registered voters” rather than “likely voters”).

This is probably why early polls have historically overstated support for Democratic presidential candidates. The closer you get to Election Day, the easier it becomes to determine who the “likely voters” are. Thus, the samples in early polls contain lots of liberal-leaning eligible voters who, in the end, won’t actually bother to vote.

This is why sampling is so important in these polls, and why Newsweek, the LA Times, and CBS traditionally provide outliers rather than predictive results.