Commentators on both sides of the aisle have been taking shots at pollsters of late. Former Secretary of Labor Robert Reich declared that polling predictions are "based on nothing," while conservative author David Limbaugh maligned poll averages as "analogous to moral equivalency arguments to me. All meaninglessness."
The truth of the matter is that polls in the final week of presidential elections do an exceptional job of predicting the eventual outcome. GWU political scientist and official friend-of-Wonkblog John Sides passes along the following chart, from Columbia's Robert Erikson and Temple's Christopher Wlezien, that makes this point:
The X axis is the share of the vote the Democratic candidate received in an average of the last week's polls. The Y axis is the eventual result. With the notable exception of 1964 — in which LBJ's 61-39 wipeout of Barry Goldwater was nonetheless smaller than the predicted 70-30 landslide — the polls rarely err by more than a point or two.
Of course, in a close race like this year's a point or two is important, and Erikson and Wlezien also found that the person leading in the national polls is disproportionately likely to underperform what the polls predict. For example, LBJ did worse than predicted, and Al Gore won the popular vote in 2000 despite trailing in the last polls.
That's bad news for MItt Romney, who, despite trailing in key swing states and most Electoral College models, still holds a 0.9 point lead in the Real Clear Politics national poll average.