The history of generic ballot polling as a House predictor is iffy.
The endlessly useful Real Clear Politics site offers numbers junkies a great deal of data about upcoming and past elections. We took a look at the site to get a sense for how closely aggregated generic ballot surveys end up reflecting what happens with the national popular vote in House races. After all, this is what's being polled: How likely Democratic and Republican candidates are to win election. One would assume that it would correlate to the actual national totals.
In each of the charts and graphs that follows, the margins are from the Republican perspective. A -10 percent spread means that the Democrats beat the Republicans by 10 percentage points on overall popular vote. A 10 percent spread means the Republicans won by 10 points.
Since 2002, the final generic ballot average was about 2.5 points off the actual national vote percentage on average. In 2002, 2004 and 2006, the RCP average underestimated support for Republicans; in the three most recent elections, it underestimated support for Democrats.
A 2.5 point miss isn't terrible, really; in two races, the average was under 2 percentage points off. But as you can see in the graph above, the generic ballot average got both 2004 and 2012 substantially wrong, predicting a tie in 2004 and a Republican triumph in 2012. Republicans won handily in the first case, and Democrats narrowly in the second. In other words, in those two elections, the average didn't really tell us much.
A look at The Post's own polling affirms that generic ballot surveys don't do a great job of predicting the actual final vote margin in House races.
But here's the thing. Even if we were able to nail the national popular vote, there's another problem.
The party that wins the national popular vote in House races doesn't always win the most seats.
Since 1974, there's been a pretty good correlation between the extent to which a party wins the overall vote in House balloting and how many seats they win. Which makes sense, of course. The black line on the graph below shows that correlation.
We marked the decades of results with different colors to make two points. First, that Republicans consistently did better for three straight decades. And, second, that the results from 2004 to 2012 have been all over the place -- and less closely correlated to the national popular vote.
Using a simple correlation tool, we looked at the overall correlation between the national vote and the change in seats in the House. From 1974 to 1994, the correlation was stronger than between 1994 and 2012.
This could be an anomaly, due to unprecedented elections over the past few cycles. But it suggests that the generic ballot probably isn't a great guide to who will control the House.
There's a third reason, besides the two above, that the generic ballot doesn't tell us much about congressional control. Quite simply, we know who will control the House, based on the slew of incumbents and targeted races. There's essentially zero chance that the Democrats retake the House, barring something phenomenally unprecedented happening over the next few months.
Which, of course, generic ballot surveys don't tell us about anyway.
This post has been updated with likely voter polling from the Post. The original version used data from registered voters, a broader measure, and included incorrect actual vote margins.