This Newsweek poll released over the weekend found some surprisingly good news for Democrats. But it probably doesn’t mean much for President Obama and his party: the sample includes too many Democrats, at least based on a lot of other recent polls.

First the good news for Obama. His approval jumps by six points in the survey (from 48 percent to 54 percent in less than a month). This would be a significant move for several reasons, but particularly because presidential approval is a key predictor of House seats loss/gain in many statistical models.

But those numbers seem suspect to me, especially since they were stuck in the 47-49 percent range since January in the last five Newsweek polls.

The Newsweek generic ballot numbers also looked good on the surface for the Democrats. They reveal a 3-point Democratic lead among likely voters and a six-point edge with registered voters. Yet these numbers also seem out of line.

The Real Clear Politics average, for example, of all generic ballot polls gives Republicans a 7.7 percentage point lead. finds that the GOP holds a slightly smaller 5.4 percentage point edge. But neither is even close to giving the Democrats the advantage, much less a six-point lead.

The explanation for the Newsweek poll’s Democratic tilt can be found on page 13, right after they ask how often people smoke Marijuana.

The party identification breakdown of the sample reveals 39 percent self-identified Democrats and only 30 percent Republicans among registered voters. It shows a similar breakdown of 40 percent Democrats and 32 percent Republicans among likely voters.

If the electorate turns out that way on November 2, Democrats will indeed do better than expected. But it’s not going to look like that – at least based on a lot of other polling.

For example, the average Democratic advantage is 3.5 points in self-identification in all the September and October polls reported on These results are also a combination of registered and likely voter samples. Most polls consistently show a more significant improvement in the GOP direction as sample screening moves from registered voters to likely voters. Gallup, for example has consistently shown such a trend.

Survey organizations hope their samples produce enough completed interviews to match the target population – in this case registered and likely voters. But sometimes the process produces too many people of one party or the other. This Newsweek poll looks like good news for Democrats because it includes too many Democrats.

Next Page