Saturday, August 28, 2010

Newsweek Poll Creates more Questions than it Answers

Please see my latest post, regarding the New Newsweek poll.

Update: According to Mark Blumenthal of Princeton Survey Research provided him with the data on the missing cases that were not included in the offical release. With that data in hand, Blumenthal was able to replicate the survey results and the 45% to 45% tie. My original post was intended to criticize Newsweek for releasing data that did not allow for replication of its reported results. I stand by the criticism. I would add as well that given the impact that the inclusion of the 5% of unaffiliated voters had on the final results they should have been included in the final release. That 5% overwhelmingly favored Democrats - a fact that should have been addressed - or given the small size of the sub sample discarded.

Original Post:
In a front page story on their website Newsweek proclaims "NEWSWEEK Poll: Democrats May Not Be Headed for Midterm Bloodbath." According to Newsweek, the poll's results suggest  "the party is better off this year than Republicans were in 2006, when the GOP lost 30 House seats, and than the Democrats were in 1994, when they lost 52 House seats."

What did the poll find?  That President Obama's approval rating is at 47% and Democrats and Republicans are tied at 45% on the generic ballot. Compared to other recent polls the Newsweek survey does show the Democrats to be in far better shape - problem is, the Newsweek poll actually confirms what other polls have found and should give Democrats great cause for concern.

Newsweek reports that Republicans and Democrats are tied at 45%. The folks at Newsweek were kind enough to break down that support by party - see Table One.

According to Table One, Democrats are winning 90% of the Democratic vote and Republicans are winning 94% of the Republican vote - Independents favor Republicans by a 45% to 33% margin - resulting a 45% to 45% tie...
Something seem fishy? It should. Consider the breakdown by party - 284 Republicans, 280 Democrats and 247 Independents - that's 33% Republican, 33% Democrat and 29% Independent (a few folks must have reported no party).  It is mathematically impossible for Democrats and Republicans to be tied at 45% if each party is receiving similar shares of their own party vote, but Independents prefer Republicans by 12 points.  If Democrats are winning a few less of their own votes as compared to the GOP, and they are losing Independents - they simply cannot be tied.
If you apply the actual sample size to the reported results then you end up with what is shown in Table Two.

The two right columns show the actual share of the respondents who supported each party - if one looks at the actual sample then Republicans are leading the generic ballot by a margin of 46% to 40%. Contrary to the headline and tenor of the story, a 6 point lead for the GOP among registered voters would signal a coming bloodbath. It would also be consistent with most recent polls.

Newsweek includes a disclaimer that reads: Data are weighted so that sample demographics match Census Current Population Survey parameters for gender, age, education, race, region, and population density.
Notice that no mention is made of party identification - but the only way to arrive at a 45% to 45% split is by significantly altering the party identification of the survey respondents. I do this in Table Three.
                                *** See Update

In order to get to a 45% to 45% race, one must change the sample such that 37% are Democrats, 30% are Republicans and 33% are Independents. There are few other combinations that get you to 45% but all involve a significant increase in Democrats and considerable decrease in Republicans. Newsweek offers no explanation for such a dramatic change to its own sample, but there is no reasonable argument for such a party breakdown (According to data provided to Mark Blumenthal by Princeton Survey Research the 5% of unaffiliated respondents supported Democrats by a 40% to 29% margin - this data was not included in the official release).
In the 2008 election Democrats were 39% of the electorate, Republicans 32% and Independents 29% - no one believes that Democrats will hold a similar advantage in 2010. In 2004, the last good Republican year, Democrats and Republicans were 37% of the electorate and Independents were 26% - roughly equal to the actual Newsweek sample.

If you want to become truly confused, look to the last page of the survey where Princeton Survey Research (Newsweek's pollster of choice) reports the results of the following question "In politics TODAY, do you consider yourself a Republican, Democrat, or Independent?" They report that respondents were 32% Republican, 36% Democrat, and 27% Independent - so the last page provides a partisan breakdown that does not match the breakdown reported on the first page. Worse - if you plug the reported breakdown into Table Two or Table Three Republicans still lead by 47.4% to 42.6% - the poll is pure nonsense.

Now it's possible that after weighting for gender, age, education, race, region, and population density the partisan ID of the sample would change - but if one reasonably assumes that the party ID shown on the final page of the survey is the weighted breakdown then the poll's results still cannot be explained. Newsweek should release fully the effects of it's weighting so that a true assessment of its poll can be made.
So Newsweek may be reporting that all is well and that 2010 will not be a bloodbath, but they had to do some serious data weighting to arrive at that conclusion. Something that Newsweek has done before and that I wrote about in National Review Online in 2008. When I "corrected" Newsweek's data back in 2008 I showed that their poll reporting an 11 point lead for Barack Obama over John McCain was more like a 4 point lead - guess who was closer to the actual election result (hint: It wasn't Newsweek then and I do not think that it's Newsweek now).