Friday, March 16, 2012

Obama At 50% Approval? You DON'T Say?!

Anybody see those new polls out this week claiming Obama is currently at 50% Approval rating or better?

Ed Morrisey at Hot Air did. I don't think there's anybody better at explaining how the Old Media 'tweaks' their polls to get the numbers they want.

First it was Reuters:

Reuters Samples 51% Dems To Get Obama To 50% Approval

At this point, you’d normally see me analyze the sample type, size, and composition for a critical analysis of whether this particular poll is reliably predictive.  Oddly, Reuters doesn’t provide a link for its readers to peruse the data set — but that’s no problem, as it turns out, because Reuters provides its own credibility-killing data at the end of its report:

The Reuters/Ipsos telephone poll of 1,084 adults included 554 respondents who identified themselves as Democrats, 421 as Republicans, and 109 as independents. The total respondents included 937 registered voters.

Er, that would mean that 51.1% of the respondents to the poll were Democrats, and 38.8% were Republicans.   Now, I grant you that Republicans have been slightly overrepresented in this sample, as the 2010 exit polls showed 35% of voters were Republicans.  However, Democrats also only comprised 35% of the vote that year, not 51.1%, and only 39% the year Obama won election.
So the big news for Obama is that in a poll that consists of 51.1% Democrats, he gets a 50% approval rating?  Yeah, that’s a boost, all right.  And Reuters needs to go back to polling school.
Then Pew Research put out a poll that same day. Again Ed took a look at the numbers and found the usual 'tweaking':

Pew Research Claims Obama At 50% Approval

How did Pew manage to find this?  Scroll down to page 38 to get to the sample composition.  The overall sample totaled 1503 adults, of which 27.4% were Republicans, and 34.5% were Democrats.  Among the 1188 registered voters, the sample has a D/R/I of 37/30/30.  The 2008 exit polls showed a D/R/I of 39/32/29, and the 2010 election 35/35/30.  Are we to believe that Republicans will comprise a lower percentage of the vote in 2012 than in 2008?  Really?

Furthermore, Pew doesn’t mention that it switches from registered voters to general-population adults when it produces the presidential approval ratings.  One only discovers that by drilling down to page 42 of the report, where the total number of voters is 1503, not 1188.  That group only has 27.4% Republicans, and it’s also the least predictive sampling type available.  They do actually calculate this among the registered-voter sample, which produces a slightly different 51/43 rating, one which would have made the same point — perhaps even more strongly — while not pulling a bit of a bait-and-switch on readers.

Ed is right:

518 Dems, 412 Repubs, a difference of 106. So they polled 106 more Dems than Reps.

Reg. Voters?

438 Dems, 358 Repubs, a difference of: 80.

At present, Repubs and Dems both make up 35% of national party ID.

This sample is laughable on it's face.  To get Obama to 50%, they had to sample 106 more Dem's than Republicans.

And today the poll follies continued with the National Journal releasing it's own poll:

National Journal Gets Obama To 50% Approval By Undersampling Repubs By 6-9 Points

How did they do it? Ed explains:

You know what else in this poll is similar to Reuters and Pew?  The composition of the sample.  NJ didn’t provide a link to the data in its article, but supplied it to me on request — and the D/R/I is about what you’d expect.  The sample gives Democrats an eight-point advantage, 34/26/35, which both overstates independents and vastly underrepresents Republicans.  Even in 2008, when Democrats surged to the polls after eight years of George W. Bush, the exit polls showed a seven-point advantage for Democrats, 39/32, which mirrored Obama’s seven-point victory in the popular vote.  In 2010′s midterms, exit polls showed a 35/35/30 split, which means that either the poll undersampled Republicans by six or nine points, depending on which turnout model one presumes this general election will most closely resemble. On top of that, the approval numbers are based on general-population adults, not the subsample of registered voters. 

Switching between Registered Voters and All Adults without telling you that is what they did is one of the oldest polling tricks in the book.

Many thanks to Ed Morrisey for exposing how the Old Media tries to push it's agenda with deceptive polling like this.

No comments:

Post a Comment