Sunday, July 22, 2012

Skewing The Polls?

I have long wondered about the manner in which American public opinion polls sometimes seem to be manipulated to achieve a predetermined result.  Recently I asked David Farrar to put up some commentary on Kiwiblog and he did.  Thank you David.

However, yesterday I came across two pieces from the US, one from The Weekly Standard and one from Huffington Post, both pretty much pointing the finger and shouting ‘BIAS.’
The thrust of argument in both pieces focused on two practices; first, the use of ‘likely’ voters versus ‘registered’ voters and second, the weighting of the poll sample to produce a predetermined ratio of Democrat/Republican/Independent respondents.

Regular visitors to No Minister will know Adolf relies heavily on the Rasmussen poll because it uses ‘likely’ voters which seems to me to provide a more accurate picture of reality. (Let’s face it, would you include Catholic nuns in a survey of preferred contraceptives?  So why include people who are not going to vote in a survey of political opinion which is supposed to be a guide to voting results?)  What I had not known before was that all the major polls switch to ‘likely’ voters a month out from election day.  It’s almost an admission of guilt.

However, the more important argument is the ability to ‘steer’ a poll by manipulating the sample.  If you want Obama to look good you boost the percentage of registered Democrats in the survey and visa versa for Romney.  It seems to me quite odd that polls can tilt the scales in this way.  The money quote comes from Jay Cost who asserts that currently polls are grossly understating support for the Republican challenger.

“….The quarter-century trend (1984 through 2008) has seen an average Democratic turnout edge of 3 points, 39 percent Democratic to 36 percent Republican. I’d expect something roughly similar this time around.

The clear answer is: they skew Democratic. In fact, every recent registered voter poll with party spreads I could identify had a Democratic advantage that exceeded the quarter-century trend. And the average of all these polls together is 35 percent Democratic to 29.5 percent Republican, or D +5.5……”
So, all the ‘in the tank for Obama’ media polls have over sampled Democrats by a range of two to five points?  They must get their sampling done by the same guys who set up the climate data to support the great global warming hoax.  The measurable historical data from exit polls over twenty five years shows a three percent advantage in Democrat voter turnout but the ‘climate models’ produced by most of the pollsters have ‘adjusted’ the advantage to five and a half percent.
Now let’s see what Mr Blumenthal has to say at Huffington Ppost.

“Are national pollsters over sampling Democrats? That's a charge made by some conservative pundits, and they may have a point: If the most recent polls could accurately predict which self-identified registered voters actually would cast ballots in November, their horse race numbers on the presidential election likely would tip a few points more Republican…….”

Just imagine what fun you can have adjusting all those battleground state polls back to reality.  Here’s how they would look.

Ohio                 Obama                         by 1.8%
Virginia             Romney                        by 1.3%
Florida             Romney                        by 1.4%
Iowa                Romney                        by 1.2%
N Carolina       Romney                        by 2.9%
Colorado          Obama                         by 0.4%
Nevada            Obama                         by 2.7%
Missouri           Romney                        by 5.5%
Wisconsin         Obama                         by 2.3%
Michigan          Romney                        by 0.7%
Pen’vania         Obama                         by 4.9%
New H’shire    Obama                         by 2.8%

No wonder the JackAss and his gangster mates are panicking.  


Iron Trike Myson said...
This comment has been removed by a blog administrator.
Iron Trike Myson said...
This comment has been removed by a blog administrator.
Anonymous said...

Gah! The 2 issues you identify are important sampling issues - polling 'likely' vs 'registered' voters, and weighting poll results to match the expected Democrat/Republican split in the registered voter population.

Sadly, the rest of your post has nothing to do with either of these issues. You talk about polls being skewed in their weighting towards Democrats, but the Huff Post quote you cite gave no evidence - just aspersions, and the Jay Cost quote only showed how little Jay Cost understands polling.

Cost notes Democrats are over-represented in exit polls, but that just means Democrat voters are more likely to answer such exit polls, not that there is a sinister pollster plot to misrepresent voter intentions.

And exit polls are meaningless - they just attempt to predict election outcomes a few hours early; everyone ignores exit polls once the actual election outcome is known a few hours or days later. Jay Cost then extrapolates the exit poll 'skew' to claim the last batch of polls are skewed in favour of Democrat voters - again without giving any evidence of that.

Sampling always carries the risk of getting a Democrat/Republican voter split that does not exactly match the D/R split you see in elections. The question is how do you deal with that? Either simply acknowledge it, or try to compensate for it by weighting the survey results to match the expected D/R split.

The pollsters did the former. No story. There would be a story if they weighted the results, as that is more dubious statistically. But the pollsters didn't do that, so Jay Cost just seems confused.

Mad Marxist.