Saturday 26 May 2012

The Games They Play

Political Polls and Wax Noses

The old saw has it that there are lies, damned lies, and statistics.  These days we might be justified in a bit of tweaking.  There are lies, damned lies, and opinion polls (which, after all, are a particular application of statistical maths). 

Many of our readers are likely too young to remember the election of Ronald Reagan, back in 1980.  Reagan, the arch-conservative Republican was running against the arch-liberal Democrat, Jimmy Carter.  Reagan won in a landslide--receiving (according to Wikipedia) the highest number of electoral votes ever won by a non-incumbent presidential candidate.  The odd thing about this, though, was that right up until the actual election, opinion polls were saying the race was neck and neck.
  "Too close to call," the pundits said. It was a salutary lesson in treating opinion polls with a truckload of salt.  At the very least one should lift the covers and have a good look underneath at the poll itself, how it was constructed, sample size, and how the results were massaged.  Yes, massaged.

Here is a contemporary example of a wax-nose poll--one that belongs in Madam Tussauds rather than in political discourse.  Once again, it is found in the United States, where political partisanship is an abstract art-form amongst media and the Chattering Classes.  According to a recent Washington Post/ABC poll, Obama is edging out Romney.  Good news for the Democrats to be sure.  Any massaging here?  You bet.

It turns out registered Republican voters were deliberately underweighted in the poll.  This from Mike Flynn of Breitbart News:
Specifically, the Post poll assumes a collapse in GOP turnout. The partisan breakdown of the poll is D-32, R-22, I-38. In other words, only 22% of the voters sampled were Republicans. If only 22% of voters in November were Republicans, it would be about the lowest turnout for the GOP in modern history. 

In 2010, 35% of voters were Republican. In 2008, the year Obama swept into the White House, 32% of voters were Republican. Even in 2006, the year Democrats took control of Congress, 36% of voters were Republican. . . .

I guess its possible that a huge chunk of the GOP simply disappeared in the last two years, but it seems unlikely. If you adjusted the poll sample to something approaching reality, it would probably show Romney with a sizable lead against Obama. Which the mandarins at the Post would dutifully report on page A-17.

Remember, the partisan screen on this poll isn't an accident or quirk of the sample. It's the direct result of specific choices made by the pollster to "weight" the sample to reflect demographic and other characteristics of the electorate. If 22% of the sample are Republicans, its because the pollster "weighted" the poll to be that way. (Emphasis, ours)
If this sort of nonsense continues, we expect that right up to polling day in the US presidential elections, the pollsters and mainstream media will be feverishly telling us that the race is "too close to call".  And, if history were to repeat itself, Romney will win with a landslide. 



No comments: