Friday, September 28, 2012

Polling Conspiracies

I once wrote a bad poem about Conspiracy Theories.  It began like this:

There are five fat men in a secret  cave somewhere.
They are naked.  They are bald.  They live in a wave of purified air.
They read the Playboy and the Joy of Cooking.
They have grown obese and sated from just looking.

They can feel us.  They can see us.
They know our petty schemes.
They have planned us.  They have manned us.
They compute our secret dreams.

And they caused it to be thus:
Nothing that we do is quite as it seems.
Dreadful, isn't it?  But also fun.

I bet you have heard about those polling conspiracies  if you live in the US.  It all started with this site, run by one Dean Chambers (whose favorite sites include Rush Limbaugh).   He argues that almost all political polls are biased in favor of the Democrats and against the Republicans because they ask too many Democrats and too few Republicans for their opinions.

In order to fix that bias, Chambers takes all those polls and then applies the method one pollster, Rasmussen, uses to estimate party membership numbers.  Because Rasmussen's method often produces more Republicans than Democrats than the methods of other pollsters*, Rasmussen's polls are the ones most likely to find  that Republicans, in general, do better.

Chambers doesn't interpret his "corrections"** just as deciding to apply the Rasmussen figures (not necessarily applicable to other polls, in any case) to everything in sight.  Nope.  He states that all the other polls are skewed or biased!  Because they are frequently carried out by that evil imaginary monster:  the leftish mainstream media.  That Fox News' own polls fail Chambers' test for unbiasedness doesn't matter.  Everybody must do as Dean says, pretty much.

This mess has at least two central tangles.  One has to do with the statistical questions, the other with the conspiracy theories.    The statistical tangle is created by the goal of the polls: to predict behavior in next November's elections while asking people about their views today. 

To get from the latter to the former requires not only careful sampling to get a representative group of American adults but also some way of predicting which of those adults are not going to vote and which are.  It is the views of those who are most likely to vote in the future that the pollsters want to tap, and it is in the way the polls arrive at that group and the variables which determine its choices that we find the disagreements.***

Those disagreements are fine and useful.  Dean Chambers is not interested in them, because he has already decided that only one poll is unskewed and the rest of them horribly biased.  And the reason?

Chambers said he suspects that big polling shops are intentionally over-sampling Democrats to help get the President re-elected.
"This year, they've been more skewed than in the past. Any poll that says NBC, CBS, or ABC is going to be skewed and invested in trying to get this President re-elected," he said.
But while Chambers' methods may appeal to conservatives, other pollsters say their samples reflect reality — not wishful thinking — and that the higher Democratic numbers are similar to those in the most recent presidential election.

Conspiracy!  I love it.  Fox News joined in right away:

It’s clear that Fox & Friends has a tricky relationship with data. But on the show Thursday morning, the hosts took that relationship to a whole new level.
After a raft of new polls showed Obama opening up leads in swing states, the Friends flew in to full-blown conspiracy mode about what’s really behind the data.
Parroting the latest Republican meme that national polls oversample Democrats, host Steve Doocy threw in to the mix the possibility that pollsters are using voter turnout from 2008 to guide who they should be asking. And why would the “left-based mainstream media” do this? Doocy had an answer.
“Well, two reasons,” he said. “One, perhaps, to keep Mitt Romney’s donors from coughing up more cash. And two, to keep people from doing early voting.”

Of course conspiracy theories can be created in reverse:  Perhaps Rasmussen stands out so much because it's a water-carrier for the extreme conservatives in this country?

I don't mean that seriously.  But anyone who believes that the owners of media outlets in this country, other than Fox News,  are overwhelmingly some sort of extreme left-wingers is seriously deluded.

*I was unable to find information on what the Rasmussen method consists of, other than some kind of sampling with something else.
** Chambers seems to just re-weight all the "skewed" polls with his own preferred party membership figures.  Because he believes there should have been more Republicans he adjusts all the findings to give Republican views a greater weight, one based on Rasmussen's calculations.
**For more discussions about what the underlying debate is all about, see here, here and  here.