1 vote

MSM Anti-Minor Party Polling Bias/BS Deconstructed

This is a slightly modified version of a post of mine to Nate Silver's
538 Blog at the New York Times. Since it will most likely never see
the light of day there, thought I would share it here.

Polls can be kind of addicting, but the GIGO (garbage in, garbage out)
principle applies - I can't see how people can get so worked up about
the smallest nuances of the results when the way the poll questions
are framed typically have no basis in reality. You are asked to choose
be between Obama and Romney when *no* voters are, in fact limited
to those choices. But Silver, Rasmussen and others seem to think it's
fine to pretend those choices don't exist/matter - but I found a
particularly graphic little example of how they do...

My (slightly edited) post to:

http://fivethirtyeight.blogs.nytimes.com/2012/11/01/oct-31-o...

"All through the general election cycle, Nate has been
accepting as valid polling that fails to ask specifically
about minor party support.

Respondents - when asked whom they plan to support -
are typically only asked about Romney and Obama by name.
In some cases, they will be given a "someone else" or "another
candidate" option but just as likely anything besides
support for Obama or Romney is classified as "undecided".

A look at the University of Cincinnatti poll here:

http://www.ipr.uc.edu/documents/op103112.pdf

Should dispel any illusions about how this distorts reality.

The same organization conducted four Ohio polls beginning
in late August. In the first three, the choices were Obama/Romney
(by name), "other" and "don't know".

The results for their October 18-23 poll?
Romney and Obama tied at 49%, "other" and "don't know" 1% each.

For their October 25-30 poll (of a larger sample), they switched to
asking about every candidate on the ballot by name. The result?

Obama 48%
Romney 46%
Johnson (Libertarian) 2%
Other minor party candidate 2%
Don't know 2%

Big shift in a week? Or a shift to methodology that
more accurately reflects reality?"

Note that the single change to asking the question in a way that
simply reflects the reality of the choice voters actually face in
filling out their ballots significantly shifts the results - minor
party support magically jumps 4x, undecided doubles.

Silver is obviously very bright - but the whole fancy edifice of
his models rests on some very shaky assumptions - guess we
should be glad he's a statistician and not an architect..