Friday, May 25, 2012

THE PRAIRIE EDITOR: Be Wary Of All Polls This Far From An Election

Most in the media today rely on polls for their analyses of upcoming
elections. This has been commonplace for many years, and lacking
other "objective" tools, it is understandable that, under the pressure of
needing to say something about an election contest, they make as
much use of them, particularly so far in advance of actual voting, as
they do.

The problem with this is the deteriorating reliability of most polling.

Some of this is the responsibility of the pollsters themselves who feel
they are required to "adjust" or "tweak" raw polling data to draw
legitimate conclusions. It is in this "fine print" of polling where most
polls fall woefully short. What are these adjustments? They include
categorizing (such as "registered voters" and "likely voters.") They
also include a subjective (and often prejudicial) decision of how many
Democrats, Republicans and independent or non-affiliated voters (or
how many men and women) to count. Sometimes, certain pollsters
(perhaps especially those who are working for a particular candidate,
political party or cause) also make "intuitive" adjustments based on
nothing more than predetermined attitudes and outcomes. In particular,
there are polls being published in some media which sample an
unjustifiable imbalance between respondents of various political

There are also increasingly factors affecting the reliability of polling
which are beyond the control of pollsters. Skepticism about polling has
led many voters either to refuse to be polled or to deliberately deceive
the pollsters. The rise of many interest groups who feel pollsters are not
reliable has increased the decline of participation in polls. When pollsters
who are gathering their data from random samples, the accuracy model
of their polls is really based on  receiving a response from the initial
voter queried. As pollster increasingly need to go back to a second,
third or more pool of voters, the accuracy of the poll is statistically
decreased, sometimes considerably.

Pollsters vehemently defend the use of smaller samples, especially on
statewide and national polls, as reasonably accurate. But with the factors
mentioned previously increasing rapidly, smaller polls are often
incredibly flawed, If you read or hear that a poll has an accuracy of
plus-or-minus 3, 4 or 5%, you should be very skeptical. Most polls with
samples under 1000, and with subjective "adjusting", really have an
accuracy much less, sometimes as much as 10% or 15%.

Most pollsters will acknowledge that their polls are  inexact and "only a
snapshot in time," but the actual use of polls, especially by many
journalists, is much more than that when they make pronouncements
about who is going to win or lose (or who is "safe" or not) far in advance
of an actual election. The best journalists (and there are many) always
make it transparent how provisional most polls are, and how easily their
numbers could change.

Those pollsters who deliberately manipulate their polls to a
predetermined outcome can and do hide their distortions behind the fact
that there is no objective way to prove they are wrong, other than by the
results of an election itself. One benefit of the plethora of polls today is
that any poll which is an "outlier" becomes easily suspect.

Polls taken just before an election, that is, two or three days before the
actual voting, thus tend to be more accurate. This is obviously
because, whatever "tweaking" a pollster has previously done in a race
must now minimized so that the final polls resemble the actual
voting. A notorious Minneapolis Star Tribune poll, taken just before the
election of statewide races in 1978, was so far off it was a national
embarrassment to the newspaper. Only later did the pollsters revealed
they had "doctored" their results. Polling is a business, and these kind of
experiences soon put irresponsible pollsters out of business.

The bottom line is that polling is a legitimate and necessary tool in
analyzing voter sentiment, but that the assumptions of a poll, often
disguised or buried, are vital to the accuracy of any poll. Polls taken
too far in advance of an election are more a curiosity than anything that
should be take seriously. Journalists who employ them too much instead
of using demographic trends, economic conditions, recent voter history,
carefully vetted biographies and developing circumstances. etc., are
simply being lazy.

A lot of polling has taken place in the 2012 election so far, at the state,
congressional and presidential levels. Much of it, especially the polling
taken early in a particular contest, has been off the mark. The volatility
of the electorate in the U.S. these days means that dramatic last-minute
changes in voter behavior take place. The recent nomination of a GOP
senate candidate in Nebraska, someone who had trailed two much
better-known opponents until the final days of the primary campaign,
is not an isolated phenomenon in current and recent campaign cycles.
This, in a variety of races and political levels, is occurring with notable

I particularly call attention to the polls of individual senate races
across the country, and of the presidential race, It seems to me that the
outcomes predicted by many of these polls are, for a variety of reasons
(many of which I have mentioned), misleading. As the summer passes,
and the autumn campaign begins in earnest, this will become
increasingly clear.

Copyright (c) 2012 by barry Casselman.  All rights reserved.

No comments:

Post a Comment