Path: bloom-picayune.mit.edu!snorkelwacker.mit.edu!americast.com!americast.com!americast-post
Newsgroups: americast.wpost
From: americast-post@AmeriCast.Com
Organization: American Cybercasting
Approved: americast-post@AmeriCast.com
Subject: Pollsters' 'Nutty' Calculations Added Up on Day That Counted;In Academia,
Date: Thu, 5 Nov 92 05:01:45 EST
Message-ID: <37.1992Nov5.050145@AmeriCast.com>

Pollsters' 'Nutty' Calculations Added Up on Day That Counted;In
Academia, However, One Well-Known Political Crystal Ball Lies
Shattered<By Richard Morin<Washington Post Staff Writer<

For those who live and die by the numbers every four years, the
day after Election Day 1992 was a time to count winners and
losers.

Yale economist Ray Fair spent most of yesterday sweeping up the
pieces of his shattered crystal ball and explaining why his pred-
iction of a big win for George Bush went bust.

Political scientist James Campbell, newly crowned king of elec-
tion prognosticators, took the day off. In August, his model
predicted a 6 percentage point win for Bill Clinton, and an elec-
toral vote total of about 350.

As for those "nutty" pollsters, most spent yesterday cheerfully
answering the telephone, knowing that, for this election at
least, the callers would not be asking, "Why were the polls so
wrong?"

In fact, the six major news media polling organizations' last
preelection presidential preference polls not only had predicted
the right winner, but the final numbers actually had come close
to the results on Election Day. Even the exit polls were reason-
ably well-behaved this year, with a few notable exceptions.

"Any dispassionate observer would have to conclude that the polls
did very well this year," said Gallup's Larry Hugick, whose or-
ganization did a much-watched tracking poll for USA Today and
CNN.

But for the small fraternity of academics who specialize in
developing mathematical models that predict the presidential
vote, 1992 will be remembered as the year when some of their
best-known systems went bust.

Yale's Fair, perhaps the dean of the academic model-builders,
took the biggest fall. He predicted Bush would beat Clinton by a
whopping 12 percentage points in a two-party race.  That is the
biggest error Fair's model has ever made in any of the 20
presidential elections to which he has applied his model -
several times bigger than the model's standard error of plus or
minus 3 percentage points.  "No question, it was wrong," he said.

Why? One possibility is that people may have based their choices
on a longer recollection period than Fair had assumed. His model
includes changes in the gross national product during the previ-
ous year and inflation over the previous two years to predict the
presidential vote. Perhaps, he suggested, "the horizon of voters
is longer than six to nine months before the election. They may
be looking back much further, and looking at the performance of
an economy that had not been doing well for quite a while."

"This huge mistake is quite interesting, from a scientific per-
spective" he said.

Months ago, top prognosticator Campbell, who works at the Nation-
al Science Foundation, and Thomas Mann, director of governmental
studies at the Brookings Institution, co-wrote a paper in which
they offered up what appears to be the most accurate forecast of
the 1992 campaign season.

Unveiled at the American Political Science convention last sum-
mer, the paper The six major news media polling organizations not
only predicted the winner, but the final numbers were close to
the results on Election Day.<

>

predicted a 6 percentage point popular-vote victory for Clinton
over Bush (Clinton's final margin was 5 points) and between 350
and 400 electoral votes. "Jim was right on target," Mann said.

But the major media polls were not far off. All six had predicted
that Clinton would win by between 6 and 8 percentage points, and
one had hit Clinton's 43 percent share of the vote right on the
button.

Numerous skeptics, however, remained unimpressed. They criticized
the media polls for once again being dominated by "horse-race"
journalism during the final week of the campaign.

Some singled out Gallup for particular criticism. "Gallup comes
away from this election year with its reputation tarnished," said
Mann, who recently collaborated on a book on the news media's use
of the polls. "Some of their behavior was inexplicable. They will
have some explaining to do to those in the polling community."

Mann was particularly critical of Gallup's decision last week to
shift from basing its results on registered voters to likely
voters. One result was that Clinton's double-digit lead was al-
most immediately erased.  Then, in its final poll, Gallup allo-
cated virtually all of the undecided vote to challenger Clinton.
As a result, Clinton's lead in the Gallup tracking poll grew from
2 points to 8 and then to 12 on Election Day.

All of this movement "really undermined the legitimacy of all
polls," Mann said. "To an outsider, it looked like they didn't
know what they were doing."

Gallup's Hugick disagreed. He said Gallup typically waits until
the final week of the campaign before it starts basing its
results exclusively on likely voters. "The questions that we use
to determine likely voters work better closer to the election,"
when people have a better idea of whether they will vote or not,
Hugick said.

He also said changes in Clinton's lead over the weekend reflected
real shifts in the electorate. But he seemed to agree that
Gallup's decision to allocate all the undecided vote to Clinton
may have, in retrospect, been unwise.

Gallup made the decision based on its analysis of previous
presidential elections, which suggested that virtually all unde-
cided voters in presidential elections go to the challenger - who
they believed would be Clinton. "What seems to have occurred is
that voters who traditionally had swung to the challenger in the
end, swung to a different kind of challenger, Ross Perot," Hugick
said.

Election Day exit polling by Voter Research & Surveys, the telev-
ision network consortium, also went generally well.

One conspicous slip-up was the premature call of Georgia for
Clinton. VRS first gave Georgia to Clinton early on election
night after its state exit poll showed a big Clinton win. When
early vote counts indicated a tight race, VRS changed its deci-
sion, saying the race was too close to call.  Later, after most
of the votes had been tallied and Clinton was maintaining a mod-
est lead, they again called Georgia for Clinton, who eventually
won there by the narrowest of margins.

Some VRS subscribers also wondered why exit poll results through
late afternoon suggested that Clinton held an 11-point lead na-
tionally when the final vote was much closer.

"The numbers during the day are only for guidance, and I tell
everybody not to be surprised if they changed, based on the
results of later exit polling and other adjustments," said Warren
Mitofsky, who heads VRS.<02:29 11-05C9999-----

Copyright 1992, The Washington Post.  This story is from the
Washington Post's Capitol Edition On-Line and is not to be ar-
chived or redistributed.

For more information, send-email to American Cybercasting Cor-
poration (usa@AmeriCast.COM)