Midterm congressional elections: ET papers

Connoisseurs of statistical practice will enjoy, in the ZBAU graph, the heaping of t-values
between 2.0 and 2.4, as regression coefficients stumble over the line to reach statistical
significance (recall the rule of thumb about the sanctification value of t-values exceeding
2.0). This curious heaping suggests that scholars might have searched through many model
specifications to root out what Bill Nordhaus, my colleague in economics, called the
“Darwinian t-statistic,” the selection of the fittest.

A major analytical issue in recent elections is the apparent loosening of the link between
economic prosperity and support for the white house party. The presidential election
models based entirely on economic conditions predicted a Gore landslide in 2000, for
example.

In the current midterm congressional election, why is it that the current prosperity
(benefiting the Republicans) seemes to be swamped by other factors? One way to answer
that question is that the Republicans would be even worse off if the
economy were in recession.

But there is something of a historical decoupling. Why?

Favorable macroeconomic conditions prevailing in election years might help the incumbent
party less than in the past because the benefits of good times now go largely to the top 10% or 20% of the overall income distribution. Thus prosperity that is not widely distributed fails to turn directly into votes, although it may turn into campaign contributions induced by tax cuts.

In short, the story in the past was that a rising tide raises all boats; nowadays a rising tide
raises all yachts.

Most of the news forecasts for the House predict a 15-20 seat Democratic gain.
These forecasts are based on district-by-district analysis in what are judged the most
competitive seats (only about 8% to 10% of the 435 House elections).

This is probably
an undersample of the number of Republican seats actually in play. In turn, my guess is
that this is an underestimate of the seat-change because the turnover rate is an esculating
function of
the size of the swing toward the Democrats; the larger the swing, the
disproportionately larger the turnover in seats.That’s because there are very few
really competitive seats (where the district is divided, say 52-48, for either side), but once
the swing toward the Democrats exceeds 3% or 4%, then a lot more Republican seats come
into
play.

So I’ll say >30 seats gained by the Democrats, which should be discounted by the unknown Republican
turnout effects, by a November surprise, and by my enthusiasms. So an interval forecast
would be 25 to 35.

This is from someone who, years ago as a consultant to the
New York Times, helped to elect Elizabeth Holtzman to the US senate in 1980 on
the first
edition of
the front page of the New York Times on election night. By the next day, Al
D’Amato was elected
senator in a
very close vote. This one of my two errors in helping to forecast probably a hundred
races on
election night for in various elections over the years for NBC, Newsweek, CBS, and
The New York Times. As I recall, we had an apparently somewhat solid exit poll
(which I had not looked at in detail or with my usual skepticism)
and some gossip that ABC had it for Holtzman. I wrote a note to myself afterwards about
the
importance of always going off quietly, away from the hectic newsroom, and looking at the
data, saying what I saw in the data, and letting the editors and
writers handle all gossip input. Calling the races on election night is part of a
bureaucracy so there were other people who participated in the mistake.

My other error was at CBS. I was working for Bill Moyers and his writers, analyzing a
midterm election, and made a prediction that was broadcast fairly early on election
night. I substantially over-estimated the
House shift toward
the Democrats (probably by 8-10 seats, it turned out). This number, when broadcast,
prompted an immediate phone call from an outraged well-known Republican honcho who
denounced
the prediction, correctly as we found out by the end of the evening. My error was to
over-estimate the swing ratio (the national swing in seats resulting from the national
swing in votes for Congress). I am using this same methodology for 2006, except without
any real data this time. On Wednesday morning, however, the swing ratio for 2006 will
have been exactly calculated..

Today’s press conference, George W. Bush: “The amazing thing about this election, and what
surprised me somewhat, which goes to show I should not try punditry, is that this economy’s
strong. And, a lot of times, off years are decided by the economy.”

As discussed above, the apparent delinkage of economic prosperity with support for the
White House party is possibly the result of the changed distribution of prosperity beginning
in the 1980s (the Rising Yachts Theory).

The other variable in the model is
presidential approval rating in the pre-election run-up (Bush’s approval rating was the second lowest in the 16 midterms since 1946) data here.

Also some 5-10 House seats were
lost by the White House party because of local district scandals. It is difficult to model, over a
series of midterm elections, the sometimes very powerful effects of election-unique or district-unique variables
such as Iraq, corruption, differential turnout, and so on. Sometimes after-the-fact dummy variables are added to the fitted equations to account for a special factor in a single election (and possibly to get better estimates of the coefficients on the more general variables in the model), a technique which never appealed to me despite its boost in R-squared.

Insider projections, apparently real not just for pitching (from Newsweek):

“Rove’s miscalculations began well before election night. The polls and pundits pointed to
a Democratic sweep, but Rove dismissed them all. In public, he predicted outright victory,
flashing the V sign to reporters flying on Air Force One. He wasn’t just trying to psych out
the media and the opposition. He believed his “metrics” were far superior to plain old
polls. Two weeks before the elections, Rove showed NEWSWEEK his magic numbers: a
series of graphs and bar charts that tallied early voting and voter outreach. Both were
running far higher than in 2004. In fact, Rove thought the polls were obsolete because
they relied on home telephones in an age of do-not-call lists and cell phones. Based on
his models, he forecast a loss of 12 to 14 seats in the House–enough to hang on to the
majority. Rove placed so much faith in his figures that, after the elections, he planned to
convene a panel of Republican political scientists–to study just how wrong the polls were.
. . .

The numbers looked a lot less rosy to the other architect of the campaign–RNC Chairman
Ken Mehlman. It was Mehlman who built the much-vaunted turnout machine. But he
feared that many inside the party were relying too much on technology, like voter
databases, and had lost sight of the bigger picture: that voters were turning against them.
“We’ve built a great new car, but the gasoline for the car isn’t us; it’s the candidates and
the issues,” Mehlman told NEWSWEEK. There was no bigger issue than the war, which Rove
had pushed as a winning theme for the GOP. As he flew back to D.C. on a private jet two
days before the elections, Mehlman scribbled his predictions on a card–not to be
revealed
until after the elections. His numbers were much closer than Rove’s: the GOP would lose
23 in the House (5 short of the final tally), 5 in the Senate (1 shy) and 6 governors (spot
on). Last week Mehlman announced he would step down and pursue opportunities in the
private sector.”

A big list of election predictions (many of them wishful thinking) at

http://thinkprogress.org/2006/11/07/2006-predictions/

Looks like Larry Sabato of the University of Virginia nailed it.

With a surprise result yesterday in a Texas run-off, it is now 30 seats as the swing toward the
Democrats in the House! There is still one contested seat, which will probably go Republican.

http://www.cqpolitics.com/2006/12/rodriguezs_upset_win_in_texas.html