February 15, 2006
A Prediction for the 2006 House Contest - Part II
By Jay
Cost
Two
days ago, I began to discuss what I believe to be the best
way to predict congressional elections. Developed first by Edward
Tufte and later refined by Gary Jacobson, the “Tufte/Jacobson
theory” is starting point I have chosen. It argues that
we can predict the outcome of a congressional election based upon
three factors: exposure, presidential job approval and changes
in real disposable income per capita (RDI/cap). It is time to
supply the details, supplement the theory with some extra concerns,
and make a prediction for 2006.
Exposure
is the extent to which the party of the President is above its
historical average. For every 1% over its historical average,
the party of the President tends to lose 0.76% of its congressional
caucus in the next election, all else being equal. The Republican
majority has averaged, since 1994, 227.33 seats. So, considering
that today it enjoys 232, one might be inclined to assign a value
of 2.01% to exposure. However, this would fail to account for
the GOP mid-census Texas redistricting. Redistricting, from a
national perspective, typically does not benefit one party more
than another. However, the 1990 census resulted in a heavy and
sustained Republican advantage – and Jacobson adjusted the
exposure statistic to account for this Republican edge (by his
estimate, 42 seats). We must do the same to account for the 2004
redistricting, which netted the Republicans four seats in the
Texas delegation. This redistricting meant that the three net
seats they gained in 2004 were not a consequence of a real expansion
of their majority, but rather a slight shrinking of their
majority counteracted by redistricting. If we control for this
“inflation,” we would say that, in real terms, the
GOP actually has one fewer seat in the 109th than it did in the
108th – 228 “real” seats. Accordingly, let us
say that the GOP is overexposed by about 0.29%. Because every
percentage over its historical average costs a party 0.76% of
its delegation, we therefore expect this slight overexposure to
cost the Republicans 0.22% of their delegation, all else being
equal.
Next is
presidential approval. History tells us that for every additional
point of popularity a president enjoys in the final Gallup poll
before the election, he increases the percentage size of his congressional
party by 0.25%, all else being equal. The Gallup polls taken this
year show Bush pulling an average of 42.0% approval. His approval
rating obviously could trend in either direction in the next nine
months. However, given the stability of his job approval since
the beginning of 2006, I think it is fair to argue that he has
largely stopped the hemorrhaging and that he will still be at
about 42% approval on November 7. If this comes true, Bush’s
popularity will yield an increase in the GOP’s congressional
delegation of 10.50%, all else being equal.
The final
important statistic is real disposable income per capita (RDI/cap).
We are interested in its percentage change in the twelve months
prior to the election. In the wake of Katrina, RDI/cap rose by
a robust 3.7%. In the final quarter of 2005, it increased by a
brisk 1.03%. Between 2001 and 2004, it increased by about 2% per
year. It seems reasonable to conclude, given the pace of its growth
in the fourth quarter of 2005 and recent history not affected
by Katrina, that it will reach the 2% mark by election time. Though
I am not an economist, this actually seems to me to be conservative.
Indeed, for RDI/cap to reach a 12-month increase of 2% by October,
it need only expand at 0.097% per month; this will not require
much exertion for an economy whose GDP is expanding by more than
3% per year. In terms of congressional elections, every 1% increase
in RDI/cap yields a 1.29% increase in the President’s party
delegation. Thus, we can expect the state of the economy to boost
the House GOP delegation by 2.58%, all else equal.
Finally,
as I indicated yesterday, there is a baseline value. In other
words, if you assume that the party of the President is neither
over- nor under-exposed, that nobody likes the President, and
that RDI/cap has neither grown nor shrunk in the last year, you
can expect the party of the President to lose seats – specifically
17.7% of its delegation.
It is now
a matter of simple arithmetic: 10.50% (presidential approval)
+ 2.58% (RDI/cap) - 0.22% (exposure) – 17.7% (baseline)
= -4.84%. In other words, we can expect that the House Republican
delegation will shrink by 4.84%. This means that, according to
this theory, the Republicans will lose about 11 seats in November.
This is
only the beginning of the story. It is not the end. Between 1946
and 2002, Tufte/Jacobson explains 70% of the variation in congressional
elections – very good for political science work,
but obviously far from perfect. So, the relevant question is:
if we vary from this 11-seat mark, whom will the variation
favor? The answer depends upon what kind of variation we are discussing
– it can either be truly random variation or systematic
variation. If it is truly random, then no theory can improve upon
Tufte/Jacobson and we must leave things where they stand. If it
is systematic variation, it means that the theory is currently
missing an important causal process.
Truly random
variation, e.g. the entire Texas Republican delegation catches
encephalitis and cannot campaign, is impossible to predict and
certainly extant to some degree in any process. If Tufte/Jacobson
is a completely true theory of how congressional elections operate,
and the only reason the real world varies from its predictions
is because of random “noise”, a Democratic recapture
of the House is improbable but not impossible.
However,
I think there are regular processes out there that this theory
does not capture, but that favor the Republicans. These should
therefore cause this theory to overestimate Democratic gains.
As I hinted yesterday, I think this theory serves as the necessary
foundation for understanding what will happen in November, but
it does not capture the entirety of the process. It excludes certain
variables that will serve to mitigate Republican losses. This
is a criticism of the theory, but it should not be taken to mean
that we should dismiss what we have already developed. All good
social science theories necessarily simplify the real world, boiling
it down to the essentials so that we can understand the process
that exists “out there”. Thus, we should understand
the theory we have just reviewed as true but incomplete.
The first
important variable that we have thus far excluded is the incumbency
advantage. In the last fifty-five years, incumbents have increasingly
shielded themselves from electoral mood swings. In other words,
the incumbency advantage has expanded in the last half century
such that most incumbent Republicans survive in anti-Republican
years, and therefore seat changes in anti-Republican years are
almost always via open seats. Scholars like Oklahoma University’s
Ronald Keith Gaddie contend that exposure is less important than
open-seat exposure – i.e. the number of net open
seats that a party sports at election time. What matters is still
the number of marginal seats a party has, but seats where incumbents
are running tend not to be marginal. What of this year? Once again
the number of open seats is minute – less than 5% of the
House. David Wissing recently noted
that there are 20 open seats (13 Republicans, 7 Democrats), only
four of which (according to Stuart
Rothenberg) are “strong takeover possibilities”.
If Gaddie is correct and retirement is today the principal engine
of partisan seat swings in Congress, we should not expect
the Democrats to beat the 11-seat margin that we have outlined.
They will likely net less than 10.
The second
important variable that we have excluded has not yet been explored
systematically by political scientists. Nevertheless, I think
it is important. This is the tight alignment of the Republican
electorate – the GOP’s set of voters is more aligned
than it ever has been and it is more aligned than the Democrats’
set of voters. Until very recently, there were a large number
of Republican-controlled congressional districts wherein voters
chose Democratic presidential candidates. Today, there are a scant
17 congressional districts that voted for Kerry and a Republican
member of Congress. Meanwhile, there are 41 congressional districts
that voted for Bush and a Democratic member of Congress. While
we cannot know for certain, this probably means that the distribution
of his anemic popularity helps Bush. It is probable that anti-Bush
voters are less likely to be located in Republican districts relative
to other Republican presidents; pro-Bush voters are about as likely
as ever to be located in Democratic districts relative to other
Republican presidents. Thus, his job approval is probably more
“efficient”, historically speaking, at protecting
Republican incumbents, but still as efficient at damaging Democratic
incumbents.
Are there
other factors that are not captured by the model that favor the
Democrats? None come to mind beside congressional job approval.
The poor marks Congress currently receives are not captured by
the Tufte/Jacobson model, but it is unclear that this will have
a significant effect. As I mentioned previously,
there has been initial work exploring whether people’s perceptions
of Congress help or hurt the President’s party. However,
it remains incomplete and there are important questions that linger.
It also seems
unlikely that the Abramoff scandal will harm Republicans in any
significant way. Beyond Congress’s poor marks remain only
Bush’s poor marks – which the Tufte/Jacobson theory
captures and (probably) overestimates. There seem to be no strong
causal processes that we have not yet discussed that would favor
the Democrats. Indeed – we could probably tick off a few
that additionally favor the Republicans: their strong money advantage,
their strong mobilization advantage, the fact that they are not
saddled with an ostensible incompetent like Howard Dean, etc.
Thus, we
should consider an 11-seat swing in November as the maximum Democratic
gain. Factoring in the incumbency pushes my estimate to less than
10. Factoring in the tight alignment of the electorate pushes
my estimate to a Democratic gain of about 8 seats. Assuming that
(a) Bush’s popularity does not drop off, (b) the economy
does not stagnate and (c) Republicans do not have to defend more
net open seats, I predict that the 110th Congress will have 224
Republicans and 211 Democrats.
Jay
Cost, creator of the Horse
Race Blog, is a doctoral candidate of political science at
the University of Chicago. He can be reached at jay_cost@hotmail.com.
©
2000-2006 RealClearPolitics.com All Rights Reserved
