By Nate Silver, The New York Times
03 November 12
f
you are following some of the same people that I do on Twitter, you may
have noticed some pushback about our contention that Barack Obama is a
favorite (and certainly not a lock) to be re-elected. I haven't come
across too many analyses suggesting that Mitt Romney is the favorite.
(There are exceptions.) But there are plenty of people who say that the
race is a "tossup."
What I find confounding about this is that the argument we're making is exceedingly simple. Here it is:
Obama's ahead in Ohio.
A somewhat-more-complicated version:
Mr. Obama is leading in the polls of Ohio and other
states that would suffice for him to win 270 electoral votes, and by a
margin that has historically translated into victory a fairly high
percentage of the time.
The argument that Mr. Obama isn't the favorite is the
one that requires more finesse. If you take the polls at face value,
then the popular vote might be a tossup, but the Electoral College
favors Mr. Obama.
So you have to make some case for why the polls shouldn't be taken at face value.
Some argue that the polls are systematically biased
against Republicans. This might qualify as a simple argument had it been
true on a consistent basis historically, but it hasn't been: instead,
there have been some years when the polls overestimated how well the
Democrat would do, and about as many where the same was true for the
Republican. I'm sympathetic to the notion that the polls could be
biased, statistically speaking, meaning that they will all miss in the
same direction. The FiveThirtyEight forecast explicitly accounts for the
possibility that the polls are biased toward Mr. Obama - but it also
accounts for the chance that the polls could be systematically biased
against him.
Others argue that undecided voters tend to break
against the incumbent, in this case Mr. Obama. But this has also not
really been true in recent elections. In some states, also, Mr. Obama is
at 50 percent of the vote in the polling average, or close to it,
meaning that he wouldn't need very many undecided voters to win.
A third argument is that Mr. Romney has the momentum
in the polls: whether or not he would win an election today, the
argument goes, he is on a favorable trajectory that will allow him to
win on Tuesday.
This may be the worst of the arguments, in my view. It is contradicted by the evidence, simply put.
In the table below, I've listed the polling averages
in the most competitive states, and in the national polls, across
several different periods.
First are all polls from June 1, the approximate start
of the general-election campaign, until the start of the party
conventions.
Next are the polls between the conventions and the first debate in Denver in early October.
Finally are the polls since that first debate in
Denver. It's been roughly 30 days since then. If Mr. Romney has the
momentum in the polls, then this should imply that his polls are
continuing to get better: that they were a little better this week than
last week, and a bit better last week than the week before. So these
polls are further broken down into three different periods of about 10
days each, based on when the poll was conducted.
What type of polling average is this, by the way?
About the simplest possible one: I've just averaged together all the
polls of likely voters in the FiveThirtyEight database, applying no
other weighting or "secret sauce."
If you evaluate the polls in this way, there is not
much evidence of "momentum" toward Mr. Romney. Instead, the case that
the polls have moved slightly toward Mr. Obama is stronger.
In 9 of the 11 battleground states, Mr. Obama's polls
have been better over the past 10 days than they were immediately after
the Denver debate. The same is true for the national polls, whether or
not tracking polls (which otherwise dominate the average) are included.
In the swing states, in fact, Mr. Obama's polls now
look very close to where they were before the conventions and the
debates. Mr. Obama led by an average of 2.3 percentage points in Ohio in
all likely voter polls conducted between June 1 and the debates; he's
led by an average of 2.4 points in Ohio polls conducted over the past 10
days. He trailed by an average of 0.5 percentage points in Florida
before the conventions; he's trailed by an average of 0.2 percentage
point in the most recent Florida polls.
Mr. Obama's polls are worse than they were in the
period in between the conventions and the debates. But they're better
than they were immediately after Denver; he's gained back one percentage
point, or perhaps a point-and-a-half, of what he lost.
What about the national polls? Aren't those still worse for Mr. Obama than they were before the conventions?
Actually, that isn't so clear. The one "trick" I've
played is to look only at polls of likely voters. Mr. Obama's national
polls looked superficially better before the conventions because many of
them were polls of registered voters instead, which do tend to show
more favorable results for Democrats. (You're welcome to say that polls
of registered voters have a Democratic bias.) We alerted you in August
to the prospect that there was a "gap" between the state polls and the
national polls, which was concealed by the fact that many of the
national polls at that time were reporting registered-voter results,
while most of the state polls were using likely voter numbers all along.
However, our method adjusted for the tendency of registered-voter polls
to be biased toward Democrats by shifting them in Mr. Romney's
direction. Some of what is perceived as "momentum" toward Mr. Romney is
in fact a fairly predictable consequence of the national polls having
flipped over to applying likely voter screens at various points between
August and October.
But now we're getting into all these complications! All these details!
I am aware - and you should be too - of the
possibility that adding complexity to a model can make it worse. The
technical term for this is "overfitting": that by adding different
layers to a model, you may make it too rigid, molding it such that it
perfectly "predicts" the past, but is incompetent at forecasting the
future. I think there is a place for complexity - the universe is a
complicated thing - but it needs to be applied with the knowledge that
our ability to understand it is constrained by our human shortcomings.
This critique fails, however, since the simplest
analysis of the polls would argue that Mr. Obama is winning. He's been
ahead in the vast majority of polls in Iowa, Michigan, Minnesota, New
Mexico, Nevada, Ohio, Pennsylvania and Wisconsin, and all the other
states where the Democrat normally wins. These states add up to more
than 270 electoral votes. It isn't complicated. To argue that Mr. Romney
is ahead, or that the election is a "tossup," requires that you
disbelieve the polls, or that you engage in some complicated
interpretation of them. The FiveThirtyEight model represents a
complicated analysis of the polls, but simplicity is on its side, in
this case.
Thursday's Polls
The polls published on Thursday ought not to have done
much to change your view of the race. The national polls showed little
overall trend toward either Mr. Obama or Mr. Romney, but they also had
Mr. Obama just slightly ahead, on average, in contrast to what we were
seeing immediately after Denver.
The battleground state polls on Thursday were
something of a mixed lot, in terms of results and quality. The most
attractive number for Mr. Romney is the poll of Ohio by Wenzel
Strategies, which had him three points ahead there. However, the polls
from this particular firm have been four or five points
Republican-leaning relative to the consensus, which the FiveThirtyEight
model adjusts for.
Or just keep it simple and average the polls together,
warts and all. You will find that Mr. Obama is the Electoral College
favorite.
