There are lots of fights about forecasting models — in 2012 and again this year. Often this is ginned-up controversy brought about by the same pathology: As with polls, people focus on whatever forecast is somewhat different and implies a competitive race. This year, it’s 538’s forecast.
Fortunately, we don’t have to decide among the individual pollsters, the various polling averages, and all the other forecasting models using different methodologies. What happens if we just average them all together?
That is the apparently radical idea of the site Pollyvote, and it’s backed up by research showing that averages of forecasts are better than single forecasts. Pollyvote has averaged dozens of forecasts using different methodologies: one prediction market, five index models, one periodic expert survey, an aggregate of citizen forecasts (see below), eight poll aggregators and 18 econometric (or fundamentals-based) models.
This is what Pollyvote currently shows:
The final forecast is for Clinton to have a five-point advantage in the major-party vote.
What about the electoral college map? There Pollyvote combines eight poll aggregators, five econometric (or fundamentals-based) models, four different sources of expert judgment, two prediction markets and one citizen forecast. The forecast is for Clinton to win 323 electoral votes to Trump’s 215.
2. Citizen forecasters get little publicity, but are among the most accurate. They also give Clinton a solid chance of winning.
A method of forecasting that’s mostly ignored in favor of econometric models and polls involves “citizen forecasters.” These are typically ordinary people asked who they think will win, not whom they are voting for. This is a good predictor of elections and, during the months leading up to the election, more accurate than any other method, including polls.
What about in 2016? These anonymous forecasters favor Clinton. At Good Judgment — a site built on scientific research showing that citizen forecasters make good predictions — Clinton has a 76 percent chance of winning.
Another data point: An Oct. 22-26 YouGov poll found that 57 percent of Americans believed Clinton would win.
3. Trump is underperforming.
All along, the statistical models of elections — predicated on the economic fundamentals — have forecast a tight race. I noted this in May and again in August. You can see a nice rundown at Pollyvote. These models would expect Clinton to have a half-point lead. If she does indeed end up with a five-point victory, then Trump will have fallen well short of what we’d expect from a typical Republican candidate. Five points is a significant cost in a presidential elections.
Indeed, Clinton’s forecast vote margin has been at least five points for much of the past several months, as the Pollyvote graph above shows. If she wins by that much or more, then here’s is a defensible summary of 2016: It was simply not a very competitive presidential election.
4. We’ve been chasing a lot of noise in polls.
One big issue — in 2000, in 2012 and again in 2016 — is whether variation in polls is due to actual shifts in preferences or to partisan differences in willingness to take polls, also known as “differential nonresponse.” Personally, I think a significant amount, though not all, of that variation is nonresponse.
Compare, for example, a poll-based forecast (538’s, in this case) to the Good Judgment forecasters.
They end up in nearly the same place, which suggests that some of the criticism of the 538 forecast is overblown. But simply following the polls leads to much wilder ups and downs. It’s just not clear that this movement as indicates fundamental shifts in Election Day odds.
5. Even if Clinton wins, she’s almost certainly going to face a divided government.
The focus on Clinton and Trump has obscured a more salient fact about what the election will likely produce: more divided government. Several weeks ago, Eric McGhee and I reported on our House forecast, which remains unchanged: We estimate that the Democrats, who currently control 186 seats, are likely to end the election with 204 seats, with a margin of error of plus or minus eight seats. (You can find our House predictions for all 435 districts here.)
This gives Democrats less than a 1 percent chance of taking the House, even though we also anticipate them winning a majority of House votes. Our forecast is in line with several handicappers — including the Cook Political Report and Larry Sabato’s Crystal Ball.
A similar model in the Senate — based only on underlying fundamentals, not on polls — generates better odds for the Democrats: a 69 percent chance of taking the Senate. This is in line with several, though not all, of the poll-based forecasts. (You can find our Senate predictions here. If you want a sleeper pick, watch Kentucky. The fundamentals suggest it should be a close race and the most recent poll has Rand Paul with only a five-point lead.)
If we factor in Clinton’s estimated vote margin — assuming it’s five points as at Pollyvote and she has typical coattails in down-ballot races — this doesn’t really change the House forecast. But it improves the Democrats’ chances of a Senate majority slightly: to 77 percent.
This might end up making it a little easier for Clinton to get her Supreme Court nominees confirmed. But these congressional election forecasts imply that Clinton will likely struggle to realize her policy agenda.
In that sense, things may not change much from what we’ve seen under the past six years of the Obama administration. Stay tuned.