Election Update: Swing State Polls And National Polls Basically Say The Same Thing

Welcome to our first Election Update, FiveThirtyEight’s regular feature where we’ll, uh, update you about the presidential election. More specifically, we’ll use this column to look at the election through the lens of FiveThirtyEight’s forecasting models, which we launched last week.

If you read FiveThirtyEight in 2008 or 2012, you might remember that we used to update our forecasts once a day, usually in the late afternoon or early evening. Then we’d write a post like this one to accompany it. The timing meant our forecasts were often half a day behind as new polls came in.

So this year, we’ve switched over to running model updates as new data becomes available, sometimes several times per day. I won’t promise you that we’ll interrupt everything to run an update if an Idaho poll drops at 1 a.m., but this method should allow us to stay more up-to-date, especially during regular working hours.

We’ll still run these Election Update columns — a couple of times a week at first and, eventually, almost every weekday. However, their focus will often be more macro than micro — before Labor Day, it usually isn’t worth sweating individual polls. We’ll tend to focus on big-picture themes instead.

Here’s one theme that I expect us to revisit repeatedly, for instance: How are Hillary Clinton and Donald Trump polling in swing states relative to their national numbers?

This is a trickier question than you might think, in part because there’s more than one way to define “national numbers.” One way is to calculate an average of national polls. As of late Tuesday evening, for instance, Clinton led Trump by 5.3 percentage points in our adjusted average of national polls, according to our polls-only forecast. But go to the polls-only homepage, and you’ll find that the forecast has Clinton beating Trump by 6.3 percentage points instead. It’s not a huge discrepancy, but what accounts for the difference?

The reason is that our forecast models use both state and national polls to estimate where the national race stands. In fact, they put most of the emphasis on the state numbers. Historically, it’s been more accurate to take a “bottom-up” approach — first, forecast the vote in each state, then sum the numbers together — than to force everything to match the national numbers. (See the users guide to the forecast for the gory details on how the methodology works.)

So that means Clinton’s swing-state numbers must be really good, right? They’re pretty good — but not great, although she’s gotten some better numbers in the past week. Instead, her biggest comparative strength — and Trump’s biggest comparative weakness — comes from red states rather than swing states.

Here, for instance, are the numbers in red states, which I’ll define as every state that was more Republican-leaning than North Carolina in 2012. (Georgia, by this reckoning, is the bluest red state.) To keep things relatively simple, these figures show FiveThirtyEight’s unadjusted polling average in states where there’s been at least one poll since November, and compare it to President Obama’s result against Mitt Romney in 2012.

Romney won these states by an average of 16 percentage points in 2012, weighted by their turnout. By contrast, Trump leads them by an average of only 7 percentage points, a net swing of 9 points toward Clinton. She’s competitive in a few of these states, such as Georgia, Arizona and — more exotically — Utah and Kansas. If you saw polls from these states only, they’d be suggestive of a double-digit landslide against Trump.

In the swing states, however, the numbers look more like the final results from 2012, when Obama beat Romney by 3.7 percentage points nationally and by slightly more than that in the swing states. Clinton is outperforming Obama in Florida, North Carolina, Virginia, Wisconsin, Minnesota and Michigan. She’s underperforming him in Colorado, Nevada and New Mexico. And she’s running basically level with him in the other states, such as Ohio and Pennsylvania.

You can get more detailed than this if you like — by noting, for instance, that the Virginia polls are fairly out of date, and that there isn’t very much polling at all in New Mexico, whereas we have lots of data about Florida. Our models have various ways to adjust for those considerations. But on average, both our fancy models and the comparatively simple analysis I’m conducting here get you to the same conclusion. Clinton is ahead by 5 or 6 percentage points in the swing states, close to where national polls have the race.

Finally, we have the blue states, which I define as everything bluer than New Mexico in 2012. (Oregon is the reddest blue state.) Clinton leads by 19.1 percentage points in those on average, so Trump is probably dreaming if he expects to put states such as California and New York into play. Still, that’s slightly behind the 21.5 percentage point advantage that Obama had on average in these states in 2012.

All of this yields some slightly complicated conclusions. On the one hand, according to our models, Clinton’s state polls tell a stronger story for her than the national polls do. On the other hand, a lot of that advantage is concentrated in traditionally red states. If Trump underperforms in states such as Texas and Mississippi, that will hurt his position in the popular vote without compromising his Electoral College math — provided, of course, that he doesn’t actually lose them. Hence, our models conclude that Trump is more likely to win the Electoral College despite losing the popular vote than the other way around, although either possibility is unlikely.

Nonetheless, Trump’s overall position has improved slightly. He has a 22 percent chance to win the election according to the polls-only forecast, as compared with a 19 percent chance when we launched last Wednesday. And in polls-plus, which also accounts for economic conditions, his chances have improved to 29 percent from 26 percent.

As I mentioned earlier, Clinton has had some pretty good state polls over the past week. By contrast, Trump has gotten some comparatively good national polls. Our models put more emphasis on state polls than national polls, which would seem to work toward Clinton’s benefit in this case.

However, our models also put a lot of emphasis on the trend within individual polls — did Trump gain or lose ground in this week’s edition of the SurveyMonkey poll, for instance? The trends in national polls have been relatively favorable for Trump over the past few days; he has often gained a percentage point or two relative to where he stood in the previous edition of the same poll. By contrast, most of the new state polls were publishing data for the first time, so while they may have been pretty good for Clinton, they don’t tell us a lot about the trend.

I know it’s a lot of detail to take in; our models have a lot of checks and balances. Usually, they work to keep the forecast fairly stable from day to day, and the overall picture of the race isn’t much changed. The next several weeks could be more volatile, with the reaction to the FBI’s denunciation (but non-prosecution) of Clinton, the VP selections, and the party conventions all looming over the polls. More about those in the days ahead.

Check out our 2016 general election forecast.