For much of December and early January, Democrats held a double digit lead in the RealClearPolitics average for the generic ballot–a poll that basically asks a national sample of voters which party they intend to vote for in the upcoming congressional elections. Today, that advantage is down to eight points, and on Wednesday a Monmouth poll showed Democrats with a narrow two point advantage. While it’s important not jump to conclusions based on one poll, the message from all the polls is clear–the Democratic advantage in the generic ballot has fallen by several points.
So how happy should Republicans be about these improved numbers? And how worried should Democrats be?
It’s easy to give short, preliminary, semi-squishy answers to these questions (e.g. the GOP should be happy whenever numbers move their way, Democrats still have a real advantage, etc.), and it’s often important to get an initial read on new numbers before they’re swept away by the next news cycle.
But it’s also important to make sure that we have good mental models and don’t overreach or underreact to new data or poll numbers. That’s why I put together this short explainer on generic ballot polls that talks through how stable they are in the long term and the short term as well as how to interpret the polls we’ll see in the coming weeks.
The generic ballot is pretty stable
In general, congressional generic ballot polls are stable. I’m not the first analyst to point this out. Harry Enten did a great regression showing that early generic ballot poll averages were strongly correlated with the eventual House popular vote result, and Nate Cohn of the New York Times‘ Upshot vertical showed that the average of generic ballot polls almost a year before Election Day typically aren’t that far off the average near Election Day.
I ran Enten’s regression with my own data (a combination of data from the Upshot, RealClearPolitics, HuffPost and political scientists Robert Erikson, Joe Bafumi and Chris Wlezien) and a slightly different timeframe (270 to 310 days until election day – from New Year’s Day to Friday, Feb 9th) and got very similar results.
The data isn’t perfect or plentiful, but you can see a real pattern. The polling average explains over three quarters of the variation in the final House popular vote, and the standard error is real but not so large that it precludes learning from the model. If you drop 2002 (an outlier election in which the aftermath of the September 11th Attacks and the ensuing foreign policy debates disrupted the usual midterm pattern) from the regression, the fit becomes tighter.
I plugged in the average of polls taken by RealClearPolitics since December 31 into this equation, it predicted that Democrats will win the House popular vote by nine points but left some room for error. Specifically, this very simple model says that although a high single digit Democratic victory in the House popular vote is the most likely outcome, anything from a mid single digit Democratic win in the House popular vote to a low-to-mid double digit win is still plausible.
These patterns aren’t laws of physics–it’s possible to think up scenarios where Trump and the Republicans to improve or worsen their position substantially. But the basic point here is that the current polling is probably telling us something meaningful about where the electorate will be in November.
There will still be bumps
Despite this stability, the generic ballot can still have its fair share of day-to-day bumps. Take a look at the RealClearPolitics generic ballot average for the 2014 election cycle.
Democrats led for almost all of 2013, and they surged during the a government shutdown in October of that year. In 2014, Republicans managed to gain ground, trading the lead back and forth with Democrats until they surged in September and won by a substantial margin. A Republican victory in 2014 made sense – Obama’s approval rating wasn’t great, the president’s party typically loses seats in midterm elections and polls often do shift away from the president’s party as the midterms approach. But, despite the broad predictability of a strong Republican showing, we still saw some volatility in the polls.
The 2006 election is another case where, despite gains and losses of a few points at various times, the overall picture was relatively clear:
This graphic shows generic ballot polls collected by the New York Times from the 2006 cycle. Individual polls showed a wide range of results–everything from no advantage at all to a 20 point or larger victory for Democrats. Moreover, the trend line (drawn using a local regression) had some bounces here and there throughout the cycle. Some of these bounces stuck, but some of them reversed themselves quickly. And overall, 2006 fit into a broad pattern–when a president is unpopular (as George W. Bush was at that time) then his party is going to end up losing ground in the midterms.
I could go into greater detail, but the point is that even in midterms that follow the expected pattern, there are still small ups and downs.
This leaves us with a puzzle–if current generic ballot polls are generally stable and helpful for predicting the eventual result, how should we interpret the day-to-day rises and falls in polling aggregates?
Learn, but don’t overreact
We can learn from these ups and downs without overreacting to them.
For example, in January the Democratic edge in the generic ballot fell from the low double digits to about eight points according to RealClearPolitics. Just a few weeks before that, the president saw a holiday bounce in his approval ratings–a quick uptick that I explore more here–that he’s been able to basically sustain through the month of January.
This uptick helped Trump and the GOP. Democrats would obviously prefer to be up by double digits rather than single digits, and I’m certain that a probabilistic model would show that Democrats took a real hit. That said, fundamentals still favor the Democrats. Their generic ballot advantage is similar to what it was a few months ago, the president is still unpopular and the president’s party typically suffers in the midterms. I’m more bullish than some, but I’d put Democratic chances of retaking the House somewhere around 60-65 percent.
Changes in the polls can also help us learn about the Trump phenomenon.
The fact that Trump’s approval rating has increased as the Democratic lead in the generic ballot diminished reinforces the idea that voters strongly mentally link Trump and the GOP.
It also suggests that Trump fares better when he keeps a lower profile.
I haven’t directly measured this, but it seems like January 2018 has (with some notable exceptions) featured fewer cultural brawls and headline-grabbing Twitter fights than previous months. The policy events of the past few weeks likely haven’t caused lasting damage to Trump’s public image either. Voters cast more blame on Trump and the GOP than on Democrats for the shutdown, but it’s not likely to do much damage – typically voters move on from shutdowns pretty quickly. Moreover, the GOP’s tax reform bill (signed in December) looks like it might be slowly becoming more popular.
In other words, I would pay attention to changes in the generic ballot, but not swing your assessment of 2018 wildly in either direction just yet (both parties still have time to change their numbers). But I would watch the numbers as they come out, because they give us a better understanding of how the public is reacting the strange political times we live in.
