What will happen to the Senate next year? This is one of the crucial questions as we start to tiptoe toward the 2018 midterms. Democrats are facing an awful map – not the worst one ever, but one under which we might expect Republicans to make gains in an adequate environment. At the same time, President Trump’s job approval remains mired in the low 40s, although it has bounced back from previous lows in the high 30s. There are signs for hope for both Republicans and Democrats.
To help flesh this out, David has created the interactive tool below that is based on a successful approach to estimating elections developed by Sean in the 2014 midterms. Now, usually when people make these sorts of tools, they describe how awesome they are, then add a few caveats at the end about limitations, which everyone then ignores. We believe that the limitations are important, so we are going to state them up front.
First, we assume that things will continue to follow previous trends. If the general rate at which factor A affects factor B changes, however, the tool won’t work as well. Trump was an unusual candidate, and has been an unusual president. We might have reason to think that “the old rules don’t apply.”
Second, these sorts of tools are only estimates. Like all estimates they have error terms, and if we operate with 95 percent confidence (think of the “plus-or-minus” that accompanies polls), that means we expect one in 20 outcomes to lie outside of that confidence range. In other words, things really do happen on the extremes of probabilities, so just because we might flag an outcome as extremely unlikely does not mean that it won’t happen. As the great statistician G.E.P. Box put it, “[a]ll models are wrong, but some are useful.” We do think this is useful, for reasons we’re about to get to.
Now, on to the fun stuff. We look at results from the 2004, 2006, 2010, 2012, 2014, and 2016 elections. We exclude the 2008 elections because we think things break down at the extremes. At a certain point, a candidate is earning disapproval from his core base, but those voters are nevertheless unlikely to vote for a member of the other party. Hence, even though President Bush’s approval rating was at 25 percent nationally in 2008, and underwater in almost every state, Texas was never in particular jeopardy. Because of this, we caution against using this tool for political environments where a president’s job approval falls below roughly 35 percent, or rises above 65 percent.
Overall, we look at three factors: The president’s job approval in the state on Election Day (as determined by the exit polls, or estimated by presidential vote share in the previous election), whether an incumbent is present, and whether a “problematic” challenger such as Sharron Angle or Christine O’Donnell is running. Notably, we exclude seats that are considered safe at the beginning of the election year, as these seats almost never flip in reality; including them will impact estimates of seat loss disproportionately for technical reasons we won’t go into here.
We then perform a regression analysis of those factors on the vote share for the Senate candidates of the president’s party is. This produces an equation, which allows us to then extrapolate how future elections might work in various circumstances.
As mentioned above, this approach has been tested over two election cycles, and has performed well. It suggested that if President Obama’s job approval were 44 percent on Election Day 2014, that Democrats should lose nine seats. That’s what happened (it also suggested that the Virginia Senate seat should probably go to Mark Warner, albeit by a razor-thin margin). It suggested that if Obama’s job approval were at 53 percent on Election Day 2016, Republicans should lose three seats; they lost two.
But rather than just giving you the outcomes of various 2016 scenarios – which we may do in a future piece – we thought an interactive might be fun. This allows you to set what you think will happen in 2018, then see what the likely results are.
So, as of today, President Trump’s job approval is 41.5 percent, which rounds up to 42 percent. Because the theory behind the approach is that elections are largely referenda on the party in power, we ignore undecideds and just look at presidential approval. So you would input 42 percent for Trump’s job approval. The model then estimates state job approvals off of the results of the previous election (again, Sean did this in 2014 and 2016, and it worked well).
You can then exclude seats that you think are unsafe – we’ve given what we think are reasonable defaults here – if you think there will be surprise retirements, and if you think a party will nominate an ineffective/problematic candidate.
Hit “simulate,” and we’ll run 10,000 simulations under the information you’ve provided. If you use the outcomes above, you’ll end up with Republicans picking up one or two seats – Dean Heller usually loses in Nevada, but Democrats usually lose in West Virginia, North Dakota, and one other state, which varies. We note that there is a reasonable chance that Joe Manchin is actually like Olympia Snowe or Susan Collins – capable of running well ahead of expectations for the party in a state. Because of this, we might suggest marking him as safe, or at least marking the state “problematic Republican,” not because the Republican will be problematic, but because the Democrat is likely to be strong.
We note that as Trump’s job approval rises, things shift rapidly for the GOP: At 45 percent, Republicans are expected to win 54 or 55 seats. At 50 percent approval, Republicans are expected to win around 57 seats, and at 55 percent approval, a filibuster-proof majority is within reach. On the other hand, we see just how difficult it is for Democrats to take control of the Senate; even at 35 percent job approval, Republicans only lose control of the chamber one time in five. Again, this should not be taken literally; this is simply a useful tool for understanding how things might work in the future.
In any event, we’ll revisit this when we get into the new year and things have come a little more sharply into focus. In the meantime, have some fun with this!