The Lay of the Land (2018 Projections)

If you're like me, you stay up nights alternately fretting about or impatient for November 6. It's hard to quell the fears that we'll have a repeat of 2016, wherein we expected to take back the Senate and win in places like Wisconsin and Pennsylvania, only to see that damn NY Times Election needle teeter, and then ultimately flip us into a dystopian hellscape.

This year, if we win the House (or come close enough to block legislation) and hold to within a couple of seats in the Senate, there will likely be the momentum to increase those gains and win the presidency in 2020. But anything shy of that and I imagine huge swathes of our voting bloc to simply give up. As a public service announcement (and as a first omnibus post), I'd like to give you my perspective on the lay of the land.


Let's start with the basics. I'm going to keep it focused on national races only today. The Dems currently hold:
  • 49 Senate Seats
  • 194 House seats (~45%), 24 fewer than a majority
Recall that we more or less tied nationally in 2016 (Trump actually lost by 2 points), but we were left with a 10% disadvantage in House.  That's messed up. It's also what you get from a very strong incumbency advantage, serious liberal concentration in cities, but most significantly, an enormous gerrymandering effect. I'll be talking a lot about that in future posts. This roughly 10% built-in disadvantage doesn't even account for the people who can't vote, but should. The Republicans have been hard at work on voter disenfranchisement for a while now.

The 49-51 split in the Senate allows us to stop some of the most odious stuff through the filibuster – and occasionally through the good will of "moderates" like Collins and Murkowski.  But only a straight-up majority will allow us to stop the parade of horrors that are Trump's judicial appointees. The House is worse, especially since the "Ryan Rule" is an even more awful take on the "Hastert Rule" (named after convicted child molester and former Republican Speaker Dennis Hastert). Paul Ryan has said, in effect, that he won't bring anything to the House floor unless it will mass with only Republican votes and that Trump will sign it. In other words, he has nothing but contempt for even the appearance of bipartisanship.

We're in pretty good shape, at least for now. To give a good sense of where we stand, I'm going to make a model – a lot of models, actually – of possible outcomes. At the center of them is a pretty useful dataset, which you should feel free to look at or download:

I've collected a handy list of the district-by-district outcomes from 2016 and 2012 (for Senate races), at both the presidential and Congressional level.  I've also included a running tally of who is running for reelection, and who has declared that they aren't. My models below are hardwired in, so you can look at my current projections at a district-by-district level. We'll be building up our model in 4 steps:
  1. The Playing field
  2. A Look at the Generic Congressional Ballots
  3. November 2017 and the Special Elections
  4. Senate Polls
The picture that they paint is not necessarily consistent, but that's how complicated systems work, I'm afraid.

1. The Playing Field

To put it bluntly, the playing field for Democrats is insanely unfair.  Trump lost by 2 points nationally and the individual House district reflect a similar underlying pattern. Nevertheless, Republicans hold 241 House seats to 194 for the Dems. 

Likewise, though we don't think of the Senate as being gerrymandered, it really is. Of the 5 smallest states, Republicans have 8/10 Senators, each of which count as much as the Senators from CA.  At present, Democratic senators represent something like 47 million more people than their Republican counterparts. And the map this year is brutal. With resignations, bad timing, and a strong 2012, we're forced to defend 26(!) seats this cycle, to the Republicans 8. 

Before figuring out individual candidates and the current environment affect the race, let's take a look at what a neutral environment would yield. We'll take a series of steps, just to see how various assumptions affect things.

No Incumbent Model

In a "no incumbent model" we could generally expect that the future follows the past. I'm using a pretty simple model to describe my expected lean for a House district:

$$Lean=\frac{(2016\ House\ Margin-Incumbency)+(2016\ Presidential\ Margin)}{2},$$

where by "margin," I mean the Democratic minus Republican vote share.  Positive numbers, in this scheme are good.  A lot of reporting tends to focus on whether a candidate, especially in a presidential election. But that seems to throw out some important information.  Inspection will demonstrate that the two measures of a district are highly correlated:

The presidential and Congressional vote margins largely track one another.  No surprise there. Perhaps more surprising is that there are a lot of uncontested races, something the Dems are seriously addressing at every level this time around.
See that discontinuity in the middle? That's the incumbency advantage.  Seated Dems on average do a bit better than in an open seat and similar for a Republican.  The effect is about 7.5%. We can subtract it out, to get a neutral district estimate:

No discontinuity!
For the Senate, it's even simpler. Base it entirely on the 2016 Presidential Race. And we assume no swing.  This model gives:
  • 40 Senate Seats (losing all of the "Trump States" including PA, but picking up Nevada)
  • 198 House Seats (some of the Republican advantage comes from incumbency)
The "No Incumbent Model" is, especially in the Senate, a worst case scenario. Political reporting tends to focus on this result since it's true that there are a lot of Democratic Senators defending their seats in "Trump States."  Nevertheless, a lot of them are a lot safer than they look.

Level Field with Incumbency Model

As I noted above, elections generally favor the incumbent. While the 7.5% estimate is based on the House, it is not clear that the same should hold for the Senate (and in the Senate, a lot can depend on the specific challenger).  

The "Incumbency Model" simply gives an additional 7.5% margin to districts where the incumbent is running. If they get knocked out in the primary, then obviously, that will change the picture. But assuming this, then a few more Senate seats become "safe": Florida, Michigan, Wisconsin, Pennsylvania, while Nevada stays red.  The House, where the Republicans hold the majority becomes safer for them. All together:

  • 43 Senate Seats
  • 193 House seats
Despite a loss of 1 House seat, this incorporates the fact that a lot more Republicans are retiring than Democrats this year.  So far, there are 55 Congresspeople retiring or stepping down to run for another office, of which there are 17 Dems, and 38 Republicans. That number, especially on the Republican side, is expected to increase.

Senate Candidate+Level with Incumbency Model

The Senate, in particular, his some information with regards to historical trends.  For instance, consider some of the most vulnerable seats.  In 2012, Obama lost West Virginia by 26.8.  Joe Manchin (D) won it by 24.1, a more than 50 point difference.  In North Dakota, Obama lost by 19.6, but Heidi Heitkamp won the seat by 0.9, a difference of more than 20.

We could imagine adding this effect – the comparative candidate advantage – to a particular incumbent if they were to run again.  There's a problem, however. The candidate advantage to some degree depends on their opponent.  In 2012, Claire McCaskill was able to easily get reelected (by 15.8 points) by running against Tod "legitimate rape" Akins, despite Obama losing the state by 10.4.  Or, more recently, Doug Jones won election in Alabama against disgraced judge and accused child molester and Republican spokesmonster Roy Moore, but he likely wouldn't have been able to win against anyone else. It was a dumb move on behalf of the Republicans to run Moore in the first place, as even before the allegations, it was a single digit race.

Consider the following model:

$$Vote\ Share=2016\ Presidential\ Margin+f\times Candidate\ Effect+Incumbency$$
where f is the fractional weighting given to history.  1 means that you assume that the candidate will outperform Clinton by exactly as much as they outperformed Obama, and 0 means that you'll assume that they won't outperform at all. On the face, unless that weighting is very strong, history seems not to matter much:
The number of Democratic seats vs. the fractional weighting, f, based on the 2012 Senatorial overperformance. Sorry for not plotting symmetrically around 50. Bad data presentation.
As a practical matter, I'll use f=0.5 for my standard model moving forward.  It's clear, though, that in some states, even with f=1 historical overperformance wouldn't be enough to keep the state blue. Trump did very well in some of the states he won.

Further Changes

There are other potential changes down the pike.  For instance, the PA Supreme Court ruled the PA maps unconstitutional. As of this morning, the districting is in the hands of a special master, hired by the courts.  That won't stop the Republicans in the state from trying their damnedest to get their way – no matter how undemocratic the process. This includes, but is not limited to, a movement to try to impeach all 5 Democratic state Supreme Court justices – all of whom were elected by popular vote at the Statewide level, like the Democratic governor, but unlike the heavily gerrymandered Republican State House and Senate. Under most scenarios, the new redistricting will likely give Dems an additional 2 seats this year.  

There are other factors. McCain may retire early, giving the Dems a very attractive pickup opportunity.  Others (perhaps lots of other) retirements may be announced, or Republicans may be successfully primaried, making pickups easier.  I'll continuously update my spreadsheet to include these changes.

The other tweak (though this is predominantly at the state level) is that a lot of races normally go uncontested historically.  Dems are fielding a lot more candidates than historically.

2. Generic Congressional Ballot

There's been a lot of concern lately that the generic ballot has gone down from about D+12 in December, to D+7 now, and as 538 (and others) have pointed out, the generic ballot is a pretty good predictor of midterm performance. Some of that tightening is likely due to noise.  Some may be the result of the tax law (and in particular, there are a few folks who've seen a bump in their takehome pay in early February). 

Whatever the value, it's not difficult to imagine that there is a nationwide trend one way or the other. This, in some sense, is the most simple-minded calculation so far.  Simply add a fixed value to every race. So simple that I'm not even going to write it down as an equation:

Democratic seat fraction in the House (Orange) and in the Senate (Blue) as a function of nationwide overall lean. The left-hand side of the plot (Lean=0) is just the "Incumbency Model" above.
On the current playing field (and assuming that there aren't pockets of overperformance) Dems need +10 to win the House (exactly as you might have guessed, based on a -10% House margin from a roughly even election), and a whopping +17 to win the Senate (and +8 to exactly hold their ground).

I noted that the generic ballot has tightened somewhat over the last couple of months from (naively) winning the House to (naively) losing the House. 

A couple of points:

  1. There may be something weird going on with party ID and poll weighting.  Some of the strongest results for the Republicans and Trump seem to overweight Republicans, and it's not clear that they're not just fixing the demographic weighting.  Following Gallup, Republican ID is down from 30% to 25% in the last year.  Assuming an 85% republican support, this means that if you renorm the Republicans, you'd add about 4 points (8 points net).  Now, in fairness, this is over the course of a year, so likely only a single point can be attributed to this effect.  Rather, some Republicans and Independents are being bought off by the tax bill, while some others are simply too exhausted to protest.
  2. Beyond a tightening of the Congressional ballot, Trump's approval rating has also been going up, from about 37% at the low, to around 41% now. This is still the lowest presidential approval ever recorded at this point, and if we're anywhere near this in November, historical data points to a landslide for the Dems.   The only 3 modern presidents to have neutral or negative net ratings at the midterms (Clinton: even, Obama: -7, Reagan: -7), all lost seats, and with an average loss of about 40 (which would easily give us the House).  And remember, even with this uptick, Trump is still at -12, far lower than even the worst 1st termer.
This tightening is concerning, but that said, there are indicators that the generic ballot is very much a lower limit to the actual D-R advantage in the coming cycle.

3. The Special Elections (and the 2017 election)

Since 2016, I've kept track of every competitive  special election:

You should check it out. I spend many Tuesdays watching the specials roll in.  And I'm not the only one with the obsessive hobby. The good folks at Daily Kos Elections have their own spreadsheet, though as I've noted, they're much more concerned with comparing presidential outcome only.

In my dataset, I've been concerned only with 2 candidate races (with minor parties ignored) as opposed to battle royales. And competitive. If a district/race has historically gone 90-10 then it's fair to say that neither party is likely to put in a lot of resources. Most specials are  low turnout affairs, but they have the advantage of not including incumbents, and also being fairly frequent. Frequent enough to do some statistics on them.

My approach is to treat the "expected" outcome of a district to be the average of the last competitive race outcome and the 2016 presidential outcome.  For what it's worth, the average of the two are within a few points of each other, and the two distributions have a correlation coefficient of 0.84 in 2017 and 0.83 in 2018. In other words, the average represents something like the "true lean" of a district.

As a summary,

January-November, 2017
  • 44 Races (16 Dem Seats, 28 Republican)
  • 11 Flipped Seats (all R->D)
  • 35 of the races (80%) saw a shift toward the Dems
  • 12% Mean D-R shift
  • 11% Weighted Mean D-R shift
  • 10% Median D-R shift
This group also included the 4 US House races (KS, MT, SC, GA) – none of which flipped, and all of which caused some gloating on behalf of the Republicans.  However, all 4 races shifted toward the Dems, by an average of 20 points, and a minimum of 12.  

November, 2017-present
  • 20 Races (4 Dem Seats, 16 Republican)
  • 5 flipped Republican Seats (including the AL Sen!), 1 flipped Dem
  • 17 of the races (85%) saw a shift toward the Dems
  • 19% Mean D-R shift
  • 19% Weighted Mean D-R shift
  • 25% Median D-R shift
The most recent flip was two days ago, and the four relevant races this week saw an average shift of 17 points (and all 4 were toward the Dems).  

By way of contrast:

2016 (When we had an actual president in the White House)
  • 19 Races (8 Dem Seats, 11 Republican)
  • 4 flipped seats (2 each)
  • 7 of the races (37%) saw a shift toward the Dems
  • -4% D-R shift (by all metrics)
In other words, Special Elections are normally bad for Dems compared to the neutral lean of a district, but in 2017 there was a big shift toward the Dems, and in 2018, the result became huge.

There are two general interpretations:
  1. The generic ballot is underestimating the Democratic advantage – perhaps by a lot.  This is likely due to an underestimate of the enthusiasm gap which may not be picked up in the "likely voter model" of most polls.  If the 10-20 point D-R advantage holds, Dems will easily re-take the House (see above).
  2. Dems are more enthusiastic, but only in the sense that you have a bunch of voters who would normally vote in the midterms, but not in specials.  And those people are now voting in specials. With ~20% turnout in the specials, it only takes a few percent of the total population to come out and vote to dramatically shift the margins. This interpretation is reasonable, but there are reasons to doubt it. The ~20 point shifts in the US House and Alabama Senate Races (even before the allegations came out against Moore) suggest that this shift isn't entirely about these races flying under the radar. Those were high-profile, high-stakes, nationalized races, with much more put into them (by both parties) than will typically be the case in November.
The caution over specials is that they are in some sense abnormal, but we do have a more typical set of elections – the ones held last November 7 – which normally would have heavily favored Republicans because (as the conventional wisdom goes) Democrats don't turn out to non-presidential elections.  We did extremely well:
Brought to you by the folks at Daily Kos Elections.
At first glance, this is amazing!  We picked up 15 seats in VA, and won all of the statewide races.  We retook the NJ governorship by 14 points and even picked up a couple of seats in the already lopsided state Senate and Assembly.  And this is addition to mayoralities,  and other state-level seats througout the country.

But, there are some troubling signs as well.  Clinton won VA by 5 and NJ by 14,  and while Northram did a little better in VA (by 4 points), Murphy's result in NJ was unchanged from the presidential margin.  Also, while Dems won in VA statewide, they still came up short in flipping the House of Delegates (it ended up 49-51, decided, in the end, by a coin flip).

In terms of polling, the RCP average of Virginia was D+3, about 6 points below his actual performance, and NJ was dead-on.  Polling was also accurate to within a few points in the Alabama senate race (about 4 points low) and the GA-06 special election (about 4 points high).  Polling averages are best, but even so, they bring with them an uncertainty of roughly 5 points.  The problem, incidentally, in 2016, wasn't that the polls were off by more than that, but that state polling averages were mostly all off in the same direction. Despite the Republican jeers to the contrary (based on a single messed up day in 2016), polls are still pretty accurate.

Oh, and while we're on the topic of polling perceptions, if anyone tries to convince you that the awful push-poll Rasmussen (who currently has Trump with a ridiculous 48% approval rating based on "likely voters") is "the most accurate for this or that election," I'd like to point out that their last poll in Virginia predicted a tie (9 points off). They were also the only pollster to pretty consistently predict a Trump popular vote win in 2016 (by as much as 5) before miraculously herding with the crowd and predicting Clinton+2 in the final week.

4. Polling by Individual Race

Primaries don't start in earnest until March (Texas is first, followed by Illinois), and as a result it seems a little premature to talk about the polling from individual races – and frankly only the Senate races need concern us here. But let's take a look anyway.

Remember that the "model" in the last column refers to our benchmark model, in which we assume a D-R shift of 10 points, and which Dems end up with exactly 49 seats (swapping Arizona and Nevada for West Virginia and North Dakota). In terms of the likely races:

D Candidate
R Candidate 
Poll Ave
Model Ave.








* Romney hasn't actually declared.  Indeed, at present, there are no declared Republican candidates in Utah.

** Corker has been making noise lately about un-retiring, but he may face steep opposition in the primary.

In terms of polling (and again, it's early, and in most cases we only have a single poll), we could win Tennesee and West Virginia, and lose Missouri.  At least early on, North Dakota looks like a lost cause. The Republican candidate is the sole House Rep, which means that he's already had to win a statewide race. And his margin in 2016 was 5 points better than Trump's.  Once again, it's total BS that we have a constitution that gives each individual Dakota the same representation in the Senate as California or Texas, but there you are.

I'll be exploring most of these issues deeper or as they develop, but for now, this is as good a start as any to the 20,000 foot view.