Assessing the state of the 2018 Senate race

(Note: this is an updated version of an article I wrote last week. The original can be found here)

This article builds on the Senate model I talked about prior to last Tuesday’s special election in Alabama. In it, I assigned the Democrats a 24% chance of taking the Senate if Jones won. But Nate Cohn, an excellent political data journalist for the New York Times, now considers the race a toss-up.

The model is updated but still highly “fundamentals” based — it considers only i) whether the candidate is an incumbent, ii) the state’s partisan lean & iii) her party’s generic ballot edge (as measured by how much the party won the national House of Representatives vote).

In each of 50,000 sims it assumes all incumbents have a 5% chance of not standing for reelection, unless PredictIt’s markets suggest the number is higher. The Democrats’ generic ballot edge is draw from a N~(10.1,4) distribution — this is based on current polling and the historic relationship between generic ballot polling 10.5 months before a midterm or general election & the observed national House of Representatives vote.

Finally a couple of “ad hoc” features — I treat residing Florida Governor Rick Scott as an incumbent the 75% of he runs (i.e the model treats Bill Nelson v. Rick Scott as if it were an open race) and make him more likely to run in a D+5 world than D+15.

I give popular former Governor of Tennessee Phil Bredesen (D) half an incumbency advantage. And the model assigns only a 5% chance of an open race in Utah, since it treats Mitt Romney as an incumbent.

Finally 50% of sims include a special election in John McCain’s seat and adds extra intra-state correlation - Democrats perform better(worse) in one Arizona/Minnesota election when they perform better(worse) in the other.

So, what sort of Thanksgiving does the model predict for Chuck Schumer? Here’s a histogram of Democrats’ net seat gain in next year’s Senate midterm elections. (They need two net wins to take control of the upper chamber.)

The model gives Democrats a 29.4% chance of retaking the Senate — realistic but a long way from a toss-up. You need to do a lot of work to get Democrats’ chances up to 50%.

This is how the model sees Democrats’ chances in the 34 currently scheduled Senate elections:

Dem’s chances in any Arizona special election is also 65.7%

It’s extremely bullish compared to “experts”/markets on Democrat chances in almost every Senate race. Other than Tennessee there’s not a single seat out of 34 where my model is less confident of a Democrat pickup/hold that the combined wisdom of political prognosticators Cook, Rothenburg & Sabato and it still only gives Dems a 26% of taking the Senate. What explains this dichotomy?

They’re running against 2012 — Why my model is bullish on Democrat’s chances in individual seats

In simple terms — how likely is it a Democrat who won in a +1.5 environment would lose in a, say, +9 or +11 environment? What changed? Three reasons spring to mind:

i) The candidate is involved in some sort of scandal.

ii) The state has become much more Republican in the past six years

iii) The candidate’s challenger is of higher quality in 2018 than 2012.

ii) Is certainly true for some states (more on that later). iii) is probably true for Missouri (Todd Akin is a low bar) and could be true in Florida and elsewhere & i) remains to be seen. But in general, we’d expect any Democrat loss to have a very good reason — the idea that Tim Kaine just randomly loses Virginia in a D+5 or better environment is far-fetched.

“Experts” presume Ohio’s Senate race must be competitive since the Buckeye state leans Republican, but there’s nothing to suggest Sherrod Brown is in danger of losing the seat — he’s popular and likely to face the same challenger as in 2012. As long as the environment doesn’t shift notably towards Republicans (which is possible), Brown likely wins reelection without breaking sweat.

In eight midterm elections from 1986–2014 only four incumbent senators lost reelection while “in opposition”. Two of these were in 1998 and two in 2002 — the Senators were running against stratospherically popular presidents, an environment unlikely to replicate itself in 11 months.

They’re running against 2006 — Why my model is bearish on Democrat chances to flip the Senate.

While incumbents losing Senate seats in midterms while in opposition is rare — so is one party defending 26 seats in one cycle.

Since the 1994 Republican Revolution, Democrats gained seats in every Class One Senate cycle. From 14 seats in 1994 to 18 in 2000, 24 in 2006 and 25 in 2012. They’re … pretty maxed out. Holding 27 Class One seats, which (special elections aside) the Democrats need to flip the Senate, extrapolates to one party holding 81 Senate seats. It’s not impossible, but it’s mightily tricky.

Further, state environments don’t shift uniformly with the nationwide ballot: West Virginia was pretty red (R+31) when Joe Manchin (D) defended his seat in 2012 — by 2016 it was R+47! Assuming Manchin or Heidi Heitkamp (North Dakota saw an even larger redward swing than WV!) have easy defences because they won in 2012 is … presumptive.

Perfect seasons are really rare for a reason

One prediction I make with certainty: some time in September 2018, NFL pundits on ESPN will speculate about an unbeaten season.

One prediction I make with near certainty: said team will not finish the season undefeated?

How can I be so certain about my first prediction? Because it happens every damn year. Why I am so confident the team won’t end 19–0? Because no team has ever finished 19–0!

And yet every year pundits fail to grasp that “This team is likely to have a perfect season” does not follow from “This team is likely to win each individual match”. 85% is a big number. 85%¹⁹ isn’t.

Almost perfect seasons are pretty common, however. The 2007 New England Patriots are went 16–0 record in the regular season before losing in SuperBowl XLII. Since the advent of the sixteen game season in 1979 six teams have gone 15–1, 20 posted 14–2.

Further, it shouldn’t be assumed any Democratic losses will come from the deep red states — Six years ago Republicans conspired to lose six-time incumbent Richard Lugar’s super safe seat in solid GOP Indiana, while holding appointee Dean Heller’s seat in purple Nevada! Democrats could, say, hold West Virginia & find Minnesota’s Tina Smith run a lacklustre campaign and lose to Tim Pawlenty. They could hold North Dakota & lose in Pennsylvania. There’s lots of way Democrats can hold less than twenty six of their seats, and only one way they can hold all twenty six.

The model gives Democrats only a 1 in 8 chance to run the Senate midterm gauntlet undefeated. Dropping just one seat is enough to make them underdogs in scenarios where McCain’s senate seat isn’t on the docket.

There’s upside for Democrats

There could be more than 34 Senate elections

Of the 66 seats currently not scheduled for election in November, 43 are held by Republicans. Any extra elections caused by allegations of sexual impropriety are likely to favour Democrats (46/51 Republican Senators are men).

It’s also the oldest ever Senate. Thad Cochran (R-MS), 80, appears in ill health. Sure, Mississippi is a deep red state but an extra race there still increases the Democrats’ chances of winning the Senate by 1–2%. And there’s no reason to expect an extra election in a place like Georgia or North Carolina — but any such occurrence would behuge boon to Democrat chances.

Pick your poison

The model assumes Democrats’ generic ballot edge drops from an estimated 11.4 points today to (on average) 10.1 points by 06 Nov 2018. That’s based on limited previous data — it is perfectly reasonable to consider this prediction too bullish or bearish based on your personal assessment. It also assigns a 50% chance of a special election in Arizona. The real probability might be 25%, it might be 90% (investigating further felt too ghoulish).

It might be more illustrative to break down the chances of Dem’s flipping the Senate:

The Senate only becomes a true tossup if John McCain retires and the political environment for Democrats improves from today’s D+11.4 to D+13. Unless…

A Senator could cross the aisle

While rare (only one Senator has done it since 2001), it’s possible an independent minded Republican Senator such as Lisa Murkowski or Susan Collins would cross the aisle and caucus with Democrats. In this scenario Democrats gain exactly one Senate seat while taking control of the House: Murkowski or Collins reason they can do more for their respective states by switching side. Making such an assumption could lead to viewing the Senate as a tossup is a D+10 environment & ~-300 favourite with D+13.

This is why I struggle to view the Senate as a tossup — you need to assume

i) The political environment doesn’t markedly improve for Republicans (Dems are kinda toast even if the GOP improves to “just” D+6).

ii) John McCain retires.

iii) At least one Republican Senator is willing to cross the aisle (or the already fantastic political environment improves further for Dems).