In May, I took a “wicked early” look at, among other elections, the three gubernatorial elections to be held on November 5, 2019 in Kentucky, Louisiana and Mississippi. I also updated my estimated effects of incumbency for Democratic and Republican United States Senators (“Senators”) and governors.

Having passed Labor Day, the traditional start of the fall campaign season, I want to take a closer look at those three 2019 gubernatorial elections, as well as a “do-over” election for the United States House of Representatives (“House”) to be held in North Carolina’s 9^{th} Congressional District (“CD”) on September 10, 2019[1].

Specifically, I want to add available public polling to the fundamentals of each election: a baseline estimate of the outcome of an election. I calculate an election’s fundamentals by summing three values:

- State partisan lean, measured by my 3W-RDM[2],
- Incumbency advantage, which I estimate to be 5.7 percentage points (“points”) for incumbent Democratic governors and 8.5 points for incumbent Republican incumbent governors, and
- National partisan lean, measured by the “generic ballot” question for the House[3]. As of the evening of September 5, 2019, FiveThirtyEight gave Democrats a 6.5-point edge on this measure.

Before I discuss the individual races, however, just bear with me while I discuss my new approach to converting fundamentals and polling averages into probabilities, as well as the weights I now use to combine those two values into a single “projection” and probability of victory.

__Probability__. When I calculated updated advantages for Senators and governor, I used the simple arithmetic difference between the “expected” result (fundamentals-only) and the actual results. This means I have an “error” distribution for 106 gubernatorial elections going back to 2011. These errors are roughly normally distributed, with a mean of 0.6 points and a very high standard deviation of 17.5 points, suggesting inordinate uncertainty in my fundamentals estimates.

Still, using the properties of the normal distribution, I can estimate the probability a given estimate of fundamentals equates to a Democratic win (*i.e.*, Democratic margin > 0).

Similarly, I can generate the probability a WAPA equates to a Democratic win using the formula

*Standard deviation = square root of ((D average*R average)/total number polled),*

using the simple polling averages for the Democratic and Republican candidates. Technically, this is calculating the margin of error (without multiplying by a “z-score” to get, say, a 95% confidence level), but with an assumed normal distribution it is the functional equivalent of a standard deviation.

__Time weighting__. This close to election day, polling averages should be given more weight than fundamentals, and that weight should increase daily. Since my polling data for these races begins on January 1, 2019, I used the number of days between that day and November 5, 2019[4]—308—as the denominator for my weight. The numerator is the number of days between January 1, 2019 and the closing field date of the most recent poll of that election. That ratio is the weight given to WAPA, with one minus that weight applied to the fundamentals.

For example, the most recent poll of the Kentucky gubernatorial election was in the field from August 19 to August 22, 2019. It is 233 days from January 1 to August 22 this year; dividing 233 by 308 yields 0.754. Thus, as of now, I weight WAPA in Kentucky 0.754 and fundamentals 0.246. The other two gubernatorial elections have similar 3-1 ratios of polling to fundamentals, which seems right.

**************

Let us begin the election in **North Carolina’s 9 ^{th} House CD**. In the 2018 midterm election, Republican Mark Harris, a conservative pastor, appeared to defeat Democrat Dan McCready, an entrepreneur and former Marine Corps captain, by just 905 votes (0.4%). However, once substantial evidence emerged of ballot tampering by a Republican operative named Leslie McCrae Dowless, the state election board refused to certify the election results. Instead, they called for a new election—including new primaries to select each party’s nominees. While McCready ran unopposed, Republicans chose State Senator Dan Bishop in a July 9 runoff election.

According to the Cook Political Report, this CD leans 8.0 points more Republican than the nation. With no incumbent running, the fundamentals suggest a generic Republican would beat a generic Democrat by 1.5 points.

However, examination of available public polling[5]–adjusted for mean partisan bias and pollster quality, as well as time to election (what I call “WAPA”: weighted-adjusted polling average)—suggests McCready is ahead by 0.5 points. That lead increases to nearly two points in the two most recent polls, both taken in late August. Take this WAPA with multiple grains of salt, however, as the average pollster rating is just C+.

Still, the time-weighed polling and fundamentals by time, gives you an aggregate of D+0.3, making this race a true toss-up, and likely within recount territory. But any outcome between D+2 and R+2 is extremely plausible…and that includes R+0.4; it would be a beautiful bit of irony if Bishop wins by the same 0.4 points Harris led by on election day 2018.

**************

That Democrats are reasonably competitive in all three gubernatorial elections this fall is extraordinary given they average 23.1 points more Republican than the nation (Table 1). A word of caution, however: the available public polling in these races ranges from an average of C+ (Mississippi) to B/B- (Kentucky).

**Table 1. 2019 Gubernatorial elections**

Incumbent |
Party |
State |
3W-RDM |
Fund |
WAPA |
Comb |
Prob D win |

John Bel Edwards | D | LA | R+22.2 | R+10.0 | D+9.1 | D+3.8 | 79.5% |

Open seat | R | MS | R+18.5 | R+12.0 | R+4.0 | R+6.1 | 6.9% |

Matt Bevin | R | KY | R+28.7 | R+30.7 | D+4.7 | R+4.0 | 76.5% |

The likeliest Democratic win is in **Louisiana**, where Democratic incumbent John Bel Edwards remains fairly popular (favorable 47%, unfavorable 33% in the most recent Morning Consult polling). This is remarkable in a heavily Republican state; even as an incumbent in a strongly-Democratic year, he would still be expected to lose by about 10 points to a generic Republican (though that would still give him about a 30% chance based on recent electoral history).

Photograph of John Bel Edwards from here

Louisiana will actually hold a “jungle primary” on October 12 in which every declared candidate will run regardless of party affiliation. If no candidate captures more than 50% of the vote, the top two vote-getters will face off in a runoff election on November 16. The available polling[6] shows Bel Edwards far ahead of his two primary Republican challengers, House Member Ralph Abraham and businessman Eddie Rispone. WAPA values across the available public polls shows Bel Edwards at 46.5%, Abraham at 24.9% and Rispone at 10.3%, with the remaining 18.3% scattered among a handful of other candidates and undecided. Bel Edwards could well fall shy of a majority on October 12.

His likeliest opponent appears to be Abraham, though he recently stirred up some self-inflicted controversy. Head-to-head matchups suggest Bel Edwards would defeat him by about 7 points. Similar polls suggest Bel Edwards would defeat Rispone by about 16 points. If we assume Abraham would be roughly a 5:2 favorite to be Bel Edwards’ runoff opponent, the weighted average is a Bel Edwards win by about nine points. Overall, Bel Edwards is about a 4:1 favorite to win reelection.

At the other end of the popularity spectrum, meanwhile, is Republican governor Matt Bevin of **Kentucky**, the least popular governor in the country (if you subtract his unfavorable rating of 56% from his favorable rating of 32%). In fact, the incumbent Bevin only defeated State House member Robert Goforth in the May 21 primary 52-39%. Still, even with a -24-point differential in favorability and a divided party, Kentucky is such a Republican-leaning state that the fundamentals have Bevin beating a generic Democrat by a whopping ** 30.7 points**. That generic Democrat would have only about a 4% chance of winning, based on recent electoral history.

However, in a closely-fought primary, Kentucky Democrats nominated state Attorney General Andy Beshear, whose father Steve was governor of Kentucky (as a Democrat) from 2008-2016. And Beshear’s name may be just enough to defeat Bevin, even in the 7^{th} most Republican state in the country. The Democrat leads by 4.7 points in the WAPA, which would make him essentially a shoo-in to win; this value is derived from only three publicly-available (two with a strong Democratic lean) polls, though[7].

Photograph of Andy Beshear from here.

The time-weighted average of fundamentals and WAPA is Bevin+4.0, reflecting the enormous disparity between the two. Still, with WAPA weighted 3-1 over fundamentals, Beshear would seem to be about a 3:1 favorite.

I remain skeptical, though, and consider this race essentially a toss-up (maybe even a slight advantage for Bevin) until I see more and higher-quality polling.

As for the open seat in **Mississippi**, where Republican Governor Phil Bryant is not seeking reelection, the fundamentals have a generic Republican defeating a generic Democrat by 12.0 points, giving that Democrat about a 1-in-4 chance of winning based on recent electoral history.

On August 6, Democrats nominated state Attorney General Jim Hood, who easily defeated seven other candidates, while Republicans nominated Lieutenant Governor Tate Reeves, who edged state Supreme Court Chief Justice William Waller in a controversial August 27 runoff election. In fact, following filmed instances of electronic voting machines switching Waller votes to Reeves, Waller refuses to endorse his party’s gubernatorial nominee.

Unlike in Kentucky and Louisiana, though, the few publicly-available Mississippi polls[8] are broadly in line with the fundamentals; the WAPA is Reeves+4.0, making him a near-certain victor. In fact, if you remove two Democratic-leaning Hickman Analytics polls, Reeves’ lead over Hood jumps to about 10 points. The time-weighted average has Reeves up about six points, with a roughly 93% chance of victory.

In sum, then, the good news for Democrats in these three races—in ** heavily** Republican southern states—is that they are unlikely to lose any ground in governor’s mansions overall (they now trail Republicans 23-27), and could even net one new seat in Kentucky, of all places.

Until next time…

[1] There will also be a special election in North Carolina’s 3^{rd} CD that day, to fill the seat vacated when GOP House Member Walter Jones died on February 13, but Republican Greg Murphy is heavily favored to win that seat over Democrat Allen Thomas.

[2] Essentially, how much more or less Democratic a state’s presidential voting has been relative to the nation as a whole over the last three presidential elections.

[3] If the election for were held today, would you vote for the Democratic candidate, the Republican candidate, or some other candidate?”

[4] November 16 for Louisiana

[5] RRH Elections, 8/26-8/28/2019, 500 LV: Harper Polling/Clarity Campaign Labs, 8/26-8/28/2019, 551 RV; Anzalone Liszt Grove Research, 7/15-7/18/2019, 450 RV; Atlantic Media and Research, 5/20-5/30/2019, 358 RV; JMC Enterprises, 5/21-5/24/2019, 350 RV

[6] Market Research Insight 8/13-8/16/2019, 4/9-4/11/2019, 600 LV; Multi-Quest International, 7/19-7/21/2019, 601 RV; Remington Research Group, 6/1-6/2/2019 (1,471 LV), 3/13-3/14/2019 (1,484 LV); JMC Enterprises, 4/25-4/29/2019, 650 LV; LJR Custom Strategies, 1/14-1/17/2019, 600 LV

[7] Garin-Hart-Yang Research Group, 8/19-8/22/2019, 501 LV; Clarity Campaign Labs, 8/12-8/13/2019, 792 LV; Gravis Marketing, 6/11-6/12/2019, 741 LV

[8] Hickman Analytics, 8/11-8/15/2019 (600 LV), 5/5-5/9/2019 (604 LV); Survey Monkey, 7/2-7/16/2019, 1.042 RV; Impact Management Group, 6/10-6/14/2019, 610 LV; Mason-Dixon Polling & Research, Inc., 1/30-2/1/2019, 625 RV; OnMessage Inc., 1/28-1/30/2019, 600 LV

## One thought on “A post-Labor-Day look at 2019 elections for U.S. House and governor”