Before Matt Bevin began his slink back toward obscurity, he made a big show of raising bogus suspicions about the circumstances of his narrow loss in last month’s election for Kentucky governor. Even after he conceded, he kept complaining about increased turnout in the state’s major cities, Louisville and Lexington, and repeatedly claimed that Democrats are “very good at harvesting votes in densely populated urban areas.”
Bevin didn’t explain what he meant by “harvesting,” though it’s a term Republicans everywhere have used to signal their belief—without any evidence, of course—that Democrats have committed some sort of electoral malfeasance.
For once in his life, though, Bevin was right about something: There had in fact been a surge in voter participation in urban areas. You can see it quite plainly when you compare the 2016 presidential election with Kentucky’s gubernatorial election this year: While turnout everywhere was naturally lower in the off-year election, it fell off much less in Louisville and Lexington than it did in the rest of the state. As a result, those two cities made up a significantly larger percentage of the total statewide vote in November.
Jefferson County, the home of Louisville, went from 18.4% of the statewide total in 2016 to 19.3% this year, while in Fayette County, where Lexington is situated, the percentage went from 7.1% to 7.8%. That may not seem like much at first glance, but in a game of inches—Democrat Andy Beshear defeated Bevin by just 5,136 votes—the fact that the only two dark-blue outposts in the entire state contributed a greater share of the vote was the difference-maker.
But Bevin was wrong, of course, that the big cities were solely responsible for his loss. Just as crucial was how poorly he fared in what Republicans have fetishized as possibly the most “real America” place of all, the impoverished and heavily white rural areas in eastern Kentucky.
Part of this region consists of the former coal-mining counties that have turned away hard from their Democratic roots. While Beshear didn’t win outright in this area, he rebounded far enough from Hillary Clinton’s low-water mark that it played a major role in helping him eke out a win statewide. In Pike County, the most populous county in this part of the state, for instance, Beshear lost by a relatively small 11 points, compared to Clinton’s 32-point drubbing.
Kentucky Benchmarks
Let’s delve in a little more detail into how Beshear narrowly won his race, and let’s do it by looking back at the predictive benchmarks that I issued before the election. Doing so has two purposes: For one, it lets us perform an investigation of a victory that hopefully can be emulated in other red states. For another, at a more meta level, it lets us think more about the benchmarks themselves and how well they work. Typically, we just discard these benchmarks in the aftermath of an election amidst either the celebrations or the wailings of despair, but this time, after such a shock win, it pays to revisit them.
What’s especially striking is that Beshear actually didn’t hit his benchmarks in most of the state's major counties, and yet he still won! This is especially the case in Jefferson and Fayette counties, where he fell far short short. According to our benchmarks, to pull off a statewide victory, Beshear would have needed 71% of the vote in Jefferson County, but he got only 67%. In Fayette County, he needed 68% but took 65.5%. Again, these might seem like small differences, but in such a tight election, with zero room for error, they were huge.
We saw the same thing play out in the Cincinnati suburbs of Kenton and Campbell counties. Most observers seemed surprised that Beshear narrowly won there, since that turf is usually reliably red. For my benchmarks, though, the only surprise was that he didn’t win by more, just performing at or slightly below par.
But, again, the percentages are only part of the story. Beshear's victory suggests that you can make up for a “meh” percentage in a given county through killer turnout instead. That’s what we saw in Jefferson County, where, as noted above, Louisville’s dark-blue vote constituted a noticeably bigger percentage of the state’s total votes.
The Kentucky results also show in the potential importance of a state’s smaller counties, which I tend to omit from the table to avoid clutter, but which can, taken collectively, make or break a candidate. We unfortunately saw that in 2016 in states such as Pennsylvania or Wisconsin, where Clinton hit her percentage-based benchmarks in most of the major counties but lost through a combination of poor turnout in cities and, above all, terrible percentages in the small counties that individually don’t have an impact but, taken together, make up a hefty share of the overall vote.
We also saw, fortunately, the opposite effect in Kentucky. On election night, Beshear dramatically overperformed the 2016 presidential numbers in many of the state’s rural counties, where they’re still open to voting blue for state-level offices. (This is mostly only true in the state’s east, though; rural counties in more agricultural western Kentucky didn’t see much of a rebound at all. See, for example, McCracken County, which is home to Paducah at the state’s western tip.)
Going forward, one way to make county-level benchmarks more useful may simply be to include more counties in the table—maybe not all of them (especially in Kentucky, which has a spreadsheet-melting 120 of them), but enough to get a more representative sample of the state's various rural areas. In the retrospective version of our benchmark table below, therefore, we’ve listed 10 counties rather than just the eight in our pre-election post.
COUNTY |
BENCHMARK %
OF STATEWIDE
|
BENCHMARK
FROM 2016 PRES.
|
ACTUAL 2019
% OF STATEWIDE
|
ACTUAL 2019
VOTE SHARE
|
STATEWIDE |
100.0 |
50/50 |
100.0 |
49/49 |
JEFFERSON |
18.4 |
71/28 |
19.3 |
67/32 |
FAYETTE |
7.1 |
68/29 |
7.8 |
66/33 |
KENTON |
3.7 |
51/47 |
3.6 |
49/48 |
BOONE |
3.0 |
43/55 |
2.8 |
41/56 |
WARREN |
2.5 |
52/46 |
2.5 |
51/48 |
DAVIESS |
2.4 |
48/50 |
2.4 |
47/51 |
HARDIN |
2.2 |
49/49 |
2.2 |
46/51 |
CAMPBELL |
2.2 |
51/46 |
2.2 |
52/46 |
MCCRACKEN |
1.6 |
46/53 |
1.6 |
42/57 |
PIKE |
1.3 |
34/66 |
1.3 |
43/54 |
Ideally, we’d also develop a way to emphasize whether turnout, not just vote percentages, is meeting benchmark levels as election night unfolds. Unfortunately, though, without access to real-time precinct-level data, it’s difficult to gauge turnout amidst the fog of war; it’s something you can really only assess retroactively once everyone has reported.
Louisiana Benchmarks
This is also a good opportunity to explore how our benchmarks performed for Louisiana's gubernatorial election. Like Kentucky, it’s another good stress test of how well the benchmarks work because it was a very close race, though not quite as close (Democratic Gov. John Bel Edwards won by 40,000 votes).
Louisiana is also worth some retrospective analysis because I tried a new sort of experiment, in the form of three separate benchmark models. In addition to the traditional model based on the 2016 presidential race (akin to the one used for Kentucky), I also created models based on the 2015 gubernatorial race and the all-party primary held in October of this year.
Louisiana: 2019 model
Perhaps unsurprisingly, given its recency, the best-performing of the three turned out to be the 2019 primary. The main drawback to using the primary numbers is that six candidates ran in that race. However, I consolidated those results down to a “D” and “R” share by combining the vote shares of the two Democrats plus the independent on one side, and the vote shares of the three Republicans on the other.
Although turnout in the primary was quite a bit lower than in the runoff, the primary data was a very realistic representation of what would happen in the runoff, both in turnout and in percentage of vote share—with two important exceptions. One of those exceptions was Orleans Parish, which is coterminous with New Orleans, the state’s most populous city and where 15% of its black population lives. As you can see in the table below, Orleans shot up from 7.4% of the state’s total votes in the primary to 8.5% in the runoff, giving Edwards 28,000 more votes in the second round than in the first.
PARISH |
BENCHMARK %
OF STATEWIDE
|
BENCHMARK FROM
2019 PRIMARY
|
ACTUAL 2019
% OF STATEWIDE
|
ACTUAL 2019
VOTE SHARE
|
STATEWIDE |
100.0 |
50/50 |
100.0 |
51/49 |
EAST BATON ROUGE |
10.2 |
65/35 |
10.4 |
66/34 |
JEFFERSON |
8.3 |
57/43 |
8.4 |
57/43 |
ORLEANS |
7.4 |
91/9 |
8.5 |
90/10 |
ST. TAMMANY |
5.9 |
41/59 |
6.0 |
40/60 |
CADDO |
4.6 |
57/43 |
5.1 |
58/42 |
LAFAYETTE |
5.0 |
40/60 |
5.2 |
40/60 |
CALCASIEU |
4.0 |
47/53 |
3.7 |
48/52 |
OUACHITA |
3.1 |
40/60 |
3.3 |
46/54 |
LIVINGSTON |
2.7 |
31/69 |
2.6 |
30/70 |
RAPIDES |
2.9 |
41/59 |
2.9 |
43/57 |
ASCENSION |
2.8 |
47/53 |
2.7 |
48/52 |
TANGIPAHOA |
2.5 |
52/48 |
2.6 |
53/47 |
BOSSIER |
2.1 |
33/67 |
2.4 |
34/66 |
TERREBONNE |
2.1 |
38/62 |
2.0 |
36/64 |
LAFOURCHE |
2.1 |
39/61 |
2.0 |
36/64 |
The other big exception was Ouachita Parish in the state’s north, where the largest city is Monroe. Here, turnout didn’t change much; instead, Edwards dramatically overperformed his benchmark percentages, unlike what we saw in every other parish.
There’s an interesting reason for that: Edwards’ runoff opponent, businessman Eddie Rispone, edged out fellow Republican Ralph Abraham for second place in the primary. Abraham is the representative of Louisiana’s 5th Congressional District, which is centered on—you guessed it!—Monroe. A significant share of Abraham’s most fervent backers didn’t show up for the unity breakfast with Rispone, instead either staying home or outright switching to Edwards—even though Donald Trump made a point of holding a pre-election rally for Rispone in Monroe.
There’s one problem with using primaries as a benchmark for a general election, though: It only works in Louisiana, and the two other states that use all-party primaries, California and Washington. The other 47 states that use traditional primaries create separate pools of each party’s voters, which you can’t merge meaningfully.
Louisiana: 2016 Model
Modeling to Clinton’s loss in 2016 also worked fairly well, though it set an unrealistically high percentage benchmark for Edwards in Orleans Parish. The problem there is that Clinton overperformed to such a degree in New Orleans relative to the rest of the state (where she fared quite poorly) that there was really nowhere higher for Edwards to go.
PARISH |
BENCHMARK %
OF STATEWIDE
|
BENCHMARK
FROM 2016 PRES.
|
ACTUAL 2019
% OF STATEWIDE
|
ACTUAL 2019
VOTE SHARE
|
STATEWIDE |
100.0 |
50/50 |
100.0 |
51/49 |
EAST BATON ROUGE |
9.7 |
64/35 |
10.4 |
66/34 |
JEFFERSON |
9.0 |
53/47 |
8.4 |
57/43 |
ORLEANS |
8.2 |
93/7 |
8.5 |
90/10 |
ST. TAMMANY |
6.1 |
34/65 |
6.0 |
40/60 |
CADDO |
5.2 |
63/37 |
5.1 |
58/42 |
LAFAYETTE |
5.2 |
43/57 |
5.2 |
40/60 |
CALCASIEU |
4.1 |
43/57 |
3.7 |
48/52 |
OUACHITA |
3.4 |
48/52 |
3.3 |
46/54 |
LIVINGSTON |
2.8 |
24/76 |
2.6 |
30/70 |
RAPIDES |
2.8 |
44/56 |
2.9 |
43/57 |
ASCENSION |
2.7 |
42/58 |
2.7 |
48/52 |
TANGIPAHOA |
2.6 |
44/56 |
2.6 |
53/47 |
BOSSIER |
2.5 |
42/58 |
2.4 |
34/66 |
TERREBONNE |
2.2 |
36/64 |
2.0 |
36/64 |
LAFOURCHE |
2.1 |
32/68 |
2.0 |
36/64 |
I also thought Edwards should dramatically overperform Clinton’s numbers in Tangipahoa Parish. You might think he has some special connection to that part of the state, in the exurbs to the north of New Orleans, and, in fact, he does: That’s where he was a state representative before being elected governor. That points to one problem with modeling based on presidential results: It doesn’t allow for the possibility that statewide candidates might have a particular strength in a particular part of the state.
Louisiana: 2015 Model
The 2015 model, based on Edwards’ first election as governor, probably wound up being the least effective benchmark, simply because the state’s politics have shifted even in that seemingly brief four-year span.
What the 2015 data helps show is that Edwards had a different path in his first win, relying much more heavily on rural areas and less on the suburbs. By contrast, his 2019 win reflects what we saw nationwide in the 2016 election and again in the 2018 midterms, which both saw Democrats continue to improve in suburban areas while getting hammered ever more badly in rural regions.
You can see that with Edwards’ dramatic improvement on his benchmarks in the state’s relatively affluent suburbs, Jefferson and St. Tammany parishes outside New Orleans and Ascension Parish outside Baton Rouge. Keep in mind, though, that Edwards narrowly won, for instance, Jefferson Parish in 2015; what the 2015 benchmark reflects is that his Jefferson Parish win in 2015 lagged his statewide numbers, as he won by a comfortable 12-point margin that year. Instead, Edwards notably overperformed his statewide numbers with his Jefferson Parish performance in 2019.
PARISH |
BENCHMARK %
OF STATEWIDE
|
BENCHMARK
FROM 2015 GOV.
|
ACTUAL 2019
% OF STATEWIDE
|
ACTUAL 2019
VOTE SHARE
|
STATEWIDE |
100.0 |
50/50 |
100.0 |
51/49 |
EAST BATON ROUGE |
10.3 |
62/38 |
10.4 |
66/34 |
JEFFERSON |
8.5 |
45/55 |
8.4 |
57/43 |
ORLEANS |
8.2 |
81/19 |
8.5 |
90/10 |
ST. TAMMANY |
6.1 |
33/67 |
6.0 |
40/60 |
CADDO |
5.4 |
56/44 |
5.1 |
58/42 |
LAFAYETTE |
5.1 |
41/59 |
5.2 |
40/60 |
CALCASIEU |
3.8 |
53/47 |
3.7 |
48/52 |
OUACHITA |
3.1 |
43/57 |
3.3 |
46/54 |
LIVINGSTON |
2.4 |
34/66 |
2.6 |
30/70 |
RAPIDES |
2.5 |
46/54 |
2.9 |
43/57 |
ASCENSION |
2.7 |
46/54 |
2.7 |
48/52 |
TANGIPAHOA |
2.9 |
54/46 |
2.6 |
53/47 |
BOSSIER |
2.1 |
31/69 |
2.4 |
34/66 |
TERREBONNE |
1.9 |
36/64 |
2.0 |
36/64 |
LAFOURCHE |
2.0 |
38/62 |
2.0 |
36/64 |
I also wanted to use the Louisiana results in 2019 to try a new data visualization method that illustrates the difference between this year’s win and the 2016 election. The maps below are known as cartograms, in that the state is broken down into component pieces (its parishes, in this case) that are sized not based on their geographical area but on a different data point, in this case the total number of votes in each parish. (You can see a larger version here.)
By sacrificing contiguity, though, we can still place the counties where they belong in relation to each other. For example, of the largest blocks, you can see East Baton Rouge parish in the center, Orleans Parish at the farthest right, and Jefferson and St. Tammany Parishes positioned just to Orleans’ left. Each block is further divided into its Democratic and Republican votes, along with yellow for third-party votes in the presidential race.
You can see two things at work here: One, there were a lot more votes overall in 2016, as shown by the greater density on the grid; and two, of the votes that are present in 2019, you can see a lot more blue. The difference is probably most notable in East Baton Rouge and Jefferson parishes, which should underscore what more traditional maps don’t show: namely, that most votes are concentrated in just a few, often dense places.
Mississippi Benchmarks
Finally, let’s take a brief look at what happened in Mississippi, the one of the three 2019 gubernatorial races that the Democrats didn’t win. Democratic candidate Jim Hood still made it a much closer race than you usually see in the Magnolia State, losing by only 5 points. The county-level results, compared with the benchmarks, show how he kept it somewhat close.
COUNTY |
BENCHMARK %
OF STATEWIDE
|
BENCHMARK
FROM 2016 PRES.
|
ACTUAL 2019
% OF STATEWIDE
|
ACTUAL 2019
VOTE SHARE
|
STATEWIDE |
100.0 |
50/48 |
100.0 |
47/52 |
HINDS |
7.9 |
81/17 |
8.3 |
77/22 |
DESOTO |
5.5 |
41/55 |
4.3 |
38/61 |
HARRISON |
5.2 |
43/54 |
5.2 |
40/58 |
RANKIN |
5.2 |
32/65 |
5.3 |
34/64 |
MADISON |
4.1 |
51/46 |
4.3 |
50/49 |
JACKSON |
4.1 |
40/58 |
3.7 |
34/64 |
LEE |
2.7 |
40/58 |
2.9 |
41/58 |
LAUDERDALE |
2.4 |
48/50 |
2.4 |
42/56 |
JONES |
2.3 |
37/61 |
2.5 |
34/65 |
FORREST |
2.3 |
52/45 |
2.3 |
48/50 |
LOWNDES |
2.1 |
56/42 |
1.9 |
49/51 |
LAMAR |
2.0 |
31/66 |
2.0 |
31/68 |
For starters, in Hinds County—home of Jackson, the closest thing to a big city in Mississippi—Hood performed somewhat like Beshear did in Louisville, falling short of his percentage benchmark but doing well with turnout. However, he didn’t hit either his percentage or his turnout marks in DeSoto County in Memphis’ suburbs. Hood also underperformed significantly in some of the state’s smaller cities, such as Hattiesburg (Forrest County) and Columbus (Lowndes County).
All in all, our benchmarks remain a very useful tool that can help everyone parse returns as they come in on election night. Beshear’s massive upset, though, teaches us to keep a close eye not just on the percentages, but also on county-by-county turnout as well—a lesson we may well see play out once again in 2020.