Note: I released this blog post early for the readers who support the site on my Patreon page. Subscribe now for a host of benefits, including early access and a premium newsletter.
Social media erupted twice on Tuesday night. The first time, it was because the New York Times’ infamous election “needle” broke (read this for an excellent description of the forecasting tool) when a county decided not to release live precinct returns to the Associated Press (we at Decision Desk HQ had our own exclusive live access).
The second explosion happened when Democrat Conor Lamb beat Republican Risk Saccone by a razor-thin 627 vote, 0.2 percentage point margin in the special election to Pennsylvania’s eighteenth congressional district. This blast has left alarm bells ringing in the Republican-held House of Representatives.
This post offers a quick rundown of the data in PA-18 and speculates about the consequences for Republicans in November. How did Lamb win? What does that mean for the 2018 midterms?
First, the obvious. Pennsylvania’s 18th Congressional district is ruby red — or so we thought. Donald Trump won it by 20 percentage points in the 2016 presidential election. Mitt Romney won it by 17 in 2012. Tim Murphy, the Republican congressman who represented the seat prior to resigning amid claims of sexual misconduct, ran unopposed in the past two election cycles. The fact that a Democrat just won in this seat is nothing to shrug off. Republicans should be panicking.
The question remains, how did Lamb pull off this massive upset victory? The answer to this question comes in two main components. First, just as we’ve seen in almost every other special election since 2016, Democrats turned out in larger numbers than Republicans did.
The graph above compares the number of votes cast for Lamb and Saccone vs those cast for Hillary Clinton/Donald Trump and Barack Obama/Mitt Romney.
At face value it looks like Democrats simply turned out more than they did in the 2016 or 2012 elections, tipping the seat in their favor. But the graph above combines both Democratic and Republican voters for the Lamb/Clinton numbers — there’s no way to say that Lamb’s success isn’t just because Republicans liked him enough to elect him as a moderate Democrat. Indeed, Conor Lamb made gains in vote share over Hillary Clinton in 591 or 592 precincts in Pennsylvania’s 18th. If Lamb’s victory was only due to turnout, not voters switching sides (or a combination of the two), we ought to see that smaller, more Republican precincts gave similar shares of the vote to Conor Lamb as they gave to Hillary Clinton while high-turnout, strongly Democratic precincts drove the ultimate Democratic victory in this seat. Instead, we see a pretty uniform swing across the entire seat.
This graph compares Republican Rick Saccone’s share of the vote at the precinct-level with the performance of other Republicans.
This graph compares Democrat Conor Lamb’s share of the vote at the precinct-level with the performance of other Democrats.
In fact, under the hood, the assertion that Conor lamb turned out enough Democratic and Rick Saccone suppressed enough Republican votes to take the cake is quite unbelievable. The table below shows Conor Lamb’s vote margin per county alongside the percentage of votes that county cast in the contest. We can see that Lamb won over a lot of votes in traditionally Democratic Allegheny County and that Republican-leaning Westmoreland County also held a lot of sway in the contest.
|County||Lamb Margin (%)||Share Of Electorate (%)|
Maybe it is geography. Where did Lamb get his votes come from? Through this lense, we might be able to uncover inferential evidence of large differential turnout. Compared with the expected votes cast per county, Democratic-leaning Allegheny cast 2 percentage points more of the share of ballots in the district than it did in 2016. Republican Westmoreland county, on the other hand, cast two points less. These differences in turnout, while larger, are not large enough to explain why Lamb won PA-18. Here’s a look at the suggested Republican dropoff by the partisan lean of PA-18 individual precincts:
Partisan Lean (%)
|Decrease in Total
Votes from 2016 (%)
|Share of Electorate (%)|
|R+100 to R+25||-43||5|
|R+25 to R+10||-40||27|
|R+10 to tied||-41||21|
|tied to D+10||-40||20|
|D+10 to D+25||-39||18|
|D+25 to D+100||-38||10|
Partisan Lean (%)
|Decrease in Total
Votes from 2016 (%)
|Share of Electorate (%)|
The differences in turnout are simply not large enough to manufacture swings as large as the one in Pennsylvania’s 18th; Democrats would have to cast far too many votes to win the race without winning over Republicans. To see that in action, let’s re-do the Pennsylvania 18 election where no partisans switch votes. Instead, turnout is just depressed in Republican-heavy precincts, according. Re-running the calculations with redder precincts casting a smaller share of their 2016 votes than bluer precincts, the outcome would have been just 1-2 percentage point better for Lamb. Of course, this makes mathematical sense if we’re just suppressing overall votes; the share given to Democrats within precincts remains roughly the same.
The graphic above presents the share of the vote that that precinct gave to Conor Lamb by the precinct-level turnout, measured as the number of 2018 ballots cast divided by the number of 2016 ballots cast
If instead I want to look at differential partisan turnout — the term we give to the phenomena by which Democrats turn out at higher rates than Republicans (or vice versa) — I have to tell my computer to spit out some more complex numbers. This process aims to determine the level of differential turnout in PA-18 based on Conor Lamb’s vote share alone and is broken down as such:
- For each precinct, estimate how many votes Conor Lamb would have gotten if the decrease in turnout was symmetrical by party. We'll call this the precincts "baseline" Lamb vote.
- Compare that to the actual share he received in the district. This is measured as the percent improvement over this baseline estimate.
- Estimate how many votes Conor Lamb would have gotten if Democratic turnout was higher than Republican turnout by x percent.
- Calculate Lamb's average precinct overperformance.
- Repeat that process for a bunch of different turnout scenarios, modifying "x" from step 3.
The results are presented in the following graph:
The graphic above shows Conor Lamb’s average precinct-level overperformance of a hypothetical PA-18 matchup at different assumptions of differential partisan turnout. Reading left to right, we see that as the Democratic turnout advantage goes up, Conor Lamb’s hypothetical vote share looks more like his actual vote share (the line at zero represents a perfect match). The two lines represent two groups of elections, one where no Trump voters voted for Lamb (dashed) and one where 9% of Trump voters cast ballots for Lamb (solid).
As it turns out, Conor Lamb wouldn’t have been able to win the special election in Pennsylvania’s 18th congressional district — which cast 106% of the ballots it did in the 2014 PA gubernatorial race and had 62% of its 2016 turnout — unless Democrats voted at a rate 35% higher than Republicans.
That is impractical in almost any reasonably imagine contest. Voters in the 2016 Virginia gubernatorial off-year election, for instance, cast about 63% of the total votes they did in the 2016 presidential election, according to voter file data provided to me by CEO of TargetSmart, a company that compiles lists of registered voters called “voter files”, Tom Bonier). And Republicans voter dropoff was actually less than the same decrease in Democratic turnout (34% vs 39%). Neither of these numbers comes anywhere close to a 35% split in partisan turnout that a Conor Lamb would need in PA-18 to win without Trump-Lamb voters.
It is very plausible that Conor Lamb actually won over typical Republican voters. Of course, we technically can’t know for sure until we have the voter file, but based on the combination of Lamb’s pro-tariff stance, hesitance to call for gun reform in the wake of the Parkland shootings, and ambivalence towards Democratic House Speaker Nancy Pelosi, I would be very surprised if he didn’t win some Trump Republicans. The idea that no Trump voters crossed over is a remove impossibility to me.
Public polling certainly corroborates the “conversion” narrative as well. A Public Policy Polling exit poll found that 51% of PA-18 voters who showed up on Tuesday said they voted for Donald Trump in the 2016 presidential election. Since the results from 2016 say about 60% actually did vote for Trump, this exit poll would suggest a roughly 9 point dropoff in Republican voting in 2018. Likewise, the difference between recalling voting for Clinton and the actual Clinton share in the district suggests a roughly 2 point increase in the share of the PA-18 electorate that is a Democratic. Of course, this analysis assumes that all Trump/Clinton votes are Republicans/Democrats — we know that’s not true, but in lieu of administrative data, these guesses give us good ammo to assess a difference in turnout by party in Pennsylvania’s 18th.
Still, if in a hypothetical PA-18 all partisans voted for their party’s candidate, the Trump/Clinton cross tabulation would suggest that PA-18 would have voted for a Republican by nine percentage points (51 Trump - 42 Clinton). What can we conclude? There had to have been some amount of persuasion in PA-18 to boost Lamb’s vote margin up nine points. Even when we consider all of the caveats of analyzing exit polls — they can be rather inaccurate, some respondents who say that they’re Trump/Clinton voters will also identify as Democrats/Republicans, recall of vote choice tends to overestimate support for winners and popular candidates, etc — the data suggest that at least 9% of Lamb’ 20 points upset came from persuading white Trump voters to select his name in the polling booth.
This is to say that differential turnout is necessarily a large part of the story in PA-18, but not all of it. The solid line in the graphic above (“How Differential Turnout…”) represents my estiamte of Conor Lamb’s overperformance at different predictions of differential turnout if we now assume that 9% of Republicans flipped sides. Then the gap between Democratic and Republican turnout is a much more reasonable sixteen points. This estimate may still be high.
This is the story of Pennsylvania’s 18th: Democratic enthusiasm evident not only in Democratic strongholds but district-wide, was not enough to give Conor Lamb the win. In high-turnout areas — where we would expect the turnout differential between Democrats and Republicans to be higher, benefiting Lamb more — Conor Lamb’s improvement over Hillary Clinton was no more pronounced than in lower-turnout areas.
The graphic above presents Conor Lamb’s improvement in vote share over Hillary Clinton as a function of turnout in the precinct.
There’s another conversation Democrats have been having since 2016: should they focus on persuading white working class Obama Democrats who voted for Trump, or focus on turnout efforts for Obama voters who didn’t turn out in 2016? If Conor Lamb’s persuasion-based strategy works well for Democrats in other heavily white, “Republican” areas like Pennsylvania’s 18th district, and efforts focused on turnout will work better in more racially diverse, low-turnout Democratic areas, the evidence from PA-18 allows them to wield a double-edged sword in the midterms this fall. They are well-equipped to do both, I argue, and PA-18 proves that.
If Democrats convert the white Republicans who rely on social services back to the party and energize the younger and more diverse cohorts of liberals who didn’t vote in 2016, they will have likely earned themselves a U.S. House majority.
The special election to Pennsylvania’s eighteenth congressional district was a huge upset for Democrats not only because they swung the district by 20 points from its 2016 Trump margin, but because they did so by persuading white Republican union voters to the Democratic ticket.
That being said, no one special election can inform a trend in voter preferences by itself. The special election to Alabama’s junior U.S. Senate seat would have us think that Democrats are doing 31 points better than they were in 2016 — but this is clearly not true. To predict a range of swings in November we need to take an average of the Democratic overperformance from all contests.
|Dem. Margin (%)|
|Seat||Partisan Lean||Vote in Special||Dem. Swing|
|South Carolina 5th||R+19||R+3||+16|
|Alabama U.S. Senate||R+29||D+2||+31|
In special elections to federal offices, Democrats have beaten the partisan lean in each district by sixteen percentage points on average. That is slightly larger than the 13-14 point Democratic overperformance we’ve seen in state legislative special elections since 2016 (notably, overperformance in contests that have only taken place in 2018 is a much larger 24%) and the current and forecast Democratic margins on the generic congressional ballot. Nevertheless, the result in PA-18 and the average federal special election swing provide a useful starting point for estimating how vulnerable some districts might be in November.
According to some measurements, there are 119 Republican seats that are more vulnerable to a Democratic flip than Pennsylvania’s 18th was assumed to be. That number doesn’t take into account the role of incumbency in House elections, however, so I used my 2018 U.S. House forecast model to do the math.
This histogram shows the number of Republican and Democratic seats that would give Democrats x% margin of victory in a neutral House environment (AKA: my forecast for the seat minus the estimated national swing). The two dark gray lines show where PA-18 and the average seat swung in special elections are positioned on this distribution. Anything to the right of the line would be won by Democrats in a national environment that leans that far to the right.
According to my model, there are 49 Republican seats that are more vulnerable than the average of special election swings that would be favored to flip in an election with a national swing that large. Democrats would pick up 80 seats if the swing in PA-18 happened across the country. Both of these meet the number of seats Democrats need to win back the House — 23, now that Lamb picked up PA-18.
Even if these estimates are optimistic for Democrats and the actual margin is the 9.1% forecast by national generic ballot polling, I estimate that the GOP would still lose control of 28 seats. You can’t spin these numbers to favor the GOP; the map just doesn’t work out.
Conor Lamb’s victory in Pennsylvania’s eighteenth congressional district was a huge victory for Democrats even if you look at the swing in the most conservative of ways. Not only have Democrats been scoring big wins in low-turnout state legislative special election, they are now doing so in high-turnout contests. If Democrats can repeat their success in PA-18 or the average special election (historically a better indicator of the swing in midterms), they will be more than favored to flip the House of Representatives come November 6, 2018.