Read my piece about the special election to PA-18 in The Economist: “Pennsylvania’s special election | Conor Lamb is likely to lose PA-18 | That is still consistent with the idea of a Democratic wave”
Like a lamb to the slaughter, Republican Rick Saccone has sprinted toward election day in a district that should not be competitive at all, much less favored by polls to be won by his opponent Conor Lamb. But can the Democrats pull it off? Or will this district that Trump won by twenty percentage points closer reflect its natural Republican lean?
The data — polls, fundamental indicators, and recent special elections — give us no clear conclusion. What can we divulge by combining these indicators?
Polling in the special election to Pennsylvania’s eighteenth district has been sparse; over the past month, only four polls have been released. Two of those have shown Democrat Conor Lamb with a 3-6 point lead, while the others have shown the exact same for Republican Rick Saccone. In all polls fielded in March, the average is a 2 percentage point lead for Lamb.
Monmouth University is the only high quality, nationally recognized pollster to release a reading in the race. Though their poll has Conor Lamb up 6 percentage points, the average of all polls has been a more predictive measurement of elections in the past than taking only the polls we think are better readings.
The polling is not perfect, however. In past special elections (I have collected data for roughly 20) the averages of polls have an average error of 5 percentage points and a root-mean-square error (more useful for people trying to predict things) is roughly 6.8%. This works out to a margin of error on the final polling average of roughly 13 points — more than enough for either Lamb or Saccone to win by double digits.
This figure above graphs the daily average of polls in PA-18 with the shaded margin of error for all polling averages of 20 special election contests.
In probabilistic terms, these numbers work out to a 62 percent change that Democrat Conor Lamb wins the race on Tuesday. But that leaves a very good chance that Rick Saccone also wins the election in his district. In fact, Saccone has a better shot at victory than you do at flipping a coin twice and getting heads (or tails) both times, a fairly common event. Go ahead and try that exercise real quick; I’m sure you’ll get two of the same side the coin a good amount (oh, roughly 25%?) of the time.
In the end, the polling tells us a fair bit about where to set our expectations for PA-18. A Lamb victory is the most plausible, according to these readings. But don’t constrain your estimates for the race too far around that point. Anything from a 15-point Lamb win to an 11-point Saccone win could happen.
2. District Geography
Make no mistake, Pennsylvania’s eighteenth district is a very Republican one. Donald Trump won by 20 points there in the 2016 presidential election. Mitt Romney won by 17 points in 2012. The past Representative, Tim Murphy (Rep.), had gone uncontested in the previous two cycles. There is no historical reading of the seat that says a generic Democrat would be favored to win a race there over a generic Republican. At a partisan benchmark more Republican than 60% of congressional districts, it is, simply, quite red.
We use these data to predict the outcome of Tuesday’s contest.
I plugged the district characteristics — one with a 19% partisan lean and an open election — into my House forecasting model that estimates seat-level outcomes under different national environments to obtain a “fundamentals-” based prediction of the district. Based on the current generic ballot polling that shows Democrats up eight percentage points, the model estimates that the PA-18 special election should break by 11-12 points for Republican Rick Saccone.
Just like in our polls-based projection, there’s a margin of error here: a (quite a bit) larger 19 percentage points. The value of the fundamentals projection is in dampening swings implied by special election and House polling, which have tended to overestimate Democrats in 2017-2018 elections. Remember that poll in the GA-06 special election that had Democratic candidate Jon Ossoff up by 7 points?? This combination of district geography and generic ballot polling produces estimates that predict outcomes better (IE: calls a district correctly for Democrats) even if the estimates of that margin are ever so slightly worse. This combination predicted that Karen Handel (Rep.) would win in the GA-06 special election last June, for example, whereas polling indicated that Jon Ossoff (Dem.) had the edge.
Of course, it might be useful to combine this fundamentals projection with the polls of the contest. This might let us combine the value of both district-level polling and district-level geography. If instead of using the House forecasting model I specify one that includes district-level polls and seat partisan lean, the projected Saccone margin shrinks to 4 percentage points. The margin of error on these predictions (+/- 14%) is still large enough to place a Lamb win within the realm of possibilities.
But perhaps this approach is flawed. It could be that using this projection system designed for regular elections produces estimates that don’t work for special elections. In that vein of thought, it might be appropriate to predict PA-18 with a model fit on all the special elections that have happened since 2016.
3. Special Elections
The results from special elections since 2016 have been sounding alarm bells for the Republican Party. The question is whether the bell they’re ringing in PA-18 is loud enough. The average swing toward Democrats in special elections since 2016 is 13 percentage points. In contests that have taken place in 2018 alone, the swing is a much larger 26 points.
What does this mean for PA-18? According to a predictive model similar to the one above — but run instead on special elections in state legislative districts (setting swing since 2016 as a dependent variable on the seat’s partisan lean) — the contest between Lamb and Saccone should shift fourteen percentage points toward the Democrats.
The figure above graphs the shift in districts from their partisan lean to their special election margins
Of course, there is a margin of error here too! This time it is quite large: 70 percentage points on the margin of Democratic victory. (Obviously, Conor Lamb is not going to win by 65 points, but the statistical uncertainty remains.) I suspect that most of this error comes from a combination of several factors: (1) a few idiosyncratic state legislative districts that saw huge shifts toward Democratic candidates; and (2) the seat- and candidate-specific characteristics that matter a lot for these local races. Whatever the cause, the data say we can’t put too much stock into this projection alone.
4. Combining Indicators
Our best shot at predicting what will happen in PA-18 is to take a combination of the measurements above. The average of all predictions is a five percentage point win for Republican Rick Saccone, though I would encourage readers to put slightly more weight on those predictions that use information from polls, a holistic interpretation is always desirable.
The figure above graphs the estimates and their margins of error for various methods of predicting PA-18.
Though it is unclear if Democrats will score a win in Pennsylvania’s 18th, it is clear that such a victory would be a huge gain for them. In federal special elections to U.S. House and Senate seats so far, Democrats have overperformed the 2016 Clinton vote share in the district by 14 percentage points. A win by Lamb on Tuesday would be at least a 20 point swing.
|Seat||Partisan Lean (%)||Vote Margin (%)||Dem. Swing (%)|
|South Carolina 5th||R+19||R+3||+16|
|Alabama U.S. Senate||R+29||D+2||+31|
County and Precint Benchmarks: How to Watch the Results
In most contests, nobody is really sure what will happen on election day. Sure there are some races where the incumbent Republican (who won by eighty points last time around) is running against a Democrat not qualified to run for dog catcher, but most races have an element of uncertainty that comes with them. In the special election to Pennsylvania’s eighteenth congressional district, that element of uncertainty is absolutely huge. At best, we’re 87% sure that Saccone will win. Really, that number is much closer to the 50-50 mark and might even favor Lamb with odds just higher than 60%. Ultimately it’s hard to say who might prevail Tuesday night.
What we can do is break down past contests in the district to estimate what Conor Lamb might need to win to be victorious. Below, I’ve broken down precinct-level election results from recent contests by each of the district’s counties. (The the 2016 U.S. Presidential Election, the 2016 U.S. Senate Election, and the 2014 PA Gubernatorial Election are included.) These benchmarks are calcualted by taking each candidate’s marginal performance in the district, subtracting out the partisan lean of that contest, and then average that ‘partisan lean’ for each of the three contests. The result is a benchmark that tells us what the county should vote for if the election is going to be tied district-wide. If Conor Lamb is beating these benchmarks in the districts, particularly for Washington and Westmoreland counties, then he’s probably on track for a victory.
|County||Dem. Margin Benchmark (%)||Clinton Margin (%)||McGinty Margin (%)||Wolf Margin (%)||Share Of Electorate (%)|
Of course, it’s hard to do the math on these benchmarks and still get a confidence projection of the race while keeping up with all the new data flowing in by the second. Then there’s the added challenge of modeling the result by districts that have large precinct-by-precinct variation within them. If so, consider this look at precinct-level benchmarks instead:
|Precinct||Clinton Margin (%)||Benchmark (%)|
|NORTH STRABANE 4||-19.81||-0.05|
|ELIZABETH TP WD 5 DIST 2||-24.67||-0.09|
|West Newton Borough, Third Ward||-24.83||-0.22|
|COLLIER DIST 1||-14.76||0.26|
|FINDLAY DIST 2||-19.95||0.32|
|BETHEL PARK WARD 3 DIST 2||-17.22||-0.35|
|S FAYETTE DIST 11||-20.57||0.37|
|Irwin Borough, Fourth Ward||-24.88||0.38|
|SOUTH PARK DIST 1||-19.76||-0.39|
|MONROEVILLE WARD 3 DIST 2||-15.43||0.43|
|MOON DIST 6||-15.32||-0.49|
|N FAYETTE DIST 4||-26.36||-0.63|
|BETHEL PARK WARD 2 DIST 2||-16.02||-0.78|
|Mount Pleasant Borough, Third Ward||-26||0.79|
To solve the problem of intra-county variation in partisan composition, I’ve built a precinct-level projection system that will use the results in precincts that have already reported to run live estimation of precincts that have yet to come in. You will be able to find these projections in the spreadsheet embedded below (linked here) with a live discussion on my Twitter feed (embedded in the sidebar of this blog). You will also be able to join my blog’s chatroom for live commentary if you’re subscribed via Patreon.
I plan on calling the election once I am 99.5% confident about the outcome. Until then, PA-18 is a waiting game.
The special election to Pennsylvania’s 18th Congressional district is sure to be an entertaining one. Polls, district geography, and the national environment all give us mixed signals as to what exactly to expect. The only thing that is certain is that this seat is just one data point in a slew of indicators on what will happen come November 6, 2018. As I contended for The Economist and on The Sit and Spin Room podcast, tonight is a likely win for Democrats either way.
Whether I end the night with a “Silence of the Lambs” or a “Saccone of shame” pun is ultimately up to the voters of southwestern Pennsylvania. Polls close at 8:00 EST/7:00 central and I will start running live projections shortly after. You won’t want to miss it.