The 2018 midterms proved a wild ride that ultimately gave Democrats control of the U.S. House of Representatives. Despite a highly unusual year and several unexpected wins for the Democratic Party, Monopoly Politics’ high-confidence projections for 2018 were 97.5 percent accurate.
This is especially notable because these projections were made two years in advance - only updated for new vacancies and mid-cycle redistricting - using a very simple model. While some models include fundraising, polling results, extrapolations from polling figures and input from election experts, FairVote’s model uses just four key factors in its calculations:
- District partisanship — based on vote shares from the last presidential election;
- A candidate’s Performance Over an Average (generic) Candidate from their party, or “POAC — based on the candidate’s performance in past elections;
- The edge given to incumbents by virtue of being incumbents — a factor we call an “incumbency bump;” and
- National two-party preference.
District partisanship and POAC are the only factors we calculate ahead of an election. What remains unknown until election results are published are the “incumbency bump” and national party preference.
National two-party preference is similar to - but not exactly the same as - the two party national vote share. However, we don’t just add up all the votes cast for Democrats and all the votes cast for Republicans - that would be skewed by various uncontested races and just the nature of a year with more Republican than Democratic incumbents. So instead we measure how well each party’s candidates perform compared to their district’s partisanship, and take the medians of various kinds of districts to isolate what the overall national mood was.
(Note: Using medians helps account for races in which one of the main parties is not represented because a candidate is running unopposed, or their opponent was not nominated by the other main party.)
The incumbency bump is calculated in a similar way, but it is party independent, and rather than focusing on all House races, only those featuring an incumbent are included.
By these measures, 2018 was about a 53.8 percent Democratic year, and incumbents were favored by a 3.7 percent boost on average.
Despite Monopoly Politics’ simplicity, our high-confidence projections were extremely accurate. There were, however, some seats our model failed to call.
Which projections did we miss?
The high-confidence projections in Monopoly Politics 2018 were incorrect for 11 seats, all of which were seats we called for Republicans that actually went to Democrats. Although it’s possible that vote splitting impacted some of the outcomes in a few very close races, there were some cases in which elements of our model led us to miss the mark.
Our projections missed for three reasons: the unusual nature of the 2016 elections (which is what we used to generate our partisanship data), what may be a general shift in voting behavior and statistical quirks for races that—during a wave year in favor of the Democrats—became competitive within our confidence thresholds.
For example, the 53.8 percent national preference for Democrats and the 3.7 percent incumbency bump would have indicated that New Mexico’s 2nd Congressional District would become a toss-up race, essentially slipping under our threshold of only making a high-confidence projection once a district’s party preference reaches 56 percent or higher.
Monopoly Politics projections take only the vote shares for the most recent presidential election into account. This means that if a district’s behavior in the 2016 election differed from its typical voting trends, using only data from the last presidential election — rather than factoring in data from 2012 or even earlier — could skew our projections. This was the case for Michigan’s 8th Congressional District, New Mexico’s 2nd Congressional District, and New York’s 11th and 22nd Congressional Districts.
However, there are some districts in which our projections were off, but their behavior in this election was closer to that of 2016. This is the case for Iowa’s 3rd Congressional District, Illinois’ 14th Congressional District, New Jersey’s 3rd Congressional District, Oklahoma’s 5th Congressional District, South Carolina’s 1st Congressional District, and Virginia's 7th Congressional District. This may indicate a general shift in these districts’ voting behavior in favor of Democrats.
Previously, our high confidence projections never missed more than one seat. Missing 11 is an outlier - but it is the exception that proves the rule. Although a higher miss rate than our past projections, it is still very small in the context of all 379 seats projected. Outcomes are still overwhelmingly set in stone long before anyone casts a single vote under our current system.
These missed projections also helped us refine our model for the next edition of our series, Monopoly Politics 2020, which was published on Friday, just three days after the 2018 elections.
Projecting winners nearly two years in advance underscores the stifling effect our two-party system has on competition. Too many voices remain unheard, and voters deserve better. Meaningful competition and a more fair system is possible. The Fair Representation Act can help us revitalize our democracy through electoral reform.
Illustration by Mikhaila Markham