What is the point of the Special Elections Index?
We wanted to develop a measure that gives us insight into the current electoral environment, based on what voters actually do in elections, and who actually turns out to vote, not what they tell some pollster they might do.
Over the last year, we have been comparing Democratic performance in special elections to the 2012 and 2016 presidential numbers, calculated by our own indefatigable Jeff Singer. But he only has 24 hours in a day, so we don’t have the presidential results for districts drawn in the 2000s or earlier. That means we’ve only been able to look at special elections going back to the last round of redistricting, starting for the most part in 2013.
With the Special Elections Index, however, we compare special election results to past election results for the same position in the same district, which means we can look back in time as far as we can find results for the special elections themselves, giving us an even more robust—and predictive—data set.
How is the Special Elections Index calculated?
Step 1: Compare the special election results to results of other elections for the same position in the same district that have used the same district lines. As a hypothetical example, suppose the Democrat, Jane Huynh (pronounced like “win”), won the special election for State House District X by 24 points in 2011. We have as many as five potential elections we might be able to compare Huynh’s victory to: 2002, 2004, 2006, 2008, and 2010 (the district lines prior to 2002 were different, and they changed again in 2012).
It turns out that in 2010, the Democrat who ran that year won by 13 points, so Huynh actually did 11 points better. But the Democrat who ran in 2006 won by a massive 33 points, so Huynh did 9 points worse. In 2002, the Democrat won by 16 points; Huynh did 8 points better than her. In 2008 and 2004, though, Republicans didn’t field a candidate at all, so we can’t actually compare the 2011 special election results to those years. (This is a very common occurrence; on average, each special election has three others it can be compared to.)
Step 2: Adjust for political environment. 2010 stank like dirty pigeon feet for Democrats, so in our example above, Huynh doing better than her predecessor did in 2010 isn’t as big of an achievement as it seems on the surface. To address this problem, we adjust the numbers using the nationwide popular vote margin for the House of Representatives.
In this case, Democrats lost the popular House vote by about 7 points in 2010. So although Huynh did 11 points better than the Democratic candidate in 2010, there’s a 7 point “penalty” since it’s easy to do better than 2010; the adjusted value shows she only did 4 points better than a Democrat in a hypothetical “neutral” environment. 2006, on the other hand, was a banner year for Democrats; they won the House popular vote by about 8 points. For 2008, Huynh gets a “bonus,” since it’s hard to do better than a Democrat in a really good year. So she did only 1 point worse (rather than 9) in the theoretical neutral environment. Here’s a table to clarify:
This adjustment assumes that the popular House vote can be used as a surrogate for the political environment at the state legislative level, averaged over the entire nation, but we have good reason to think it works well, as we’ll discuss.
Step 3: Average all the comparisons. In Huynh’s case, we have three pairs of elections we can look at: 2011 vs. 2010, 2011 vs. 2006, and 2011 vs. 2002. Taking into account the adjustments above, she did 3 points better, 1 point worse, and 4 points better, respectively. Together, this tells us that Huynh did an average of 2 points better than other Democrats did running for the same seat.
But we’re not really interested in Huynh alone. We want to average all pairs of special elections and regular elections we can find in a given year (or two years, in the case of the biennial version of the index)—the more data points, the better. It turns out we have data on a total of 647 special elections and 1,724 pairs of elections over 29 years, which were used to make the graph above.
Where did the data come from, and which data were included?
Special elections for the House of Representatives and state legislatures were identified using Our Campaigns, Ballotpedia, and official state election results websites, and most results were obtained from Our Campaigns. Our goal was to use elections that represented standard Democrat-versus-Republican dynamics. Only special elections in states with regularly scheduled even-year elections were used (so numbers could be standardized using the national popular House vote), but not those which took place during the regularly scheduled November midterms or presidential elections. Runoff elections were also included.
In order to be properly comparable to other elections, we only included races where the top two candidates were major party candidates from opposite parties, each affiliated with only one of the major parties (so no recent party-switchers), and together received more than 90 percent of the vote. Special elections where a close family member ran following the death of an office holder were excluded. One race where a candidate ran in a special election to replace himself following expulsion from the state senate was also excluded. Elections held after Election Day in an election year were collated with the following year’s data.
Finally, not all special elections in the time period covered can be found online. The further back in time, the less information is available, so our data set only goes back to 1989. (If you know of sources for older special election data, please contact us!) A spreadsheet of all the data we used can be found here.
Does this really mean anything?
The purpose of the Special Elections Index was to act as a substitute for comparing special election results to the presidential election margins. So how well does it do when the index is stacked up against actual presidential results (which, as we noted, we have for legislative districts going back to 2012)? Pretty well, it turns out:
You can see the dark green line (Special Elections Index, adjusted for 2012 popular House vote) is a decent approximation for the average performance of Democrats in special elections compared to Obama’s 2012 margin (teal line), with three of the five points matching almost exactly. That’s good enough to serve as a substitute.
So the index looks like it does what we wanted it to, more or less. We can now also see that this year has a really high Special Elections Index value, as high as any we see as far back as we can go. (Consult the graph at the very top of this post once more to see.) Which must be great, right? But then again, what exactly does it mean for Democrats to outperform the presidential numbers in special elections and have a high Special Elections Index value?
We’ll delve into that topic next time.
Update: Part 2, showing strong correlation with House election results, is here. Part 3, using the Index to forecast the House, is here.