The 2020 elections are still 18 months away and yet pollsters are out in force, giving us just enough information to break out our crystal balls and make wildly irresponsible predictions. This is the first post in a four-part series looking at the generic ballot and its utility as an election predictor.
Sign up for ESY below! We won’t bother you except to let you know about fresh, new content. We promise!
Democratic presidential primary polls have been dominating election headlines, but some congressional polls have been been released too, albeit to much less fanfare. We should probably ignore these polls — it’s too early for them to be predictive. Prognosticating off these preliminary numbers is rash, reckless and generally unwise. And so that’s exactly what we will do here in a three-piece series. This post will focus on how predictive the “generic ballot” is on, or close to, election day. The next post will look at how valuable these numbers are now, given that the election is still a year and a half away. The last will look at the relationship between popular vote and seat apportionment.
In “generic ballot” polls, respondents are asked to choose between a nameless Republican and Democrat for Congress. Gallup asks it this way: “If the elections for Congress were being held today, which party’s candidate would you vote for in your congressional district — the Democratic Party’s candidate or the Republican Party’s candidate?” It measures national support for the two parties without some of the baggage carried by their polarizing national figures (think Donald Trump, Bernie Sanders, Mitch McDonnell, Nancy Pelosi, Alexandria Ocasio-Cortez, etc.). It can also bring forward nuanced voter trends. For example, some voters might support a Democrat for President but Republican for House because they want the grinding impotency of split government. Another might vote the same way but because they support Trump but want continued House oversight. Another might choose to reward the current power dynamic because the economy is strong.
Final generic ballot polls — the ones taken within a few days of the actual election — are a better indicator of election outcomes in midterms years than in presidential years. Since 1948, the final generic ballot has missed the real midterm vote by an average of only 2%. The fallibility of the generic ballot in presidential years, though, is clear in the chart below.1Galup data from 1988 was not available.2Data is based of ‘likely voters’ for 1976 and 1996-2006 and ‘registered voters’ otherwise3The two-party generic ballot was used when available (2004-2016)4Data from Real Clear Politics was rounded to nearest integer to make it consistent with Gallup’s data
Since 1968, the generic ballot has missed the real House popular vote by an average of 4% and until 2008, it consistently overestimated Democratic support. Both of these problems have been ameliorated in recent years, resulting in a more accurate and balanced generic ballot since the late 1990s. And even though the generic ballot is less accurate in presidential years, it is still correlated with election results.7Data is based on the chart above. However, data from Real Clear Politics (2004-2016) is rounded to two decimal places rather than the nearest integer as in the chart
A party that performs better on the generic ballot will generally earn more votes in the election. The R2 in the bottom left corner indicates how much of the variance in the dependent variable (Democratic Margin in the Two-Party Vote) can be explained by variance in the independent variable (Democratic Margin in Generic Ballot). In English: R2 shows how well Variable 1 can explain or predict Variable 2. Here, R2 equals almost exactly 0.5, meaning that about half of the variance in the popular vote can be explained by the generic ballot.
Looking at the equation above the R2 value, the trend line has a slope of 0.54. This means that in general a 1% increase in a party’s margin in the generic ballot translates to about a 0.54 % bump in the House popular vote. There is still plenty of variation, though (see how the data points are not clustered too closely along the trend line), so this is not at all a perfect measure for any individual election. Even with this variation, and the knowledge that a lead in the generic ballot usually overestimates electoral success, a party is better off leading the generic ballot than trailing. And their better off leading with a larger margin than a smaller one.
Two caveats to this data. First, the makeup of the House of Representatives is not determined by the popular vote. Seat apportionment, gerrymandering, demographic & partisan sorting and other structural imbalances mean that a party’s share of the national House vote can differ substantially from the share of seats they actually win. Second, this data is from polls within a few days of the elections. We’re still a year and a half away from November 2020. The next two posts will focus on understanding these qualifications to the generic ballot.