Election tracking 2020, Part 1: The U.S. House

June 17, 2020 by Sam Wang

In the coming weeks, PEC will roll out new features and a new design. Most prominent will be an emphasis on local action. Our editorial stance this year is to leverage your local efforts locally for the Presidency (4 years), the Senate (6 years), and redistricting (10 years).

This week, we start up the previous PEC federal trackers: U.S. House, Senate, and Presidential. The math under the hood is mostly the same: the cleanest possible snapshot of polls with transparent assumptions. The notable exception is an increase in the minimum amount of uncertainty in the final stretch.

Today let’s start with the simplest tracker: the U.S. House. As of today, Democrats appear to be 5.0 percentage points above threshold from what they need to retain control of the chamber.

Here’s how it works.

Despite the complexity of having 435 House races, a single survey quantity does a good job of tracking the House: the generic Congressional ballot (“do you support the Democrat or Republican for your local Congressional race?”). There are two reasons supporting this. First, districts are of equal population, and a national survey only has to weight them all equally. The first-past-the-post (plurality vote-getter wins) rule makes this simple. Second, regional differences average out in terms of how votes translate into number of seats won.

The black trace is a polling median. Its “averaging” rule is to take an N-week median of polls, one poll per pollster, the date defined by the last day that the poll was being conducted. For our House tracker, N=2 for the entire campaign season.

The next task is to convert this voting tendency to representation. This is strongly dependent on how districts are drawn. In the U.S., districts are redrawn every 10 years. Because they have to be contiguous, they are constrained by how voters arrange themselves. Voters favor Democrats by 30 points in urban areas, and Republicans by 15 points in rural areas. That 30%-vs-15% asymmetry creates a natural tendency for Democrats to be packed into fewer districts. It also provides the raw material for drawing lines to place them at a further disadvantage. In other words, natural geography is the starting point for partisan gerrymandering that we’ve seen explode since 2000.

We had to estimate what it would take for partisan control to switch between the parties. In 2018, I estimated the Democrats needed to win the popular vote by 6% (“D+6%”). This year, because of state court decisions in Pennsylvania and North Carolina, and because Democrats have more incumbents, I estimate that the threshold is lowered to D+3%. I base this on comparison with the historical pattern since 1948, plus analysis of current single-party gerrymanders. Think of it as analogous to a golf handicap, where the abstract ideal of majoritarian rule collides with the tradition of drawing districts.

This handicap is included in the graph above. The right axis shows the actual polling median, and the left-hand axis is the same median shifted by 3 percentage points, to show the actual advantage in terms of representation.

Seat margins are generally about 6-8 seats per percentage point of popular-vote margin. So the current effective lead of D+5% translates to a margin of 30-40 seats. That’s fairly approximate.

There is a second feature. The orange line indicates the outcome of special elections since 2018. For the last few cycles, special elections have been a good predictor of how the national vote will go. This measure and the generic Congressional ballot are in approximate agreement. The same was true in the 2018 election. Of course, conditions can change. The black trace will always show a snapshot of current conditions. We will see whether the two measures continue to agree.

An open question in our democracy is whether the “handicap” above can be made close to zero, achieving similar treatment of the major parties, while still observing traditional districting principles. Achieving that outcome requires bipartisan or non-partisan control over redistricting. Another question is whether candidates of either party have to work for re-election, or are protected by a well-designed map. This is done best by independent commissions. To learn more about how to achieve those goals, see our work at gerrymander.princeton.edu.

In future weeks we’ll get into how you can influence that process in this year’s elections. Certain state legislative races will have an outsized effect on representational fairness in 2021 redistricting. They are reflected in the ActBlue and WinRed links in the right margin. Coming soon, we’ll do a deep dive into these “Moneyball” analytics.

Contributors to this feature: Lucas Manning, Ben Deverett. The code is at https://github.com/Princeton-Election-Consortium/data-backend. Outputs: tables and charts.


David Elk says:

Hi Prof. Wang,
Thank you for including local races this year! W/ no competitive races nearby me, and COVID-19 upending GOTV, this is a hugely useful resource for staying involved.
I’m curious, how did you end up adjusting the metamargin to account for more uncertainty? I remember in 2016 the margin itself was quite accurate to the populate vote, but the histogram of possible EV outcomes was very over-confident. How does that happen?

Sam Wang says:

I’ll write about it more. Basically the November uncertainty is about 5-6 points now and diminishes to Election Day. Previously I set the floor kind of low in May 2016, forgot about it – and paid for it in ways that you know about. This year that floor is 2 percentage points, which gives probabilities in the Upshot/FiveThirtyEight range. This 2 points is what a lot of people call the “correlated error,” though that is not the term I use for it.
I am also going to de-emphasize probabilities. I know it drives traffic, but I want to refocus the site on citizen action. There’s going to be a lot more on that front. It’s where we will grow the most.
A preview:

Leave a Reply

Your email address will not be published. Required fields are marked *