Calibration or Contrition
One of the things that is sorely lacking in the practice of evidence-based campaigns is a coherent theory of politics. We’ve spent decades optimizing what is easy to measure and in many ways have lost sight of what’s important for winning elections. I’ve made my career in evidence-based campaigning, and I can say that I’ve contributed to a large chunk of this problem. In many ways this post is a first step towards contrition.
For the better part of the last decade I’ve worked on quantifying the hard-to-measure parts of politics so that we can have weak signals that tell us about what to do for what’s important rather than what to do among a set of easy-to-measure options. These include trying to quantify the effects of earned media, disentangling the nuances of candidate effects, and most recently estimating diminishing returns to paid campaigning with far too little data.
My hope is to share the totality of this research over time, but for today I want to start with a framework that will allow us to slot in research moving forward.
So to begin– what’s important for winning a general election?
The three factors in order of importance– which we’ll dive deeply into in subsequent posts– are:
The national environment
Candidate quality
Campaign effects
The National Environment
Much has been said about the importance of uniform swing, but enough can’t be said about how much this colors election outcomes. Good years for either party have a massive effect on election outcomes; especially in close races. Let's put some actual numbers on this.
If we look at every election since the WW2 era (since President Eisenhower) we get a standard deviation on uniform swing of ~6 percentage points (pp) in margin or ~3pp in vote share. This holds if we narrow to every election this century.
Now let's run a simulation to look at how that affects House races. I’m choosing House races here because the peculiarities of Senate maps can obscure the point, since some Senate cycles favor Democrats and others favor Republicans only because of what states are in play.
I’ll adjust the 2024 results to start at a neutral year by adding 1 pp to Dems across the board. If we simulate the election with a 3% uniform swing and a 1.5% random district-level swing, we wind up with a distribution that looks like this:
In practical terms a 1 standard deviation swing in either direction from a neutral year looks like this:
In a year with 1 standard deviation swing towards Democrats, they win ~228 seats in the House. A neutral year puts them at 219. (Democrats won 215 seats in the 2024 elections.) A 1 standard deviation swing away from Democrats puts them at ~208 seats. For context, a swing of >1 or <-1 standard deviations should happen about 1 in every six elections. With the average House race totaling ~300k votes a 1SD swing is worth ~9k votes in a House race. And importantly, it’s worth 9k votes in every house race.
A thing that I’d like to linger on here is that most evidence-based campaigning focuses on tactical interventions like whether or not your campaigns should canvass or what the ads should say on TV, and recently a lot of focus has been on candidate issue positioning. This is the lamppost problem in a nutshell. We focus on what is measurable and what is in our control and soothe ourselves by talking about the process.
Toiling over the measurable / malleable
To put a finer point on this, the difference between a good and a bad TV ad is measured in hundreds, and in rare cases, thousands of votes[1]. The difference between spending money on TV or canvassing has roughly the same order of magnitude impact. The difference between moderate or progressive candidates, while larger, still pales in comparison to the magnitude of national swing. Campaigners and political pundits are focused on things that feel like they’re within our control – where someone might be making a mistake – rather than humbly submitting to the power of random forces bigger than us.
I get it. We want to explain the world. We want to feel like we are in control. We want to believe that this one weird trick could change everything. That’s natural, it’s human. We’re all storytellers who got to modernity by building a map of cause and effect: using it to survive and propagate. That there are incomprehensibly large factors which are seemingly random and dictate our fate is both heartbreaking and unsatisfying.
But if you’ll bear with me, I’d like to take you on a journey into the abyss; away from what’s measured and into speculation, reasoning, and hypothesis construction. We’ll widen the aperture for what we can consider evidence, going away from just what fits neatly into a spreadsheet.
Through a broader approach to evidence, we can start to speculate on how we might affect this much more impactful phenomenon. In my next post, I’ll take a look at a few different approaches to understanding the national environment and what (if anything) can be done to increase your chances of landing on the right side of the distribution.
What’s next
In the coming posts, we’ll go deep into how opinion is formed, how it propagates through communities and ultimately materializes as public opinion.
We’ll look at a handful of studies and start to build a comprehensive view of media, opinion formation, and how we can start to take shots at shaping national swing.
[1] A single flight of paid media will net a few hundred votes, campaigns can typically double this effect by generating many ads, evaluating them with a sample of voters, and selecting the ad that persuades voters most.
