To put these numbers into context, an average 0.6 percentage point gain across all campaigns using Deck in 2022 would represent a total of 750,000 additional votes, with each campaign and organization on average gaining 1,017 more votes.
Previous evaluations of Deck found that Deck users performed better than non-Deck users, on average, but these analyses could not exhaustively control for factors that could have explained why Deck users overperformed. Given how important understanding our impact is, we attempted a new design that would control for as many outside variables as possible by factoring in district characteristics and candidates’ projected performance before any campaigns began using Deck.
Deck worked with 738 campaigns and organizations in the 2022 election. After the election, we analyzed our performance using Ordinary Least Squares, controlling for district features, such as age, race, Spanish language preference, religion, urbanicity, and education; candidate incumbency; Biden’s 2020 performance; and our initial forecast before the campaign began using Deck. An added benefit of controlling for our initial forecast is we can implicitly control for the other features we used to create our forecast. We also have to control for our initial forecast because our forecasts update throughout the election, which means that if we affected campaigns’ performance, then our forecasts would capture our effect. That would induce post-treatment bias. We also added fixed effects for state and contest. Of course, endogeneity, or confounding factors, may still exist so we should interpret these estimates conservatively. We use HC3 robust standard errors.
The table below displays the relationship between being a Deck user and different electoral outcomes. The first column shows the relationship between Deck usage and vote share when only controlling for Biden performance and initial forecasts. Deck usage is associated with 1.6 percentage points higher vote share. We believe this is an optimistic estimate because other features about the district, candidate, and contest may explain Deck usage and vote share. In the second column, we add controls for district features and fixed effects. Deck usage is associated with 0.6 percentage points higher vote share.
We’re now interested in the degree to which campaigns outperformed benchmarks. The third column evaluates the relationship between Deck usage and how much the campaign outperformed Biden 2020 performance. We find that Deck usage is related with 0.8 percentage points higher vote share relative to Biden 2020 performance. The fourth column evaluates the relationship between Deck usage and how much the campaign outperformed our initial forecasts. We find that Deck usage is associated with 0.5 percentage points higher vote share relative to our initial forecasts.
Although we have provided several different results to show that Deck usage is associated with better electoral performance regardless of the outcome, endogeneity could still bias our results. For example, experienced campaigns may be more likely to use Deck but experienced campaigns also likely perform better. We try to control for some of this by controlling for candidate incumbency. Incumbents should be more experienced than challengers, but we cannot rule out all sources of experience, such as campaign staffers and consultants. Therefore, experience might explain the positive relationship. Because of this, we want to be careful about how we interpret the results. Short of an experiment in which we would randomly assign access to Deck, we cannot be sure the effect is causal. However, it appears that Deck usage is associated with better electoral performance.
After exploring different outcomes and using different controls, we find that Deck usage is associated with better electoral performance.