Did Protests Harm Biden’s Performance?
By Johannes Fischer
In the months leading up to and following the 2020 election, in the wake of a brutal insurrection and years of racial justice protests, commentators debated whether events like the protests in Kenosha, Wis., jeopardized Joe Biden’s ability to win key swing states in order to reach the necessary Electoral College votes. Evidence was sparse in those debates, and as we near the 2022 midterm elections, it’s crucial for us to contextualize protests and clarify their role with regard to Democratic performance at the ballot box (especially as protests are likely to return over gun violence, abortion access, and LGBTQIA+ rights).
Discussion
In order to address the question of the impact of protests on Biden’s vote share empirically, we’ve assembled the first national dataset of election results and protests at the granular geographic level. We compare Biden’s vote share to Hillary Clinton’s, using the change between the two as the “swing.” We combined multistate datasets on electoral results, census demographic data, and a national dataset of all the 2020 protests from the Princeton Crowd Counting Consortium.
We rely on census tracts (relatively permanent subdivisions of a county with an optimal size of 4,000 people) for their uniformity and ability to use accurate census demographic data. Our tract analysis reveals different trends in the relationship between education and Biden swing inside and outside Kenosha that aren’t captured by larger or different geographies. Additionally, using geographic proximity as a variable reveals that the “within-Kenosha” trend is likely spatial confounding: The most important variable is proximity to Kenosha, and not much else.
To begin, we replicate the analyses from the Economist and the “Political Kiwi” blog which illustrate what we find, too — that protests in Kenosha likely decreased Biden’s vote share in Kenosha. First, we examine if there are any obvious effects in the aggregate by grouping census tracts by their 2020 swing, and counting the number of tracts with and without protests.
Effects seem to be minimal, and if anything tracts with protests appear to swing slightly in Biden’s favor. We then compare the swing in Democratic presidential vote share to the education level of each census tract. Nationally we see virtually no effect, but in Wisconsin we see slight differences, with protest tracts overperforming at low educational levels, and underperforming at higher educational levels (tracts in Kenosha County are highlighted with black x-marks).
There do seem to be Kenosha-specific effects that pull down vote share, but these do not seem to extend across the entire state. We can demonstrate this by plotting the vote-share swing against the log distance, in miles, of each census tract from downtown Kenosha.
To complete the analysis, we analyze tracts in a series of regressions. To capture both protests and their “disruptiveness,” we use the log count of protests (since most tracts only had one or two protests) and allow them to interact with the log count of protests with property damage.
Across all of our regressions, what seems clear is that protests had virtually no effects nationally, and Kenosha’s effects are limited to Kenosha. Given these results, we conclude that there is some heterogeneity among voters or media coverage that determined how protests affected their vote, specific to Kenosha.
A factor absent from this analysis, because it is so difficult to quantify, is the role of media coverage and general salience of the protests. It’s possible that these protests received broader positive coverage than those of the past. Alternatively, it’s possible the protests and the previous administration’s aggressive response had effects that canceled each other out. This is the point of Professor Omar Wasow’s analysis of protests and their impact on the 1968 election — media coverage and the public’s perception of protests are what ultimately appear to matter when looking at causal effects on election outcomes.
It does appear that vote choice was at least partially influenced by racial attitudes and attitudes toward Black Lives Matter, Shom Mazumder estimates between a 4 and 6 percentage point boost in voters for Democrats in 2018. In PerryUndem’s analysis of the 2020 electorate, “favorability toward Black Lives Matter” was the fourth highest predictor of vote choice. This was not, however, the strongest factor — lagging behind “President Trump’s lies;” “President Trump cares about people like me;” and “President Trump is a great example of the American Dream” — suggesting that attitudes toward Trump and perhaps his economic status were larger motivators in the electorate on both sides of the aisle. While racial salience and media coverage certainly influence how voters pick a candidate, it is not obvious that protests themselves had impacts on how Americans voted in 2020.
Conclusion
There appears to be no obvious data backing the claim that protests harmed Democrats’ electoral performance, outside of Kenosha. However, we would be remiss in neglecting to state that even if our analysis did show a slight negative impact by protests on Biden’s vote share, that wouldn’t mean protests should all be discouraged. Protests fill the space between elections and outcomes by pushing elected officials to honor the commitments they made to voters. As we near the 2022 midterms, progressives should feel empowered to continue to hold elected officials accountable and remain vocal about the important issues at stake!
Methodology
Election results are generally reported at the precinct level, but Data for Progress acquired or produced official election results disaggregated to the census block level from a political vendor and using the geomander package. We aggregated the results to the census tract level to estimate the change in Democratic vote share from 2016 to 2020 for each census tract in states where election results were available (every state, but not Washington, D.C.).
This was then combined with mean-centered, five-year estimates from the Census Bureau's 2019 American Community Survey and protest data assembled by the Crowd Counting Consortium. Protest data was geotagged and placed in the relevant census tract by using the consortium’s provided latitude and longitude data, and an API from Geocodio.
Finally, a linear regression was run in R, using the glm function, on the swing in Democratic vote share as a product of key demographic variables.
Johannes Fischer (@pollhannes) is Survey Methodology Lead at Data for Progress.