The Recent Protests Won’t (and Haven’t) Changed Political Behavior
The tragic death of George Floyd has offset racial protests across the US and around the world. It has sparked much needed discussion of difficult topics in the US around systematic racism and its deep costs.
Many have also pointed out that the majority of Americans support these protests and that Trump’s strong condemnation of them will likely cost him support in November. The truth is, however, that these protests are unlikely to have a sustained impact on his election odds, nor is he really incentivized to take meaningful policy steps based on them. Indeed, the thing likely to decide the election is still the state of the economy heading into November.
Historically similar incidents of civil disobedience in the US have had a non-significant impact on presidential approval polls, suggesting that whatever impact these protests will have is likely already baked into voter perceptions due to strong partisanship and the concentration of protests in Democratic urban centers.
To investigate this in detail we collected historical data on incidents of civil unrest in the US and Presidential approval polls since 1980. Wikipedia records around 67 major incidents of civil unrest in the last 40 years. We categorized these into 7 groups and show the breakdown below.
Over half of major incidents of civil unrest in the US are race riots generally sparked by similar incidents as the recent death of George Floyd (tragically most of these race riots do start because of individuals being killed by police). The next most common category are policy protests related to protesting specific organizations like the WTO or specific administrations like the protests during Trump’s inauguration in 2017.
When we look at the occurrence of these incidents over time, we see they are generally rare with an average of 1–2 each year. However, in the last decade, incidents of civil unrest have become increasingly common, especially in 2016 and 2017, when Trump campaigned and got elected. This is further evidence that this particular president is highly polarizing, leading to increased levels of protest nationwide but also strong support from his loyal base.
To see whether these incidents have a relationship with Presidential approval polls historically, we can use a regression to estimate the impact that race riots have had on future approval polls T weeks after an event for T ranging from 0 to 10. This allows us to understand the relationship between events of unrest and approval polls historically and how it changes over time. If the coefficient from this regression is negative and statistically significant, then race riots tend to lead to lower presidential approval ratings. But if the coefficient is either positive or not statistically significant, then there is no historical impact of race riots on approval ratings.
For comparison we do the same thing with the national unemployment rate, showing the relationship between the unemployment rate in a given month and future approval polls.
We can visualize the impact by plotting the regression coefficients for each regression, representing the impact T days after either the race riot occurred (left) or the unemployment rate month (right). These represent the marginal impact on approval polls of an incident occurring or a 1% increase in the unemployment rate respectively. We also show 95% confidence intervals for the coefficients.
In the plot below we see that the estimated impact on approval polls of race riots is negative but quite noisy; confidence intervals always include 0, suggesting that they do not seem to have a significant impact on future approval polls. On the contrary, unemployment rate is a consistently strong predictor of future approval polls: a rise in the unemployment rate tends to forecast lower approval polls for the incumbent president.
Historically, we see that the data suggests race riots (and other incidents of civil unrest) do not significantly impact presidential approval polls. This is because these incidents themselves are polarizing where the loyal supporters of the incumbent and opposed voters tend to each react in expected ways, making them have very little overall impact. However, if the economy gets worse and the unemployment rate rises, we do tend to see the incumbent’s approval rating decrease, as less loyal supporters change their opinion.
A natural question though is whether this historical data applies to 2016 given how different things are right now during the pandemic. We do know Trump’s approval rating has cratered in recent weeks, yet its decline is fully explainable by the declining US economy in April and May, starting its descent well before the recent incidents.
Below we plot Trumps’ approval rating in 2020 in orange (left y-axis) and US unemployment rate in 2020 in blue (right y-axis). We see that in April the unemployment skyrocketed because of the economic shutdown due to COVID. Meanwhile, Trumps’ approval rating peaked in April as well and only started falling until a few days after the start of the month. It consistently fell the following weeks as the unemployment rate stayed at elevated levels, and the economy slowed, reaching its currently all time low with the unemployment rate still above 10%.
Therefore, rather than Trump’s approval rating falling due to the recent protests, it appears to have closely followed the unemployment rate and state of the economy. As states begin to reopen, and the unemployment rate falls, we should expect his approval rating to revert closer to its historical average, increasing his election odds, and producing little change in his behavior because there is no cost to maintaining the status quo.
If supporters want these protests to actually have an impact, they need to work on changing the incentives of the politicians they target. The best way to do that is to start locally with politicians whose fates can directly be impacted by protester turnout. It also means doing things to elect candidates for the Senate that will back things like police reform and other policies to promote racial equality.
Vinod Bakthavachalam (@vinod__b) is a data scientist who writes about economics, politics, and policy. He has previously written for the Harvard Business Review, the World Economic Forum, and the New York Times.