The Feedback Effects of Controversial Police Use of Force

Throughout this paper, we refer to “controversial” instances of police use of force. While we have personal views of specific encounters, we do not label particular situations as “excessive” or “justifiable.” For decades, researchers have discussed what constitutes legitimate use of force (e.g. Bittner 1970), but the criteria have also changed over time. Because police departments, courts, and the public have different beliefs about what constitutes excessive force, we refer to them here as “controversial.”

The informed consent stated, “It is possible that reading or viewing information about these events may cause some level of emotional or psychological discomfort. Please be aware that you are free to end your participation in the survey and exit at any time.” The screen before the video stated, “The video shows a controversial altercation between a police officer and a civilian, and this may be uncomfortable for some people to watch. If you do not feel that you can watch the video in its entirety, just continue forward in the survey.” The caption under the video stated “Please click the play button to watch the video. The video contains graphic content.”

Dynata participants rate their overall satisfaction with surveys they complete. Their average satisfaction rating across all recent studies is 75 (out of 100); our study received a satisfaction rating of 86.

For statistics about police contacts, use of force, and differences by race in the U.S., see Davis et al. (2018).

This design does not account for two other important situations: non-lethal/high salience and lethal/low salience. Our design, instead, focuses on the conditions that provide the broadest range of potential effects. The low-salience non-lethal situations are abundant in the real world, and there are reasons to question how impactful they might be (relative to the other options). They also help us avoid issues associated with pretreatment exposure that might be tied to the high-salience case. That we largely find the same pattern of results for both situations (at the opposing ends of the spectrum) leads us to believe that the conditions we did not include would likely have similar effects. With respect to pretreatment, toward the end of the survey, we asked, “How familiar are you with circumstances surrounding the shooting of Philando Castile?” If we restrict our analysis to the control group (which is uncontaminated by treatments), 59.44% of respondents said “not familiar at all.”

The original non-lethal video was approximately nine minutes, but included the officer exiting his vehicle and walking to the civilian’s car and a period after the main incident, in which nothing of note is visible. We cut just over two minutes to focus the treatment video on the time from the beginning of the interaction until the scene was secured.

The sample (collected in June 2019) was 57% female, 66.61% non-Hispanic White, 12.29% non-Hispanic black, and 38.65% liberal. The median education level was Associate’s degree, and median income was 50–59 k. The sample was matched based on age, gender, ethnicity, region, and partisanship.

The data and replication code for all analyses is available at the following link: https://​dataverse.​harvard.​edu/​dataset.​xhtml?​persistentId=​doi:​10.​7910/​DVN/​Q7SFJW.

Treatment effects may be mediated through any number of variables (e.g., emotional responses). We did not establish mediation hypotheses and did not design our experiment to test mediation. Tests for mediation using our design require many assumptions that are often untestable and unjustified.

The wording of these questions allows for consistency across conditions, but we recognize that they do not tell us precisely at whom or what the emotions are directed.

In the model with controls, the effect of the Non-Lethal Video for confidence in appropriate legal consequences is not significant.