close× Call Us
Title Allies or Agitators? A Survey Experiment on Perceptions of Nonviolent Protest
Post date 06/25/2019
C1 Background and Explanation of Rationale

Protests have been a major force in American history for bringing about political and social change. Yet public opinion about protests in U.S. history is best described as conflicted, and different traditions frame protests as conducive to advancing social progress or threatening to stability and the position of dominant groups. One common rationale used by elites for endorsing or suppressing social movements is whether they are predominantly (non)violent.

In recent years, scholars have argued that nonviolent protests are more likely to succeed than violent ones due to the perceived legitimacy of nonviolence (Thomas and Louis 2013; Chenoweth and Stephan 2011; Schock 2005). However, research in cognitive psychology provides grounds to question whether an assessment of the intensity of tactics used in a protest—that is, how (non)violent they are—is independent of people’s prior political or ideological views. Experiments have shown that judgments about observable “facts” cannot be detached from preexisting biases and perceptual processes (Haidt 2001; Goff et al. 2014). Examples from different substantive areas illustrate how cognitive distortions can lead to pre-rational judgments and have real-world consequences.

We propose a theory of protest perceptions that incorporates psychological findings on implicit cognitive biases, and argue that the assimilation of information about a protesting group precedes and shapes perceptions of how (non)violent it is, which in turn affects judgments about its merits. People’s first exposure to information about a mobilizing group—its demands, actions, and demographic makeup, for example—triggers an affective charge that is experienced as warm or cold feelings toward the group (Lodge and Taber 2005; Morris et al.2003; Redlawsk 2002). If prompted to render a judgment about the group’s interaction with the physical world, they will engage in motivated reasoning consistent with their initial charge (Lodge and Taber 2013).

If a positive/negative response is triggered in a given instance, this effect can have implications for views on the legitimacy of protest more generally. We propose a causal chain in which the perceived level of (non)violence is endogenous to the interaction of a) information about a protesting group and b) partisan predispositions, which also shape policy views and abstract opinions about protests. Specifically, through the mediation of perceived violence, the interaction of tactics and partisan sentiments should be associated with c) immediate determinations of how to respond to protesters, d) attitudes on the importance of protest for democracy and e) support for punitive anti-protest policies in general.

C2 What are the hypotheses to be tested?

H1: The severity of protest tactics should be associated with the perceived level of violence
H2: Respondents whose partisan preference is aligned with the protest group should perceive the level of violence of the protest as lower
H3: Perceived level of violence should have a positive relationship on whether respondents think it is appropriate to arrest the organizers of the protest
H4: Perceived level of violence should be negatively associated with whether respondents think protests are important for democracy
H5: Perceived level of violence should be positive associated with support for punitive anti-protest policies in general

C3 How will these hypotheses be tested? *

We use an original fictional vignette within a survey experiment. Within the vignette—a description of a protest in the guise of a news story—we randomly assign varying protester group identity and tactical choices. To test our hypotheses, we intend to field a survey with an embedded experiment. It will be conducted on 1100 respondents on the Mechanical Turk platform, from which respondents are directed to the survey website.

The vignette describes a fictitious protest and randomly assigns one of two protesting groups and one of three tactics, in a 2x3 fully crossed design. The vignette is as follows:

A crowd of 200 people calling themselves [“Americans Against Racist Policing” / “Americans Against All Abortion”] gathered at city hall this morning. Larry Carter, 49, was among the participants. “Our political leaders are supposed to protect us, but they don’t. We are angry and feel our voices are not heard!” Carter said. When local officials refused to meet with the organizers, the group [held up placards and shouted slogans, some laced with profanity / blocked a nearby highway, bringing traffic to a standstill / threw rocks and other objects at the building].

All the elements in the vignette were chosen to isolate critical variables and minimize bias. Both “Americans Against Racist Policing” and “Americans Against All Abortion” were devised to refer to salient political issues that have generated real-life social movements. The former is reminiscent of Black Lives Matter and is associated with Democrats. The latter is not represented by a single major movement but is associated with Republicans. We use fictitious groups in order not to activate conscious associations with existing movements, yet we provide enough information for readers to be able to form an impression and render a judgment.

The tactics in the vignette are typical of the repertoire of protests in the U.S. and are not associated with any party or faction. They represent gradations of direct action. Because the vignette does not state what the placards say or what damage, if any, is caused by throwing “rocks and other objects,” respondents must fill in the gaps in the narrative by imagining the contextual details. It is within the realm of possibility that certain kinds of profanity could be perceived as violent, while rocks that cause no injuries could be seen as nonviolent.

There are five dependent variables that correspond to different steps in the theorized causal chain: First, we ask how the respondent would describe the intensity of the protest, on a scale from completely nonviolent to completely violent. Second, we ask how much the respondent agrees with the goals of the protest, which acts as a validity check on the correspondence between attitudinal factors and the information in the vignette, providing a measure of approval for the protesters apart from their tactics. Third, we ask whether it would be appropriate to arrest the protesters, which represents a “law-and-order” response. This measure is related to, but does not automatically follow from, perceptions of violence, as it may be possible to disapprove of the tactics used but favor a non-punitive response.

Additional questions gauge how views about protests in general are shaped both by the treatments and by intervening variables that lie more proximate to the vignette. Thus, the fourth post-treatment question asks how much people “agree that protesters provide a useful service to our democracy,” which tests the extent to which exposure to a single episode of protest can tap into ostensibly deep-rooted philosophical beliefs. Fifth, we ask whether people would support a law punishing certain types of protest actions, the text for which was adapted from a bill proposed by Arizona legislators in early 2017 (Christie 2017). Views on this law indicate the extent to which perceptions of tactics or other aspects of the immediate episode can have broader public policy implications (Lodge and Taber 2013).

The main independent variables are the six combinations of treatments. To test the effect of partisan bias, we interact party affiliation with the protest group treatment. In addition, we include psychological and demographic variables that may be associated with attitudes toward protest: child-rearing values (Stenner 2005), age, education, race, sex, and income. The survey instrument also includes two manipulation checks.

Our analytical strategy is as follows. First, we present descriptive statistics and face validity checks. Then, since the dependent variables are ordinal Likert scales (e.g. strongly disagree to strongly agree), we test our claims using an ordinal logit model. We use weighted least squares with means and variance adjusted estimation (WLSMV), which performs better for categorical data (Beauducel and Herzberg 2006; Muthén and Asparouhov 2002). Finally, as the outcomes are sequential, we use a structural equation model to test paths of direct and indirect effects. We use the statistical software R for the ordinal logit models and Mplus 8 for the structural equation model.

C4 Country United States
C5 Scale (# of Units) 1100
C6 Was a power analysis conducted prior to data collection? No
C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval? Yes
C8 IRB Number STUDY00003602
C9 Date of IRB Approval November 14, 2017
C10 Will the intervention be implemented by the researcher or a third party? Researchers
C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research? not provided by authors
C12 If relevant, is there an advance agreement with the implementation group that all results can be published? not provided by authors
C13 JEL Classification(s) not provided by authors