Download data and study materials from OSF
New York University
Sample size: Wave 1: 2689; Wave 2: 2131
Field period: 02/19/2016-07/06/2016
In the first wave, to test whether people can "self-polarize," subjects were randomly assigned to try to persuade others via an essay-writing task. In the second wave, subjects were shown pro, con, or no information on the relationship between gun control policies and firearm-related incidents. Overall, we find qualified support for the self-polarization hypothesis and some limited evidence of attitude change on this contentious issue. One possible reason for discrepancies between our results and findings of persuasion on other topics is that the national debate on this issue was particularly vigorous in the aftermath of the mass shooting in Orlando, which occurred three days before the survey was fielded. We uncover one robust effect: Evidence purporting to show gun regulation to be counterproductive reduces support for stricter gun laws in general, but only among those who are initially in favor of gun control. Policy-specific opinion, by contrast, does not seem to move much. As these findings suggest, even in the midst of a heated national debate, polarization did not increase among our subjects.
H1: A persuasive essay treatment will make gun control proponents more supportive of gun control and gun control opponents less supportive of gun control.
H2a: Pro-gun-control information (relative to control) will have a positive effect on respondents' Wave 2 attitudes toward gun control, regardless of Wave 1 treatment condition or other pre-treatment attributes.
H2b: Anti-gun-control information (relative to control) will have a negative effect on respondents' Wave 2 attitudes toward gun control, regardless of Wave 1 treatment condition or other pre-treatment attributes.
H3: The Wave 1 persuasive essay treatment will not moderate the effects of either pro or con information.
In Wave 1, we first determined subjects' "proponent" or "opponent" status. Proponents support stricter gun control laws while opponents do not. Subjects were then randomly assigned to either a control condition or a treatment condition in which they were asked to write a short essay that would persuade the "other side" of the gun control debate.
In Wave 2, subjects were randomly assigned to one of three conditions: no information (control), pro-gun-control information, or anti-gun-control information. Subjects were shown graphical evidence of a relationship between gun control policies and four outcome variables: gun homicides, gun suicides, gun accidental deaths, and gun assaults. The evidence was presented as if it were the central finding in "Kramer and Perry (2014)", a fictitious academic article.
Upon completion of the essay in Wave 1, subjects were asked four questions about their preferred gun control policies. We also combine all four dependent variables into a composite index using factor analysis in order to improve power. These four questions also serve as our main dependent variables in Wave 2.
We can see that the immediate effect of the essay treatment is small: mostly null effects with the exception of gun-control opponents, who seem to become more opposed to assault weapons bans as a result. This effect is not symmetric.
The only significant effect of information that we find is on the handgun ban: "con" information makes proponents less supportive. This is important in that the sign is in the wrong direction from a motivated reasoning perspective predicting attitude polarization but the right direction from a Bayesian perspective. However, the overall picture across measures is generally one of little movement.
Despite the modest findings, we do find that support for "stricter gun control laws in the United States"---the same question we initially used to determine subjects' "proponent" or "opponent" status---changes significantly as a result of the information treatments. The signs of all estimated coefficients are in accord with a rational updating account, and the effect of "con" information is significant among proponents (making them less supportive). This is a potentially interesting finding in that we see a more robust effect on a generalized measure of support compared to specific, policy-related opinions. (We note that this particular analysis---testing the effect on this question---was not preregistered.)
Guess, Andrew and Alexander Coppock: “Does Counter-Attitudinal Information Cause Backlash? Results from Three Large Survey Experiments.” 2017. Working Paper. URL: https://acoppock.github.io/projectpages_GC_prevalence.html