The Effect of Firearm Media Exposure on Political Attitudes

Participants in an online survey experiment are randomly assigned to watch one of two short videos, each approximately two minutes long and drawn from popular social-media content. The treatment video showcases an AR-15-pattern rifle being handled and fired at an outdoor shooting range. The control video showcases Apple consumer products. Both videos are consumer-product content of similar length, format, and popularity, and neither contains explicit political messaging. A random half of the participants assigned to the treatment video additionally read a brief, balanced written summary of the Democratic and Republican parties' positions on gun policy immediately before the outcome questions. All participants then complete a survey measuring attitudes toward gun regulation, affective evaluations of the two major parties, intentions to purchase a firearm, and intentions to engage politically on gun policy.

2026-09-02

2026-10-31

Research questions
RQ1. Does exposure to popular gun-related media change attitudes toward gun regulation?
RQ2. Does it change affective evaluations of the Democratic and Republican parties?
RQ3. Does it change stated intentions to acquire a firearm and intentions to engage politically on gun policy?
RQ4. Does information about the parties' gun-policy positions amplify the translation of gun media exposure into partisan affect, and does it sort gun-policy attitudes by party?
RQ5. Do the effects differ by gun-ownership status, partisanship, and gender, and does exposure change the gender gap in partisan support?

Three primary outcomes, each measured in the post-treatment survey module. (1) Net partisan affect: the difference between the respondent's 0-100 feeling-thermometer rating of the Republican Party and of the Democratic Party. (2) Gun-policy attitudes index: a summary index of seven survey items on gun regulation (the preferred strictness of laws covering the sale of firearms and six specific policy proposals), oriented so that higher values indicate more pro-gun-rights, less-regulation positions. (3) Gun purchase intention: the respondent's stated likelihood of buying a gun for themselves in the future, measured on a five-point scale.

Orientation convention. Every gun-attitude measure is coded so that higher values are more pro-gun-rights (less regulation), and every partisan-affect measure so that higher values are more favorable to the Republican Party relative to the Democratic Party. Under this convention the hypothesized sign of every primary effect is positive.

Net partisan affect is the Republican Party thermometer minus the Democratic Party thermometer, ranging from -100 to 100. The difference captures relative partisan affect, which is the construct linked to vote choice and political contributions in our observational work, and it is robust to respondent-specific use of the thermometer scale.

The gun-policy attitudes index combines seven items. The first asks whether laws covering the sale of firearms should be made more strict, kept as they are now, or made less strict (five categories from much more strict to much less strict, coded 1 to 5 so that less strict is higher). The remaining six ask whether the respondent favors or opposes (five categories from strongly favor to strongly oppose): high capacity magazines ban; banning assault-style weapons; permit to purchase firearms; making private gun sales and sales at gun shows subject to background checks; allowing people to carry concealed guns in more places; and shortening waiting periods for people who want to buy guns legally. The four restriction items are coded so that opposition is high (1 strongly favor to 5 strongly oppose); the two expansion items (concealed carry, shorter waiting periods) are reverse-coded so that support is high. Following Kling, Liebman, and Katz (2007), each oriented item is standardized by subtracting the mean and dividing by the standard deviation of the pooled control group; the seven standardized items are averaged with equal weights; and the average is re-standardized against the control group so that the index has mean zero and standard deviation one among control respondents. The index is computed for respondents who answer at least five of the seven items and is missing otherwise.

Gun purchase intention is the response to "Do you think you will buy a gun for yourself in the future, or not?", coded 1 definitely will not, 2 probably will not, 3 unsure, 4 probably will, 5 definitely will, and standardized against the control group. A binary indicator for "probably will" or "definitely will" is reported alongside.

Treatment effects on all standardized outcomes are expressed in control-group standard deviation units and are also reported in native units (thermometer points, scale points, percentage points). The constructions above are the ones used for hypothesis tests and headline estimates. The following alternative constructions are pre-specified and will all be reported as robustness checks. For net partisan affect: each thermometer separately; the binary Republican-preference indicator (Republican rating exceeds Democratic rating); an indicator for a Republican rating of 50 or above; a rank-transformed version; and standardization within ownership stratum. For the gun-policy index: an inverse-covariance-weighted index (Anderson 2008); the first principal component with loadings estimated in the control group; the simple mean of items rescaled to 0-1; a count of pro-gun-rights responses; a six-item version excluding the strictness item; separate restriction and rights-expansion sub-indices; item-by-item estimates with Romano-Wolf adjustment; ordered logit per item; and standardization within ownership stratum. For purchase intention: binary indicators for a response of 4 or above, 3 or above, and equal to 5; an ordered logit; and estimates among non-owners only. If estimates differ materially across constructions, the pattern will be reported rather than a construction selected.

Partisan affect: the Republican Party feeling thermometer; the Democratic Party feeling thermometer; a Republican-preference indicator equal to one when the Republican rating exceeds the Democratic rating; and the treatment effect on the gender gap in partisan support, defined as the male-female difference in net partisan affect and in the Republican-preference indicator.

Gun policy: the gun-law strictness item; each of the six gun-policy proposal items; a restriction sub-index (four items) and a rights-expansion sub-index (two items); and perceived threat to the right to own guns in the United States (four-point scale from not at all threatened to very threatened).

Political engagement: the importance of gun policy in vote choice (five-point scale from not an important issue to deal-breaker); likelihood of contacting a public official to express an opinion on gun policy; likelihood of joining or contributing money to an organization that opposes making gun laws more strict; likelihood of joining or contributing money to an organization that supports making gun laws more strict (each on a seven-point scale from very unlikely to very likely); and net pro-gun mobilization, the difference between the last two.

Purchase: a binary indicator for stating that one probably or definitely will buy a gun.

Composite: a gun-rights orientation index combining the seven gun-policy items, perceived threat to gun rights, and net pro-gun mobilization.

The gender gap in partisan support is treated as an outcome defined at the group level. For a measure of partisan support S (net partisan affect, or the Republican-preference indicator), the gender gap in an arm is the mean of S among men minus the mean of S among women in that arm. The estimand is the gap among treated respondents minus the gap among control respondents, which equals the coefficient on the treatment-by-male interaction in a regression of S on the treatment indicator, a male indicator, their interaction, the ownership stratum indicator, and the pre-specified covariates. It is estimated for the pooled treatment indicator and separately for each treated arm, for both the continuous and the binary measure; the binary version expresses the change in the gap in percentage points. The gap in each arm and the treatment effect within each gender are reported alongside so that a change in the gap can be traced to its source. Respondents who select "Other" for gender are included in all pooled analyses but not in gender-specific estimates. Gender is taken from Verasight's panel profile, which predates the survey; the in-survey item is used only when the profile value is missing.

This is a randomized online survey experiment fielded by Verasight to 2,000 U.S. adults from its national panel. The sample is stratified by personal gun ownership: 1,000 respondents who personally own a firearm and 1,000 who do not. Within each ownership group, every respondent is individually randomized to one of three arms.

Control (40% of each ownership group; 800 respondents in total): the respondent watches an approximately two-minute popular social-media video about Apple consumer products.

Treatment Condition1, gun video (30% of each ownership group; 600 respondents in total): the respondent watches an approximately two-minute popular social-media video showcasing an AR-15-pattern rifle being handled and fired at an outdoor shooting range.

Treatment Condition 2, gun video plus partisan nudge (30% of each ownership group; 600 respondents in total): the respondent watches the same gun video as Arm 1 and then, immediately before the outcome questions, reads a brief, balanced written summary of the Democratic and Republican parties' positions on gun policy.

The three conditions differ only in the assigned video and the presence of the nudge; everything else in the survey is identical. Survey flow: Consent -> Gun-ownership screener (assigns the respondent to the owner or non-owner group) -> Pre-treatment items: party identification, ideology, and two filler items -> Random assignment to Control, Treatment 1 or 2 within the ownership group -> Assigned video, played to completion -> Comprehension questions about the video -> Outcome questions: attitudes toward gun regulation, importance of gun policy in vote choice, perceived threat to gun rights, gun-related political action intentions, and, future gun purchase intention -> Partisan nudge (Treatment2 only) -> feeling thermometers for the Democratic and Republican parties -> Demographic questions -> Debriefing.

Heterogeneity by gun ownership, partisanship, and gender, and the effect on the gender gap in partisan support, are pre-registered. A pre-analysis plan is registered before data collection begins.

Not available

Treatment is assigned at the individual level by the survey software Qualtrics at the moment the respondent reaches the treatment block, after all pre-treatment questions have been answered.

Individual respondent. There is a single level of randomization, stratified by personal gun-ownership status (two strata); there is no group-level or cluster assignment.

No

Not applicable. The design is not clustered: randomization is at the level of the individual respondent, so the number of randomization units equals the number of observations, 2,000 individual respondents.

2,000 individual respondents (1,000 personal gun owners and 1,000 non-owners), each contributing one completed survey.

800 respondents in the control-video arm, 600 respondents in the gun-video arm, and 600 respondents in the gun-video-plus-partisan-nudge arm. Within each of the two gun-ownership strata (1,000 gun owners; 1,000 non-owners): 400 control, 300 gun video, 300 gun video plus nudge.

Assumptions: two-sided test at the 5% level, 80% power, individual-level randomization with no clustering, and outcomes standardized against the control group. Minimum detectable effects (MDEs) are stated in control-group standard deviations (SD), with and without covariate adjustment. The adjusted figures assume that the stratum indicator and pre-treatment covariates explain 50% of outcome variance (R-squared of 0.5), which is realistic for the partisan-affect outcomes given that party identification and ideology are covariates; the unadjusted figures are conservative. All figures were verified by Monte Carlo simulation of the primary specification with HC2 standard errors (2,000 replications per scenario; rejection rate at the computed MDE between 79% and 81%; size 5% under the null). Any gun video versus control (1,200 versus 800 respondents): 0.128 SD unadjusted, 0.090 SD adjusted. One treated arm versus control (600 versus 800): 0.151 SD unadjusted, 0.107 SD adjusted. Gun video plus nudge versus gun video alone (600 versus 600): 0.162 SD unadjusted, 0.114 SD adjusted. Any gun video versus control within one ownership stratum (600 versus 400): 0.181 SD unadjusted, 0.128 SD adjusted. Difference in treatment effects between gun owners and non-owners, or between men and women (two groups of about 1,000 each): 0.256 SD unadjusted, 0.181 SD adjusted. Difference between Republicans (about 950) and Democrats (about 850): 0.270 SD unadjusted, 0.191 SD adjusted. In native units, assuming control-group standard deviations typical of national surveys: for a single party feeling thermometer (SD about 33 points), the pooled MDE is 4.2 points unadjusted and 3.0 points adjusted; for net partisan affect, the Republican minus Democratic thermometer (SD about 58 points), 7.4 points unadjusted and 5.2 points adjusted, and 10.5 points (7.4 adjusted) within one ownership stratum; for the gun-policy attitudes index (SD of 1 by construction), 0.13 SD unadjusted and 0.09 SD adjusted; for the five-point purchase-intention item (SD about 1.2), 0.15 scale points unadjusted and 0.11 adjusted. For the binary purchase-intention indicator, assuming a control-group rate of 30%, the pooled MDE is 5.9 percentage points; within the non-owner stratum at an assumed control rate of 15%, 6.5 percentage points; within the owner stratum at 45%, 9.0 percentage points. For the Republican-preference indicator at a control rate near 45%, the pooled MDE is 6.4 percentage points and the MDE for the change in the gender gap is 12.7 percentage points.