Wielding a gun increases judgments of others as holding guns: a randomized controlled trial

The gun embodiment effect is the consequence caused by wielding a gun on judgments of whether others are also holding a gun. This effect could be responsible for real-world instances when police officers shoot an unarmed person because of the misperception that the person had a gun. The gun embodiment effect is an instance of embodied cognition for which a person’s tool-augmented body affects their judgments. The replication crisis in psychology has raised concern about embodied cognition effects in particular, and the issue of low statistical power applies to the original research on the gun embodiment effect. Thus, the first step was to conduct a high-powered replication. We found a significant gun embodiment effect in participants’ reaction times and in their proportion of correct responses, but not in signal detection measures of bias, as had been originally reported. To help prevent the gun embodiment effect from leading to fatal encounters, it would be useful to know whether individuals with certain traits are less prone to the effect and whether certain kinds of experiences help alleviate the effect. With the new and reliable measure of the gun embodiment effect, we tested for moderation by individual differences related to prior gun experience, attitudes, personality, and factors related to emotion regulation and impulsivity. Despite the variety of these measures, there was little evidence for moderation. The results were more consistent with the idea of the gun embodiment effect being a universal, fixed effect, than being a flexible, malleable effect.


Supplementary Materials
The supplementary materials contain the outcomes from the multiverse analysis.

Multiverse Analysis Plan
To eliminate effects of experimenter degrees of freedom, we report a multiverse analysis (Steegen, Tuerlinckx, Gelman, & Vanpaemel, 2016). For a multiverse analysis, the data are presented for each analysis that would have been conducted. For the analysis on the gun embodiment effect, we report a multiverse analysis for which we systematically varied the determination of outliers. We determined outliers at three points in the analysis: raw reaction times (RTs), the bias or difference scores for each hold condition, and the gun embodiment effect (which is the difference in bias or the difference in difference scores between the two hold conditions). For the raw RTs, we used four different criteria. For the none criterion, we did not exclude any RTs. For the fixed criterion, we excluded RTs faster than 100ms and slower than 1200ms. This criterion matches that reported in Witt and Brockmole (2012), although in retrospect, neither the fixed nor the none criteria are sensible. Thus, the primary focus will be on outlier criteria that are specific to each participant and each stimulus condition. For the 1.5xIQR criterion, we excluded RTs that were beyond 1.5 times the interquartile range (IQR) for each participant for each stimulus condition. The 3xIQR was the same except we excluded RTs beyond 3 times the IQR. For the difference scores and the gun embodiment effect, we used three criteria: none, 1.5xIQR, and 3xIQR. Altogether, there were 36 different combinations of outlier criteria (4 x 3 x 3). This means that we conducted our critical analysis (e.g., a pairedsamples t-test comparing the hold gun condition to the hold spatula condition) 36 times under each of the different outlier criteria.
For the gun embodiment, we also analyzed various dependent measures including a measure of bias, RT, and a measure that combines proportion correct and RT. The outcomes of all are reported as part of the multiverse. For the analysis on individual differences and the gun embodiment effect, we systematically varied the criteria for determining outliers using the 3 criteria of none, 1.5xIQR, and 3xIQR. To the extent that the effect is similar across the multiverse, this is evidence that the effect is robust to outlier exclusion. Note: The statistical measure for each grid is shown in respective boxes in the top-left corners. Columns show the four outlier criteria for the raw RTs, and rows show the three outlier criteria for the difference scores (Diff) and the gun embodiment effect (Gun). In 3 of the grids, darker colors signify p < .05, dz > .20, or Bayes factor > 3.

Reliabilities
To do individual differences, it is critical to ensure the measure has good reliability.
Reliability was calculated by computing a gun embodiment score for the odd trials and the even trials and calculating the correlations between them (see Table 3). Note: Columns show the four outlier criteria for the raw RTs, and rows show the three outlier criteria for the difference scores (Diff) and the gun embodiment effect (Gun). The shading of each cell corresponds to the magnitude of the correlation.

Individual Differences
Gun Embodiment Effect 5 We conducted a multiverse analysis for the correlation between the gun embodiment effect, as measured with RTs, and each of the individual differences variables. Each analysis was conducted with all participants (none), and by removing participants with scores beyond 1.5 times or 3 times the IQR (1.5xIQR, 3xIQR, respectively) on either of the two measures used within each correlation. For example, if someone had a score identified as an outlier on the Big 5 agreeableness measure but not the Big 5 neuroticism measure, their data would be excluded when comparing the gun embodiment effect with agreeableness but not when comparing with neuroticism. The correlations, p-values, and Bayes factors are shown in Table 4. Given that people who have used a gun at least once did not show the effect, it seemed reasonable to re-evaluate the correlations with only participants who reported having never used a gun. The corresponding multiverse is in Table 5.