Download data and study materials from OSF
Principal investigators:
Tanika Raychaudhuri
University of Houston
Email: traychaudhuri@uh.edu
Homepage: https://www.tanikar.com/
Tali Mendelberg
Princeton University
Email: talim@princeton.edu
Homepage: https://scholar.princeton.edu/talim/home
Sample size: 1550
Field period: 04/07/2020-07/24/2020
Valence Hypotheses:
Racialization Hypotheses:
Moderator Hypotheses:
To test these expectations, we conducted a between-subjects survey experiment. We sought to vary valence and race independently and estimate their separate and interactive effects. In addition, we sought to vary valence fully, from unsympathetic to sympathetic, rather than examining only one valence or the other. Finally, we aimed to create strong valence treatments, allowing us to test the impact of a ‘full dose’ of sympathy.
Respondents were randomly assigned to one of four treatments or a non-story control. In the treatments, respondents read a hypothetical news story about opioid use, varying the race of users (White or Black) and the valence of the frame (sympathetic or unsympathetic). The story draws on actual news stories and their reader commentaries, and, importantly, resembles them in length, narrative style, and the use of vivid photos. The news story begins with statistics about opioid use. Next, it presents the personal account of a fictional drug user named Mike, recounting how he began using opioids and the impact of drug use on his life. This full-length narrative format generalizes to the type of news content many people consume. It represents an advance over many studies where information is stylized, abbreviated, and lacks imagery, with stimuli too pallid to allow the full development of a frame.
We find support for the sympathy and full valence hypotheses but little support for the antipathy hypothesis. As the sympathy hypothesis predicts, the sympathetic White condition increases support by statistically and substantively significant amounts, on three of the four outcomes relative to the control: treatment policies (21 points, p < 0.001), a candidate who favors treatment (8 points, p < 0.001), and sympathetic emotion (4 points, p < 0.01). Likewise, the sympathetic Black condition increases support, for all four outcomes relative to the control: treatment (13 points, p < 0.01), candidate (6 points, p < 0.01), taxes (5 points, p < 0.05), and emotions (5 points, p < 0.01). However, the antipathy hypothesis is not supported. Specifically, the unsympathetic White and Black conditions each only mildly affect emotions relative to the control (5 points, p < 0.05). As the full valence hypothesis predicts, relative to the unsympathetic White condition, the sympathetic White condition generates strong, statistically significant, support on three of the four outcomes: policy (18 points, p < 0.001), candidate (7 points, p < 0.001), and emotions (8 points, p < 0.001). In addition, relative to the unsympathetic Black condition, the sympathetic Black condition increases support for policy (8 points, p < 0.05) and emotions (8 points, p < 0.01).
Of the racialization hypotheses we only find support for racial sympathy. The anti-Black bias hypothesis is not supported, as the sympathetic story about Black users increases support for all four outcomes relative to the control, as described above. The unsympathetic White treatment has no statistically significant effects on outcomes relative to the control (except emotions, as described above) in line with Pro-White bias, but neither does the unsympathetic Black condition. As the racial sympathy hypothesis predicts, the sympathetic condition generates less support for treatment policy with Black than White users (8 points, p < 0.05). There is no evidence for the racial antipathy hypothesis: Unsympathetic coverage does not generate more antipathy with Black than White addicts. Likewise, there is limited evidence for the racial main effect hypothesis. When comparing the pooled Black and White conditions, the Black conditions decrease support for treatment policy, but this effect is somewhat uncertain (5 points, p < 0.10).
We find some support for the racial moderator hypothesis. First, we find that racial resentment moderates racial sympathy. Resentment moderates the impact of sympathetic Black versus sympathetic White coverage, on candidates and taxes (p < 0.05). This racial sympathy effect carries over to an overall racial main effect on candidate support (p < 0.05), and for taxes (p = 0.052). Racial resentment does not moderate valence effects or motivated racial effects. Second, racial stereotypes also moderate the racial frame effect, but less consistently. In line with racial sympathy, treatment policy gets less support with the sympathetic Black than White ‘face’ among the top half of the stereotype distribution (p < 0.05). Racial antipathy is also supported: unfavorable coverage also elicits less policy support with a Black than White ‘face’ among those high in racial stereotypes (p < 0.05). These two effects add up to a racial main effect on policy support from Black vs. White coverage among stereotyping Whites (p < 0.01). Racial stereotypes do not moderate other effects.
The results largely support this ideological null hypothesis, with one exception: The Black ‘face’ of sympathetic coverage negatively affects conservatives more than liberals regarding taxes (p < 0.001).
Raychaudhuri, Tanika, Tali Mendelberg, and Anne McDonough. Forthcoming. “The Political Effects of Opioid Addiction Frames.” The Journal of Politics. doi: 10.1086/720326