Durably reducing partisan animosity through multiple scalable treatments

Fig. 1 presents treatment effects on general partisan animosity at T1–T5. Consistent with research examining discrete treatment effects (1), participants in the treatment (versus control) condition reported lower general partisan animosity following exposure to the first treatment at T1 (b = −6.77; P < 0.001), second treatment at T2 (b = −6.79; P < 0.001), and third treatment at T3 (b = −9.07; P < 0.001; H1). The effect at T3 was significantly larger than the effects at T1 (Δ: b = −2.20; P = 0.007) and T2 (Δ: b = −2.24; P = 0.004). For reference, Americans’ partisan animosity has increased by approximately 0.6 percentage points per year between 1978 and 2016. Therefore, this T3 effect was the equivalent of reversing more than 15 y of increasing partisan animosity. Further, participants in the treatment (versus control) condition reported lower general partisan animosity 2 wk (T4: b = −3.76; P < 0.001; H2) and 4 wk (T5: b = −3.92; P < 0.001; H3) after the last treatment. Effects at T4 and T5 were significantly weaker than effects at T3 (ΔT4: b = 5.31; P = 0.007; ΔT5: b = 5.05; P = 0.007), but the rate of decay appears to be nonlinear (RQ1): There was no decline in effect size between T4 and T5 (Δ: b = 0.27; P = 0.752).

Fig. 2 presents the treatment effects on partisan animosity toward personally known outparty coworkers and acquaintances at T1–T5. While general partisan animosity was higher than 60 out of 100 at every wave in both conditions, interpersonal partisan animosity was always less than 40. Nonetheless, assignment to the treatment also reduced interpersonal partisan animosity (RQ2) at T1 (b = −1.74; P = 0.025), T2 (b = −2.74; P = 0.002), and T3 (b = −3.45; P < 0.001). The effect at T3 was significantly larger than the effect at T1 (Δb = −1.83; P < 0.001) but not versus T2 (Δb = −0.75; P = 0.337). Finally, the effect on interpersonal partisan animosity was also significant 2 wk (T4; b = −3.27; P < 0.001) and 4 wk (T5; b = −2.60; P = 0.007) after the last treatment. The effects at T3–T5 were not statistically different (all |Δb|s < 0.75; all Ps > 0.433).

Treatment effects were significantly stronger for general (versus interpersonal) partisan animosity at T1–T3 (all bs < −4.07, all ps < 0.001). However, at T4 and T5, the treatment effects did not differ significantly between the two outcomes. Thus, while the treatments had stronger initial effects on animosity toward general outparty members, the decay of the effect on animosity toward outparty members in preexisting, ongoing relationships was significantly weaker from T3 to both T4 (b = −5.06, P < 0.001) and T5 (b = −4.24, P < 0.001). While more than 50% of the effect on general animosity decayed after 2 wk, only about 5% of the effect on interpersonal animosity decayed over the same period (and only 25% of the effect decayed after 4 wk). These differing results highlight the potential value of studying interpersonal as well as general partisan animosity.

Partisan animosity in the United States has been linked to numerous negative interpersonal and societal outcomes (15). Much remains unknown, however, about effective ways to meaningfully reduce partisan animosity over time using scalable treatments. We found that exposure to multiple scalable treatments over a week, employing multiple previously validated strategies, produced substantial and durable reductions in partisan animosity, persisting even a month later. While designed to reduce animosity toward general outparty members, these treatments also reduced animosity toward specific outparty coworkers and acquaintances. Thus, treatments designed to reduce interpersonal partisan animosity might do even more to remedy the negative effects of polarization in everyday relationships. Future work should also directly compare the effects of employing multiple (versus singular) treatments and treatments using different (versus redundant) strategies. Moreover, researchers should consider designs that can rule out possible consistency bias created by repeatedly measuring partisan animosity in the same manner. Because the current treatments require so little time, they could be feasibly administered in educational, civic, or professional contexts, akin to other forms of bias training that circumvent self-selection. Accordingly, our results provide an empirical basis for practically deploying strategies to ameliorate partisan polarization in society.