Combining interventions to reduce the spread of viral misinformation

Abstract

Misinformation online poses a range of threats, from subverting democratic processes to undermining public health measures. Proposed solutions range from encouraging more selective sharing by individuals to removing false content and accounts that create or promote it. Here we provide a framework to evaluate interventions aimed at reducing viral misinformation online both in isolation and when used in combination. We begin by deriving a generative model of viral misinformation spread, inspired by research on infectious disease. By applying this model to a large corpus (10.5 million tweets) of misinformation events that occurred during the 2020 US election, we reveal that commonly proposed interventions are unlikely to be effective in isolation. However, our framework demonstrates that a combined approach can achieve a substantial reduction in the prevalence of misinformation. Our results highlight a practical path forward as misinformation online continues to threaten vaccination efforts, equity and democratic processes around the globe.

Fig. 1. Overview of our data-processing and generative model of viral misinformation spread.

a, Example of an event segmented from a larger incident (dashed lines). b, Our model fit to the time series for a single event. The dashed line indicates the expected value; the shaded region denotes the 89% CI. c, Cumulative engagement as a measure of total misinformation. The lines and shading are as in b. d, Model-simulated platform interventions for a single event. The lines indicate median cumulative engagement over 100 simulations. Grey indicates the baseline, blue indicates a 10% ‘nudge’, orange indicates banning, yellow indicates a virality circuit breaker and brown indicates the outright removal of content.

Fig. 2. Simulated impacts of content removal and virality circuit breakers.

a, The impact of the outright removal of all misinformation-related posts following a delay specified in minutes. b, As in a if only 20% of events are removed. c, The impact of applying a virality circuit breaker that reduces virality by 10% to all misinformation events after a specified period. d, As in c if the virality circuit breaker is applied to only 20% of events. The horizontal grey bars in each plot represent the baseline conditions, with the line indicating the mean and the shaded area highlighting the 89% CI. The violins indicate the simulated distribution of total posts across all events. The dots and lines within the violins indicate the median and interquartile range.

Fig. 3. Simulated efficacy of nudging and account removal interventions.

a, The effect of nudges that inoculate a percentage of the population against spreading misinformation. Shown is the cumulative total engagement across all events. b, Number of accounts that either are currently removed or would have been removed under a three-strikes policy. Brown bars indicate the threshold being applied to all users above a given threshold of followers in thousands, K. c, The effect of account removal for those that are currently banned (orange) or those banned following a three-strikes rule applied solely to verified accounts (blue). d, As in a and c, but showing the impact of enacting three-strikes policies with varying thresholds. The horizontal grey bars in a, c and d represent the baseline conditions, with the line indicating the mean and the shaded area highlighting the 89% CI. The violins indicate the simulated distribution of total posts across all events. The dots and lines within the violins indicate the median and interquartile range.

Fig. 4. Simulated impact of combining interventions.

a, The impact of a modest combined approach to intervention (described in the text, green) and each intervention applied individually. b, The impact of a more aggressive combined approach (described in the text, green) and each intervention applied individually. c, Relationship between engagement within the largest viral event for a given incident and subsequent engagement. d, Expected post-event engagement given the action taken during an event. The dots and lines within the violins indicate the median and interquartile range.