Publications

Complex contagion rests on the idea that individuals are more likely to adopt a behavior if they experience social reinforcement from multiple sources. We develop a test for complex contagion, conceptualized as social reinforcement, and then use it to examine whether empirical data from a country-scale randomized controlled viral marketing field experiment show evidence of complex contagion. The experiment uses a peer encouragement design in which individuals were randomly exposed to either one or two friends who were encouraged to share a coupon for a mobile data product. Using three different analytical methods to address the empirical challenges of causal identification, we provide strong support for complex contagion: the contagion process cannot be understood as independent cascades but rather as a process in which signals from multiple sources amplify each other through synergistic interdependence. We also find social network embeddedness is an important structural moderator that shapes the effectiveness of social reinforcement.

‍

Is there a relationship between depression and political evaluations? Building on existing work, we argue that experiencing depressive symptoms will positively correlate with supporting a populist politician and negatively correlate with supporting a nonpopulist officeholder. We evaluate these predictions with data from the United States, focusing on Donald Trump and Joe Biden. Our data are consistent with our hypotheses, and, as expected, we find particularly strong relationships for Democratic respondents. The results highlight the importance of considering mental health when studying the approval of politicians both in and out of office. We conclude with a discussion of next steps for a research agenda on depression and political evaluations.

Complex contagion rests on the idea that individuals are more likely to adopt a behavior if they experience social reinforcement from multiple sources. We develop a test for complex contagion, conceptualized as social reinforcement, and then use it to examine whether empirical data from a country-scale randomized controlled viral marketing field experiment show evidence of complex contagion. The experiment uses a peer encouragement design in which individuals were randomly exposed to either one or two friends who were encouraged to share a coupon for a mobile data product. Using three different analytical methods to address the empirical challenges of causal identification, we provide strong support for complex contagion: the contagion process cannot be understood as independent cascades, but rather as a process in which signals from multiple sources amplify each other through synergistic interdependence. We also find social network embeddedness is an important structural moderator that shapes the effectiveness of social reinforcement.

The transmission of communicable diseases in human populations is known to be modulated by behavioral patterns. However, detailed characterizations of how population-level behaviors change over time during multiple disease outbreaks and spatial resolutions are still not widely available. We used data from 431,211 survey responses collected in the United States, between April 2020 and June 2022, to provide a description of how human behaviors fluctuated during the first 2 y of the COVID-19 pandemic. Our analysis suggests that at the national and state levels, people’s adherence to recommendations to avoid contact with others (a preventive behavior) was highest early in the pandemic but gradually—and linearly—decreased over time. Importantly, during periods of intense COVID-19 mortality, adaption to preventive behaviors increased—despite the overall temporal decrease. These spatial-temporal characterizations help improve our understanding of the bidirectional feedback loop between outbreak severity and human behavior. Our findings should benefit both computational modeling teams developing methodologies to predict the dynamics of future epidemics and policymakers designing strategies to mitigate the effects of future disease outbreaks.

Importance  Screening measures of depressive symptoms (eg, 9-item Patient Health Questionnaire [PHQ-9]) are increasingly used in surveys and remote applications, where shorter versions would be valuable.

Objective  To derive shorter versions of the PHQ-9 that maximize the variability in total depressive symptom severity captured.

Design, Setting, and Participants  This survey study used data from 4 waves of a 50-state nonprobability web-based survey conducted between November 2, 2023, and July 21, 2024. Survey respondents were aged 18 years or older. The first survey wave data were used to identify shortened question subsets capturing variance in the PHQ-9 and estimating a PHQ-9 score of 10 or higher. Resulting models (eg, 3-item version of the PHQ [PHQ-3]) were validated in subsequent survey waves.

Main Outcome and Measure  Performance of PHQ-3 in the full sample and across subgroups of age, gender, race and ethnicity, and educational levels. Depressive symptom severity was measured with the PHQ-9 (total score range: 0-27, with a score ≥10 indicating moderate or greater depressive symptoms).

Results  In the 4 survey waves, there were 96 234 total participants (mean [SD] age, 47.3 [17.1] years; 55 245 [57.4%] identifying as women). In the full sample, 4401 participants (4.6%) identified as Asian American, 12 699 (13.2%) as Black or African American, 9776 (10.2%) as Hispanic or Latino, and 65 309 (67.9%) as White individuals, with 4049 (4.2%) who identified as having other race or ethnicity. Among these participants, the mean (SD) PHQ-9 score was 6.5 (6.6), and 25 411 (26.4%) met the criteria for moderate or greater depressive symptoms (PHQ-9 score ≥10). The optimal 3-item version, PHQ-3, used items 2 (subject: depressed mood), 6 (self-esteem or failure), and 1 (interest), yielding a Cronbach α of 0.88 (95% CI, 0.88-0.88) and Pearson correlation with the PHQ-9 total score of 0.93 (95% CI, 0.93-0.94). At a threshold of 3 or greater, the PHQ-3 sensitivity was 0.98 (95% CI, 0.97-0.98) and specificity was 0.76 (95% CI, 0.75-0.76) for moderate or greater depressive symptoms. Performance was consistent across sociodemographic subgroups and survey waves.

Conclusions and Relevance  This survey study of US adults identified a 3-item scale that remained highly correlated with the full PHQ-9 instrument. The reduced set of questions could enable more widespread and efficient incorporation of depressive symptom measurement in general population samples.

‍

Social media play a pivotal role in disseminating web content, particularly during elections, yet our understanding of the association between demographic factors and information sharing online remains limited. Here, we introduce a unique dataset, DomainDemo, linking domains shared on Twitter (X) with the demographic characteristics of associated users, including age, gender, race, political affiliation, and geolocation, from 2011 to 2022. This new resource was derived from a panel of over 1.5 million Twitter users matched against their U.S. voter registration records, facilitating a better understanding of a decade of information flows on one of the most prominent social media platforms and trends in political and public discourse among registered U.S. voters from different sociodemographic groups. By aggregating user demographic information onto the domains, we derive five metrics that provide critical insights into over 129,000 websites. In particular, the localness and partisan audience metrics quantify the domains’ geographical reach and ideological orientation, respectively. These metrics show substantial agreement with existing classifications, suggesting the effectiveness and reliability of DomainDemo’s approach.

Following the leadership transition on October 27, 2022, Twitter/X underwent a notable change in

platform governance. This study investigates how these changes influenced information quality for

registered U.S. voters and the platform more broadly. We address this question by analyzing two

complementary datasets—a Twitter panel and a Decahose sample. Our findings reveal a subtle yet

statistically significant decline in information quality across both datasets, stemming from an increase in

content from low-quality sources and a decrease in content from high-quality sources. These results

suggest that the ownership change and subsequent policy adjustments were associated with shifts in the

platform’s information ecosystem. Our results underscore the broader significance of ownership and

governance for information quality in dynamic sociotechnical systems, highlighting the determinantal

power that platform owners may have in shaping the information ecosystem.

‍

How does social network structure amplify or stifle behavior diffusion? Existing theory suggests that when social reinforcement makes the adoption of behavior more likely, it should spread more—both farther and faster—on clustered networks with redundant ties. Conversely, if adoption does not benefit from social reinforcement, it should spread more on random networks which avoid such redundancies. We develop a model of behavior diffusion with tunable probabilistic adoption and social reinforcement parameters to systematically evaluate the conditions under which clustered networks spread behavior better than random networks. Using simulations and analytical methods, we identify precise boundaries in the parameter space where one network type outperforms the other or they perform equally. We find that, in most cases, random networks spread behavior as far or farther than clustered networks, even when social reinforcement increases adoption. Although we find that probabilistic, socially reinforced behaviors can spread farther on clustered networks in some cases, this is not the dominant pattern. Clustered networks are even less advantageous when individuals remain influential for longer after adopting, have more neighbors, or need more neighbors before social reinforcement takes effect. Under such conditions, clustering tends to help only when adoption is nearly deterministic, which is not representative of socially reinforced behaviors more generally. Clustered networks outperform random networks by a 5% margin in only 22% of the parameter space under its most favorable conditions. This pattern reflects a fundamental trade-off: Random ties enhance reach, while clustered ties enhance social reinforcement.

‍

Belief in conspiracy theories has significant social and political consequences. While prior research has focused primarily on psychological predispositions as drivers of conspiracy beliefs, relatively less is known about the role of social networks. Here, we examine how information received from different sources is linked to the endorsement of conspiracy theories, using the 2024 attempted assassination of presidential candidate Donald Trump as a case study. In surveys conducted days after the attack, social media was the most commonly reported source of conspiracy theories about the event. At the same time, information consumption on social media was not consistently associated with stronger conspiracy beliefs. In contrast, information received through interpersonal ties was more closely linked to belief in both left-leaning and right-leaning conspiratorial narratives. These findings highlight the importance of examining the social dimensions of conspiracy belief formation. Understanding how interpersonal communication shapes conspiracy beliefs is critical for explaining their spread and persistence. Future research would benefit from further investigating the social contexts that sustain conspiratorial thinking.

Much of the research quantifying volume and spread of online misinformation measures the construct at the source level, identifying a set of specific unreliable domains that account for a relatively small share of news consumption. This source-level dichotomy obscures the potential for users to repurpose factually true information from reliable sources to advance misleading narratives. We demonstrate this potentially far more prevalent form of misinformation by identifying articles from reliable sources that are frequently co-shared with (shared by users who also shared) "fake" news on social media, and concurrently extracting narratives present in fake news content and claims fact-checked as false. Specifically in this study, we use Twitter/X data from May 2018 to November 2021 matched to a U.S. voter file. We find that narratives present in misinformation content are significantly more likely to occur in co-shared articles than in articles from the same reliable sources that are not co-shared, consistent with users using information from mainstream sources to enhance the credibility and reach of potentially misleading claims.