Source-linked AI summary
The Dynamics of Health Behavior Sentiments on a Large Online Social Network
Marcel Salathé, Duy Q. Vu, Shashank Khandelwal, David R. Hunter
TL;DR
The paper examines which social-network factors drive the spread of vaccination sentiments, using temporal statistical models of H1N1-related sentiment on Twitter. It finds that neighborhood size and exposure effects differ by sentiment, with positive exposure sometimes predicting more negative expression.
Problem
Identifying the main drivers of health-behavior spread in social networks remains limited, despite its importance for understanding vaccination sentiment.
Method
The study statistically models temporal effects of network variables on users’ positive, neutral, or negative H1N1 vaccination sentiments.
Results
Larger neighborhoods generally inhibit sentiment expression, while positive exposure can predict increased negative-sentiment expression.
Takeaways & Limitations
Peer influence and social contagion in behavioral spread are strongly content-dependent, complicating communication strategies based on positive exposure.
Takeaways & Limitations
Some actual contagion may have gone unnoticed, limiting interpretation of observed sentiment spread.
Abstract
from arXiv · showhide
Modifiable health behaviors, a leading cause of illness and death in many countries, are often driven by individual beliefs and sentiments about health and disease. Individual behaviors affecting health outcomes are increasingly modulated by social networks, for example through the associations of like-minded individuals - homophily - or through peer influence effects. Using a statistical approach to measure the individual temporal effects of a large number of variables pertaining to social network statistics, we investigate the spread of a health sentiment towards a new vaccine on Twitter, a large online social network. We find that the effects of neighborhood size and exposure intensity are qualitatively very different depending on the type of sentiment. Generally, we find that larger numbers of opinionated neighbors inhibit the expression of sentiments. We also find that exposure to negative sentiment is contagious - by which we merely mean predictive of future negative sentiment expression - while exposure to positive sentiments is generally not. In fact, exposure to positive sentiments can even predict increased negative sentiment expression. Our results suggest that the effects of peer influence and social contagion on the dynamics of behavioral spread on social networks are strongly content-dependent.
1 Center for Infectious Disease Dynamics, Penn State University · 2 Department of Biology, Penn State University
The paper lists Penn State University affiliations spanning infectious disease dynamics, biology, and computer sciences and engineering.
- 1 Center for Infectious Disease Dynamics, Penn State University: The Center for Infectious Disease Dynamics is affiliated with Penn State University.
- 2 Department of Biology, Penn State University: The Department of Biology is affiliated with Penn State University.
- 2 Department of Biology, Penn State University: The Department of Computer Sciences and Engineering is affiliated with Penn State University.
4 Department of Statistics, Penn State University
The study examines temporal dynamics of vaccination sentiments on Twitter and finds that network effects are strongly content-dependent. Negative sentiment exposure predicts future negative expression, whereas positive exposure generally does not and can predict more negative expression.
- Study design: The study analyzes the temporal dynamics of a readily quantifiable intent to get vaccinated against a novel pandemic virus using categorized positive, neutral, and negative H1N1 vaccination sentiments.Although the data do not directly measure health behavior, they explain a large fraction of spatial variance in CDC-estimated H1N1 vaccination rates.
- Network effects: Larger opinionated neighborhood sizes generally inhibit expression of both the same and opposite sentiments.Larger positive and negative neighborhoods both predict diminished expression of the corresponding sentiment and the opposite sentiment.
- Network effects: Negative reciprocal neighborhood size predicts increased negative and decreased positive sentiment expression, while positive reciprocal neighborhood size generally has no significant predictive effect.These content-dependent effects are consistent with a role for homophily that differs across sentiment types.
- Exposure effects: 64% of network realizations showed increased negative exposure intensity predicting increased negative sentiment expression, while 44.5% showed positive exposure predicting increased negative expression.A further 33.5% showed higher positive exposure intensity predicting decreased positive sentiment expression.
- Interpretation: Overall, negative sentiment exposure is predictive of future negative expression, whereas positive sentiment exposure is generally not and effects overall favor negative sentiment spread.The authors define contagion here as predictive association rather than demonstrated causal transmission.
- Limitations: The observational framework may miss actual contagion and may misclassify personal opinions when peer pressure drives online sentiment expression.These limitations constrain interpretation of expressed sentiments as direct measures of underlying behavior or opinion.
FIGURE LEGENDS
Figure 1 depicts time-varying covariates capturing users’ sentiment histories, social contacts, and network-neighborhood structure, while coefficient panels summarize estimated effects on negative and positive sentiment expression. The panels show 95% confidence intervals across 200 network realizations and identify statistically significant positive or negative estimates.
- Figure 1 covariates: Figure 1 organizes covariates around users’ past sentiment expression, social contacts, and neighborhood network structure.Nodes represent users, arrows indicate follower relationships and information flow, and node numbers record positive and negative sentiment-expression histories; neutral sentiments are counted but not depicted.
- Figure 1 covariates: The covariates include followee counts, followee tweet volume, reciprocated relationships, follower information, user tweet counts, clustering, and followee-neighbor degrees.The figure identifies these measures as f1, f2, f5, u2, u1, f6, f7, f3, and f4, respectively.
- Figure 1 covariates: Covariate values can change over time as users post tweets and gain followers.The legend notes that the depicted covariates may change with the advance of time.
- Coefficient panels: The left panels estimate negative-sentiment-expression likelihoods, whereas the right panels estimate positive-sentiment-expression likelihoods.Panels A, C, E, and G correspond to negative sentiment; panels B, D, F, and H correspond to positive sentiment.
- Coefficient panels: The coefficient plots use circles for estimates, lines for 95% confidence intervals, and a zero-effect reference line across 200 network realizations.Panel percentages indicate the fraction of realizations with statistically significant positive or negative coefficient estimates.
FIGURE 1
Figures 2 and 3 present estimated coefficients for covariates related to social contagion and homophily, respectively. Each panel in Figure 2 shows coefficient means.
- Social contagion: Figure 2 reports estimated coefficients of covariates related to social contagion.Its panels show the means of the estimated coefficients.
- Homophily: Figure 3 reports estimated coefficients of covariates related to homophily, following the format of Figure 2.The passage identifies it as like Figure 2, but focused on homophily.
SUPPLEMENTRAY MATERIAL
Supplementary material is available on the Salathé Group website under Publications, listed with the article reference.
- Supplementary material is available at www.salathegroup.com under Publications, where readers can find the reference to this article.