Activate Networks Inc., Newton, Massachusetts, United States of America.
PLoS One. 2013;8(2):e55234. doi: 10.1371/journal.pone.0055234. Epub 2013 Feb 13.
Studies of social networks, mapped using self-reported contacts, have demonstrated the strong influence of social connections on the propensity for individuals to adopt or maintain healthy behaviors and on their likelihood to adopt health risks such as obesity. Social network analysis may prove useful for businesses and organizations that wish to improve the health of their populations by identifying key network positions. Health traits have been shown to correlate across friendship ties, but evaluating network effects in large coworker populations presents the challenge of obtaining sufficiently comprehensive network data. The purpose of this study was to evaluate methods for using online communication data to generate comprehensive network maps that reproduce the health-associated properties of an offline social network. In this study, we examined three techniques for inferring social relationships from email traffic data in an employee population using thresholds based on: (1) the absolute number of emails exchanged, (2) logistic regression probability of an offline relationship, and (3) the highest ranked email exchange partners. As a model of the offline social network in the same population, a network map was created using social ties reported in a survey instrument. The email networks were evaluated based on the proportion of survey ties captured, comparisons of common network metrics, and autocorrelation of body mass index (BMI) across social ties. Results demonstrated that logistic regression predicted the greatest proportion of offline social ties, thresholding on number of emails exchanged produced the best match to offline network metrics, and ranked email partners demonstrated the strongest autocorrelation of BMI. Since each method had unique strengths, researchers should choose a method based on the aspects of offline behavior of interest. Ranked email partners may be particularly useful for purposes related to health traits in a social network.
使用自我报告的联系人绘制的社交网络研究表明，社交联系对个人采用或保持健康行为的倾向以及对他们采用肥胖等健康风险的可能性具有强大影响。社会网络分析对于希望通过确定关键网络职位来改善其人群健康的企业和组织可能证明是有用的。已经表明健康特征在友谊关系中相关，但在大型同事人群中评估网络效应存在获得足够全面网络数据的挑战。本研究旨在评估使用在线通信数据生成综合网络地图的方法，这些地图再现离线社交网络的与健康相关的属性。在这项研究中，我们使用基于以下三种技术从员工人群的电子邮件流量数据中推断社交关系：（1）交换的电子邮件数量绝对数量，（2）离线关系的逻辑回归概率，（3）排名最高的电子邮件交换伙伴。作为同一人群中离线社交网络的模型，使用调查工具中报告的社交关系创建了网络地图。根据捕获的调查联系比例，比较常见的网络指标以及跨社交联系的体重指数（BMI）的自相关，对电子邮件网络进行了评估。结果表明，逻辑回归预测了最大比例的离线社交联系，基于电子邮件交换数量的阈值产生了与离线网络指标最佳匹配，并且排名最高的电子邮件伙伴表现出 BMI 的最强自相关。由于每种方法都有其独特的优势，因此研究人员应根据离线行为的感兴趣方面选择方法。排名最高的电子邮件伙伴对于与社交网络中的健康特征相关的目的可能特别有用。
