Abstract
In November 2020, we conducted a cross-sectional study to implement and test the method of acquaintance sampling (randomly sampling friends of randomly sampled individuals) in detecting students with a higher probability of COVID-19 positivity. Overall, 879 students were randomly sampled and participated in this study. In an online survey, the randomly sampled participants nominated a friend and reported their own and their nominated friend's COVID-19 status. Nominated friends were about 1.64 (95% CI: 1.33, 2.00) times more likely to have ever been infected with COVID-19, compared to randomly sampled students. Our study corroborates the effectiveness of acquaintance sampling for identifying members of networks with higher COVID-19 risk. These findings could be useful for university policy makers when developing mitigation testing programs and intervention strategies against COVID-19 spread.
Introduction
Effective pandemic management, particularly within university settings, necessitates the identification of individuals at the highest risk of infection for targeted testing and intervention. Traditional mass surveillance is resource-intensive, while contact tracing is often reactive. Human social networks are structured such that individuals with higher social connectivity (high-degree nodes) are more likely to be exposed to and transmit viruses. Identifying these "hubs" typically requires complex network mapping.
This study explores a potential shortcut: the "Friendship Paradox." This mathematical principle states that, on average, your friends have more friends than you do. Consequently, a "friend of a random person" is statistically more likely to be a hub in the network. We aimed to determine if this principle could be leveraged through "acquaintance sampling" to identify high-risk individuals during a pandemic, even when social distancing measures had disrupted typical contact patterns.
Methods
We conducted a cross-sectional study in November 2020, utilizing data from an endline survey of a larger randomized controlled trial involving Indiana University undergraduate students. The study employed a two-step acquaintance sampling method:
- Index Group Selection: A random sample of 879 students was selected from the population.
- Nomination Group Selection: These index participants were asked to nominate a friend they had spoken to recently.
Participants reported their own COVID-19 test history and the perceived status of their nominated friend via an online REDCap survey. We used Poisson regression models with robust error variance to calculate the prevalence ratio between the two groups, adjusting for the crude association between sampling method and test positivity.
Results
The quantitative results strongly support the utility of acquaintance sampling. The prevalence of COVID-19 was 14.0% in the random sample (index group) compared to 23.0% among the nominated friends.
As shown in Table 1, nominated friends were 1.64 times (95% CI: 1.33, 2.00) more likely to have been infected with COVID-19 than the randomly sampled students. These findings confirm that acquaintance sampling is an effective tool for identifying higher-risk segments of a population without requiring full knowledge of the underlying network architecture.
Discussion
The "Friendship Paradox" serves as a structural sensor within social networks. Because hubs are preferentially targeted by epidemics, acquaintance sampling allows researchers to find these individuals simply by asking random participants to name a friend. This study demonstrates that even in a disrupted network where students practiced social distancing, these structural properties remained predictive.
Limitations
The study's opportunistic design introduces several limitations:
- Proxy Reporting: Data for nominated friends was reported by the index participants, potentially introducing recall bias or underestimation of infection status.
- Selection Bias: The study's focus on COVID-19 might have primed participants to nominate friends known to have had the virus.
- Temporal Scope: The study assessed "ever infected" status rather than active, current infections.
Conclusion and Future Work
Acquaintance sampling offers a low-cost, network-agnostic strategy for identifying high-risk individuals. Future research should evaluate the feasibility of this method for real-time surveillance and targeted immunization strategies. Further investigation is also needed to determine if the observed 1.64x multiplier remains consistent across different demographic groups and various infectious diseases.
Summary updated