Bootstrapping under constraint for the assessment of group behavior in human contact networks

The increasing availability of time –and space– resolved data describing human activities and interactions gives insights into both static and dynamic properties of human behavior. In practice, nevertheless, real-world datasets can often be considered as only one realisation of a particular event. This highlights a key issue in social network analysis: the statistical significance of estimated properties. In this context, we focus here on the assessment of quantitative features of specific subset of nodes in empirical networks. We present a method of statistical resampling based on bootstrapping groups of nodes under constraints within the empirical network. The method enables us to define acceptance intervals for various Null Hypotheses concerning relevant properties of the subset of nodes under consideration, in order to characterize by a statistical test its behavior as ``normal’’ or not. We apply this method to a high resolution dataset describing the face-to-face proximity of individuals during two co-located scientific conferences. As a case study, we show how to probe whether co-locating the two conferences succeeded in bringing together the two corresponding groups of scientists.

💡 Research Summary

The paper addresses a fundamental challenge in social network analysis: assessing the statistical significance of observed properties when only a single empirical realization of a network is available. Traditional random bootstrapping methods often ignore the underlying structural constraints of the network, leading to misleading null‑model distributions. To overcome this, the authors introduce a “bootstrapping under constraint” framework that generates resampled groups of nodes while preserving selected structural characteristics of the original group of interest.

The methodology proceeds in several steps. First, the target node set G (e.g., a specific community, demographic group, or event participants) is defined, and a set of constraints is chosen. Typical constraints include the size of the group, its average degree, internal edge density, and the proportion of edges that connect to the rest of the network. These constraints can be tailored to the research question, ensuring that the null model respects the most relevant aspects of the original group’s topology.

Second, the authors develop an efficient Markov‑Chain Monte Carlo (MCMC) sampling algorithm to draw a large number of random groups that satisfy the constraints. Starting from an arbitrary feasible group, the algorithm repeatedly proposes elementary moves—such as swapping a node with an external neighbor or adding/removing a node—while checking that all constraints remain satisfied. Convergence diagnostics and appropriate thinning intervals are employed to guarantee that the sampled groups are approximately independent.

Third, for each resampled group a set of “observable” statistics is computed. In the case study these include the average contact duration per pair, the frequency of cross‑group contacts, and clustering coefficients. The collection of statistics across all bootstrap samples forms an empirical null distribution for each observable. The observed value for the original group G is then compared to the corresponding acceptance interval (e.g., the 5th–95th percentile). If the observed statistic falls outside this interval, the null hypothesis—that G behaves like any other group with the same constraints—is rejected, indicating that G exhibits atypical behavior.

The authors validate the approach on synthetic networks (Erdős‑Rényi, Barabási‑Albert, hierarchical modular graphs) and demonstrate that constrained bootstrapping dramatically reduces false‑positive rates compared with unconstrained random sampling. They also show that tightening constraints narrows the acceptance intervals, thereby increasing the power of the test, albeit at the cost of higher computational effort due to reduced sampling efficiency.

The core application involves a high‑resolution RFID data set collected during two co‑located scientific conferences (approximately 300 participants each) over three days. The research question is whether the joint venue fostered meaningful interaction between the two scientific communities. By imposing constraints on group size and average degree, the authors generate null distributions for intra‑ and inter‑group contact metrics. Results reveal that while each conference’s internal contact patterns fall well within the expected range, the cross‑conference contact frequency is significantly lower than the null expectation. Consequently, the study concludes that the co‑location did not substantially increase face‑to‑face interactions between the two groups, suggesting that logistical co‑hosting alone may be insufficient to promote interdisciplinary networking.

In the discussion, the authors acknowledge limitations: overly strict constraints can make the sampling process inefficient, and the choice of constraints may introduce subjectivity. They propose extensions to incorporate temporal constraints (e.g., preserving contact patterns within specific time windows) and to apply the framework to multilayer or multiplex networks where different types of interactions coexist.

Overall, the paper contributes a robust statistical tool for evaluating the behavior of specific node subsets in empirical networks. By respecting key structural features during resampling, the constrained bootstrapping approach offers more reliable hypothesis testing than conventional randomization techniques, with broad applicability across fields such as epidemiology, organizational studies, and urban mobility analysis.