Introduction
In social science and epidemiology, researchers frequently aggregate individual data into "household" units for analysis. This practice, while convenient, rests on the implicit assumption of "entitativity"—the belief that a group of individuals functions as a single, coherent social actor. However, this paper argues that treating households as monolithic nodes can lead to significant methodological errors, fundamentally altering network metrics and misidentifying key influencers within a community.
The Problem of Node Contraction
The process of simplifying a network by grouping individuals into higher-level units is known as "node contraction." While this reduction in complexity can make large datasets more manageable, it often obscures the underlying social dynamics.
When individuals are contracted into a single household node, the internal structure of their relationships is lost. If a household does not share information perfectly or act with a unified purpose, this contraction creates an "illusion" of connectivity that does not exist in reality. The authors define specific contraction rules—Basic, Weighted, and Gendered—to mathematically describe how these transformations occur and where they fail to capture social reality.
Metric Distortions and the "Inversity Flip"
The authors demonstrate that node contraction can lead to "staggeringly different conclusions." One of the most critical findings is the "inversity flip." At the individual level, a village network might appear assortative—where popular individuals tend to connect with other popular individuals. However, once contracted into households, the same network can appear disassortative. This reversal occurs because households often form dense "cliques" (completely connected subgraphs) that skew aggregate data when collapsed.
Furthermore, clustering coefficients—measures of how likely nodes are to form triangles—are significantly higher in individual networks than in household networks. This discrepancy suggests that the perceived "tight-knit" nature of a community may be an artifact of family structures rather than broad social cohesion.
Influence and Gendered Bottlenecks
The choice of representational level also impacts "influence maximization" strategies. When identifying the top 10 most influential nodes to initiate a viral information campaign, household-level models often select entirely different targets than individual-level models. In datasets like the Banerjee et al. (2013) study of Indian villages, the overlap between these two sets of "influencers" was remarkably low, suggesting that researchers using household models might be targeting the wrong people for interventions.
This is particularly evident in "gendered bottlenecks." In many contexts, such as maternal health education, information flows through specific gendered layers. A household node masks these barriers; if women primarily share health advice with other women, a male-headed household node might suggest a path for information flow that is physically or socially blocked in reality.
Choosing the Right Level of Analysis
To address these issues, the authors propose a systematic decision tree based on Donald Campbell’s (1958) psychological criteria for entitativity: proximity, similarity, common fate, and internal diffusion.
By evaluating whether a group of individuals truly shares a "common fate" or exhibits high "internal diffusion" of information, researchers can make principled decisions about whether to use individual or household nodes. Applying this framework to previous high-profile studies, the authors suggest that in several cases, researchers likely should have utilized weighted individual or gendered networks rather than the unweighted household models they actually published.
Conclusion
The "illusion of the household" serves as a cautionary tale for network scientists. While aggregation is a powerful tool, it must be applied with rigorous attention to the social reality of the subjects. Future work must continue to explore the dynamic nature of household boundaries and the cultural nuances that define "entitativity" across different global contexts to ensure that social network models reflect reality rather than convenience.
Summary updated