Author Correction: Node-based generalized friendship paradox fails

Anna Evtushenko, Jon KleinbergScientific 2023 · DOI · OpenAlex

Abstract

This document provides a necessary correction to the numerical values presented in Figure 4 of the original article "Node-based generalized friendship paradox fails." The correction clarifies the "delta" values associated with specific nodes in an anti-SGFP graph. Crucially, the authors confirm that these adjustments do not alter the study's primary conclusion: the Node-based Generalized Friendship Paradox (SGFP) is sensitive to attribute sampling and can fail even in networks with zero degree-attribute correlation.

Introduction: The Limits of Social Paradoxes

The "Friendship Paradox" is a well-known sociological phenomenon stating that, on average, a person's friends have more friends than they do. This concept was later expanded into the Generalized Friendship Paradox (GFP), which suggests that your friends are also likely to surpass you in other attributes, such as wealth, popularity, or productivity.

In their original research, Evtushenko and Kleinberg investigated the Node-based Generalized Friendship Paradox (SGFP) to determine if this paradox holds true at the level of individual nodes across all network structures. By identifying "anti-SGFP" graphs—specific network configurations where the paradox fails—the authors demonstrated the mathematical limits of these social observations.

The Nature of the Correction

The primary focus of this correction is a calculation error in the illustrative Figure 4. This figure serves to demonstrate how different attribute samples on the same graph structure can produce varying "gaps" ($g$) between a node and its neighbors.

Figure 4: An anti-SGFP graph illustrating zero, positive, and negative gaps depending on attribute sampling.

The correction specifically updates the "delta" ($δ$) values for nodes with degrees 1 and 3. These values are critical for calculating the "gap," which represents the difference between a node's own attribute and the average attribute of its immediate neighbors.

Key Insights: Context and Sampling

The corrected figure reinforces several key theoretical insights:

  1. Structural Sensitivity: The same graph structure can exhibit different paradoxical behaviors depending on how attributes are distributed across its nodes.
  2. The Role of Correlation: Even when the correlation between a node's degree and its attribute is zero ($r_{d,a} = 0$), the perceived "gap" can be positive, negative, or zero. This proves that the paradox is not a structural certainty dictated solely by network topology.
  3. Gap Dynamics:
    • Zero Gap: Attributes are balanced such that the paradox is neutralized.
  4. Positive Gap: The neighbors have a higher average attribute, supporting the paradox.
  5. Negative Gap: The node itself has a higher attribute than its neighbors' average, causing the paradox to fail.

Conclusion

While the numerical "delta" values in the original publication were misreported, the underlying mathematical framework and the resulting conclusions remain sound. The existence of anti-SGFP graphs highlights that the Generalized Friendship Paradox is a context-dependent observation rather than a universal law of social networks. The authors emphasize that this correction is purely technical and does not impact the integrity of the research findings.

Summary updated

Abstract

Figure 4.An anti-SGFP graph such as this may produce differently-signed gaps for r d,a = 0, depending on the attribute sample.
BibTeX
@inproceedings{evtushenko2023author,
  author = {Anna Evtushenko and Jon Kleinberg},
  title = {{Author Correction: Node-based generalized friendship paradox fails}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41598-023-34980-5},
}
Figure