TY - CHAP

T1 - The Role of Network Size for the Robustness of Centrality Measures

AU - Martin, Christoph

AU - Niemeyer, Peter

N1 - Conference code: 8

PY - 2020/1/1

Y1 - 2020/1/1

N2 - Measurement errors are omnipresent in network data. Studies have shown that these errors have a severe impact on the robustness of centrality measures. It has been observed that the robustness mainly depends on the network structure, the centrality measure, and the type of error. Previous findings regarding the influence of network size on robustness are, however, inconclusive. Based on twenty-four empirical networks, we investigate the relationship between global network measures, especially network size and average degree, and the robustness of the degree, eigenvector centrality, and PageRank. We demonstrate that, in the vast majority of cases, networks with a higher average degree are more robust. For random graphs, we observe that the robustness of Erdős-Rényi (ER) networks decreases with an increasing average degree, whereas with Barabàsi-Albert networks, the opposite effect occurs: with an increasing average degree, the robustness also increases. As a first step into an analytical discussion, we prove that for ER networks of different size but with the same average degree, the robustness of the degree centrality remains stable.

AB - Measurement errors are omnipresent in network data. Studies have shown that these errors have a severe impact on the robustness of centrality measures. It has been observed that the robustness mainly depends on the network structure, the centrality measure, and the type of error. Previous findings regarding the influence of network size on robustness are, however, inconclusive. Based on twenty-four empirical networks, we investigate the relationship between global network measures, especially network size and average degree, and the robustness of the degree, eigenvector centrality, and PageRank. We demonstrate that, in the vast majority of cases, networks with a higher average degree are more robust. For random graphs, we observe that the robustness of Erdős-Rényi (ER) networks decreases with an increasing average degree, whereas with Barabàsi-Albert networks, the opposite effect occurs: with an increasing average degree, the robustness also increases. As a first step into an analytical discussion, we prove that for ER networks of different size but with the same average degree, the robustness of the degree centrality remains stable.

KW - Centrality

KW - Measurement error

KW - Missing data

KW - Noisy data

KW - Robustness

KW - Sampling

KW - Business informatics

UR - http://www.scopus.com/inward/record.url?scp=85076713879&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-36687-2_4

DO - 10.1007/978-3-030-36687-2_4

M3 - Article in conference proceedings

AN - SCOPUS:85076713879

SN - 978-3-030-36686-5

VL - 1

T3 - Studies in Computational Intelligence

SP - 40

EP - 51

BT - Complex Networks and Their Applications VIII

A2 - Cherifi, Hocine

A2 - Gaito, Sabrina

A2 - Mendes, Jose Fernendo

A2 - Moro, Esteban

A2 - Rocha, Luis Mateus

PB - Springer Schweiz

CY - Cham

T2 - International Conference on Complex Networks and their Applications - 2019

Y2 - 10 December 2019 through 12 December 2019

ER -