The Centre for Social Sciences Institute for Sociology cordially invites you to
Jour Fixe 116.
Zoltán Brys: Estimating the effect of inaccurate edge information on centrality ranking order - case study of a university classroom network
Presenter: Zoltán Brys (CSS Institute for Sociology)
Co-presenter: Júlia Galántai (CSS Institute for Sociology)
Date: 25 May, 2023, 1 p.m.
Opponents: László Hajdu (University of Primorska), András Bóta (Lulea University of Technology) and Bence Kollányi (Centre for Social Sciences, RECENS)
Venue: Centre for Social Sciences, room B.1.15 (HU-1097 Budapest Tóth Kálmán utca 4.)
Online:
https://us06web.zoom.us/j/87213752214?pwd=OXkrNFBDV1lVWG95eGEycVRhRUFJdz09
Meeting ID: 872 1375 2214
Passcode: 104159
Abstract:
Collecting complete data on classroom networks is rarely possible. Absences, recall bias, response bias, compliance bias, and mistyping all result in incomplete input data. These noises affect the reliability of various centrality measures. Earlier studies highlighted the relative stability of degree centrality and eigenvector centrality and showed the instability of closeness centrality and betweenness centrality. Also it is known that the topology of the graphs also influences the robustness of centrality measures.
As in classroom network studies researchers typically know the number of nodes, we argue that specific studies are required which only estimate the effect of edge noise in classroom networks.
In this case study we used non-surveyed data of a university classroom (Fire et al. 2012) and we tested the sensitivity of various centrality measures against random edge deletion and random edge addition. The proportion of discordant pairs between the original and sampled ranking was used as sensitivity/robustness measure. Using various exponential and sigmoid approximations we estimated the robustness of centrality measures against random edge deletion and to random edge addition.
Albeit our findings are limited to analyzed graph, our results suggest that edge information might have stronger effect on centrality measures of classroom networks than we thought.