BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//POPNET - ECPv6.15.20//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-ORIGINAL-URL:https://www.popnet.io
X-WR-CALDESC:Events for POPNET
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20210328T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20211031T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20220327T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20221030T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20230326T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20231029T010000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Europe/Amsterdam:20220718T230000
DTEND;TZID=Europe/Amsterdam:20220722T225959
DTSTAMP:20260504T025339
CREATED:20220704T111256Z
LAST-MODIFIED:20220711T113451Z
UID:905-1658185200-1658530799@www.popnet.io
SUMMARY:Capturing the social fabric: population-scale socio-economic segregation patterns
DESCRIPTION:Conference poster presentation by Yuliia Kazmina at Harper Center of the Booth School of Business at the University of Chicago (online talk) \n\n\n\nAuthors: Yuliia Kazmina\, Eszter Bokanyi\, Eelke Heemskerk and Frank Takes \n\n\n\nSegregation is a widely studied issue traditionally explored from the point of the spatial distribution of different groups\, defined by individual attributes such as race\, religion or social class. Instead\, in this work we argue that the issues of persistent segregation\, specifically socio-economic segregation\, are in fact networked phenomena and should thus be studied as such [1\,2]. We present a methodological contribution that moves away from a traditional spatial understanding of segregation\, and instead considers segregation measurement within the direct social network of individuals.  \n\n\n\nThe study is based on Dutch population register data sourced from multiple existing registers that contain information on formal ties of ~17 million residents. Data covers multiple social contexts (layers): kinship\, household\, neighborhood\, school\, and work. With the multilayer network of geospatially embedded formal ties in hand\, we study to what extent social segregation is clustered in social networks\, and how each network layer contributes to it. Specifically\, we measure to what extent people are exposed to individuals of different socio-economic statuses (SES) for each of the social contexts. Moreover\, we look at what social contexts provide diverse social contact opportunities with respect to the socio-economic status and\, inversely\, what social contexts play a role in sustaining so-called  “socio-economic bubbles”.  \n\n\n\nTo capture socio-economic segregation patterns on a population scale\, we introduce a concept of “social opportunity structures” that builds upon the Opportunity Structure Theory proposed by Ken Roberts [3]. Individual ego networks we observe in this study are assumed to be a realization of a particular aspect of Roberts’ opportunity structures – they represent the anatomy and composition of social circles within which individuals exist\, evolve\, and are required to make successive choices. We aggregate household-level social opportunity structures in each of the selected contexts to the level of a municipality to measure to what extent households of a certain socio-economic status (captured by the equivalised household income) are\, on average\, exposed to households across all income deciles. In this abstract\, we focus on the municipality of Amsterdam. \n\n\n\nEstimated social opportunity structures for each of the selected contexts are represented by what we call social opportunity matrices\, in which the vertical axis represents analyzed households divided into ten income deciles\, sorted in ascending order. Then\, the horizontal axis indicates income deciles of connected households in the increasing order. Each cell at the intersection of two income deciles displays the share of contacts a household of a certain income bracket (on the vertical axis) shares with the households in the income decile on the horizontal axis. Values are normalized by row. The diagonal elements represent the share of contacts each income decile has within its own income bracket. To capture the overall segregation for a particular context\, we measure the extent of link assortativity [4]  with respect to income.  \n\n\n\nFigure 1 presents social opportunity structures with respect to income for the households in the city of Amsterdam (~460k households) in the kinship\, school\, work\, and neighborhood (both administrative neighborhoods typically containing several hundred to thousands of households as well as the ten closest neighbors) contexts. The estimated social opportunity matrices present a number of interesting findings.  \n\n\n\nFirst\, in Fig. 1a we see that all income brackets are highly exposed to the neighbors that belong to the two lowest income deciles in the context of being in the same administrative neighborhood. Second\, once the context is narrowed down to the subset of the ten closest neighboring households only (Fig. 1b)\, the matrix reveals a significantly different pattern: close neighborhood social context is much more assortative with respect to income\, as evidenced by the assortativity value of  0.12 vs 0.04 in the case of the administrative neighborhood.  \n\n\n\nThird\, the family layer (Fig. 1c) exhibits similar income assortativity pattern\, with a high prevalence of within income bracket connectivity with 25-30% of family members living separately from an observed household belonging to the same income bracket.  \n\n\n\nAlthough the overall assortativity in the school layer is again comparable\, the distribution of the preference for the own income class along income range is significantly dissimilar: the strongest preference to be classmates with children and adolescents that belong to the same socio-economic class is observed in the lowest income decile as well as in the richest 10% of the households. Finally\, the workplaces’ (Fig. 1d) assortativity is relatively high\, however\, we do not observe an apparent prevalence of diagonal elements\, likely due to several very large workplaces being present in the data.  \n\n\n\nConcluding\, we find that the analyzed social contexts are highly dissimilar in terms of socio-economic assortativity. The most assortative layer is the family network. Other layers\, while being less assortative overall\, reveal interesting patterns. Close neighbors and small workplaces exhibit highly assortative mixing patterns with respect to income that limits the exposure to individuals from different socio-economic backgrounds. On the other hand\, school networks display relatively lower income assortativity and provide individuals with diverse social contact opportunities.  \n\n\n\nThe broad implication of the present study is the potential to capture and quantify social segregation patterns on a large scale with the ability to distinguish between different social contexts\, advocating the study of multi-layer administrative data for the purposes of obtaining a more global policy-relevant insight into population-scale social cohesion. \n\n\n\nReferences\n\n\n\nFreeman\, L. C. (1978). Segregation in social networks. Sociological Methods & Research 6 (1978): 411 – 429.Dimaggio\, P.\, & Garip\, F. (2012). Network effects and social inequality. Annual Review of Sociology 38:1 (2012): 93-118.Roberts\, K. (1977). The Social Conditions\, Consequences and Limitations of Careers Guidance. British Journal of Guidance & Counselling 5:1 (1977): 1-9.Newman\, M. E. J. (2002). Assortative mixing in networks. Physical Review Letters Vol. 89 (20): 208701.\n\n\n\nFigure 1. Social opportunity structures of the households in Amsterdam\, each subfigure displaying a different context: \n\n\n\na) administrative neighborhood  (assortativity: 0.035)               \n\n\n\n\n\n\n\nb) close neighbors (assortativity: 0.118) \n\n\n\n\n\n\n\nc) family (assortativity: 0.124) \n\n\n\n\n\n\n\nd) school (assortativity: 0.114)                                                   \n\n\n\n\n\n\n\ne) workplace  (assortativity: 0.123)
URL:https://www.popnet.io/events/capturing-the-social-fabric-population-scale-socio-economic-segregation-patterns/
CATEGORIES:Conference talk
ATTACH;FMTTYPE=image/png:https://www.popnet.io/wp-content/uploads/2021/07/ic2s2.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/Amsterdam:20220720T192000
DTEND;TZID=Europe/Amsterdam:20220720T202000
DTSTAMP:20260504T025339
CREATED:20220704T103548Z
LAST-MODIFIED:20220905T080720Z
UID:901-1658344800-1658348400@www.popnet.io
SUMMARY:Anonymity of Multi-hop Neighborhoods in Social Networks
DESCRIPTION:Conference poster presentation by Rachel de Jong at Harper Center of the Booth School of Business at the University of Chicago. \n\n\n\nAuthors: Rachel de Jong\, Mark van der Loo and Frank Takes \n\n\n\nIntroduction & Goal. Sharing large-scale social network datasets is advantageous for the development of computational social science\, since studying and replicating findings on such datasets is key to understanding and modeling various social phenomena [1\, 2]. Following the principles of widely implemented privacy laws such as GDPR\, such datasets need to be anonymous\, which means that people should not be identifiable by someone with a realistic amount of background knowledge. This work focuses on a method to assess this so-called risk of disclosure\, by measuring the anonymity of individuals in networks based on their structural position within the network. \n\n\n\nPrevious work has focussed on measuring anonymity using only the direct surroundings of a node [3]. However\, in [4] it is shown that when a possible attacker has information about a larger neighborhood beyond these direct surroundings\, this could drastically decrease the anonymity of the individual. Therefore\, in this work\, we present a novel approach that extends these two earlier works into a parametrized measure that can serve as a lower bound for the expected anonymity at different levels of knowledge of the attacker. On both modeled and real-world social network data\, we demonstrate that if an attacker has perfect information about what we call multi-hop neighborhoods\, the anonymity of individuals in the social network is severely compromised. This has serious implications for any social science researcher sharing social network data with other parties. \n\n\n\nApproach. We measure the anonymity by partitioning the set of nodes of a given social network into equivalence classes. We define equivalence by using the measure of d–k-anonymity\, where two nodes are d-equivalent if 1) their respective d-hop neighborhoods (i.e.\, neighborhoods up to distance d of the node) are isomorphic\, and 2) there is an isomorphism mapping the two compared nodes onto each other. Next\, following [3]\, we define a node as unique if it has no equivalent nodes in the network.  \n\n\n\nTo understand anonymity of individuals in real-world networks\, we measure structural anonymity in various known graph models (Erdős–Rényi (ER) and Watts Strogatz (WS)) and a range of empirical network datasets. We investigate anonymity for increasingly larger hop neighborhoods\, and therewith different attacker knowledge scenarios. This improves upon [3] because we allow for larger-hop neighborhoods\, and upon [4] because we assume perfect information about connectivity of individuals up to a certain distance. \n\n\n\nResults. Figure 1 shows the fraction of unique nodes as a function of the number of nodes n and the average degree. Blue indicates a small fraction of unique nodes\, thus\, high anonymity\, and red indicates a large fraction of unique nodes\, thus\, low anonymity. In the case where d=1\, so in the leftmost column of Figure 1\, our work reproduces precisely the findings in [3]. However\, most importantly\, for larger d-hop neighborhoods\, shown in the middle and rightmost columns of Figure 1\, we see that the uniqueness landscape changes completely. The number of unique nodes\, and its dependence on n and the average degree\, both change drastically. This holds for both models: the fraction of unique nodes becomes high for networks with lower average degrees\, and increasing the network size has less effect on the fraction of unique nodes than for d=1. We conclude that increasing the distance therefore radically decreases the overall anonymity of nodes in the network. \n\n\n\nIn Figure 2\, we summarize our findings for various empirical networks with sizes ranging from 167 to 19.7K nodes. For 10 different real-world networks\, we observe behavior in three categories: 1) high anonymity at d ≥ 1\, 2) high anonymity at d=1\, low anonymity at d ≥ 2 and 3) low anonymity at d=1. Despite currently being publicly available for research\, for most network datasets a large fraction of nodes is uniquely identifiable when information about the 1-hop neighborhood is known. When information about 2-hop neighborhoods is known\, this fraction increases drastically; more entities represented in the network can be uniquely identified and are thus not anonymous. \n\n\n\nConclusions. Our results show that if an attacker has perfect information about multi-hop neighborhoods\, even just at distance two\, then this can drastically reduce the anonymity of nodes in networks\, as observed for the network models and the empirical networks analyzed in our experiments. Since it is realistic for an attacker to obtain some (but not always all) information about larger-hop neighborhoods\, one cannot dismiss the de-anonymizing effects of network structure surrounding a node for d ≥ 2. In future work\, we will explore the effect of possible incomplete knowledge of neighborhood structure. Moreover\, we will investigate how by using small perturbations\, networks can in fact be made fully d-k-anonymous.  \n\n\n\nReferences\n\n\n\nLazer\, D.\, et al. (2020). Computational social science: Obstacles and opportunities. Science 369.6507: 1060-1062.van der Laan\, J. and E.\, de Jonge (2017). Producing official statistics from network data. In Proceedings of the 6th International Conference on Complex Networks and Their Applications\, pp. 288-289.Romanini\, D.\, Lehmann\, S. & Kivelä\, M. (2021). Privacy and uniqueness of neighborhoods in social networks. Scientific Reports 11: 20104.Hay\, M.\,  Miklau G.\, Jensen\, D.\, Towsley D.\, Weis P. (2008). Resisting Structural Reidentification in Anonymized Social Networks. In Proceedings of the VLDB Endowment\, 1.1\, pp. 102-114.Jérôme Kunegis (2013). KONECT – The Koblenz Network Collection. In Proceedings of the International Conference on World Wide Web Companion\, pp. 1343–1350. Ryan A. Rossi and Nesreen K. Ahmed. (2015). The Network Data Repository with Interactive Graph Analytics and Visualization. In AAAI Conference on Artificial Intelligence\, pp. 4292-4293.Sapiezynski\, P.\, Stopczynski\, A.\, Lassen\, D. D. & Lehmann\, S. (2019). Interaction data from the Copenhagen networks study. Scientific Data 6.1: 315.\n\n\n\n   Figure 1. Fraction of unique nodes in artificial network models. Top: Erdős–Rényi (ER)\, bottom: Watts Strogatz (WS). Size: 100-20\,000 nodes. Average degree 2-100. Distance d from left to right: 1\, 2\, 5.\n\n\n\nFigure 2. Fraction of unique nodes in real-world networks [5\, 6\, 7]\, for different values of distance d.
URL:https://www.popnet.io/events/anonymity-of-multi-hop-neighborhoods-in-social-networks/
CATEGORIES:Conference talk
ATTACH;FMTTYPE=image/png:https://www.popnet.io/wp-content/uploads/2021/07/ic2s2.png
END:VEVENT
END:VCALENDAR