Spatial patterns of close relationships across the lifespan Hang-Hyun Jo Dept. of Physics, Pohang University of Science and Technology, Republic of Korea Dept. of Computer Science, Aalto University School of Science, Finland

H.-H.J. (johanghyun@ postech.ac.kr)

crucial for understanding the complex dynamics of close relationships and their effect on the migration patterns of human individuals.

uman societies have successfully been described in the framework of social networks based on dyadic relationships1. In recent years, a number of social networks have been characterised in terms of smallworld properties2, broad distributions of the number of neighbours3, assortative mixing4 or homophily5, and community structure6. This is partially due to recent access to large-scale highly-resolved digital datasets on human dynamics and social interaction7. Mobile phone datasets in particular have provided a unique opportunity to study the structure and dynamics of human relationships8–13. Although the lack of detail about individuals often undermines the importance of large-scale studies based on anonymised datasets, limited geographic and demographic information of mobile phone users has successfully been used, e.g., in studies of age and sex biases in social relationships14,15. It is important to stress that humans are embedded not only in social networks but also in geographic OPEN space 16,17 . People move and migrate for a number of reasons. For example, people leave their parental home for education or employment, to get married, to rear a family, or they can move due to divorce and separation18. In these life-course events, close relationships play a crucial role in shaping migrational patterns. OPEN SUBJECT AREAS: Human mobility patterns have recently been studied by using the datasets of time-resolved location of 19–22 SCIENTIFIC DATA mobile phone users These ¨datasets limited to1,3periods of Kaski a few1,3,4,5 years at most, so far allowing only Hang-Hyun Jo1,2, Jari. Sarama ki1, Robinare I. M. Dunbar & Kimmo COMPLEX NETWORKS cross-sectional analysis. In contrast, the longitudinal approach adopted by social scientists has been used to SUBJECT AREAS: long-term migration patterns over the human lifespan223,24, but suffers from the fact that sample COMPUTATIONAL SCIENCE investigate 1 BECS, Aalto University School of Science, Box 12200, FI-00076, BK21plus Physics Division and Department of 1,2 SCIENTIFIC DATA Jolimited. , Jari Sarama Robin I. M. Dunbar1,3 Finland, & Kimmo Kaski1,3,4,5 ¨ ki1, P.O. size Hang-Hyun is invariably 3 Physics, Pohang University of Science and Technology, Pohang 790-784, Republic of Korea, Department of Experimental COMPLEX NETWORKS Large-scale mobile phone datasets canRoad, be used toOX1 understand role Complexity of close relationships in the life-course Psychology, University of Oxford, South Parks Oxford 3UD, UK, 4the CABDyN Centre, Saı¨d Business School, Received migration, andOX1 demographic information mobile phone Department users. Frequency Universityby of exploiting Oxford, Park geographic End Street, Oxford 1HP, UK, 5Center for ComplexofNetwork Research, of Physics,of contact COMPUTATIONAL SCIENCE 1 2 BECS, Aalto University School of Science, P.O. Box 12200, FI-00076, Finland, BK21plus Physics Division and Department of 21 July 2014 between Northeastern Boston,has MA been 02115, USA. a pair University, of individuals established as a reliable index of emotional 3closeness in relationships, and Physics, Pohang University of Science and Technology, Pohang 790-784, Republic of Korea, Department of Experimental the frequency contact telephone andRoad, otherOxford digitalOX1 media andComplexity text message is known to correlate 4 Accepted Psychology, of University of by Oxford, South Parks 3UD,like UK, email CABDyN Centre, Saı¨d Business School, 25–27 5 with the frequency of face-to-face . Thus can assume that mostDepartment of important relationReceived 13 October 2014 significantly University of Oxford, Park End Street, Oxford OX1 contact 1HP, UK, Center for one Complex Research, of Physics, The dynamics of close relationships is important for understanding theNetwork migration patterns of individual 21 July 2014 shipsNortheastern of individuals are Boston, captured by mobile phone communication records, andlimited that the of size, emotional University, MA 02115, USA. life-courses. The bottom-up approach to this subject by social scientists has been by level sample Published closeness relationship is reflected in the strength of communication. Wefrom also assume theabout life-course while in theamore recent top-down approach using large-scale datasets suffers a lack ofthat detail 11 November 2014 Accepted the human individuals. We incorporate the sex-dependent geographic and geographic demographic information millions of words, migration patterns are reflected in the age- and correlations of of users. In other 13 October 2014 dynamics of close relationships important understanding thecross-sectional, migration patterns of use individual mobile phone users with their communication patterns to study the dynamics of close relationships andthem to evenThe though our mobile phone dataset of is users of all agefor groups is necessarily we can The bottom-up approachmigration to subject byhow social been limited dyadic by sample size, its effect in their life-courselife-course migration. Wethis demonstrate thescientists close age-has and sex-biased Published gainlife-courses. insights into longitudinal, patterns. relationships are recent correlated with theapproach geographic proximity of thedatasets pair of individuals, young and while the more top-down using large-scale suffers frome.g., a lack of couples detail about 11Correspondence November 2014 tendhuman to live further from each other than old In addition, we find that information emotionally closer pairs of the individuals. We incorporate thecouples. geographic and demographic of millions requests for materials are living geographically to each other. These findings implythe that the life-course is and mobile phone users withcloser their communication patterns to study dynamics of closeframework relationships should be addressed to SCIENTIFIC REPORTS | 4 : 6988 |its DOI: 10.1038/srep06988 1 crucial forinunderstanding the complex dynamics of close relationships and their on the migration effect their life-course migration. We demonstrate how the close age-effect and sex-biased dyadic H.-H.J. (johanghyun@ patterns of human individuals. relationships are correlated with the geographic proximity of the pair of individuals, e.g., young couples Correspondence and postech.ac.kr) tend to live further from each other than old couples. In addition, we find that emotionally closer pairs requests for materials

H

Spatial patterns of close relationships across the lifespan Spatial patterns of close relationships

across the lifespan

www.dashboardinsight.com

home.web.cern.ch www.stcorp.no

Gartner’s 2014 hype cycle

Mobile phone data at Aalto •

Source: A European operator

•

January 1 to July 31, 2007

•

1.9 billion calls among 33 million mobile phone users

•

Sex, age, and location for 5.1 million company users

Why mobile phone data? •

Mobile phones carried by people almost always

•

Almost 100% of coverage in many countries

•

Good proxy of the real social networks

A

100 10 1

B

Onnela et al., PNAS (2007)

2 | A sample of local network between best friends. A part of the network with a gender and age correlations. Blue circles correspond to m es to female subscribers. Circle sizes reflect subscriber ages: the bigger the circle, the older the subscriber. Grey circles correspond to sub Palchykov et al., Sci. Rep. (2012) ender and age information is not available in our data set.

Do your close friends live around you? Jo et al., Sci. Rep. (2014)

A few assumptions •

Call frequency ~ emotional closeness [Saramäki et al., PNAS (2014)]

•

•

Top-ranked alters (BFs) ~ family members •

Age difference ≤ 10 years: spouses or partners

•

Age difference > 10 years: parents or kids

Age- and sex-dependent geographic correlations of users ➞ Life-course migration patterns

switch their focal interest from their partners to their children as

show the age distributions of top-ranked alters

ive with their parents until leaving home in their 20’s; this is these become adult; however this would only affect the geographic n the local minimum at around mid-40’s as well. For older difference indices for female egos old enough to have grown-up dren have left home and live elsewhere, increasing the geo- children. For younger women, it may relate to the need to have a ifference index with ego’s age. close female friend in whim to confide; women are more likely to ographic difference indices between egos and their 2nd- have a same-sex intimate friend than men are. For ego-alter pairs lters behave roughly similarly to the top-ranked alters with age difference . 10 years, ego and alter sexes do not matter, d)). The main difference is that the indices peak somewhat except for female egos in their 30’s–40’s whose 2nd-ranked alters are hermore, for age difference # 10 years, the F:M and M:F slightly more often located in the same municipality, compared to longer overlap, but the 2nd-ranked alters of females live corresponding male egos. This effect might be because of motherore often in the same municipality with the egos. There are child relationships, considering the age difference. ossible reasons for this behaviour and no definite conclube drawn from demographic data alone. For example, this Correlations with alter age. These geographic correlations can be r might arise from the partners of female egos being ranked more clearly understood by considering age correlations between often than the partners of male egos. In particular women highly-ranked alters and their geographic difference indices. We F25partners = 25-year-old the age distributions of top-ranked alters and their eir focal interest from their to their children female as show egos

Age distributions of BFs

one generation one generation

Figure 2 | Geographic difference indices of top-ranked alters who live in a different municipa 25, 40, and 60 (top to bottom), and of females (left) and males (right). The fraction for each age trends. Error bars show the confidence interval with significance level a 5 0.05. In each panel, lo of top-ranked alters, which are magnified 2.5 times for clearer visualization. Vertical black lin SCIENTIFIC REPORTS | 4 : 6988 | DOI: 10.1038/srep06988

Proximities of BFs settling down with BFs

www.nature.com/scientificreports grown-up kids moving away

farther from ego BFs moving for college or job independent of sex! closer to ego

age diff. ≤ 10 yrs. spouses? ego’s sex : BF’s sex

age diff. > 10 yrs. parents and kids?

Proximities of 2nd BFs farther from ego

closer to ego

Figure 1 | Geographic difference indices of top-ranked and 2nd-ranked alters who live in a different municipality to ego (top and bottom, respectively). age diff. ≤less 10than yrs. age diff. > than 10 10 yrs. The left panels are for ego-alter pairs with age difference 10 years, while the age difference is larger years in the right panels. We use the notational convention that, for example,spouses? ‘‘M:F’’ denotes male ego and female alter. Error bars showand the confidence parents kids?interval with significance level a 5 0.05.

ages of both egos and alters, and divide ego-alter pairs into two Results ego’s sex : BF’s sex We analyse a large-scale mobile phone call dataset from a single categories where the age difference between egos and alters is # 10 mobile service provider in a European country8,9. This dataset spans years and . 10 years. This division is motivated by the observation the first 7 months of 2007 and contains around 1.9 billion calls that the age distribution of top-ranked alters is bimodal, with one

in Fig. 2. We use the notational convention that, for example, ‘‘F25’’ denotes the group of 25-year-old female egos. For the F25 and M25 groups (18675 and 21018 pairs, respectively), the curves for geographic difference indices are overall high and relatively flat, showing little correlation between demographic and geographic information. For 40-year-old females (F40, 8550 pairs) (Fig. 2(c)), the male top-ranked alters are on average slightly older than the egos, and tend to live close to egos; it is reasonable to expect these to be their partners. In contrast, female top-ranked alters of similar age live more often in another municipality. The age distribution of topranked alters shows a small peak at around 20 years of age, more pronounced for female egos. As these alters commonly live in the same municipality, this peak can likely be attributed to the children of egos. For the M40 (10933 pairs), the overall pattern is fairly similar. For the F60 and M60 groups (4360 and 5115 pairs, respectively), the data display a lot of variation because of smaller samples. However,

sexes are more likely to call and get called by alters who live further away than either 40-year-olds or 60-year-olds. This presumably reflects the fact that once married, egos are more likely to focus their attention on people who live closer, irrespective of whether these are other family members or friends. We have repeated all the above analyses by calculating the geographic difference indices based on provinces instead of municipalities, and with distance thresholds of 10 km and 50 km. The overall patterns are not affected by the choice of geographic difference index (see Figs. 5, 6, and 7).

Emotional vs. geographical Discussion We have analysed a large-scale mobile phone dataset to study the geographic correlations of emotionally close human relationships in the life-course framework. For this, we have made a number of assumptions: (i) Important relationships of mobile phone users are

farther from ego

closer to ego

Figure 4 | Geographic difference indices of alters who live in a different municipality to ego for several demographic groups of egos: F25 and M25 (a), F40 and M40 (b), and F60 and M60 (c). The rank of an alter is determined in terms of the number of calls to/from the ego. Error bars show the confidence interval with significance level a 5 0.05. SCIENTIFIC REPORTS | 4 : 6988 | DOI: 10.1038/srep06988

Out of sight, out of mind?

4

Summary •

Life-course framework for understanding spatial patterns of close relationships

•

Young couples live further to each other than old couples.

•

Emotional closeness ~ geographical closeness

•

Thank you! [email: [email protected]]