Source-linked AI summary
The persistence of social signatures in human communication
J. Saramaki, E. A. Leicht, E. Lopez, S. G. B. Roberts, F. Reed-Tsochas, R. I. M. Dunbar
TL;DR
The paper asks how changes in ego-network membership affect the distribution of communication across relationships. It combines 18 months of mobile-phone call and survey data from students transitioning from school to university or work, finding that individually distinctive social signatures persist despite substantial alter turnover.
Problem
The paper addresses the limited understanding of how ego-network membership turnover affects the distribution of an individual’s network ties.
Method
The study combines mobile-phone call records and survey data to track ego networks and communication patterns during students’ transition from school to university or work.
Results
Social signatures remain stable and retain their characteristic shape over time, while being only weakly affected by network turnover.
Takeaways & Limitations
Individuals differ in how they allocate available communication time, with a small number of emotionally close, top-ranked alters receiving a disproportionately large share of calls.
Takeaways & Limitations
The study is constrained by limited data and focuses on participants’ transitions, offering only limited kinds of evidence about social circumstances.
Abstract
from arXiv · showhide
The social network maintained by a focal individual, or ego, is intrinsically dynamic and typically exhibits some turnover in membership over time as personal circumstances change. However, the consequences of such changes on the distribution of an ego's network ties are not well understood. Here we use a unique 18-month data set that combines mobile phone calls and survey data to track changes in the ego networks and communication patterns of students making the transition from school to university or work. Our analysis reveals that individuals display a distinctive and robust social signature, captured by how interactions are distributed across different alters. Notably, for a given ego, these social signatures tend to persist over time, despite considerable turnover in the identity of alters in the ego network. Thus as new network members are added, some old network members are either replaced or receive fewer calls, preserving the overall distribution of calls across network members. This is likely to reflect the consequences of finite resources such as the time available for communication, the cognitive and emotional effort required to sustain close relationships, and the ability to make emotional investments.
I. INTRODUCTION
Human ego networks face finite constraints on communication and relationship maintenance, yet it is unclear whether these constraints produce persistent patterns as network membership changes. Using an 18-month study combining call records and surveys of students transitioning from school to university or work, the paper identifies robust, individually distinctive social signatures that persist despite alter turnover.
- Finite time, cognitive capacity, and emotional effort constrain how many relationships people can maintain and how closely they can sustain them.
- The study asks whether such constraints shape ego networks similarly across circumstances, producing characteristic features that persist despite network turnover.
- Individuals show robust social signatures describing how communication is distributed across alters, with much communication focused on a small number of alters.
- The dataset tracks 24 students for 18 months during their transition from school to university or work, combining automatically recorded mobile phone calls with survey data.
- The call records provide complete time-stamped records of calls to alters, while questionnaires capture emotional closeness, face-to-face contact, and alters’ phone numbers.
- These signatures retain distinctive individual variation over time even as network members turn over, suggesting that new alters can replace or receive fewer calls than former members.
A. Emotional closeness of ego-identified relationships and calling behavior
The study tests whether call frequency reflects egos’ perceived relationship quality by combining phone records with repeated surveys of active personal networks. More calls are associated with greater emotional closeness and more frequent face-to-face contact, supporting call patterns as a proxy for relationship importance.
- The analysis combines call records with questionnaires completed at study entry, 9 months, and 18 months to assess relationship quality.
- Multilevel modelling accounts for the hierarchical structure in which alters are clustered within participants.
- The number of calls significantly predicts emotional closeness, with positive coefficients indicating that closer relationships receive more calls.
- More calls are associated with fewer days between face-to-face contacts, indicating that frequently called alters are also encountered more often in person.
- The results support phone call data as a reliable estimate of relationship importance to the ego.
B. Social signatures and turnover in close relationships
Across three consecutive six-month intervals, participants’ social signatures remain broadly stable even though their ego-network membership changes substantially. Communication remains concentrated among a few top-ranked alters, while turnover affects both membership and alter ranks, including close relationships.
- Measurement: The 18-month observation period is divided into three consecutive six-month intervals for tracking communication and network change.
- Measurement: Social signatures are constructed by counting calls to each alter, ranking alters by call count, and calculating each rank’s fraction of total calls.
- Social signatures: For almost all participants, signatures have a heavy tail, with communication concentrated among a few top-ranked alters.
- Persistence: Average signature shape remains stable over time despite substantial turnover in network membership.
- Turnover: Entire-network Jaccard indices are 0.22 ± 0.09 from I1 to I2 and 0.27 ± 0.09 from I2 to I3, indicating substantial turnover.
- Turnover: Only 42% of top-ranked alters retain their specific rank from I1 to I2, compared with 54% from I2 to I3.
C. Persistence of individual social signatures
Individual social signatures tend to persist across consecutive intervals: each ego’s signature is more similar to its own earlier signature than to other egos’ signatures, despite network turnover.
- Despite alters entering or leaving networks, signatures retain much of their shape and a substantial fraction of communication remains persistent.
- Jensen-Shannon divergence measures the distance, or shape difference, between social signatures.
- Self-distances are averaged across consecutive intervals, while reference distances compare each ego’s signature with those of other egos.
- Self-distances showed a moderate correlation with network turnover (r=-0.41, p=0.0034), yet most remained below reference distances.
- 82% ± 12% of each ego’s distances to other egos were greater than its self-distance.
- The average self-distance was 0.036 ± 0.014, compared with 0.086 ± 0.055 for other egos (p < 10^-4, Welch’s t-test).
- Turnover among only the top 5 or top 10 alters did not significantly correlate with self-distance variation.
III. DISCUSSION
The study identifies broad, robust social signatures: individuals differ in how they distribute communication across alters, and these characteristic patterns remain stable despite network turnover. The findings suggest that constrained time, cognitive capacity, and emotional investment limit how relationships are maintained and replaced.
- The authors identify a consistent, broad, and robust pattern in how people allocate communication across their social-network members.
- Individuals vary in their characteristic communication-allocation patterns, with a small number of emotionally close alters receiving a disproportionately large share of calls.
- Individual social signatures remain stable and retain their characteristic shape over time, only weakly affected by network turnover.
- Call frequency correlates with emotional closeness and face-to-face interaction frequency, suggesting call patterns can serve as proxies for relationship quality.
- The data cannot determine whether social signatures primarily reflect time constraints, cognitive constraints, or other mechanisms, requiring different kinds of data.
- Because communication resources are constrained, adding new high-intensity alters requires downgrading or dropping some existing alters.
A. Personal network survey and call records
The study followed 24 students who completed three network questionnaires and provided 18 months of mobile-call records during the transition from school to university or work.
- Transition setting: The sample comprised students aged 17–19 in their final year of secondary school, followed as they moved into work or university.Some stayed in the same city, while others moved elsewhere in England.
- Sample and timeline: Questionnaires were administered at the beginning of the study, after 9 months, and after 18 months.These corresponded to time points t1, t2, and t3.
- Sample and timeline: 24 participants completed all three questionnaires and used their mobile phones throughout the 18-month study.The completed sample comprised 12 males and 12 females.
- Call records: Electronic monthly invoices recorded each outgoing call’s recipient, time, and duration, and were combined with questionnaire data for analysis.Participants received study phones with 500 monthly voice minutes and unlimited text messages.
B. Constructing ego-centric call networks
Researchers combined survey-defined personal networks with phone invoices to construct ego-centric call networks and rank alters by their communication share.
- Network construction: Survey-generated lists of kin and friends or acquaintances were combined with electronic phone invoices to construct ego-centric call networks.Phone records were used independently of whether alters were recalled in surveys.
- Network construction: Calls to any listed mobile or fixed-line number were recorded as calls between the ego and that alter.Unlisted non-service numbers were treated as unique alters, while service numbers were filtered out.
- Temporal segmentation: The 18-month invoice record was divided into three consecutive 6-month intervals: I1, I2, and I3.The intervals were March–August, September–February, and March–August.
- Social signatures: For each ego and interval, researchers counted calls to every alter, ranked alters from most to least called, and calculated the fraction of calls at each rank.This rank-based distribution constituted the ego’s social signature.
C. Analyzing social signatures
Social signatures were compared across alters and time intervals using set overlap and Jensen–Shannon divergence applied to rank-based call distributions.
- Validation: The analysis quantified variation between the sets of alters called by an ego in two time intervals.Jaccard overlap provided an additional validation of pairwise comparisons.
- Set overlap: The Jaccard coefficient measured overlap between the sets of alters called in two time intervals.It equals 1 for identical sets and 0 when the sets share no alters.
- Distributional comparison: Jensen–Shannon divergence compared social signatures between egos or between two intervals for one ego.Each signature records the fraction of calls to the alter at rank r.
- Distributional comparison: The divergence combines the Shannon entropies of the two signatures and is zero exactly when their distributions are identical.JSD was selected over KLD because it handles zero probabilities.
- Distributional comparison: Because the maximum number of alters could differ between intervals, shorter rank distributions were zero-padded to the larger length.The maximum rank k is the total number of alters called.
1. The relationship between mobile phone calls and emotional closeness
Multilevel models tested whether phone-call frequency, duration, and alter rank related to emotional closeness and face-to-face contact, while additional analyses examined weighting choices and network turnover.
- Modeling: Multilevel models treated alters as nested within 24 egos and examined call number, call duration, and their rank-based versions.Call variables were log transformed to reduce outlier effects.
- Modeling: When random slopes prevented convergence, some models were fitted with random intercepts only.This occurred in models examining emotional closeness and face-to-face contact.
- Weighting robustness: Call-number and call-duration weights were highly correlated, and using duration weights did not qualitatively change the results.The main analyses used call counts to rank alters.
- Signature persistence: Across intervals, average self-distances were smaller than distances to other egos, indicating persistent individual signature patterns.Using call counts, average self-distance was 0.045 ± 0.023 versus 0.122 ± 0.094 for other egos, with p < 10^-4.
- Survey recall: Non-recalled alters made up 0.63 ± 0.21, 0.68 ± 0.18, and 0.70 ± 0.17 of alters across I1, I2, and I3, respectively.They were biased toward low-frequency or low-duration links.
- Network turnover: Only 42% of the highest-ranked alters retained their top rank between intervals, despite substantial turnover across ranks.Low-ranked alters were more likely to leave, while new alters generally entered at low ranks.