Social network analysis provides an insight on the complex influences that relationships between individuals may have on their health and social wellbeing throughout the lifecourse. ALSPAC has a unique data resource in this respect; a mapped friendship network of adolescents linked to detailed records of behavioural traits and health/social outcomes.This will allow the investigation of risk taking behaviour in adolescents within friendship 'communities' and how the clustering of these communities relate to the location of the friends residence and school. The investigation will be conducted as a 'Complexity Project' run by Uiniveristy of Bristol Computational Sciences Department. The complexity student will be supervised by Computational Science and School of Social and Community Medicine Staff.

This project has two components, a methodological one - to apply novel computational methods to analyse the ALSPAC friendship dataset, and an analytical one - to analyse the friendship data in this way to provide additional useful information to identify / target high risk groups within the population, which may otherwise not be obvious based only on individual risk profiles/standard epidemiological techniques.

Specifically the project aims to look at the following hypothesees:

Hypothesis 1: Can community detection algorithms (CDA), a novel computational method of network data analysis, be used to identify groups of closely connected individuals within a population?

Required Variables: CCXB friendship matrix data and friendship attribute variables.

Hypothesis 2: Within these groups of connected individuals is disease transmission (in this instance Chlamydia infection) more likely than between different "communities"? Note: community in this sense is defined purely by the computer algorithm rather than in the real world e.g. according to location of residence.

Required Variables: Chlamydia infection status from 17 year clinic.

An important element of the 'complexity study' project is to simplify complex data. In this project the ALSPAC participants propensity for risk taking behaviour (e.g. substance use, alcohol use, anti-social behaviour) will be categorised using statistical methods (e.g. principal component analysis) to define an ALSPAC risk profile. This profile can be used to look at "diffusion of behaviour", or preferential mixing of individuals based on behaviour - over and above sexual risk.

Hypothesis 3: Do the individuals within a "community" detected in this way share other characteristics such as risk behaviours?

Required Variables: Risk taking measures from the 15 and 17 year clinics and the 16 year questionnaire. Attainment data from Key Stage 4 (Linkage).

The risk profile data will be used to investigate the extent to which like with like mixing (homophily, e.g. if two individuals who smoke are more likely to become friends due to their shared smoking habit) drives the formation of friendship groups detected through CDAs. We will further use these data, together with computer simulation techniques to analyse whether interventions aimed at communities could be more effective than interventions at reducing disease transmission. The analysts will look at ways to visualise these data in a more compelling way than a list of densities/closeness measures using specialist network

ALSPAC Research Proposal Form page 7 of 8 December 2010

drawing software.

Hypothesis 4: Does the clustering of friendship communities within schools/geographical areas influences homophily of friendship networks.

Required Variables: Pseudonymised school codes and Lower Super Output Area of home residence.

This strand investigates the clustering of network community effects within the key georgaphical spaces for adolescents; schools and residential areas. We will consider if the geographical clustering appears to 'validate' the network communities and whether individuals with links outside of these shared geographical/network communities have differing risk profiles.

The team:

Katy Turner, Steve Gregory, John Macleod, Andy Boyd, Matthew Hickman, Rhiannon Pinney. Rhiannon is a Bristol Complexity Sciences student with Maths background, who will be supported by the rest of the team, Katy and Steve to be her main supervisors. 3 month project, leading to mini-dissertation write up, ideally peer review paper and potentially expansion/development to full PhD project proposal.