If you want to influence the behaviour of a group of individuals, whom should you target to influence the group as a whole? This topic is considered in one of the most interesting papers to be published this year. It is based on the idea of social networks.
The paper reports a cluster randomised study to test different ways of targeting individuals in a group, so as to influence the whole group. The groups were villages in Honduras. The behaviours that the investigators wanted to influence were use of chlorine for water purification and use of micro-nutrients – good ideas in those villages where diarrhoeal diseases are an important cause of death and stunting of child growth. The plan was to select (target) individuals who might then influence others to adopt the desired behaviour – e.g. to purify drinking water before use.
Clearly, if the plan is to target a limited number of individuals in order to influence the whole group to which they belong, then consideration needs to be given to how to select the group that will have the greatest influence in prospect. So the intervention here was three different methods to select the optimum mix of target individuals to produce the maximum effect in terms of behaviour change overall. In each case, the target group was set to 5% of the group – in this case 5% of the villagers. In the control villages, the 5% consisted of random members of the village. The first experimental group was based on a 5% target group selected on the basis of a full mathematical analysis of the connectedness of all villagers. A complex algorithm was used to identify the 5% sample with the greatest collective ‘reach’ across the cluster. Obviously this is a poor method of selection since the need to interview all participants vitiates the point of taking a (5%) sample. The second experimental group was therefore based on a sampling strategy that might approximate the formal network analysis methodology without needing to interview everyone in the group. Sampling for this third group was based on the friendship paradox – the observation that the friends of a randomly selected individual are more widely connected on average, than the randomly selected individuals themselves – colloquially, “your friends have more friends than you do”.
So what did they find? The control group showed an increase in the desired behaviour, but to a lesser extent than the other two methods. ‘Full-fat’ (so called ‘in-degree’) targeting was expected to be the optimal (most effective) method, but was intermediate. The experimental ‘friendship method’ yielded the best result. Since it is scalable, this method is recommended. CLAHRC WM will be considering the implications of this novel, intellectually intriguing study for its work programme.
— Richard Lilford, CLAHRC WM Director
- Kim DA, Hwong AR, Stafford D, et al. Social network targeting to maximise population behaviour change: a cluster randomised controlled trial. Lancet. 2015; 386: 145-53.