User:Aadamecz/sandbox

Several results from network science research have shown recently that patterns of human behavior may spread through social networks, or, in other words, human behavior is affected by the structure of the social network in which the individual is embedded. This pattern has been verified for example regarding smoking, alcohol consumption, and drug use.

The prevalence of obesity has been exploding worldwide. The exploration of the driving factors of the obesity epidemic is still far from being completed, but all sources agree that the main cause behind obesity is unhealthy lifestyle, ie. lack of physical activity and high calorie intake. Assuming that to some extent obesity is the result of individual choices and behaviors, Christakis and Fowler (2007) show empirical evidence that obesity is spreading in social networks.

The spread of obesity in a large social network over 32 years

The authors use a longitudinal survey covering 32 years, 5,209 individuals and their 38,611 social ties from the Framingham Heart Study. Study participants were surveyed every 3 years so the dynamic network database consists of 7 observations of the participants' networks. There is information available about their BMI (Body Mass Index), first order relatives and at least one friend of theirs in each observation period. Figure 1. shows how this network evolved and how obesity spread during these three decades.

Figure 1. []

Source: http://www.nejm.org/doi/full/10.1056/NEJMsa066082#t=articleTop

As they argue, three possible mechanisms may be present:


 * 1) the network structure affects obesity, so if there are obese people in your social environment you are more likely to eventually get obese;


 * 1) there is something unobservable going on in your social network that makes you and your friends obese at the same time;


 * 1) obesity affects your network, so if you are obese, you are more likely to connect with obese people. This is a type of homophily, and it is based on the fact that you prefer those people who are similar to you.

It can be clearly seen from the data that the network ended up containing separate clusters of obese and non-obese people. (see Figure 2.) The research question is: are you more likely to become obese if you have obese connections? Is the network structure driving the spread of obesity? The empirical analysis conducted by the authors shows that people having obese contacts in their social environment are more likely to gain weight than those who do not have obese friends. Moreover, this result is true up to 3 degrees distance, so even the friend of a friend of your friend can have an affect on your weight. Some other factors possibly affecting obesity either had lower (i.e. relatives) or no effect at all (neighbours).

Figure 2. []

Source: http://www.nejm.org/doi/full/10.1056/NEJMsa066082#t=article

The influence and critiques of the paper

The paper of Christakis and Fowler has been highly influential and it also induced a huge dispute. On the one hand, this paper and some other works of the authors on the spread of smoking, happiness etc. addressed whole new ideas about how individuals behave in a society using the tools of a quickly emerging academic field, network science. On the other hand, several critiques arose regarding the statistical methodology of the paper. The basic problem, which is in some form there in every single applied non-experimental econometrics work is that the effect may run in both directions, and the cause and the effect cannot be separated from each other. This is a form of endogeneity, thus the identification fails. Social science is usually interested in causal relationships, and for identifying causality exogenous variation of the variable of interest is needed. If we are interested in the causal effect of the network structure on the prevalence of obesity, we have to make sure that the network structure is exogenously changing through individuals and through time. Neither obesity can have an effect on the network nor can some unobserved characteristics drive both obesity and the network at the same time. Christakis and Fowler (2007) developed a way to conduct causal identification in their estimation strategy but others have questioned their methodology. Shalizi and Thomas (2011) have concluded that it is not possible to separate the effects of the network on obesity and the effects of obesity on the network from each other.

See also

Social Contagion Theory

References