Tuesday, March 16, 2010

Suicide and Social Learning

I found an interesting article on the Public Library of Science web site called The Cultural Dynamics of Copycat Suicide. (And thanks to the author for being willing to share his information under a Creative Commons license---this is how medical research should be!)

The author, Alex Mesoudi, used a computer model to study the effects of social learning and mass media influence on suicide clusters. He used a statistical method to see if suicides were clustered in time and space at an unexpectedly high frequency. This method is called an agent-based similiation, and is commonly used to model transmission of infectious disease. I'm not going to pretend to understand the statistics behind this! If you're curious you can read that part of the article.

He started by explaining the difference between point suicide clusters and mass clusters. Point clusters are suicides that are grouped together in time and space, while mass clusters are suicides that are grouped together in time, but are separated geographically. Suicide clusters have been thought to be due to social learning or mimicry, but it's also possible that they occur through homophiliy (the tendency for similar people to pool together in groups). Mass suicides are thought to be due to the influence of prestigious individuals (eg. celebrity suicide deaths) combined with coverage by the media. This leads to a one-to-many transmission model.

The computer model was run using three different assumptions: that suicides were totally random and unclustered, that clusters were due to social learning, and that clusters were due to homophily. He used different formulas to generate "suicides" under each model, and looked at the kind of clusters (spatiotemporal versus just temporal) that resulted.

What he found was that social learning caused spatiotemporal point clusters while homophilic clusters were more likely to be spacial rather than temporal. In order to understand this better, imagine the difference between teenage suicide epidemics versus correctional suicides. Teen suicides clusters happen among individuals who know each other, they happen in the same geographic area and within a short time frame of one another. These are the "social learning" clusters. Correctional suicides happen at a rate higher than in free society (in other words, they're geographically 'clustered' in a jail or prison), but are spread out over time. These are the homophilic suicides, in other words deaths by high risk people who happen to be grouped together. Based on this study, the correctional deaths are less likely to be due to social learning or mimicry.

Finally, the author studied the factors influencing mass suicide clusters: deaths that happen at the same time over a broad geographic area. These the kinds of suicides you see when a celebrity commits suicide. They are generally associated with a lot of media coverage. The computer model found that social learning played almost no role in these deaths.

It's a really complicated paper and I'm sure I didn't do it justice, but I thought it was pretty fascinating that someone could basically recreate the kind of suicide death patterns we see in the real world based on a theoretical mathematical framework. And I liked the term this author used for this kind of experiment: in silica. If in vivo experiments are done on animals or humans and in vitro experiments are done in test tubes or petri dishes, then "in silica" is a great term for computer model experiments.

And if none of this post made sense, hang on and I'll resurrect it in one of our podcasts.