An op-ed from Paul Steinberg in the New York Times a few days ago discusses the fact that families of soldiers who have committed suicide do not receive presidential condolence letters. The concern is that actions taken to commemorate or honor people who commit suicide will encourage more suicides. Steinberg writes

“The hard truth is that any possible glorification of suicide — even reports of suicide — make the taking of one’s life a more viable option. If suicide appears to be a more reasonable way of handling life’s stresses than seeking help, then suicide rates increase.”

The theory that the publicity of suicides can cause more suicides is known as “the Werther effect”, after Johann Wolfgang von Goethe’s The Sufferings of Young Werther. According to Robert Cialidini’s book Influence, when the Goethe’s novel about the suicide of Werther was published, it set off a wave of copycat suicides.

Evidence for the theory was first offered by David Phillips in 1974 (gated version of the paper here). He found that suicides dramatically increase after a story about a suicide makes the front page New York Times. His estimate was that after each front page story, 58 more people kill themselves than otherwise would have. The idea is that finding out that someone else has killed themself gives people considering suicide the impetus to do it themselves. He later found that suicides also increased the number of deaths from car accidents. The theory here is that drivers are wrecking their cars as a way to kill themselves.

This theory is believable enough. And it makes sense to take such things into consideration when deciding how to commemorate and honor soldiers who have taken the own lives. Where things become slightly unbelievable is another of Phillips findings; that commercial airplane crashes increase after publicity of suicides. My instinct would be to interpret this statistical result the exact opposite way that Phillips did; I would take it as calling into question the validity of the original model, and thus the original result that publicity of suicides cause more suicides.

A common approach economists take to testing an empirical model sensible is to test whether they can find a relationship by plugging in something similar to the dependent variable they are investigating, but that couldn’t possibly be affected by the explanatory variables at hand. If you find an effect when it is not plausible, then something may be wrong with the model.  For instance, say someone has what they think is a good exogenous proxy measure for Wal-Mart store openings, and they find it has a negative effect on retail employment. A good test of this would be to see if the instrument effects manufacturing employment, which, like retail employment, would be related with overall economic growth but would not plausibly effected by a Wal-Mart opening (This exact study was done in this paper by Emek Basker). If you find your proxy variable for Wal-Mart openings has a strong effect on manufacturing employment, then something is probably wrong with the model or the proxy measure, and your previous results are called into question.

So if I found that publicized suicides were causing more suicides, and more car accidents, as Phillips does, I would test it against something similar that it wouldn’t likely be affecting, like commercial airplane crashes, or even better, boat crashes. If I found a positive effect, I would question my model’s specification, not assume that I had found some new effect of suicides.

The reason I find it so unbelievable that suicides cause commercial plane crashes is the (what seems to me) lack of any specific plane crashes that can be blamed on a pilot’s suicide. If this was occurring so frequently, then surely there must be specific cases where an after-the-fact investigation cited suicide as a probable or even possible cause for the plane crash. In one of his papers, Phillips calls for interviews with family members of people who have recently committed suicide to see if the story holds up. This kind of evidence may well exist for plane crashes, and if I did I would reconsider the hypothesis. But a result like this can’t stand on the statistical evidence alone.

According to Jane Pirkis, of the University of Melbourne, over 100 studies have been conducted to examine the impact of media reporting on suicides, and that the evidence overwhelmingly supports the causal relationship. I am curious if any of these look at commercial plane crashes, and if that relationship is robust. Robert Cialdini certainly finds the evidence persuasive. It seems like the question of whether or not publicized suicides cause plane crashes is something we would like to have an answer to, especially when considering policies regarding how honor those who have killed themselves.  In addition, if the result is sound, it seems we should be a lot more critical of the media for sensationalizing suicides.

It may well be that this result is long settled and the hypothesis rejected, and that I am simply unaware of it. If so, Robert Cialdini is incorrect. If this is not the case, then I think awareness of this potential problem is far too low.