Carbon Mitigation, Causality vs. Correlation, El Nino, Global Warming Hiatus, Inflammatory Chain Reaction, Marshall Burke, Nature Journal, P-Value Problem, Rare Events, Suicide Rate, Suicides and Global Warming
The latest entry in the scare-mongering literature of global warming, published this month in Nature, purports to show that warming will lead to more suicides! I’m not sure whether these researchers deserve an award for naiveté or cynicism, but they should get one or the other. The lead author is listed as Marshall Burke of Stanford University.
The basic finding of their research is that an increase in average monthly temperature of one degree Celsius (1.8 degrees Fahrenheit) increases the monthly suicide rate by 0.68% (or 0.42% when accounting for the previous month’s temperature as well). They might have used the preferred verbiage “is associated with”, rather than “increases”, because they surely know that correlation is not the same as causation, but perhaps they wished to impress the news media. Let’s put their result in perspective: the annual U.S. suicide rate per 100,000 persons was 11.64 over the years 1968-2004 in their sample. A 0.68% increase in the suicide rate would have brought that up to roughly 11.72. Of course, the average U.S. surface temperature did NOT increase by 1 degree Celsius over that period — it was about half that, and temperatures have been relatively flat since then.
The real problem here is that most of the variation in temperatures across the sample used by Burke and his co-authors is seasonal and geographical. While they claim to have accounted for such confounding influences using non-parametric controls, they give few specifics, so I am unconvinced. It has long been known that suicides tend to be seasonal and are higher in the warmer months of the year. The reasons cited vary, including a boost provided by warmth in the energy needed to execute a suicide plan, “inflammatory chain reactions” from high pollen counts, seasonal peaks in bipolar disorder, and stress from greater social interactions during warm weather. These are seasonal phenomena that are not even incidental to the question at hand. And let’s face it: if warmer weather gives you the energy to kill yourself, the temperature is probably not the problem.
The authors also report a positive “effect” of temperatures on suicides using annual data, but with a rather large variance. This result probably captures geographical variation in suicide rates, though again, the authors claim to have made adjustments. Southern states tend have high suicide rates, but no one has suggested that warm, southern climates are to blame. Instead, there are other socioeconomic factors that probably account for this regional variation. I suspect that this is another source of the correlation the authors use to project forward as a likely impact of global warming. (While the inter-mountain West tends to have high suicide rates relative to other regions, many of those states are lightly populated, so they would receive low weights in any analysis of the kind discussed here.)
Finally, the trend toward slightly warmer temperatures between 1968 and the late 1990s was spurred largely by a series of strong El Nino events, especially in 1997-98. Suicide rates in the U.S., on the other hand, reached a high in the mid-1970s, ran slightly lower until hitting another peak in the mid-1980s, and then tapered through the late 1990s even as temperatures spiked. Since 1998, suicides have trended up as temperature trends flattened during the so-called “global warming hiatus”, which is ongoing. This sequence not only contradicts the authors narrative; it reinforces the fact that the variation exploited in the samples may well be seasonal and geographical, and not related to climate trends.
An issue over which Burke, et al demonstrate no awareness is the exaggerated statistical significance of meaningless effects in very large samples. This has been called the “p-value problem” because large samples can lead to vanishingly small p-values (which measure statistical significance). In a very large sample, any small difference may appear to be statistically significant. It’s a well-known pitfall in empirical work. A suicide is what’s known in the statistical literature as a “rare event”, given it’s annual incidence of about 0.01% of the population. I submit that the estimated impact of a 1% change in that rate, a change of 0.0001%, is well-nigh meaningless.
But the authors, undaunted, do their very best to make it seem meaningful. First, they pick a sub-sample that yields a somewhat higher estimated effect. Then they apply it to a future climate change scenario that is considered extreme and “extremely unlikely”, by climate researchers. They calculate the cumulative increase in suicides implied by that estimate out to 2050 — 32 years — for the U.S. and Mexico combined: about 22,000 extra suicides (they give a confidence interval of 9,000 to 39,000). That would be a lot, of course, but aggregating over many years using a high-end estimate and an extreme scenario can make an otherwise tiny effect appear large. And remember, their confidence interval is tightened considerably via the use of many observations on essentially irrelevant seasonal and geographic variation.
Burke and his co-authors have succeeded in publishing a piece of research that is not just flimsy, but that they apply in a way that is grossly misleading. They made it as ripe and plump as possible for promotion by the news media, which seems to love a great scare story. I might just as easily claim that as declines in income are associated with higher suicides, efforts at carbon mitigation requiring high taxes and punitive consumer rates for electric power will lead to an increase in suicides. And I could “prove” it with statistics. Then we would have a double-warming whammy! But I have a better idea: let’s expose bad research for what it is, and that includes just about all of the literature that warns of catastrophe from global warming.