Lies, Damn Lies, and Medical Research

According to a recent finding published in the New England journal of Medicine and reported by Reuters, eating more chocolate increases the likelihood of winning a Nobel Prize. With the amount that I consume, I expect to be receive mine in the near future.

I’m not holding my breath however, since the study, at least as reported, has made some obvious errors in their use of statistical methods and the interpretation of the results. (Though perhaps it was intended to be facetious) The author of the study plotted the number of Nobel prizes per capita against consumption of chocolate per capita. What he found was that they appeared to form a line. A simple regression showed that the relationship was positive with a probability of error (that there was no relationship) equal to 1/10,000. Compelling evidence to be sure, until one delves into the world of statistical analysis.

The underlying theory behind the relationship between chocolate and Nobel Prizes has to do with the effects of flavonoids (whatever they are) on cognitive abilities. The more chocolate (or wine) consumed, the higher is cognitive function and this increases the probability of being awarded the Nobel Prize. There is no indication, however, that the Nobel Prize winners ever consumed chocolate. Nor did the study consider those that did consume chocolate and did not win the Nobel Prize. Forrest Gump comes to mind.

Without doing any analysis of my own, I suspect that chocolate consumption per capita is correlated with Nobel Prize winners per capita, but there is no causal relationship. There is likely a causal link between chocolate consumption and income, as chocolate is a normal good. This could be confirmed by finding, or determining the income elasticity of demand for chocolate. Alternatively, one could regress chocolate per capita against GDP per capita (PPP estimates) and the GINI index – to control for income distributions. The coefficient on income should be positive and I suspect the coefficient on the GINI Index to be negative if significant.

Next, run a regression of average education or literacy rates as a proxy for education, against income per capita; again, checking the effect of income distribution. Since education is a normal good, the coefficient on income should be positive. This shows correlation, not causation. Higher income leads to higher spending on education and higher education leads to higher income.

Finally, regress average education or literacy rates against Nobel Prize winners. More education leads to more research. More research leads to more Nobel Prizes.

The relationship observed between chocolate consumption and Nobel Prizes likely has very little, if anything to do with chocolate consumption. There is a causal relationship between income and both education and chocolate consumption. Thus, there is a correlation between chocolate consumption and education and therefore, between chocolate and Nobel Prizes.

Consider this an open invitation for any reader to undertake the proposed research. Perhaps the New England Journal of Medicine will publish it.