Statistical inference and spatial patterns in correlates of IQ

“

This paper, co-authored with Tom Sherratt at Carleton University, received a bit of attention in the press (see "Resources" on the right). As pretty much everyone has offered their views on the paper based on the press coverage (which has been pretty good but not perfect…), this is my attempt to set the record straight. Why we did the study International IQ scores have been around for about a decade, beginning with a book called “IQ and the Wealth of Nations” by Richard Lynn and Tatu Vanhanen in 2002. They published a list of IQ scores that were recorded or estimated for the vast majority of the world’s countries. This data has been mined ever since to look for explanations for the variation (which is great) and this has resulted in a range of hypotheses with varying degrees of support. Of particularly interest to us was an study by ecologists at the University of New Mexico linking variation in IQ to the impact of infectious and parasitic diseases (Eppig et al., 2010). Upon reading Eppig et al’s study in a journal club, we were initially impressed with the strength of the relationship between disease measure and IQ. However, it quickly became clear that there was a problem in their analysis: they had not accounted for a phenomenon known as “spatial autocorrelation“. This occurs when you can know something about one datapoint based on its spatial proximity to another datapoint in your analysis. For example, in the analysis that Eppig et al. conducted, they included Benin, Togo and Ghana as well as the Netherlands, Belgium and Germany (among 120 other countries). Now, it is clear that the climate of each of the three African countries will be more similar within that region than to the climate of any of the three European countries. Here, then, we know that if we measure the climate of Togo then we will know approximately what the climate of Benin is, and likewise for Belgium and the Netherlands. This is a problem because virtually all statistical tests require that all of the datapoints be independent – in other words, you should not be able to guess at values for one datapoint based on values of another. However, what you see when you plot the data is that the African countries are grouped together, most of the European countries are grouped together, and most of the Asian and Middle Eastern countries group together. This is a clear demonstration of spatial autocorrelation. What we did No previous analysis had explicitly controlled for spatial autocorrelation in their analyses, despite the fact that a paper had already been published on “the geography of IQ” (Gelade, 2008) stating that this was a problem. We re-analysed all of Eppig et al.’s data, extracting their variables from the original sources, to test a series of six hypotheses: Temperature – this hypothesis states that humans evolved on the African savannah where temperatures are pleasant for humans to live. As we moved further from those temperatures, we needed to develop cognitive tools to produce clothing and shelter to survive harsher climates. Therefore, IQ is negatively correlated with temperature. Distance from the environment of evolutionary adaptedness (DEEA) – this hypothesis suggests that, like above, our cognition had evolved to deal with the African savannah, which they define as the environment of evolutionary adaptedness (EEA). In order to persist in other habitats, we developed enhanced cognitive tools, hence IQ is positively correlated with the distance from Africa. Literacy and/or education – some have suggested that performance on IQ tests is simply a function of the quality of teaching in a country. Hence, more teaching and reading leads to better performance on IQ tests. GDP – this is a general measure of development that lacks a definite proximate relationship with IQ. It can be interpreted as saying that higher GDP is associated with a greater proportion of the population out of manual labour and into jobs where they are better educated. Therefore IQ is positively correlated with GDP. Nutrition – Richard Lynn, author of the book that inspired all of these hypotheses, promoted a link between nutrition and IQ a few decades ago. He suggested that poor nutrition led to poor growth, including in the brain. Since brain size is (weakly) correlated to IQ and intelligence, there was a mechanism by which poor nutrition could affect IQ. Parasites – finally, Eppig et al. promoted their theory that parasites and infection diseases could influence IQ. Like Lyn’s nutrition hypothesis, they propose a mechanism by which the body is stressed to the point that it can no longer invest in the cognitive architecture that is necessary for high intelligence. What we found Analysing all of these potential hypotheses, we found very strong support for Eppig et al.’s parasite hypothesis, with some evidence still for an effect of temperature. This was not much of a surprise – the parasite hypothesis provides the most reasonable mechanism by which the environment could influence IQ. Not only this, but the general improvement in global health (however slow it is in some places) explains what is known as the Flynn Effect – the pattern of global increases in IQ over time. We also reframed the temperature hypothesis. Rather than interpreting it as an evolutionary hypothesis based on climatic similarity with the environmental of evolutionary adaptedness, Tom and I proposed that temperature may act through epidemiological channels to modulate the effect of disease and parasites on IQ. Summary We noticed a problem with statistical analyses in a previous study. Correcting for this problem allowed us to test more rigorously the findings of a previous analysis and propose our own interpretations for the patterns observed. As ecologists, we were in a better position to carry out these kinds of analyses, since geographical patterns are our bread-and-butter problems. As such the paper was written as a methodological note with emphasis on providing the psychology researchers with the tools to avoid these problems in the future. While we have demonstrated correlation, we cannot comment on causation apart from to say that a plausible mechanism exists by which disease could affect cognitive development. It is interesting to think that all the pseudo-racist research and commentary concerning IQ variation between races might be down to a few bugs and worms, though… _______________________________________________________________ References Eppig, C., Fincher, C.L. and Thornhill, R. (2010) Parasite prevalence and the worldwide distribution of cognitive ability, Proceedings of the Royal Society of London: Series B, 277: 3801-3808. Gelade, G.A. (2008) The geography of IQ, Intelligence, 36: 495-501. Hassall, C. and Sherratt, T.N. (2011) Statistical inference and spatial patterns in correlates of IQ, Intelligence, 39: 303-310.

”