Egenis seminar with Dr Omri Tal: 'Lewontin's Fallacy Revisited: Keep it simple, but not too simple'
SpeakersDr Omri Tal, LSE
Egenis,Byrne House,University of Exeter,St Germans Road,Exeter, EX4 4PJ
Room no: GF7
Time: 3:30 - 5.00pm
In a highly-cited paper the Cambridge genetic statistician AWF Edwards has provided a clear explanation of the effect of additional markers on classification accuracy, by use of a very simple model (Edwards, 2003). This model formed part of his critique of Lewontin's (1972) inferences from analysis of variation across populations. Edwards' key message was that, (a) while within-population variation is indeed greater than between-population variation, as demonstrated by differentiation measures in human populations, (b) this has only little bearing on the potential for classification of individuals to their origin population, since differentiation and diversity measures are effectively single-locus while classification power improves with additional loci. Edwards had famously referred to the argument from analysis of variation as "Lewontin's Fallacy", stressing that "there is nothing wrong with Lewontin's statistical analysis of variation, only with the belief that it is relevant to classification."
My talk will focus on the merits and pitfalls of Edwards' oversimplified model: while the model may lead to a correct general conclusion and provide some insight, its underlying assumptions may be too superficial to reflect real state of nature. I also show why the explanation for the effect of multiple loci is misguided (and has confused both philosophers and biologists). The potential for classification arises not due to 'correlation of allele frequencies across loci' as is often claimed, following Edwards, but from themultidimensional nature of genetic data.