Figure 2.

To what extent can variation in human microsatellite length be predicted by demographically induced variation in heterozygosity and allele length skew?. Separate general linear models were fitted to data from each of 783 microsatellites genotyped in 53 worldwide populations, with mean allele length as the response and distance to Africa, distance from Africa squared, skew in allele length and log(modern population size) plus all second order interactions as predictor variables. Distance from Africa is taken as the land-only distance from Addis Ababa in kilometres [22,23]. Data are from [29]. Each model was simplified by backward elimination to achieve the minimum adequate model, MAM. The significance of the three primary terms (skew, log population size, distance + distance²) were then estimated by dropping each in turn and using ANOVA to compare the resulting MAM with the MAM produced when all terms were fitted. Figure 2a: number of models achieving a given level of significance for the full models (black bars) and for dropping skew (grey bars), population size (white bars) or distance + distance² (striped bars). Values on the X axis refer to lower bin boundary; i.e. '1' indicates non-significant models with P > 0.05, '0.05' indicates models with P-values lying between 0.05 and 0.01. Figure 2b: number of models explaining a given proportion of the null deviance. Colour coding of the bars is the same as in Figure 2a but X axis values are upper bin boundaries