Fig. 2. GME of COST full-member countries as a function of area per capita GDP.
a,b, Generalized linear models for the GME of COST full-member countries, represented by international postal codes, as a function of area (a) and per capita GDP (b). The equations of the lines are shown, along with 95% confidence intervals in shading. The models were fit as negative binomial distributions with the log link function. Model fit is given as Veall-Zimmermann R2. Turkey is of substantially greater geographic extent than the displayed countries, but it has no documented GME and is omitted as an outlier and influential observation. Both the linear area term and the quadratic GDP term are significant in the multiple generalized model corrected for spatial autocorrelation (two-tailed tests; area: z = 2.269, P < 0.0233; GDP quadratic: z = −2.969, P = 0.00299; see the Methods for the details). A significant quadratic term remains upon the omission of any one of the three high-GDP countries.