Abstract
A three-parameter lognormal model is used to estimate the city-size distribution of the world and of eight UN-defined macroregions. The model is found to fit the data better than the Pareto function, and to provide a powerful means of comparing distributions among regions. Although system concentration (measured by the standard deviation index) is relatively stable in Europe and in the world at large, it is decreasing in North America, Africa, and East Asia, and increasing in Latin America and South Asia. Cities in the 250000–500000 size class are somewhat more numerous than predicted, suggesting the possibility of some kind of ‘optimum’. The theory of extreme values is used to predict the most populous city of a region and to compare predictions with actual maxima, demonstrating that the largest cities in the world are well within systematic possibilities.

This publication has 26 references indexed in Scilit: