A distribution-free spatial scan statistic for marked point processes

A new spatial scan statistic is proposed for identifying clusters in marked point processes. Contrary to existing methods, it does not rely on a likelihood ratio and thus is completely distribution-free. It applies whatever the nature of the marks: binary, discrete or continuous. This spatial scan test seems to be very powerful against any arbitrarily-distributed cluster alternative. I apply this method first to a classical epidemiological dataset and then to the spatial distribution of incomes in France.