A DOUBLE-SCAN STATISTIC FOR CLUSTERS OF 2 TYPES OF EVENTS

We develop a scan-type statistic to measure the unusualness of the clu stering of two types of events over time. The statistic allows for a l agged effect between the two types of events. We derive the expected n umber of nonoverlapping clumps of clusters under retrospective and pro spective chance models of no association. Results are derived and appr oaches are given to handle both uniform and more general distributions of events over time. We investigate the power of the statistic agains t an alternative where the observed data is a mixture of linked and un associated clusters. The statistic is applied to data on homicide/suic ide clusters over a 7-year period for several counties and several sex /race combinations.