Spatial autoregressive models for scan statistic

Spatial scan statistics are well-known methods for cluster detection and widely used in epidemiology medical studies detecting evaluating the statistical significance of disease hotspots. For sake simplicity, classical spatial statistic assumes that observations outcome variable different locations independent, while practice data may exhibit a correlation. In this article, we use autoregressive (SAR) models to account correlation parametric/non-parametric statistic. Firstly, parameter is estimated SAR model transform into new independent over all locations. Secondly, propose an adapted based on detection. A simulation study highlights better performance proposed than one presence data. The latter shows sharp increase Type I error false-positive rate but also decreases true-positive when increases. Besides, our retain have stable true false positive rates with respect illustrated using economic dataset median income Paris city. application, show taking leads identification more concentrated clusters those identified by