Reducing MAUP bias of correlation statistics between water quality and GI illness

Abstract This research investigates the role of spatial aggregation and the modifiable area unit problem (MAUP) on the correlation between drinking water quality and gastrointestinal (GI) illness. Using water quality estimates from hydraulic modeling of a water distribution system and a linear dose–response model, we simulate illness point patterns with a theoretically determined correlation to average pathogen concentrations. We then assess the sensitivity of the Pearson’s correlation statistic ( r ) to different aggregation units. Because public health data are often geocoded illness events, we assess the importance of their network-clustered structure by comparing two spatial scenarios. The first scenario uses a random spatial distribution for illness point patterns. Randomly located points are then compared to network-clustered illness event patterns where the set of possible illness locations is limited to network nodes. We then analyze multiple illness simulations with various sets of commonly used areal units, such as census units, regular grids, and Voronoi tessellations. A systematic bias on r due to the MAUP is estimated by showing an average reduction in r of 0.65. Consideration of the spatial network constraint of illness data during aggregation reduces this MAUP bias estimate 41% from 0.65 to 0.38.