A Power Comparison of Generalized Additive Models and the Spatial Scan Statistic in a Case-Control Setting

• Activity Number: 82
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Robin Young*+ and Janice Weinberg and Veronica Vieira and Al Ozonoff and Thomas Webster
• Companies: Boston University and Boston University and Boston University and Boston University School of Public Health and Boston University
• Keywords: Generalized Additive Models (GAMs) ; Permutation Test ; Spatial Scan Statistic ; Simulated Data ; Power

Abstract

Generalized additive models (GAMs) are advantaged over generalized linear models by allowing inference about associations between outcomes and predictors without parametric restrictions. The predictor of interest can be smoothed with a locally weighted regression and the degree of smoothing can be determined by minimizing the Akaike Information Criterion. GAMs can be applied in spatial statistics with a bivariate smoother accounting for geographic location as a predictor of disease status. A natural hypothesis is whether location is associated with disease, i.e. is the smoothing term necessary? Permutation tests are reasonable testing methods and provide adequate power under a simple alternative hypothesis but have yet to be compared to other spatial statistics. This research uses synthetic data to compare power and sensitivity of GAM permutation tests and the spatial scan statistic.