Spatio-Temporal Adaptive Fused Lasso for Proportion Data

Population-corrected rates are often used in statistical documents that show the features of a municipality. In addition, it is important to determine changes of the features over time, and for this purpose, data collection is continually carried out by census. In the present study, we propose a method for analyzing the spatio-temporal effects on rates by adaptive fused lasso. For estimation, the coordinate descent algorithm, which is known to have better estimation accuracy and speed than the algorithm used in genlasso in the R software package, is used for optimization. Based on the results of the real data analysis for the crime rates in the Kinki region of Japan in 1995-2008, the proposed method can be applied to spatio-temporal proportion data analysis.