SAS IML 軟體在地理加權廣義線性模式之應用

SAS IML Software for Computing Geographically Weighted Generalized Linear Models

劉智欣(Chin-Hsin Liu)

指導教授 ： 陳怡如

淡江大學/管理學院/統計學系碩士班/碩士(2012年)

https://doi.org/10.6846/TKU.2012.00796

摘要

由於地理資訊系統技術的進步, 以及其空間資料的取得容易, 使得空間計量經濟(Spatial Econometrics) 方法已廣泛的應用於各領域。其中, Nakaya et al.(2005, 2009)提出的地理加權廣義線性迴歸模式(Geographically Weighted Generalized Linear Regression model, GWGLM) 分析方法主要應用於研究空間異質性(Spatial Nonstationarity)的問題, 其模式的建立根據資料類型的差異, 有邏輯斯(Logistic) 、卜瓦松(Poisson)與常態(Normal) 三種選擇。隨著此地理加權廣義線性迴歸方法的備受重視, 研究人員期望使用一有效率、方便且快速的軟體或程式來執行空間異值性分析的需求亦與日遽增。近年來, 已有部分套裝軟體(GWR 軟體) 或程式(R package 、Stata 模組) 應運而生; 當這些工具被廣泛使用之此時, 卻未有相關SAS 統計程式的設計與開發, 直至Chen and Yang 於2012年提出一SAS 巨集程式(Sas Macro) 才有效的將其應用於地理加權廣義線性迴歸。為了提高資料分析的彈性與實用性, 本研究有別於Chen and Yang(2012) , 所提出的巨集程式, 利用SAS IML 軟體設計一套完整的使用者介面, 並將其應用於實際資料, 與以上提到的現行分析工具做比較。此介面除了可做到現有工具所提供之主要功能外, 亦額外增加了部分新功能或選擇, 以彌補現有分析技巧上的缺點與不足,進而使其他領域研究者(如社科人員、地理分析人員) 在分析空間資料時更加方便。此外, 本研究的提出也進一步拓展了SAS 的應用範圍, 讓後續有興趣的開發者, 能根據所提供的程式碼進行的推廣與改進。

關鍵字

空間異值性 ； 地理加權迴歸 ； 地理加權廣義線性模型 ； 拔靴法 ； SAS/IML Studio ； 地理加權廣義半參數模型

並列摘要

Due to the rapid development of techniques on Geographic Information System (GIS) and with the geographical database being accessible easily, spatial statistical analysis tools have been widely applied in empirical studies of various
disciplines. Of them, the Geographically Weighted Generalized Linear Model (GWGLM) introduced by Nataya et al. (2005, 2009) is mainly designed to explore
spatial nonstationarity of the data. The GWGLM fits ”local” regression models where the response variable could go beyond continuous measures and the error terms are allowed to follow nonnormal distributions such as binomial and poisson. As the GWGLM has received increasing attention in recent years, researchers demand user-friendly analysis programs to analyze georeferenced data.Such inquiries have generated several specialized software programs (e.g. GWR 4.0, packages in R, modules in Stata); none of them, however, can be integrated natively into SAS environment. Very recently, Chen and Yang (2012) proposed
a macro program to fill the gap. Their work are among the first to fuse strength of SAS into the GWGLM framework.
The objective of this study is to expand the work of Chen and Yang using the new SAS/IML (or SAS/IML Studio) programming software. We not only design and develop effective user interfaces to conduct GWGLM but also distinguish our work with great flexibilities in the modeling process and data analysis.We illustrate the capability of SAS/IML by applying the proposed programs to
several empirical datasets, and then demonstrate the advantages by comparing the GWGLM results with those obtained from other existing software programs.It is concluded that this software program provides researchers a relatively new,yet powerful and flexible, computational environment to investigate spatial nonstationarity in empirical studies. The programs can be also easily modified or restructured by SAS users for the purpose of developing advanced analysis tools related to GWGLM.

並列關鍵字

Spatial nonstationarity ； Geographically weighted regression ； Geographically weighted generalized linear regression model ； Bootstrap Method ； SAS/IML Studio ； S-GWGLM

參考文獻

[1] A.S. Fotheringham, C. Brunsdon, and M. Charlton (2002). Geographically Weighted Regression: The Analysis Of Spatially Varying Relationships. John Wiley & Sons Inc.

[3] P. Congdon (2003). Modelling Spatially Varying Impacts Of Socioeconomic Predictors On Mortality Outcomes. J Geogr Syst 5(2):161-184

[4] Daryl Pregibon (1981). Logistic Regression Diagnostics. The Annals of Statistics, 9(4):705-724

[5] T.J. Hastie, R.J. Tibshirani (1990).Generalized Additive Models. Chapman Hall : London.

[6] Zhang Huiguo, and Changlin Mei (2010). Local Least Absolute Deviation Estimation Of Spatially Varying Coefficient Models : Robust Geographically Weighted Regression Approaches International Journal of Geographical Information Science . 25(9):1467-1489