以時空加權迴歸模型探索地域犯罪機制的時空變異性

Abstract

為了更有效地理解地域建成環境中犯罪影響因子與犯罪熱點的時空動態，本研究嘗試以多維度核密度估計與時空加權迴歸模型為工具，試圖在時間地理學的脈絡下討論利用此方法偵測犯罪熱點與機制移轉的可能性與實務上意義。 由於空間資料本身具有空間自相關與誤差異質性，直至今日地理加權迴歸模型已被驗證為處理空間資料變數不穩定性的有效取徑。然而地理加權僅能從時間剖面分析的本質，亦限制其探索多維度社會現象的可能性。從而本研究試圖發展其時間延伸模型，以探索時空資料之變數不穩定性。 基於過往研究證實住宅竊盜與環境因子具有深度的互動關係。本研究利用臺北市大安區1999至2008年之住宅竊盜空間對位資料，驗證該延伸模型之分析結果不僅足以在三維時空中探索犯罪群聚的轉移，更可以深入地檢視其誘發機制的動態過程，並為後續犯罪機制時空動態移轉之研究提供基礎。

For a more effective understanding of spatial-temporal dynamic of criminal factors and hotspots in local-scale built environment, this study employs multi-dimensional kernel density estimation (KDE) and extended weighted regression (STWR) to uncover future possibility of detecting the displacement of hotspots and factors in a context of time geography. Due to the spatial autocorrelation and heterogeneity of spatial data, geographically weighted regression (GWR) has been proven as a significant approach to address the spatial heterogeneity. However, the cross-sectional nature of GWR constrains it to explore the multi-dimensional phenomena simultaneously. Thus, this study develops a temporal variant of GWR to detect the spatial-temporal heterogeneity of structural measures in space-time cube. Using a geocoded database of residential burglary in Da-an district of Taipei City from 1999 to 2008, this study examine that proposed framework allowing interactively 3-D geovisualization of hotspots by volume rendering. This thesis also represents the spatial heterogeneity of estimations of social structural measures by spatial-temporal weighted regression. Emphasizing the supplementary aspect of this embedded framework, the author concludes that interactive spatial-temporal data analysis and weighted regression could extract and interpret the spatial-temporal heterogeneity of residential burglary as well as uncover the possibility of detecting criminal displacement in future study.