贝叶斯方法在犯罪时空格局研究上的应用——以长春市为例

对比分析了2008年和2018年长春市犯罪空间格局的变化特征，进而构建了贝叶斯时空分析模型，整合了犯罪时空格局演化中的固定效应、空间随机效应和时间随机效应，基于R环境中的INLA程序包对模型的各个参数进行了拟合，结合GIS制图，识别出异于总体趋势的犯罪相对风险高值区，并进一步解析了犯罪格局形成和演化的过程和规律。研究发现，犯罪总量在10 a间显著下降，犯罪数量较高的警区数量明显减少。长春市周边地区犯罪率有所提高，而城市中心区域的多数警区则明显下降。贝叶斯时空模型表明，虽然城市犯罪相对风险的平均水平较低，但其总体上却呈现出显著的增加趋势。空间效应的高值区主要集中在城市中心核心区域，特别是传统的商业网点或经济活动较为集中的警区。时间效应的高值区主要集中在城市外围地区，尤其是国家级开发区所在的警区。综合空间效应和时间效应，城市中心区域存在既是空间效应的高值区也是时间效应高值区的时空共同高风险区。贝叶斯方法在数据整合、区域异质性识别以及灵活性方面具有明显的优势，对于犯罪时空格局形成和演化规律的理解和把握上均有所助益。

Bayesian Modeling for Analyzing Spatial and Temporal Pattern of Crimes: A Case Study in Changchun,China

This study firstly compares and analyzes the changes in terms of the spatial and temporal patterns of crimes in Changchun between 2008 and 2018. A Bayesian spatio-temporal model integrating the fixed effects, the spatial random effects and the temporal random effects is built and fitted using the R-INLA package in R program. Combining with the maps of different effects of the model which are made used the software of ArcGIS 10.5, we identify the areas with higher relative risks of crime different from the general level or trend and further analyze the process and spatio-temporal patterns of crimes in Changchun. The study shows that the total amount of crimes displayed an obvious decreasing trend and the numbers of police precincts with higher crimes declined apparently. The crime rates showed an obvious increasing trend in the peripheral areas of the city while it went down dramatically in the inner city during the two years. The results of the Bayesian spatio-temporal model shows that the average trend of relative risk of crimes shows a significant increase in spite of the lower basic average relative risk of crimes. The police precincts with higher spatial effects were predominantly concentrated in the core area of the inner city. Especially, the highest spatial effects were mostly located in the precincts with those major traditional commercial area or the intensive economic activities. The higher differential time effects were mainly located in the peripheral areas. Particularly, those precincts where the three major national development zones located possess the higher relative risks represented by the differential time effects. After integrating the spatial and differential time effects, several higher relative risks regions are identified through both spatial and temporal effects within the inner city, deserving more attentions in the work of crime control and countermeasures in the future. Bayesian method show obvious advantages on data integration, area-specific heterogeneity identification and degree of flexibility and can provide more insights into the formation and evolution of the crime pattern.