ObjectivesThe residential population of an area is an incomplete measure of the number of people that are momentarily present in the area, and of limited value as an indicator of exposure to the risk of crime. By accounting for the mobility of the population, measures of ambient population better reflect the momentary presence of people. They have therefore become an alternative indicator of exposure to the risk of crime. This study considers the heterogeneity of the ambient population by distinguishing residents, employees and visitors as different categories, and explores their differential impact on thefts, both on weekdays and weekends. Methods We analyze one-year of police recorded thefts across 2104 1 km 2 grid cells in a central area in Beijing, China. Controlling for the effects of attractiveness, accessibility, and guardianship, we estimate a series of negative binominal models to investigate the differential effects of the three groups (residents, employees and visitors) in the ambient population on crime frequencies, both on weekdays and during weekends and holidays. Results Overall, larger ambient populations imply larger theft frequencies. The effect of visitors is stronger than the effects of residents and employees. The effects of residents and employees vary over the course of the week. On weekdays, the presence of residents is more important, while the reverse holds true during weekends and holidays. Discussion The effects of ambient population on thefts vary by its composition in terms of social roles. The larger role of visitors is presumably because in addition to being potential victims, residents and employees may also exercise informal social control. In addition, they spend more time indoors than where risk of theft is lower, while visitors might spend more time outdoors and may also bring about greater anonymity and weaken informal social control.

No abstract

China is generally considered a safe place: among the safest for foreigners to visit. For local (longterm) residents and for Chinese criminology scholars, China as a country and its individual cities might be safer compared to many of their foreign peers; however, they might not be as safe as some travel agents claim. To show why this is the case, we crawled and geovisualized the 2015-2016 crime records (n ¼ 24,803) available to the public online. The geovisualization shows that the seven crimes (n ¼ 12,516) that were most likely to be influenced by space in Beijing were aggravated assault, blackmail, cheating and bluffing, dangerous driving, picking quarrels and provoking trouble, robbery and theft. It also shows a Beijing that many might not have known before.

To advance the interpretability of machine learning for long-term crime prediction in China, we compared the performance of multiple machine learning algorithms in predicting the spatial pattern of theft in Beijing. Gradient boosting decision tree emerged as the algorithm with best predictive accuracy. After identifying the importance of criminogenic features, we extended the interpreter SHAP to reveal nonlinear and spatially heterogeneous associations between environmental features and theft and we summarized six relation types of such associations at the global scale. At the local scale, we clustered six area types according to the contribution of environmental attributes to theft prediction in each grid. Policy makers should adopt place-based crime prevention measures based on the specific type of each grid belongs to.