Scholarlink

<p>This study systematically examines how spatial resolution, operationalized as grid cell size, influences the predictive performance of Risk Terrain Modeling (RTM) across 32 Swedish urban areas with varying demographic and criminogenic characteristics. Using performance metrics such as the Predictive Accuracy Index (PAI), Predictive Efficiency Index (PEI*), and the F1-score, the analysis explores how methodological factors (prediction thresholds), demographic and geographic conditions (population density, study area size), and contextual factors (social vulnerability, crime density) mediate or moderate the effect of cell size on model performance. Results show that spatial resolution exerts a strong effect on predictive performance, though the direction of this effect varies by performance metric. Smaller grid cells yield higher PAI values, reflecting greater localized accuracy, but perform poorly on PEI* and F1, indicating limited coverage of crime events. Larger grid cells, by contrast, improve PEI* and F1-scores by capturing more crime incidents, but at the cost of spatial specificity. Taken together, the study demonstrates that no single grid resolution is universally optimal. Instead, the effectiveness and fairness of RTM depend on balancing accuracy, efficiency, and coverage in relation to local conditions and intended applications. These findings provide methodological guidance for crime forecasting and contribute to the broader discussion of spatial scale and The Modifiable Areal Unit Problem in predictive policing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scaling Risk: Evaluating the Impact of Spatial Resolution on RTM Predictive Performance in 32 Swedish Urban Areas

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月27日

查看全文
  • Mia Puur,
  • Maria Camacho Doyle,
  • Nicklas Guldåker,
  • Manne Gerell

摘要

This study systematically examines how spatial resolution, operationalized as grid cell size, influences the predictive performance of Risk Terrain Modeling (RTM) across 32 Swedish urban areas with varying demographic and criminogenic characteristics. Using performance metrics such as the Predictive Accuracy Index (PAI), Predictive Efficiency Index (PEI*), and the F1-score, the analysis explores how methodological factors (prediction thresholds), demographic and geographic conditions (population density, study area size), and contextual factors (social vulnerability, crime density) mediate or moderate the effect of cell size on model performance. Results show that spatial resolution exerts a strong effect on predictive performance, though the direction of this effect varies by performance metric. Smaller grid cells yield higher PAI values, reflecting greater localized accuracy, but perform poorly on PEI* and F1, indicating limited coverage of crime events. Larger grid cells, by contrast, improve PEI* and F1-scores by capturing more crime incidents, but at the cost of spatial specificity. Taken together, the study demonstrates that no single grid resolution is universally optimal. Instead, the effectiveness and fairness of RTM depend on balancing accuracy, efficiency, and coverage in relation to local conditions and intended applications. These findings provide methodological guidance for crime forecasting and contribute to the broader discussion of spatial scale and The Modifiable Areal Unit Problem in predictive policing.