Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study adapts a fine-tuned LLM classification pipeline to analyze nearly 3,000 UK police incident logs for indicators of vulnerability, such as mental ill health and homelessness. The findings reveal that while LLMs can provide meaningful prevalence estimates, they are prone to biases and instability, necessitating significant human oversight and statistical correction to ensure defensible measurements. Ultimately, the research underscores the potential of LLMs in extracting insights from unstructured data, while also highlighting the challenges of reliability and accuracy in operational contexts.
LLMs can identify vulnerability indicators in police data, but their outputs are fraught with biases that require extensive human intervention to correct.
Purpose: Understanding how much of routine policing involves vulnerable people could inform resourcing, training, and multi-agency response, yet administrative data provide limited insight. We explore whether an LLM-based classification pipeline, developed on open-source US police data, can be adapted to estimate the prevalence of four vulnerability indicators - mental ill health, substance misuse, alcohol dependence, and homelessness - in UK police incident narratives, and when outputs can be treated as defensible measurements. Methods: We analyse nearly 3,000 de-identified incident logs from a UK police force, using a multi-stage pipeline combining repeated model inference, label aggregation, structured human review, and statistical correction. The pipeline runs on a locally hosted open-weight LLM, reflecting the secure environments police must work in. Results: LLMs can produce meaningful, if imperfect, prevalence estimates at scale. Mental ill health indicators are present in approximately one in five incidents, with lower prevalence for other indicators. However, naive LLM deployment is unreliable: single-pass classifications are unstable, and aggregated outputs systematically over-assign indicators relative to human judgement. Correcting these biases required substantial human input and statistical adjustment, leaving considerable uncertainty. Conclusions: While LLMs can extract information from unstructured police data, their outputs cannot be treated as valid measurements without careful methodological support. At the population level, defensible estimates are achievable but resource-intensive; at the individual level, errors remain frequent and unpredictable, limiting suitability for operational decisions. This study highlights both the potential and the constraints of LLM-based measurement in applied settings.