DRCOG analysts find widespread reporting gaps in regional crash data, urge clearer officer reporting and system fixes

Catherine Rush, senior safety planner at the Denver Regional Council of Governments (DRCOG), told attendees at a regional crash-data consortium meeting that analysts found significant data‑quality problems in crash records from 2020–2024 that hamper efforts to identify high‑priority corridors and design safety interventions.

Rush said routine cleaning of the Colorado crash dataset revealed repeated problems: geocoding errors that concentrated many crash points at a single location, widespread underreporting of speeding in the driver‑action field, missing narrative text in many serious crashes and inconsistent coding of the first harmful event or crash type.

"Only 7 percent of these severe crashes were marked as speeding related in the driver action field," Rush said, adding that targeted searches of narrative text turned up roughly another 6 percentage points of cases that referenced excessive speed. "Even with this additional cleaning, one could still think that speeding or speed is not a major cause of severe crashes. We know that's not the case," she said.

Rush outlined several consequences of the gaps. Without accurate latitude/longitude and corroborating location descriptions, analysts spend considerable time manually relocating crash points. Missing narratives (about one quarter of fatal and serious crashes in the dataset she examined) remove essential context for understanding how a crash occurred and whether countermeasures are warranted. Inconsistent first‑harmful‑event coding — for example, a single left‑turn collision coded variously as front‑to‑side or front‑to‑front — makes automated aggregation and mitigation prioritization less reliable.

To address these gaps, Rush recommended a mix of procedural and technical changes: encourage law‑enforcement agencies to include estimated vehicle speeds in narratives (via video or skid‑mark analysis when available), seek vehicle event data where justified, train officers on selecting a specific driver‑action (for example, "failed to stop at signal" for red‑light running) and require narratives for fatal and serious injury crashes. She also suggested adding illustrations to the crash‑reporting manual and adding validation logic to electronic reporting systems to reduce inconsistency and manual cleaning.

Attendees from law enforcement and agencies raised practical concerns. Cammy and Alyssa noted officers often cannot conclusively prove speed at the initial crash scene and that journal guidance and training shape what fields are filled in; Alyssa said the crash report is a snapshot and amendments are used to update details later. Jason Okers of the Boulder County Sheriff's Office said many deputies lack formal accident‑investigation training and the capacity to estimate speed accurately when crashes are not investigated by specialized traffic units.

Rush acknowledged those constraints and suggested incremental approaches such as emphasizing narrative content, offering templates that avoid personally identifying information, and building logic validations into records‑management systems or a statewide reporting form so that required fields are easier for officers to complete.

The consortium agreed to continue conversations around training and reporting guidance and to pursue technical fixes that would reduce manual cleaning and improve the utility of the crash dataset for engineering and enforcement decisions.

The consortium will revisit these topics at its next meeting; DRCOG staff offered to circulate examples and templates and to share more detailed breakout findings with agencies that asked for their agency's data.