Fifty stations over 20 months
West Oakland opened as the prototype station on December 28, 2023. BART used the pilot to test the gate design before starting the broader rollout at Fruitvale in August 2024. Installations then moved across the system in waves. Milpitas was completed last on August 27, 2025, four months ahead of BART's original deadline.
Those different completion dates matter. Each station needs its own before-and-after window.
Loading project fact
Stations completed by month
Counts use BART's published completion dates. Months with no completed stations remain visible.
West Oakland was the pilot station
BART opened the prototype there in December 2023 and tested three barrier materials before selecting the final design. The broader station rollout began at Fruitvale in August 2024. That pilot period explains the seven months with no station completions.
Loading installation dates
Source: BART station completion dates.
Reported incidents before and after installation
For each station with enough data, we compared 26 full weeks before installation with 26 full weeks after. We excluded two weeks on either side of the switchover. Full weeks keep weekdays and weekends balanced across both periods.
Calculating matched periods
Source: BART Police daily activity reports collected by Transit Crime Data Platform. Analysis uses station-specific installation dates from BART. Methodology and data checks.
Does the result change with a different time window?
We repeated the same comparison using 13 weeks, 26 weeks, and 52 weeks before and after installation. If the direction changes sharply across windows, the headline result is less stable.
Calculating sensitivity checks.
Source: The same BART Police daily activity reports and rollout dates used in the primary comparison. Only the window length changes. Methodology and data checks.
Results vary by station
A systemwide result does not describe every station. Downtown hubs, airport stations, and suburban park-and-rides differ in ridership, surrounding activity, policing, and installation date. Choose a station to see its own before-and-after result and compare it with the rest of BART during the same dates.
What these station counts represent
The counts include only incidents that appeared in BART Police daily activity report emails. Even a major station can have a low count here because this is not a complete crime or calls-for-service database.
Loading station comparison
Source: BART Police daily activity reports. The selected station is compared with its own prior period and with the rest of BART over the same dates. Methodology and data checks.
What the result means
Loading the systemwide result.
The comparison cannot isolate the effect of the gates
The rollout happened while BART was also increasing its visible safety presence, cleaning, completing its Fleet of the Future transition, and introducing Tap and Ride. Ridership, patrol patterns, seasonality, and reporting practices may also affect incident counts.
BART publishes station-level ridership reports, but this analysis does not yet incorporate them. It therefore cannot estimate risk per rider. Missing reports also cannot always be separated from days with no incidents.
Method details
- Equal 26-week periods with the same weekday mix
- A separate installation date for each station
- Two transition weeks excluded on each side
- An approximate 95% uncertainty interval
- A limited-data label for sparse results
- A same-period rest-of-system comparison for individual stations
Sources and method
Methodology and data quality
This is a descriptive before-and-after analysis, not a causal evaluation. The steps below define exactly which records are counted and how the comparison is calculated.
Data universe and cleaning
The source is BART Police daily activity reports received by email. The parser extracts one record per reported incident, normalizes dates, assigns a canonical station ID, and removes duplicate case-time records.
Station matching checks verified aliases first. Fuzzy matches are accepted only when the best score is at least 0.88 and exceeds the next candidate by at least 0.08. Ambiguous labels are withheld for review instead of guessed.
These emails contain incidents published by BART Police and are not a complete inventory of crime, calls for service, or safety events at each station.
Station comparisons include only records with a canonical station ID inside the reviewed coverage period. Dates outside that period remain stored but are excluded from the analysis.
Comparison design
Each station is aligned to its own published installation date. The primary result counts reports during 26 complete weeks before and after installation, excluding 14 transition days on each side.
The pooled rate divides reports by station-days. A station is excluded when either side of its requested window falls outside reviewed data coverage.
This is not a difference-in-differences or event-study design. Stations that had not yet received gates are not used as a formal control group.
Uncertainty model
The 95% range is an approximate Poisson rate-ratio interval. A 0.5 continuity correction is used when a period has zero reports. Results with fewer than 20 combined reports are labeled limited data.
The model does not adjust for overdispersion, repeated station measurements, concurrent programs, ridership, or multiple station-level comparisons.
Exact count rule
Count incidents where:
agency = selected agency
station_id = selected station
datetime >= period start
datetime < period endSystemwide results sum those station-specific counts and station-day exposures. Dates without stored incidents contribute zero incidents. They may be legitimate zero-incident days, and the source feed does not separately label them as missing.
The operational data audit remains available through the public audit API and repository tooling. Review the source code and audit implementation.