In the summer of 2023, the SFMTA launched an 18-month pilot program to test a center-running bike lane along Valencia Street. This redesign replaced the street’s traditional side-running bike lanes—located between parked cars and moving traffic—with a single, bidirectional bike lane positioned in the center of the road. Northbound and southbound bicycle traffic shared this lane, separated by a painted white line, while vehicle traffic ran on either side, divided from the bike lane by rubber curbs and flexible bollards.
The project aimed to improve safety for cyclists while preserving both parking and vehicle through-traffic. The center-running design was expected to eliminate several common hazards: cyclists being “doored” by people exiting parked cars, drivers making right turns across bike lanes, and vehicles merging out from curbside parking. The addition of physical barriers—bollards and rubber curbs—also offered a level of protection that the previous side-running lanes lacked entirely.
This analysis seeks to evaluate the impact of the Valencia center-running bike lane by addressing the following questions:
This analysis draws on multiple data sources to assess changes in bike safety and ridership during the Valencia center-running bike lane pilot.
Traffic Crash Data
Comprehensive data on traffic collisions was obtained from the SF Gov “Traffic Crashes Resulting in Injury” dataset. This dataset includes police-reported crashes citywide involving injuries to motorists, pedestrians, or cyclists. It provides details such as crash type, location, date/time, and parties involved—making it essential for identifying trends in bicycle-related crashes before, during, and after the pilot.
Bike Volume Data (SFMTA Counters)
Ridership trends were analyzed using the SFMTA’s automated bike counter data. These counters track the number of bicycles passing fixed locations throughout the city, including Valencia Street.
Bay Wheels Trip Data
Additional insights into cyclist activity were drawn from the Bay Wheels system data, which provides anonymized trip-level records from the city’s bike-share program. This dataset includes start and end locations, timestamps, and bike types, helping capture short-term shifts in ridership that might not be reflected in fixed counters alone.
The traffic crash dataset only includes crashes reported by SFPD. There are likely more bicycle related crashes that went unreported so it should be assumed that this analysis overestimates actual bicycle safety. The dataset includes traffic crashes reported between January 1st, 2005 to May 31st, 2025. Crashes are restricted to San Francisco city limits.
Due to the significant size of the baywheels ride dataset, only rides that start or end on Valencia Street are included, with the assumption that these rides traveled via the center bike lane. Additionally only data from the following years is included: [2018, 2019, 2020, 2021, 2022, 2023, 2024, 2025].
Bicycle crash data was extracted from the full crash dataset by identifying records where a bicyclist was listed among the involved party types. Crashes specific to Valencia Street were indexed by matching the primary road name to “Valencia Street.” To isolate crashes occurring within the center-running bike lane corridor, spatial filtering was applied to include only those with latitude coordinates between 15th and 23rd Streets.
Although the center-running bike lane was in place from August 1, 2023, to February 1, 2025, this analysis primarily uses crash data from calendar year 2024. Data from August to December 2023 was excluded to avoid potential bias introduced by road users adapting to the new street configuration. By January 2024, it is assumed that most users were familiar with the layout, making crash patterns more representative of typical conditions. Center lane data from 2025 was also excluded due to temporary road obstructions and construction-related disruptions during the lane’s removal. To account for the small sample size of center lane data (just 2024), the analysis emphasizes population-weighted and trend-based metrics, rather than raw counts.
Before diving into the center bike lane impact, it's useful to contextualize bicycle crash metrics on Valencia street against the rest of San Francisco. Several visualizations are presented to explore crash types, crash severity, driver behavior preceeding a crash, party involvement, and time of day between bicycle crashes on Valencia and those citywide.
Crashes on Valencia Street are similar to the rest of the city in terms of severity and road users implicated (although during the time span analyzed there have been no fatal bike-related accidents reported on Valencia). However, Valencia has a significantly higher fraction of crashes invovling parked cars -- indicative of a cyclicst getting "doored" -- as compared to the top categories of crashes prevelant city-wide. Valencia also has a much higher percentage of crashes stemming from a vehicle making a left turn.
Valencia also shows a disproportionately higher concentration of crashes during the evening rush hour (5–8 p.m.) versus the morning rush hour. Peak morning crashes reach only 30.6% of the peak evening volume. This contrasts with citywide trends, where morning rush hour crashes reach approximately 77.8% of peak evening levels.
Valencia Morning Peak vs Evening Peak: 30.61% All bike crashes Morning Peak vs Evening Peak: 77.81%
Analyzing the spatial and temporal distribution of bicycle crashes on Valencia Street helps place individual months and locations in the context of broader patterns. This perspective is useful for identifying trends as well as flagging potential outliers or anomalies.
Over the sampled time period, the average number of crashes per month on Valencia Street was 1.44, with a variance of 1.14. The maximum number of crashes observed in a single month was six, which occurred only twice—an event with an estimated probability of just 0.82%, suggesting these were statistically rare occurrences.
Crash locations along Valencia Street were not evenly distributed. The blocks near 16th, 17th, and 18th Streets recorded the highest individual crash counts. Within this central segment of the corridor, crash locations were relatively balanced between intersections and midblock segments. However, certain areas stand out as having distinct patterns. For example, Brosnan Street stands out: 92% of crashes there occurred within 20 feet of the intersection, indicating a localized risk pattern.
| Statistic | Value |
|---|---|
| Total months with data | 245 |
| Mean crashes per month | 1.44 |
| Median crashes per month | 1.00 |
| Standard deviation | 1.29 |
The crash data on Valencia Street is divided into two periods: crashes that occurred during the center-running bike lane pilot in 2024, and crashes that occurred before the pilot when side-running lanes were present. To ensure consistency, both datasets are limited to the same geographic corridor between 15th and 23rd Streets. Absolute crash metrics are reported for each period. Notably, only 13 crashes were recorded during the center bike lane pilot, compared to 187 crashes outside the pilot period. As such, it’s important to recognize that the center lane data represents a much smaller sample size and is therefore more vulnerable to statistical noise.
This analysis addresses the investigation question: How did the types and severity of crashes under the center-running design compare to those under the previous side-running bike lanes? As expected, the center bike lane in 2024 had zero crashes resulting from vehicles making right turns into cyclists, and zero crashes from parked cars -- which infers no cyclists were "doored". Both of these crash mechanisms were highly prevelant with side running bike lanes. There was a relative increase in bike-on-bike and bike-on-pedestrian crashes during the center lane period, though this shift should be interpreted with caution due to the small sample size. Overall, crash severity remained broadly similar between the two periods.
In terms of location, Sycamore Street had the highest number of crashes during the 2024 pilot, followed by 16th Street and the segment spanning 17th to 20th Streets. The elevated crash count near 16th Street is consistent with pre-pilot patterns. However, the spike in crashes at Sycamore Street stands out as a potential anomaly: in 2024 alone, there were four crashes at or near Sycamore, compared to only seven crashes at that location over the previous 17.5 years.
Center lane crashes in 2024: 13 Valencia crashes before center lane: 186
| Date/Time | Type of Collision | Severity | Car Action | Bike Action |
|---|---|---|---|---|
| Sep 16, 2024 08:30 AM | Sideswipe | Injury (Complaint of Pain) | Making Left Turn | Proceeding Straight |
| Mar 17, 2024 10:23 PM | Head-On | Injury (Severe) | Proceeding Straight | Not Stated |
| Jun 28, 2024 07:25 PM | Broadside | Injury (Other Visible) | Making U Turn | Proceeding Straight |
| Sep 04, 2024 06:00 PM | Broadside | Injury (Complaint of Pain) | Making Left Turn | Proceeding Straight |
To evaluate the safety impact of the center-running bike lane, it’s useful to measure crashes relative to the number of bike trips—essentially, crashes per ride. This metric and accompanying analysis will aim to answer the investigation questions, did the center-running bike lane make cycling on Valencia Street safer? and, Did the number of cyclists using Valencia Street increase after the redesign?
Answering these questions requires reliable estimates of how many cyclists used the center lane during the pilot. The SFMTA operates several automated bike counters across the city, including two on Valencia Street. However, these sensors can be unreliable due to downtime and incomplete coverage. To supplement this, Bay Wheels trip data—specifically rides starting or ending on Valencia—was used as a proxy for broader cycling activity.
Bike counter data at Duboce–14th Street shows only a slight increase in traffic during the pilot as compared with the previous two years. In contrast, Bay Wheels trips to and from Valencia rose sharply during the center-running lane period: in 2024, rides increased by 25.6% compared with 2022. While this suggests greater cycling activity, the Bay Wheels dataset has limitations based on the assumptions stated previously, and furthermore, the increase could partly reflect growing use of the bike-share service itself rather than the new infrastructure. This analysis does not attempt to separate those effects.
BayWheels trip counts and SFMTA bike counter data were used separately to calculate a “crashes-per-10,000-rides” metric. While the absolute values are subject to the limitations and assumptions of each counting method, the relative change in this metric before vs. during the center bike lane can help infer the lane’s impact on bicycle safety. Low crash counts make the analysis sensitive to noise, which limits the statistical strength of monthly fluctuations. For this reason, the analysis includes on two perspectives: (1) peak monthly crash-per-ride values to indicate edge-case/extremes (2) yearly averages, which draw from a larger sample size to reduce statistical noise. The comparison spans 2018, 2019, 2022, and 2024.
During the center bike lane period (2024), peak crash-per-ride rates occurred in September and October, reaching roughly 1.5 and 2 crashes per 10,000 SFMTA counted rides and BayWheels trips respectively. These rates exceeded that of peak month in 2022 (0.86 and 1.5, both in November) but are comparable to peaks observed in 2018 and 2019.
When averaged across the pre-center-lane years (2018, 2019, and 2022), the crash-per-ride rate was 0.5415 crashes per 10,000 SFMTA counted rides and 0.8302 crashes per 10,000 BayWheels trips. In 2024, these rates dropped to 0.4476 and 0.5964 respectively. While the center lane dataset covers only one year, the yearly averages—being based on far more rides than any single month—provide a stronger, less noise-driven indicator of change. On this basis, the data suggests a meaningful improvement in bicycle safety from a per-ride perspective.
| Ride Count Source | Year | Total Rides | Total Crashes | Crashes per 10,000 Rides |
|---|---|---|---|---|
| Bay Wheels Valencia | 2018 | 123,781 | 15 | 1.212 |
| Bay Wheels Valencia | 2019 | 148,366 | 13 | 0.876 |
| Bay Wheels Valencia | 2022 | 173,532 | 9 | 0.519 |
| Bay Wheels Valencia | 2024 | 217,915 | 13 | 0.597 |
| SB Duboce-14th | 2018 | 200,850 | 15 | 0.747 |
| SB Duboce-14th | 2019 | 194,515 | 13 | 0.668 |
| SB Duboce-14th | 2022 | 287,862 | 9 | 0.313 |
| SB Duboce-14th | 2024 | 292,295 | 13 | 0.445 |
| Totem NB 16th-17th | 2018 | 496,113 | 15 | 0.302 |
| Totem NB 16th-17th | 2019 | 478,822 | 13 | 0.271 |
| Totem NB 16th-17th | 2022 | 354,240 | 9 | 0.254 |
| Totem NB 16th-17th | 2024 | N/A | N/A | N/A |
Note: 2024 data highlighted in gold represents the center bike lane period. Crash data covers the same geographic area (15th-23rd St) for all years. N/A indicates data not available for that period.
Did the center-running bike lane make cycling on Valencia Street safer?
How did the types and severity of crashes under the center-running design compare to those under the previous side-running bike lanes?
On a per-ride basis (accounting for the limitations of ride-count data), the center-running bike lane reduced the number of crashes—indicating a measurable improvement in cyclist safety. This reduction was largely driven by the elimination of certain high-risk scenarios: in 2024, there were no recorded crashes involving right turns into cyclists or “dooring” incidents. However, when crashes did occur, their severity mirrored that seen with the previous side-running bike lanes, meaning that the design did not improve injury outcomes.
Did the number of cyclists using Valencia Street increase after the redesign?
SFMTA’s bike counter just north of the lane’s starting point showed a slight increase in rides, while Bay Wheels data indicated a significant uptick in trips to and from Valencia. However, given the uncertainties and limitations of these counting methods, the data cannot definitively confirm an increase in overall bike lane usage.
One notable finding was the high frequency of crashes near Sycamore Street during the center-running lane’s operation. This location was a point where vehicles could cross the lane uninhibited—either by making U-turns or turning left from Sycamore onto Valencia. Of the four crashes recorded here, three could potentially have been prevented if the bike lane protection were extended through the intersection. Considering that only 13 crashes occurred in 2024, eliminating those three incidents would have significantly improved the safety metrics.
Valencia is a vibrant street for all road users, and the safety of pedestrians should also be considered when proposing and analyzing new road infrastructure projects. In a follow-on to this analysis, the impact of the center bike lane on pedestrain safety, as well as pedestrian safety on Valencia street in general will be analyzed.
Author: Carlos Sama
Notebook and datasets: https://github.com/los-sama/valencia_bike_lane