Transit Simulator
Downtown San Francisco’s vitality is strongly tied to regional mobility. As a major hub served by a wide range of regional transit operators, activity in the district tends to move closely with the systems that connect it to the broader Bay Area.
A Data-Driven Foundation
The Downtown SF Partnership maintains the transit dashboard to provide a data-driven foundation for advocacy and urban planning. By quantifying how regional mobility trends align with downtown visitation, we can better support transit investments and policy discussions that strengthen the urban core. Visitation here is measured using Placer.ai’s estimated foot traffic, which represents aggregated device-based visit counts within the Downtown SF Partnership district boundary. This metric includes employees, residents, and visitors and reflects total visit activity rather than unique individuals.
Explore the Historical Associations
To provide a more detailed view of the relationship between transit and visitation, we developed the Downtown Visits: Transit Sensitivity Simulator. This tool models historical associations, allowing users to explore how changes in transit ridership corresponded with changes in downtown foot traffic.
Use the simulator below to adjust BART, Ferry, and Bridge activity and see the model-implied change in downtown visits based on the historical elasticities estimated in the analysis.
Important Note on Interpretation:
These results represent associational elasticities derived from historical data. They describe patterns of alignment and do not imply a direct causal relationship or a guaranteed forecast of future visitation.
Because visitation is shaped by many forces (employment patterns, tourism, major events, and broader economic shifts), the model should be interpreted as a measure of historical co-movement, rather than an isolated transit effect.
How-to-interpret:
If BART ridership rises 10% in a given month, Ferry ridership rises by 2%, and Bridge Crossings rise by 4%, the model estimates that the historical relationships suggest visits tend to rise by approximately 3.5%, holding seasonal patterns constant.
Considerations
- MUNI ridership was excluded from the model because it lacked independent explanatory power. Most of its variation is explained by other transit variables, thus it was dropped by the LASSO regression.
- These are associational elasticities; not causal, based on historical data from 2020 to 2025.
- Sliders stop at −99% (not −100%) because the model is log-based and requires positive values (a full −100% would imply zero ridership/crossings, which is outside the model’s valid range).
The Value of Modeling and Data Monitoring
Recovery Benchmarking
Our transit dashboard uses 2019 levels to benchmark the district’s transit recovery. The simulator specifically highlights the modes that provided the strongest statistical signals in our model.
Regional Connectivity
Modeling these historical trends further highlights downtown’s role as a central destination and a vital link within the greater Bay Area transportation network.
Downtown Accessibility
Transit brings people into the district, while walkable streets and destinations may influence how long visitors stay. The close alignment between ridership and visits underscores the importance of maintaining an accessible urban core.
Methodology: Robust Variable Selection and Validation
The simulator is based on monthly ridership and visitation data from 2020 through 2025. Transit ridership is used here as an indicator of broader regional movement into and through downtown.
For data collection, this analysis combines publicly available regional mobility data with downtown visitation estimates from Placer.ai, restricted to the Downtown SF Partnership district boundary.
Key inputs include:
Monthly ridership data from BART, limited to the two downtown core stations: Embarcadero and Montgomery.
The average daily boardings by route and month from SFMTA, restricted to lines operating within the district, were converted by weekday and weekend counts to monthly totals for the purpose of this analysis.
Monthly ridership statistics from regional ferry operators, including: SF Bay Ferry and Golden Gate Ferry
Monthly traffic volumes from major regional bridges, including: Bay Bridge and Golden Gate Bridge
Data Analysis
Downtown visits were treated as the dependent variable, with transit ridership and bridge crossings modeled as explanatory variables. To account for recurring seasonal patterns in travel and visitation (e.g., summer peaks and winter declines), we included month indicator variables (“monthly dummies”) using January as the reference month.
We applied LASSO regression to identify which transit modes provided the strongest independent explanatory power. This approach prioritizes variables that contribute a unique signal while shrinking or excluding those that are highly redundant. All primary variables were modeled in log form, allowing coefficients to be interpreted as elasticities (percent change relationships). Model specification was selected using cross-validation, and results were evaluated on held-out chronological test data to confirm stability.
Under this process, MUNI ridership was excluded from the final model. This does not suggest MUNI is unimportant; rather, its month-to-month variation closely overlaps with other regional mobility measures and does not provide an independent signal once BART, ferry ridership, and bridge crossings are included.
To further test robustness and reduce the risk that results simply reflect post-pandemic recovery trends, we estimated an additional specification using log differences (month-over-month growth rates). This formulation shifts the analysis from long-term recovery levels to short-term fluctuations, isolating how changes in transit activity align with changes in visitation.
Across both the levels-based regularized regression model and the growth-rate specification, results were consistent: BART, ferry ridership, and bridge crossings continued to show positive historical alignment with downtown visitation. The out-of-sample performance of the growth-rate model remained strong (test R² ≈ 0.72), reinforcing that these relationships persist beyond the initial recovery period and are not driven solely by shared upward trends.
Model Validation: Predicted vs. Actual Changes in Visits
The figure below compares predicted and observed month-over-month changes in downtown visitation during the holdout period, using estimates from the log-difference specification.
Overall, the predicted series closely tracks the observed direction and timing of monthly fluctuations. While some months show deviations, reflecting factors not captured in the model such as major events, weather disruptions, or sudden shifts in employment activity, the alignment across the holdout period indicates that changes in transit ridership and bridge crossings provide a meaningful explanatory signal for short-term variation in downtown visitation.
This validation supports the central finding of the analysis: even after accounting for seasonality and removing long-run recovery trends, month-to-month changes in regional mobility remain positively and consistently associated with changes in downtown visitation.
Interested in learning more about the Downtown SF Partnership's Data Dashboards and Economic Development initiatives?
Built by Annoushqa Bobde,
Research and Data Analyst,
annoushqa@downtownsf.org