AHSC Ridership Dashboard - Recalibrate Stop-Level Ridership Model with Expanded Observed Ridership Data and Enhanced Specifications

[shweta487]

Relevant PRs:
#1246
#1482
#1674
#1731

Question or Goal:
Update the existing stop-level ridership regression model used in the AHSC Ridership Prediction Dashboard by incorporating newly available observed ridership datasets (19 agencies as of 1/8/2026), and assess whether adding variables or modifying the model form improves predictive performance and robustness relative to the current log-linear specification.

Data Required:
Currently have:

• Preprocessed stop-level observed ridership datasets for 19 transit agencies in the transit-ridership-analytics repository. These datasets may include updated versions of LA Metro, Santa Barbara MTD, and Monterey-Salinas Transit, plus additional agencies
• Existing GTFS-Schedule-derived service variables (daily trips and routes by stop for weekday, Saturday, Sunday)
• ACS-derived demographic and socioeconomic variables aggregated to stop buffers
• LEHD-based job density variables aggregated to stop buffers
• Current model code and coefficients (log-linear OLS, Fall 2022 - updated in 2025)

Research Required:

• Integrate all 19 observed ridership datasets into a unified training dataset at the stop level
• Evaluate how expanded agency coverage affects coefficient estimates and model stability
• Test candidate enhancements to the model, such as:
  • Additional service variables derived from GTFS (e.g., service span, frequency distribution)
  • Alternative functional forms (e.g., updated log-linear variants or semi-log specifications)
• Document any changes to assumptions (e.g., exclusions, annualization, calibration approach)

Expected Outputs / Findings:

• An updated ridership regression model trained on expanded observed data (19 agencies)
• Quantitative comparison between the Fall 2022 baseline model and updated specifications
• Clear documentation summarizing methodology changes for downstream dashboard updates

[hhmckay]

Following up on a few items related to this:

• FTA STOPS documentation: Pages 7-13 provide a good overview of how the model works. We don't want to replicate this model, but it is useful to understand at a high-level.
• Direct Ridership Models (DRMs): This is broadly the type of modeling we are interested in exploring, in addition to the work that has already been done for the AHSC tool. This report provides a good example of one of these models in practice.
• Ridership Model Scoping Document: Putting my thoughts on what is needed from the model here. Feel free to add.

[shweta487]

SIMPLIFIED TRIPS-ON-PROJECT SOFTWARE VERSION 2.52 – 2.53
This tool is developed by FTA to estimate ridership for transit projects:

• Forecasts ridership for various transit project types, including fixed guideway starter lines, extensions, new lines added to existing systems and gap-filler projects.
• MPO models
• Nested logit mode choice model to split person trip tables into separate tables for different travel modes based on socioeconomic class and purpose.
• Assigns travel trips to specific transportation facility such as transit routes, stations or bus stops to determine total ridership.
• Also generates estimates of transportation system user benefits including travel time savings

Uses:
• Census data
• GTFS- Timetable information
• Demographic forecasts for MPOs

Modes
• Synthetic mode: Uses census Transportation Planning (CTPP) Journey to Work Flows. Applies mode choice model to estimate how many of these trips will use transit. Primarily used when detailed user surveys aren't available.
• Incremental mode: Used in regions where detailed transit user surveys are available to provide more precise travel information. Uses survey records to estimate existing transit trips and then applies the model to predict how changes in service (a new rail line) will incrementally affect ridership.
• Special Market type: That combines both of the modes.

Models used:
• Uses a nested logit model to determine which travel mode a person will choose.
- Top level : splits the total person trips to different travel and access modes like (transit vs auto and walk-to-transit vs drive-to-transit)
- Bottom level: Determines which specific transit path a rider will take (bus vs rail)
- Variables: Model considers travel time, cost, and socioeconomic factors like household automobile ownership)

• Path building model
- Actual GTFS data.
- Calculates travel times: precise in vehicle travel times, transfer times, wait times. For peak and off-peak
- Access Calculations: Uses either straight-line distances or detailed Walk shape files(street grids) to estimate the time it takes a traveler to walk from their home or work to transit stop.

[shweta487]

Direct Ridership Model

• Specialized tool used to estimate transit ridership for Bus Rapid Transit alternatives in Albuquerque metropolitan area.
• Main goal is to predict ridership at individual stations based on local land use and system characteristics.
• Generates ridership estimates for both Opening day and 2035 to account for population and employment growth.

Modeling

• OLS regression analysis to establish empirical relationships between station-area variables and daily boardings.
• Model is calibrated to Albuquerque context using an error adjustment equation derived from existing rapid bus data
• Incorporates regional travel demand data to determine the size and demand for park-and -ride facilities at specific stations.
• Model is sensitive to the "Auto transit speed differential" which measures how competitive transit travel times are compared to driving.

Uses

• Ridership data was collected from a variety of BRT and rapid ride systems throughout the country.
• Urban density data: Detailed counts of household and employment density within a 1/2 mile walk of each proposed station
• Transit Service metrics: Data on the number of intersecting feeder bus routes, daily feeder trains, and the percentage of the route with dedicated transit lanes
• Socioeconomic indicators: Information on poverty density and the presence of major trip generators like colleges or universities within the station area.
• Metrics such as the number of available park-and-ride spaces and distance to the nearest stop.
• Variables Comprehensive List:
  Boardings, End of Line, Buses Per Day, Route Length, Feeder Bus Routes, Parallel Bus Routes, Rapid Bus Routes, Feeder Rail Routes, Daily Feeder Trains, Residential Density, Employment Density, Urban Density, Park And Ride, Bus Only Lane, Pct Transit Only Lane, Dist To Nearest Stop, College, Diversity, Poverty Density, No Vehicle HH Density, ACC Households, ACC Employment, Peak Frequency, Span, Speed Ratio

Service Levels

• Rapid ride: provides express bus service with limited stops, similar to current local rapid lines
• Some BRT improvements: Includes express service plus limited dedicated right-of-way and operational features like queue jumps
• Significant BRT improvements: High investment mode featuring dedicated right-of-way on 80% of the corridor significantly increasing travel time competitiveness with automobiles.