2018 WILLIAM W. MILLAR AWARD: Outstanding Paper in Public Transportation Transportation Research Record 1–11 Ó National Academy of Sciences: Transportation Research Board 2018 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0361198118796940 journals.sagepub.com/home/trr Bus Load Inference and Crowding Performance Evaluation through Disaggregate Analysis of Fare Transaction, Vehicle Location, and Passenger Count Data Gabriel E. Sa ´nchez-Martı ´nez 1 , Laurel Paget-Seekins 2 , Christopher W. Southwick 1 , and John P. Attanucci 1 Abstract Comfort is an important aspect of the transit passenger experience. Crowding can significantly decrease passengercomfort and disrupt service delivery, causing passenger travel times to increase and even resulting in passengers being unable to board an arriving vehicle. This research explores the use of automatically collected vehicle location data, fare transaction data, and passenger origin–destination inference to measure crowding on buses. Three model components are involved: scaling vehicle trip-level origin–destination transfer data, measuring crowding as perceived by passengers through performance measures defined for this purpose, and determining the sources of crowding. The latter is important to identify the most effective means of addressing crowding in each case. The models are tested on data from the Massachusetts Bay Transportation Authority, and examples of graphical applications already being used by planners are presented. Together with trip time, cost, and reliability, ride comfort is among the factors that influences people’s choice of mode for commuting and other urban transportation needs (1–6). Comfort is especially relevant in the context of urban public transportation systems, which may strug- gle to provide the required capacity at peak times. Riders of public transportation services operating near capacity experience highly crowded vehicles, and are less likely to find a seat or ride with sufficient space to read or use a mobile device, making their ride not only uncomfortable but also less productive. It has been shown that increas- ing frequency to decrease crowding can increase social welfare, in spite of the additional operating costs and traffic that additional vehicles can cause (7). When vehicles reach capacity, people waiting at a stop may not be able to board the next vehicle to arrive, resulting in additional waiting time and inconvenience that is not captured by traditional performance mea- sures. High loads also slow the boarding and alighting process, resulting in longer stop dwell times, slower oper- ating speeds, and longer cycle times, which increase oper- ating costs, as well as increased headway irregularity, which further increases waiting times and decreases the reliability as experienced by riders (8–13). Given the importance of crowding to passengers’ experience, mode choice, capacity, reliability, and oper- ating cost, it is important for transit agencies to continu- ously monitor vehicle loads, identify routes that need service improvements, determine what changes are needed to mitigate high crowding, and implement those changes. Measuring loads is a long-established practice in transit agencies (14–16), but the aggregation level at which this is typically done, spatially at the route level and temporally for the whole day or by time periods, which may suffice for reporting and meeting the require- ments of a service delivery policy, may hide how passen- gers experience their trips and thus the real impact of high loads on demand and performance (17). For exam- ple, the 2010 MBTA Service Delivery Policy (18) speci- fies that the average maximum load for a bus route over a peak time period should not exceed 1.4 times the num- ber of seats on a vehicle. It is probable that among the 1 Massachusetts Institute of Technology, Cambridge, MA 2 Massachusetts Bay Transportation Authority, Boston, MA Corresponding Author: Address correspondence to Gabriel E. Sa ´nchez-Martı ´nez: gsanmar@mit.edu