Performance metrics have typically focused at two main scales: a microscopic scale that focuses on specific locations, time-periods, and trips; and, a macroscopic scale that averages metrics over longer times, entire routes, and networks. When applied to entire transit systems, microscopic methodologies often have computational limitations while macroscopic methodologies ascribe artificial uniformity to non-uniform analysis areas. These limitations highlight the need for a middle approach. 

This dissertation presents a mesoscopic analysis based around timepoint-segments, which are a novel application of an existing system for many transit agencies. In the United States, fix-route transit is typically defined by a small subset of bus stops along each route, called timepoints. For this research, routes are divided into a consecutive group of bus stops with one timepoint at the center. Each timepoint-segment includes all data collected in that segment during one hour of operation. 

The utilized data sources are widespread and generally available to transit agencies. A methodology for merging and cleaning the data sources is proposed that: first, identifies broken data collection system to flag missing and inaccurate data; second, defines parameters of probability distributions, representative of specific locations, times, and routes, using sufficient statistics; and third, replaces flagged values with a random, but probabilistically representative value. The merged and stochastically cleaned data is aggregated by timepoint-segment to reduce subsequent computational requirements, yet maintains high granularly for statistical analysis after aggregation.

The results of linear and non-linear regressions for service durations, at and between bus stops, are presented and discussed. Independent variables were chosen based on previous published literature, but also included several updated classes of variables to provide comparisons for stop types, traffic signals, vehicle interactions, and time-of-day. The coefficients and performance of aggregated models are compared to previously published methods. The results show that factors identified at the microscopic scale (e.g. passenger movements, bus interactions at stops, travel times, travel speeds, unplanned stops, bus bunching, etc.), can be examined in aggregate without lost utility and without the heavy computation burden required to process large microscopic datasets, while also capturing double the variability in the data. 

Visuals for congestion and headway performance, based on the aggregated datasets, are designed to examine transit performance along a route, between routes, and for specific segments. These visuals are a potentially useful tool for evaluating performance along routes and for identifying areas that may require a closer examination. Additionally, the methods are not computationally intensive and may be easily customized to examine specific locations, times, or feature sets. 

The methodologies for data cleaning, regression modeling, and performance visuals, provide a foundation for how timepoint-segments may prove useful to researchers and agencies. The aggregated analysis reduces variability caused by singular atypical events, but still preserves enough detail for a robust statistical analysis. Overall, this approach improves realism, which is beneficial for evaluating the key trade-offs ridership, service, accessibility, and costs. Mesoscopic performance measures may help to understand relationship between key factors influencing transit operations, evaluate iii uncertainty, examine variations in service, determine points sensitive to disruption, quantify congestion costs for users and agencies, and compare travel patterns between different routes, days of the week, and peak versus off-peak travel.

Photo by jorgeantonio/iStock