Assessment of Models to Estimate Bus-Stop Level Transit Ridership using Spatial Modeling Methods

The objective of this research is to develop and assess bus transit ridership models at a bus-stop level using two spatial modeling methods: spatial proximity method (SPM) and spatial weight method (SWM). Data for the Charlotte (North Carolina) area are used to illustrate 1) the working of the methods and 2) development and assessment of the models. Features available in Geographic Information System (GIS) software were explored to capture spatial attributes such as demographic, socioeconomic, and land use characteristics around each selected bus stop. These, along with on-network characteristics surrounding the bus stop, were used as explanatory variables. Models were then developed, using the generalized estimating equations (GEE) framework, to estimate riders boarding (dependent variable) at the bus stop as a function of selected explanatory variables that are not correlated to each other. Results obtained indicate that Negative Binomial with log-link distribution better fits the data to estimate ridership at the bus-stop level (for both SPM and SWM) than when compared to linear, Poisson with log-link and Gamma with log-link distributions. Although SPM models demonstrated distance decay behavior, statistical parameters indicate that SWM (based on functions 1/D, 1/D2, and 1/D3) does not yield better or more meaningful estimates than when compared to SPM using 0.25mile buffer width data. Journal of Public Transportation, Vol. 15, No. 1, 2012 34 Introduction Transit systems support a broad range of goals that include air quality improvement, energy conservation, congestion reduction, provision of mobility to the disadvantaged, access to employment or attraction centers, the promotion of economic development, sustainability, and enhanced livability. Understanding the factors that influence transit ridership is very important to achieve these goals and increase transit market potential. The Bureau of Transportation Statistics (2010) reports that 6,922,000 people (~5% of the nation’s overall trips) used public transportation as their principal means of transportation to work each day during 2009. Transit system managers and planners often rely on statistical models that are cost effective, developed in a reasonable amount of time using available data inventories, and provide a good understanding of the relationship between the dependent variable and explanatory variables. This research aims to develop statistical models to estimate riders boarding at a bus stop using spatial data and on-network characteristics around/near the bus stop. The data required to develop these models are typically available with most state Departments of Transportation and local agencies as well as in many open source data inventories. The use of bus transit depends on accessibility to a bus stop. Accessibility could be defined in terms of walking time (say, 5, 10, 15, or 20 minutes from an origin to a bus stop) or walking distance (say, 0.25, 0.5, 0.75, or 1 mile from an origin to a bus stop). To better comprehend the substantial effect and area of influence of spatial attributes (includes explanatory variables such as demographic, socio-economic, land use, and on-network characteristics) on ridership (dependent variable), a spatial analysis needs to be conducted at several different buffer widths (say, 0.25, 0.5, 0.75, and 1 mile) to identify the ideal spatial proximity distance to extract data for modeling. The maximum buffer width for consideration depends on the acceptable maximum walking distance to access a bus stop (generally, 1 mile). In general, the number of riders who use bus transit system decreases as the distance from the bus stop increases. Integrating data pertaining to demographic, socio-economic, and land use characteristics from different buffer bandwidths (say, 0–0.25, 0.25–0.5, 0.5–0.75, and 0.75–1 mile) based on this distance decay effect, and using it to develop ridership models may yield better and accurate estimates. The objective of this research is 1) to identify explanatory variables and distribution functions and 2) to develop and assess ridership models at a bus-stop using two spatial modeling methods: spatial proximity method (SPM) and spatial weight