Transit ridership forecasting at station level: an approach based on distance-decay weighted regression

• Published in: Journal of transport geography 2011-11, Vol.19 (6), p.1081-1092
• Main Authors: Gutiérrez, Javier, Cardozo, Osvaldo Daniel, García-Palomares, Juan Carlos
• Format: Article
• Language: English
• Subjects: Direct ridership models at station level; Distance-decay weighted regression; Geographic information systems; GIS; Madrid; Mathematical models; Metro; Networks; Regression; Satellite navigation systems; Stations; Thresholds; Weighting
• ISSN: 0966-6923; 1873-1236
• DOI: 10.1016/j.jtrangeo.2011.05.004

► The article develops a rapid response ridership forecast model. ► It is based on the use of distance-decay functions and multiple regression models. ► Weighting the predictors according to that functions provides better results. This article develops a rapid response ridership forecast model, based on the combined use of Geographic Information Systems (GIS), distance-decay functions and multiple regression models. The number of passengers boarding at each station in the Madrid Metro network is estimated as a function of the characteristics of the stations (type, number of lines, accessibility within the network, etc.) and of the areas they serve (population and employment characteristics, land-use mix, street density, presence of feeder modes, etc.). The paper considers the need to evaluate the distance threshold used (not the choice of a fixed distance threshold by assimilation from other studies), the distance calculation procedure (network distance versus straight-line distance) and, above all, the use of distance-decay weighted regression (so that the data from the bands nearer the stations have a greater weighting in the model than those farther away). Analyses carried out show that weighting the variables according to the distance-decay functions provides systematically better results. The choice of distance threshold also significantly improves outcomes. When an all-or-nothing function is used, the way the service area is calculated (straight-line or network distances) does not seem to have a decisive influence on the results. However, it seems to be more influential when distance-decay weighting is used.