Integración de modelos para la estimación de la demanda en el transporte público. Aplicación de sistemas de información geográfica, análisis de regresión múltiple y funciones distance-decay al metro de madrid

Abstract

The analysis of the potential travel demand in order to build new infrastructure is an essential task in the process of transport planning. Despite widespread criticism about its reliability and accuracy, the classical method (four-stage model) is still widely used, perhaps not as much for its efficacy as for the absence of other valid alternatives. In recent years, the need to know as accurately as possible the future demand, aroused a renewed interest to improve the predictive models, either reformulating the existing ones or promoting innovations. Within this ambit emerge the so called alternative models, which in opposition to the traditional perspective are not sequential, and usually show a variety of—often combined—methods, use disaggregated data and are not always of practical use. Perhaps it is because of these characteristics that they are little-known. Despite its scarce dissemination, these models offer significant advantages compared to the traditional approach in terms of simplicity, ease of result interpretation and low cost. For example, greater disaggregation of data around the transport stations allows to surpass the typical analysis of coverage and potential demand for transport by more realistic estimates of demand through the penetration curves. Another newer alternative is the use of direct response models, based on the combined use of Geographic Information System tools (GIS1) and statistical analysis, which enhances the analytical capacity and data modelling, by allowing to integrate in a multiple regression, the variables considered essential to explain the behaviour of the demand. In this framework of ideas we attempt to propose valid alternatives to classical transport modelling. It is proposed to generate a series of predictive models of travel, integrating into the multiple regression analysis, the pedestrian accessibility, the spatial treatment of the variables with distance-decay2 functions and the local analysis at station level, to capture in detail the socio-demographic and urban features of its environment. To demonstrate its real applicability, it was decided to take the Madrid Metro as an example. Taken into account that it absorbs nearly 50% of the total trips by public transport, it is fully justified the interest in analyzing its demand. The Household Survey on Mobility conducted in the year 2004 (HSM-04) and a measuring in the access to the transport networks, commissioned by the Transport Regional Consortium (TRC) in Madrid, are the main input to know the use by the population. No less important to the process used was the geo-referenced network of street information provided by the Statistical Institute of the Community of Madrid and the microsectioning of the Madrid Transport Consortium. From the methodological point of view it is worth mentioning the intensive use of GIS and statistics tools for treating the variables. Mobility data disaggregated at the local level (stations) allowed to calculate the distances travelled to the station according to population groups, as well as building the curves or distance-decay functions. To estimate the variables in the surroundings of stations, service areas were calculated on the network every 100 meters. Then the distancedecay curves were used to ponder the variables in each distance strip and to collect the decreasing effect of distance. Finally, the weighted data led to the direct estimation of demand, supported by multiple linear regression models. To correct the problems of parameter instability and spatial autocorrelation in the traditional regression, advanced GIS functions (Moran’s local and global indices and geographically weighted regression) were used. Practical applications show encouraging results in several ways and appear as an alternative or a complement to traditional modelling. The distance-decay functions for young and adults, men, immigrants and public transport captives, point out that they are willing to travel longer distances. The coverage area of stations using distance traveled on the network is lower but more real than those obtained from a standard radio. Besides, the overestimation has a different effect on groups using the Metro with different frequencies and having different degrees of vulnerability. Regression-based models sometimes show adjustments similar to those obtained with the fourstage model, but also allow to assess the impact of variables such as employment density, diversity of land use or road density when explaining the use of the Metro. They also provide information on the spatial variation of the adjustments, elasticity and global and local statistical significance, allowing to achieve more realistic results. In short, we can state that several methodological improvements of great interest to transportation planning were achieved, such as a more convenient treatment of distance, a direct estimation of the demand for new stations by means of regressions, and the integration of GIS tools and statistical modeling.