A classification tree approach to identify key factors of transit service quality

A key aspect to take into consideration when developing indices to evaluate transit service quality is to determine how much weight passengers give to each attribute when making a global assessment of service quality (SQ). The simplest method of a direct question in customer satisfaction survey (CSS) poses a number of problems, and therefore statistical regression methods have been developed to infer attribute importance on the basis of CSS or stated preference surveys. However, most regression models have their own model assumptions and pre-defined underlying relationships between dependant and independent variables. If these assumptions are violated, the model could lead to erroneous estimations. This paper proposes using a classification and regression tree (CART) that does not require any pre-defined underlying relationship between dependent and independents variables, to identify the key factors affecting bus transit quality of service. The paper uses the data gathered in a CSS conducted on the Granada metropolitan transit system in 2007, which was a non-research oriented survey. Two CART models were developed to compare the key attributes identified before and after making passengers reflect on the main aspects of the system. The outcomes show that, in a preliminary evaluation, passenger perception of SQ is basically influenced by frequency. After being asked to evaluate all the attributes, however, other attributes (e.g. proximity, speed and safety) become more important than frequency.

► The paper presents a new approach to evaluating SQ in PT systems using decision trees.
► CART is a good alternative for analyzing metropolitan PT quality of service.
► CART can be used to extract valuable information from non-research oriented surveys.
► This is a good method for the public transport industry.