Segmenting Fare Evader Groups by Factor and Cluster Analysis

Abstract

In proof-of-payment transit systems, fare evasion represents a crucial topic and undermines financial viability for public transport companies (PTCs). Two studies segmented all transit passengers by using qualitative research and explorative two-step cluster analysis as well as web-survey data. However, as far as the authors know, no study segmented exclusively the fare evader passengers, with the aim to know deeply characteristics of distinct groups, and to use data gathered from an intercept survey. Moreover, the combined use of a factorial and a cluster analysis was not adopted to solve this issue. By using 850 on-board personal interviews, gathered from an Italian PTC, a Multiple Correspondence Analysis and a Hierarchical Cluster Analysis have been performed to group the fare evader passengers. Three clusters of evaders' passengers are derived: 1) the chronic, 2) the calculated-risk and 3) the accidental fare evader. The chronic cluster includes passengers who always evade the fare, and represents a medium numerous groups of fare evaders. The calculated-risk cluster includes passengers who usually evade the fare and represents the most numerous group of evaders. The accidental cluster includes passengers who rarely evades the fare and represents the less numerous groups. These outcomes are very relevant and useful for both research and practice. Indeed, this paper contributes to the empirical understanding of the determinants of being a fare evader for a selected group. Moreover, it helps PTCs understand how evaders' groups differ each other, adopt suitable strategies for each group and evaluate their effects.

Fare evasion in public transport systems: a review of the literature

“…Delbosc and Currie (2016b) refined their previous segmentation and merged fare evaders into deliberate, unintentional, and never-evaders. Salis et al (2017) obtained a similar result.…”

Fare Evasion in Public Transport: Grouping Transantiago Users’ Behavior

“…Accidental evaders and circumstantial non-evaders are similar groups, meaning those who sometimes pay and those who sometimes do not pay the bus fare, according mainly to their level of planification and access to charge farecards. These groups may be compared or related to what Salis et al [25] define as accidental evaders, denoting those who rarely evade and, in this case, do it because of a structural problem (lack of places to refund) that is experienced by users as an unplanned/accidental, "not my fault," issue. This is probably the easiest group to control by structural measures-in this case, new places to charge farecards and ways to pay for the bus fare.…”

“…Few studies, however, have investigated the evasion phenomenon with a view to identifying evader types and the reasons that motivate users to pay or not pay transport fares. Investigations that have looked at this aspect [17,[20][21][22][23][24][25] differ in the methodologies they utilize for taking samples and analyzing the data obtained.…”

“…They also observed high levels of evasion among those who took trips under 15 min, were regular users and were dissatisfied with the service. In a later paper, Salis et al [25] performed a cluster analysis to identify the different evader types with a view to establishing the determinants of each one and how they differ. Their results segmented evaders into three different groups-chronic (those who always evade), calculated risk (those who usually evade), and accidental (those who rarely evade).…”

Multiple Correspondence and Hierarchical Cluster Analyses for the Profiling of Fresh Apple Customers Using Data from Two Marketplaces

“…The p eigenvalues in MCA are extracted from a matrix developed from the Burt’s matrix and the latent variables (components) reproduce, in descending order, the highest variation (inertia) among the observations in the Burt’s matrix (as also discussed by Salis et al [ 15 ]). The total number of eigenvalues, p, is determined by considering the dimension of the dataset, as indicated in Equation (1).…”