Modeling Participation and Consumption

This article investigates the Greek household travel demand for domestic interurban public transportation using cross-sectional micro-data from a countrywide Household Budget Survey. A number of limited dependent variable models, including the Heckit, Two-Part, and Double-Hurdle models, are implemented to jointly estimate the probability of selecting a specific mode and the amount of using it. The results provide useful insight into the existence of feedback relationships between the decision-making mechanisms of mode selection and amount of usage, and they demonstrate the predominant effect of income on the demand for coaches, railways, airplanes, and coast-wise sea ferries. Introduction The principal objective of this article is to investigate the consumption demand of Greek households for interurban public transportation services using cross-sectional data from a countrywide Household Budget Survey (HBS). The amount of money spent by households for trip making can provide a useful metric of their Journal of Public Transportation, Vol. 11, No. 2, 2008 86 travel demand for different passenger transportation modes. In particular, the magnitude of (e.g., monthly average) travel expenditures can incorporate such information as those related to the frequency and amount of trip making as well as the monetary cost of travel by each mode. The use of expenditure micro-data can facilitate the task of identifying major social and economic determinants of household budget allocation for different transportation modes. In particular, such data can allow determining different trip decision-making structures of travelers belonging to diverse economic, social, and demographic population groups and dissimilar geographical settings. The present study concentrates on simultaneously identifying the main factors influencing the choice and usage of interurban public transportation services in Greece. The data about travel demand for air travel, coast-wise sea travel, and intercity coach and rail travel services refer to the household level and they are based on the (most recent) 2004–05 Greek HBS. The article provides an overview of the current econometric models used in the literature to estimate travel demand using micro-data on transportation expenditures. The methodological approaches employed in the current study to analyze intercity travel demand are examined, and the study data and the variables used in the model estimation are described. Results of the model estimation are presented and conclusions and policy implications are discussed. Travel Demand Models Using Micro-data on Expenditures Although the effects of several economic and sociodemographic factors on household travel choices have been well documented and studied using cross-sectional data from a variety of sources, very few studies have investigated the mechanisms of household spending behavior for trip making. These studies are principally based on the use of limited dependent-variable models. The most familiar type of such models is the Tobit model (Tobin 1958), which has been applied to a wide range of consumer demand studies using micro-data (Deaton 1997). Among them, Hagemann (1981) studied the household expenditures for vacation travel using the 1972–73 U.S. Consumer Expenditure Survey (CES). Nolan (2003) investigated urban household travel decisions in the Dublin area, separately with regard to petrol, bus, and taxi fares using the 1994–95 Irish HBS. Also, Thakuriah and Liao (2006) examined the household decisions on the daily short-distance (urban) overall travel expenditures using the 1999–2000 U.S. CES. Modeling Participation and Consumption 87 The use of Tobit models is based on the rather restrictive assumption that the same variables affect in the same way both the decisions of travelers to use or not to use a specific transportation mode and the intensity of using it. In this way, Tobit models can only capture corner solutions, which imply that all households are potential users and choose to use or not to use a particular mode due to lack of affordability (low income) or the high price of petrol or public transportation fare, or both (low income and high price). However, zero observations can sometimes be attributed to other factors, such as habitual (or true) nonusage of a specific transportation mode, infrequency of traveling with the particular mode, and reduced accessibility (e.g., due to lack of appropriate transportation infrastructure and public transportation services and geographical constraints, particularly in remote regions with low connectivity, such as islands). The literature related to the treatment of potential bias in the travel demand estimation using micro-data due to the above factors is limited. Existing studies mostly concentrate on the effect of the habitual nonusage of a particular mode through application of the Heckman’s two-stage sample selection estimator, also know as Heckit model or adjusted Tobit model, or the Limited Information Maximum Likelihood selection estimator (Heckman 1979). According to this methodology, the dichotomous choice of households to travel or not to travel with a specific mode dominates their decision about the intensity of using the particular mode. The Heckit model has been employed for analyzing the car petrol consumption behavior by Kayser (2000) using the U.S. Panel Study of Income Dynamics (PSID), Asensio et al. (2003a), who employed the 1990–91 Spanish HBS, and West and Williams (2007), who used the 1996-98 U.S. CES. Asensio et al. (2003b) also used the Heckit methodology for the estimation of household expenditures for urban public transportation services using the 1990–91 Spanish HBS. The solution of instrumental variable (IV) systems of travel demand equations based on microdata can also be included in the category of the methods used for controlling the effect of selection bias (Bergantino 1997). The present study implements the Heckit methodology for the disaggregate analysis of household expenditures for intercity passenger transportation. In addition, a different type of limited dependent variable model—the Double-Hurdle model— is employed here for comparison purposes. According to this model, the problem of potential users who do not use a transportation mode is examined with regard to factors concerning the habitual nonusage (first hurdle), as well as other factors that inhibit them from realizing a trip, such as personal or intrahousehold Journal of Public Transportation, Vol. 11, No. 2, 2008 88 constraints and reduced accessibility (second hurdle). Different versions of the Heckit and Double-Hurdle modeling methodologies are considered in this study (see below), in accordance with the existence of a feedback relationship between the decision-making mechanisms on choosing to use or not to use a specific mode and the amount of using the particular mode. Methodological Approaches of the Study The Heckit Model and the Two-Part Model The consumption demand equation that describes the amount of expenditures y i made by household i for using a specific transportation mode (e.g., for purchasing petrol or buying tickets or travel cards for using public transportation services), can be described as a latent dependent variable equation, which is typically solved using Ordinary Least Squares (OLS) among the subsample y > 0, as follows: