«Technical Report Documentation Page 1. Project No. 2. Government Accession No. 3. Recipient's Catalog No. SWUTC/11/161127-1 4. Title and Subtitle 5. ...»
The most positive effects of such a life is the enlarged geographical scale of job locations while the most negative effect is the missing out on the daily family life for a significant time(van der Klis & Karsten, 2009a). The perception of the commuting partnership could vary among different individuals under different situations. For instance, by interviewing 43 spouses, Gross (1980)concluded that older couples, couples married longer or freed from childrearing responsibilities, and those among whom at least one partner had an established career might consider the lifestyle less stressful.
Furthermore, Bunker et al. (1992)compared the quality of life of 90 commuting couples with that of 133 single residence dual career couples. Their study demonstrated that commuting couples expressed less satisfaction with their partner relationship and family life, yet commuting couples, especially men were more satisfied with their work life and appreciated the additional time they could reserved for themselves.
4.1. Intercity Travel Demand Model Intercity travel demand model has different types that focus on different geographic units including major intercity corridor, statewide, regional, and national models. The motivation of developing intercity travel model in addition to the urban travel model is that transportation researchers believe that people travel according to a different set of rules over longer distances and between metropolitan areas from inside a metropolitan region (TRB, 2006). The ability to analyze intercity travel demand relationships and forecast future intercity travel demand is important to assist public agencies and private carriers in making intercity transportation service decisions, such as investment in HSR technologies (Koppelman & Hirsh, 1986).
The earliest intercity model were developed in the 1960s, and in the 1980s, Rice et al. (1981) and Koppelman et al. (1984) conducted detailed reviews of intercity model development efforts. Yet, the intercity models were often associated with an academic exercise, making use of fewer, more carefully chosen origin-destination pairs and generally presenting situations that were a little more
in nature (TRB, 2006). The most implemented intercity travel models are those statewide models that developed by different state transportation agencies starting in the 1990s.
The existing intercity travel modeling approaches can be categorized into two major classes – aggregate approach and disaggregate approach. (Koppelman, et al., 1984; Rice, et al., 1981; TRB, 2006). The aggregate approach relied on aggregate data that describe the averages or totals of the socioeconomic status of a city or a region, such as population, employment, economic activity, while disaggregate approach introduced disaggregate data that go further exploring the behavioral motives and characteristics of individual trip makers.
4.1.1. Aggregate Approach
The examination of the Northeast Corridor initialed the intercity travel modeling effort in the 1960s. Most of the early intercity travel model applied the aggregate approach. These models can be further grouped into direct-demand model and sequential models in terms of their structures – a direct-demand model “calculates all of the desired travel information in one, singly calibrated step” (TRB, 2006, p. 79) while a sequential model divides the calibration process into multiple stages. The typical example of the sequential model is the traditional urban four-step model. As Koppelman et al. (1984) summarized, the direct demand model either focused on direct origindestination traffic volume for one or all travel modes, or focused on modal share, and sequential models included both intercity traffic volume and mode share.
The early aggregate intercity travel mode revealed that variables that were statistically related to travel volume included city activity and attraction variables such as population, employment and average/medium income, as well as city pair level of service such as travel time, travel cost and service frequency, and it was important to segment intercity travel market by trip purpose (at lease business and non-business) and trip distance (Koppelman & Hirsh, 1986; Koppelman, et al., 1984).
Quandt & Baumol (1966) developed one famous intercity travel model in the 1960s – the abstract mode model. In the abstract model, the choice of a mode by a traveler depended on the absolute performance level of the “best” mode on each criterion (i.e. travel time, travel cost, service frequency) and the performance level of each mode on each criterion relative to the “best” mode. The travel modes were defined in terms of the type of service they provided to the travelers but not in terms of the physical equipment they employed (i.e. whether it is airplane or railroad). The authors claimed the advantage of possibility to use the abstract model to predict travel on a new mode or a mode with no historical data. However, other researchers did not obtain ideal results when applying this abstract mode approach, and believed that the “the use of the best attribute approach representing competitive effects…[was] …a weak representation of intermodal competition” (Koppelman, et al., 1984).
The aggregate intercity models were criticized for lack of behavioral basis and hence insensitive to important policy variables, estimation bias caused by data aggregation, and unsuccessful functional form (Koppelman, 1989; Koppelman & Hirsh, 1986). These deficiencies led to poor model performance and encouraged researchers divert attention to the disaggregate approach.
4.1.2. Disaggregate Approach
The disaggregate approach analyzes intercity travel at the level of the decision marker – the individual or household. The most advantage of disaggregate approach is the inclusion of a wide range of policy-sensitive variables. Thus disaggregate model is regarded by researchers as more accurate in representing the behavioral response of travelers to changes in economic activities and to changes in intercity transport services (Koppelman, 1989).
Watson (1974) developed both a aggregate model and a disaggregate model using the same data from the Edinburg-Glasgow Area. The aggregate model contained 158 zone-to-zone pairs between the two cities, and the disaggregate model used a binary logit model (rail versus car) based on a sample of 2,546 individuals. Watson concluded that the disaggregate model provided a better statistical explanation of mode-choice behavior. In Watson’s study, the predictions of modal split derived from the aggregate models were inferior to those obtained from the disaggregate models. Thus, Watson believed that disaggregate models have “extremely desirable performance characteristics”.
Understanding the intercity passenger decision-making process is an important step to develop disaggregate intercity travel model. Koppelman & Hirsh (1986) constructed a intercity travel behavioral framework as show in Figure 1.
Figure 1: Intercity Decision Making (Source: Koppelman & Hirsh, 1986) Koppelman (1989) then developed a multidimensional model system for intercity travel choice behavior. Koppelman’s model used the 1977 NTS data which contained l00 miles or more trips during a 3-month period for randomly selected households in 34 metropolitan areas. Seven cities (Atlanta, Baltimore, Boston, Buffalo, Chicago, Los Angeles, and Washington, D.C) were selected as either origin or destinations of city pairs to limit the number of city pairs and hence reduce the burden of collecting intercity level-of-service data. Thus 130 city pairs were chosen and information of travel time, fare and service frequency was obtained for available modes and fare classes between these city pairs.
This model system contained four sequential disaggregate models: the choice of trip frequency, destination, mode, and for fare/service class for air travel. The trip frequency model applied a linear regression method to predict the expected trip frequency for each traveler, and the other three models applied multinomial logit method to predict the possibility that the traveler will choose each alternative in the available choice set. Koppelman described the four steps of decision making as an interrelated process which can be reflected in a hierarchical structure. In the hierarchy, each choice decision is made conditionally on the higher level choices and influenced by the lower levels choices. For example, the choice of travel mode is based on the selection of the selected city, and a traveler will make decision about service class only after he/she decides to travel by air. This hierarchical structure was realized by a nested logit model approach.
Koppelmand reported the importance of level of service variables and demographic variables were supported by the significance of the corresponding parameters; the hierarchical mode structure was also supported by the estimation results for the composite variables. Moreover, Koppelman pointed out since this model system relied on travel service data obtained from published schedules rather than actual performance, and access travel time and cost were excluded from the model due to lack of precise origin and destination locations, his approach was not a fully disaggregate approach that would produce even better predicted results.
Based on the hierarchical model framework proposed by Koppelman, Yao & Morikawa (2005) developed an integrated intercity travel model for the Tokyo – Nagoya – Osaka corridor in Japan to forecast the travel need of a proposed HSR project. Yao & Morikaw’s model included trip generation, destination choice, mode choice, and route choice. In addition to applying the nested logit model structure to capture the relationship between each choice, Yao & Morikawa introduced an accessibility measure to capture the short term induced travel. Since this model is to evaluate a proposed HSR project, in the mode choice step, Yao & Morikawa constructed three sub-models using the revealed preference data, stated preference data, and aggregate Origin/Destination trip data. They estimated there is a general preference for the HSR service relative to other modes, and by 2020 when the HSR would be put into operation, the induced travel accounted for 16.5% of the travel demand.
4.2. Mode Choice modeling
The mode choice modeling represents the single most critical component of the overall intercity demand forecasting process (Miller, 1992), and the disaggregate approach is most applied and developed in modeling intercity travel mode choice. The choice among a set of mutually exclusive available intercity travel mode including auto, bus, rail, air, and/or HSR is referred as discrete choice. Discrete choice analysis is commonly used to model such choice based on principles of utility maximization – travelers are assumed to select the mode with the highest utility. The utility of a choice contains a deterministic portion which can be explained by a set of variables including characteristics of travelers and the transport mode, and a random component which represents the unknown or unobservable effect (Ben-Akiva & Lerman, 1985). The distribution of the random component of the utility largely decides the functional structure of the mode choice model.
The most widely used operational intercity passenger mode choice model is the multinomial logit (MNL) model, which assumes the random term is Gumbel distributed. The MNL model has the advantage of a closed form mathematical structure to simplify computation in both estimation and prediction (Koppelman & Wen, 2000). Stephanedes et al. (1984) calibrated a MNL model for business travel in the Twin Cities-Dulutn, Minnesota Corridor considering the bus, auto and plane mode in the 1980s. Stephanedes et al. defined this model as “fully” disaggregate comparing to the previous models which were only partially disaggregate because of the utilization of average values for travel time/cost and level of service variables for each trip mode and corridor. This model used data from non-random observation of 90 intercity travelers at the Twin Cities air and bus terminals and outlying gas stations. In order to ensure full data disaggregation, the researchers estimated the out-of-vehicle trip characteristics for the nonchosen alternatives.
The MNL model is based on the assumption of independence of irrelevant alternatives (IIA) of random term, which implies the alternatives being considered in the model are independent of each other and have the same variance. This assumption represents the biggest weakness of the MNL: the cross elasticity of one mode to all other modes remains constant, which means improvement in one mode, or introducing a new mode will result in trip diversions to the changed mode/new mode in fixed proportions from all other modes. When the reality violated such condition, the MNL model will results in incorrect predictions.
The weakness of the MNL can be strengthened by relaxing the IIA assumption. The first and the most widely used relaxation of the IIA assumption is the nested logit (NL) model by grouping similar alternatives into nests (Ashiabor, et al., 2007). Other models that relaxed the IIA assumption include cross-nested logits, ordered generalized extreme value model, paired combinatorial logit (PCL), generalized nested logit (GNL), and mixed logit (Ashiabor, et al., 2007), etc. Figure 2 (Koppelman & Sethi, 2005) summarized a conceptual overview of the different random utility based discrete choice models. In this report, the PCL model (Koppelman & Wen, 2000), the heterogeneous GNL(Koppelman & Sethi, 2005), and mixed logits model (Ashiabor, et al., 2007; Srinivasan, et al., 2006) will be reviewed.