You are on page 1of 18

Transportation Research Part A 40 (2006) 621638 www.elsevier.

com/locate/tra

Understanding repetitive travel mode choices in a stable context: A panel study approach
John Thgersen
*,1
Aarhus School of Business, Department of Marketing & Statistics, Haslegaardsvej 10, 8210 Aarhus V, Denmark Received 21 October 2004; received in revised form 10 October 2005

Abstract It is argued that most travel mode choices are repetitive and made in a stable context. As an example, the everyday use of public transport is analyzed based on a panel survey with a random sample of about 1300 Danish residents interviewed up to three times in the period 19982000. The use of public transport is traced back to attitudes towards doing so, beliefs about whether or not public transportation can cover ones transport needs, and car ownership. The inuence of these variables is greatly attenuated when past behavior is accounted for, however. For subjects without a car, behavior changes are in the direction of greater consistency with current attitudes and perceptions. For car owners, current attitudes are inconsequential. The temporal stability of transport behavior is also higher for car-owners than for non-owners. 2005 Elsevier Ltd. All rights reserved.
Keywords: Public transport; Theory of planned behavior; Habit; Panel analysis; Structural equation modeling

1. Introduction Because of its both local and global environmental impacts, the rapid growth in automotive transportation is a source of increasing concern worldwide (e.g., Bonsall, 2000; Cameron et al., 2004; Garon, 2003; Jensen et al., 1998; Mackenzie, 1997; Milj-og Energiministeriet, 1999; Naturvardsverket, 1996; OECD, 1996; Tolba and El-Kholy, 1992). Person transport makes up the bulk of total transport, and increasingly and dominantly private cars are used (e.g., Cameron et al., 2004; Jensen et al., 1998; Milj-og Energiministeriet, 1999; Tolba and El-Kholy, 1992). Policy interventions targeting private transport decisions typically aim to change the modal split, preferably moving medium and long trips to public transport and short trips to walking and biking (e.g., Bonsall, 2000; Garon, 2003; Milj-og Energiministeriet, 1999).
*

Tel.: +45 8948 6440; fax: +45 8615 3988. E-mail address: jbt@asb.dk 1 This paper is the spin-o of a collaborative research project with Folke Olander, Aarhus School of Business, in the framework of Center for Social Science Research on the Environment (CeSaM), which was a multidisciplinary center based on collaboration among scholars from a number of Danish universities and other research institutions. I am grateful to Paul Harland, Linda Steg, and especially Henk Staats for helpful comments on an earlier draft. 0965-8564/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.tra.2005.11.004

622

J. Thgersen / Transportation Research Part A 40 (2006) 621638

Private consumers of transport products and services are also the majority of voters in every country. Hence, in a democratic society, whatever solution one can think of to relieve the environmental and societal burdens produced by transport, it stands or falls with the acceptance by these individuals, either as buyers of transport products and services or in the voting booth (Claxton et al., 1983; Emmerink et al., 1995; Jacobsson et al., 2000; Jones, 1995; Olsen, 1983; Schlag and Teubel, 1997). In this paper, I focus on individuals in their former capacity. Since the bulk of person transport is made by individuals as private consumers, and since their dual capacity as consumers and voters makes it virtually impossible to eectively force people to change their transport behavior, a change in the overall modal split depends on persuading them to accept the more environment-friendly transport solutions (Olsen, 1983; Steg and Vlek, 1997; Tertoolen et al., 1998). And eective persuasion depends on understanding the targeted behavior and especially why it is chosen. Descriptively, travel mode choices may be characterized in a multitude of ways, depending on the purpose. Two characteristics, which are especially important from a regulation perspective, are that they are usually (a) repetitive and performed in a stable context and (b) volitional, but with a limited (often severely limited) number of realistic options. Although few will dispute these characteristics, they are often ignored in transport research. As regards the issue of volition and constraints, there is a tendency in transport research to concentrate either on how possible facilitators and constraints condition choicesthe traditional preoccupation of economists, planners, and behaviorist psychologists (see, e.g., Bresson et al., 2003; Cameron et al., 2003; Geller et al., 1982; Jovicic and Hansen, 2003; Kim et al., 2003; OFallon et al., 2004)or on the evaluations and motives that lead some to adopt and others to reject a specic travel modethe domain of social psychologists, sociologists, and social marketers (see, e.g., Ellaway et al., 2003; Garvill et al., 2003; Golob and Hensher, 1998; Holland et al., 2002; Steg et al., 1995). Few studies have investigated how these two types of possible determinants interact in inuencing travel mode choices (Steg et al., 2001). As a consequence, a rather fragmented understanding of the causes of travel-mode choices (and of how to change undesirable ones) is conveyed to the political system. Just as there is a tendency to ignore behavioral determinants that are outside the conventional focus of ones discipline, there is a tendency to ignore the repetitive nature of travel mode choices, which is reected in the type of data used in most studies: cross-sectional surveys or simple experiments (cf., Steg et al., 2001).2 An important consequence of this ignorance is that it tends to entail unrealistic assumptions about the reasoning preceding travel mode choices. It is an old insight, formulated early by James (1890), that when an action has been repeated frequently in stable contexts in the past, only minimal, sporadic thought is required to initiate, implement, and terminate it (see Wood et al., 2002). As a consequence, when performing repetitive behaviors, people tend to ignore new information (e.g., Aarts et al., 1997; Verplanken et al., 1997), even though it could be highly relevant for their choices, including information about changes in the relative prices of available options. In the following, I (a) propose a simple, but comprehensive conceptual framework for studying repetitive transport behavior while unifying the facilitators and restrictions approach and the evaluations and motives approach, and (b) apply the framework on an analysis of the temporal dynamics of the use of public transport, as it unfolds under normal circumstances, in everyday life. In the concluding section, I discuss implications for policy intervention aimed at inuencing the modal split in the direction of public transport. 2. Conceptual framework Many possible determinants of travel mode choices are discussed in the literature. Some are general characteristics of the individual or the context, which means that their eect on choices is, at best, indirect, mediated through more behavior specic variables (see, e.g., Stern and Oskamp, 1987). This is not the case here. I limit my discussion to determinants that have a fairly direct inuence on travel mode choice. I assume that travel mode choices are partly volitional (i.e., inuenced by the travelers evaluations and motives), partly determined by individual (e.g., transport habits, car ownership) and contextual (e.g., the availability of public

There are noteworthy exceptions, though (e.g., Bamberg et al., 2003; Fujii and Garling, 2003; Golob, 2001; Golob et al., 1997).

J. Thgersen / Transportation Research Part A 40 (2006) 621638

623

transportation) constraints. Hence, I trace travel mode choices back to three generic co-determinants of behavior: the travelers evaluations and motives, individual abilities (and constraints), and contextual oppor tunities (and constraints) (Hoyer and MacInnis, 1997; Olander and Thgersen, 1995; Van Raaij, 2002). Further, consistent with the repetitive nature of travel mode choices, it is assumed that at least some of these determinants are themselves inuenced by (past) behavior. When performing a behavior frequently and in a stable context it may become habitual (e.g., Ouellette and Wood, 1998; Ronis et al., 1989; Wood et al., 2002) and the experience extracted from past behavior may also lead to changes in attitudes (e.g., Fishbein, 1967; Van Raaij and Verhallen, 1983) and in perceived behavioral control (e.g., Ajzen, 1991). In the following, the key concepts in the model will be explained in more depth. 2.1. Motivation Among the psychological theories developed to account for goal-directed behavior beyond complete volitional control, the most widely used is the theory of planned behavior (TPB) (Ajzen, 1988, 1991), also in transport research (e.g., Bamberg and Schmidt, 1999; Harland et al., 1999; Magelund, 1997; Verplanken et al., 1998). The TPB belongs to a class of theories that attempts to account for the psychological implications of behavioral constraints (which may be real or imagined). It distinguishes between three types of evaluations, which co-determine how strongly a person is motivated to pursue a course of action, such as using public transport: an evaluation of expected outcomes (i.e., the attitude towards performing the behavior), an assessment of the social pressure for or against performing the behavior (i.e., the subjective norm), and an assessment of whether or not the behavior seems feasible to carry out (i.e., perceived control). In accordance with the TPB, I expect that the use of public transport is rooted in a favorable attitude towards doing so, a supportive subjective norm, and/or a favorable assessment of the feasibility of covering ones transport needs this way.3 2.2. Past behavior and habits When an action has been performed repeatedly in a stable context, which is obviously the case for most travel mode choices, only minimal thought is required in order to initiate, implement, and terminate the action (Wood et al., 2002). Consistent with this observation, it is often argued that travel mode choices are usually performed in a habitual rather than a reasoned or planned way (e.g., Aarts et al., 1997; Aarts et al., 1998; Garling et al., 2001; Mller and Thgersen, in press; Verplanken and Aarts, 1999; Verplanken et al., 1994; Verplanken et al., 1998). The strongest evidence in support of this claim is based on a script-based measure of habit (i.e., presumed to capture its automatic response nature) developed by Verplanken et al. (1994). This habit measure has been shown to signicantly increase the TPBs explanatory power regarding travel mode choice (e.g., Verplanken et al., 1998; but see also Bamberg et al., 2003). As one would expect, the script-based habit measure is strongly correlated with the frequency of past behavior. The standard measure of habit in survey research is the frequency of which the behavior has been performed in the past (e.g., Ouellette and Wood, 1998). However, as pointed out by Ajzen at several occasions (e.g., Ajzen, 1991, 2002), it is a weakness of this measure that frequent performance of a behavior is only a necessary, not a sucient condition for developing a habit. While a signicant contribution from past to the prediction of future behavior, over and above the variables included in the TPB, indicates that important variables are omitted from the TPB, it is not sucient to prove an inuence of habitual processes. On the other hand, denying any relationship between past behavior and habit is equally wrong, as the signicant relationship between past behavior and the script-based habit measure demonstrates. Substantively, a strong link from past to future behavior speaks about a considerable degree of behavioral inertia and thereby about conditions that are favorable for the development of a habit (e.g., Conner and Armitage, 1998). Further, when the unique impact of past on current behavior varies inversely with that of current intentionality, depending on how much the context facilitates the formation of a habit (see Ouellette and Wood, 1998; Wood et al., 2002), it

The TPB makes no predictions about the relative importance of each of these motivational factors.

624

J. Thgersen / Transportation Research Part A 40 (2006) 621638

strengthens the case for claiming that the residual inuence or past behavior (at least partly) reects non-reasoned processes, such as habit. I will return to this point when discussing the inuence of car ownership on the use of public transport in the following. The empirical study reported later in this paper is based on a panel survey, which makes it possible to assess the inuence of past on current behavior. Based on previous research suggesting that transport choices are often habitual, I expect a strong eect of past on current behavior, over and above the eect of TPB variables. 2.3. Opportunity Because the data material of the present study comes from a large national survey, it has not been possible to assess objective opportunities and constraints regarding the use of public transport. Hence, only the persons own assessment of opportunities and constraints is included in the analysis (i.e., the measure used to represent the TPBs perceived control concept). Of course, a persons assessment may be more or less accurate. The TPB is mainly concerned with the motivational impact of (perceived) restrictions, which is assumed to be mediated through behavioral intentions. It is acknowledged, however, that constraints can have a direct eect on behavior as well. In fact, such an eect was modeled explicitly in the rst version of the TPB, represented by the concept actual control (Ajzen, 1985). However, in order to obtain a more parsimonious model, Ajzen (1988) later decided to let this eect be represented by a direct eect of perceived control on behavior, over and above the eect mediated through intentions; an eect which has been found in several empirical studies (Ajzen, 1991). Hence, in a TPB context it is assumed that the measure of perceived control reects actual constraints as well as motivational eects. 2.4. Car ownership Whereas habits limit a persons de facto choice options regarding travel modes, ownership of a car extends them. Hence, the availability of a private car in the household is conceived here as a contextual variable, which facilitates the choice of car transport and thereby reduces the likelihood of choosing other modes. Indeed, studies have found that ownership of a car is a good predictor of car driving, also when a number of other variables are controlled (e.g., Bjrner, 1994; Bresson et al., 2004). It seems likely that at least part of the impact of car ownership reects attitudes towards available travel mode options (preferences) and/or the belief that public transport cannot fulll ones transport needs. However, as I will argue in the following, there is reason to expect a residual inuence of car ownership on behavior over and above these two factors. First, car owners have more options to choose from, and are therefore more able to transform a preference for using a car for a particular trip into action, than people who do not own a car. Hence, due to the mere availability of a private car, car owners, on average, should be expected to use public transport less than non-owners, and more so than can be explained by dierences in attitudes and perceived control. Secondly, there are reasons to expect that habitual processes are not equally important for the travel mode choices of car owners and non-owners. Several empirical studies have suggested that driving tends to become habitual (e.g., Garling et al., 2001; Verplanken et al., 1994; Verplanken et al., 1998). Further, a panel study in which the use of several travel modes was measured found that the temporal stability of behavior (and also attitudinal variables and behavioral intentions) was substantially higher for car driving than for the use of public transport (Bamberg et al., 2003). Based on a 10 year panel, Dargay and Hanly (2003) found that whereas on average 18% of panel members switched commuting mode from one year to the next, only 8% of car drivers did so. Dierences with regard to the context of the behavior point in the same direction. The car is more exible (i.e., less dependent on schedules, route, distance, weather conditions, and health conditions) than most other means of transportation. This means that car driving is likely to be repeated at higher frequency and to be perceived as a more stable context for behavior than other means. In addition, research indicates that transport by car is experienced as more rewarding than the use of public transport (Ellaway et al., 2003). Hence, car-driving scores higher than most other means of transportation on the three most important prerequisites for developing a habit (cf. Ouellette and Wood, 1998). Everything else being equal, stronger habits will lead car owners to drive more and use other means of transportation less than non-owners, which is a second reason to expect a negative eect of car ownership on the use of public transport, over and

J. Thgersen / Transportation Research Part A 40 (2006) 621638

625

above the TPB variables. The direct eect of car ownership on behavior, which is due to habitual processes, will disappear when habit is explicitly controlled, however. Finally, a residual eect of car ownership should be expected whenever the specic attitude towards using a car is not measured. For example, in the empirical study reported in the following only the attitude towards using public transport is measured. To the degree that attitudes towards using dierent means of transportation are not completely redundant (i.e., do not correlate perfectly), the use of a specic mode depends not only on the attitude towards using this mode, but also on attitudes towards using other, more preferred, modes. 2.5. Possible interactions Until now I have only discussed possible additive eects of the proposed determinants of travel mode choices. However, the eects are not necessarily additive. There are at least two reasons to expect interactions between car ownership and the attitude towards using public transport. 1. A strong habit reduces the likelihood that a person will consider alternatives to what he or she usually does (Aarts et al., 1998; Garvill et al., 2003; Verplanken et al., 1997; Wood et al., 2002). Hence, if they hold stronger transport habits, one should expect that car owners attitudes towards a means of transportation are less predictive of behavior than those of non-owners. On the other hand, one should expect a stronger behaviorbehavior relationship for car owners than for non-owners. 2. If car owners hold more positive attitudes toward car driving than toward using public transport, the attitude towards using public transport may be inconsequential and therefore relatively unimportant. Since unimportant attitudes tend to be weakly correlated with behavior (e.g., Bizer and Krosnick, 2001), this is a second reason to expect that the correlation between the attitude towards using public transport and behavior is weaker for car owners than for non-owners. Notice that these two mechanisms are not mutually exclusive. Hence, in practice they may work together in producing the interaction eect suggested here. 3. Method The study reported here is based on a panel survey consisting of three waves of interviews, separated by one-year intervals. Subjects are a random sample of Danish consumers. Structural equation modeling (SEM) is used for the statistical analyses.4 At the construct level, SEM has the advantage that it is possible to explicitly account for measurement error5 when a latent variable of interest is represented by multiple manifest variables. In addition, SEM makes it possible to estimate complex models, consisting of several equations, simultaneously. Measures of how well the implied variancecovariance matrix, based on the parameter estimates, reects the observed sample variancecovariance matrix can be used to determine whether the hypothesized model gives an acceptable representation of the analyzed data. I used AMOS 5 (Arbuckle and Wothke, 1999) for SEM in this study. AMOS is one of the rst applications that oered Full Information Maximum Likelihood (FIML) to deal with item non-response. Non-response (full as well as item) reduces statistical power and may lead to biased parameter estimates. Recent years extensive research into ways of dealing with missing data suggests that currently FIML is the most ecient method to deal with missingness due to item-non-response and incomplete panel participation, not only because it minimizes the loss of information and, hence, statistical power, but also because it leads to the most unbiased parameter estimates (Arbuckle, 1996), and this even in the case of non-normal data (Enders, 2001). The usual assumptions about uncorrelated error terms and a simple structure factor pattern in the measurement model are applied in the cross-sectional analyses. How and why the assumptions are relaxed in the panel analysis will be explained when presenting the results. When there is only one item representing a latent
For a review of the use of SEM in travel behavior research, see Golob (2003). Strictly spoken, what is accounted for is unique variance in the manifest variables, i.e., variance not reecting the hypothesized latent category, but something else, including measurement error.
5 4

626

J. Thgersen / Transportation Research Part A 40 (2006) 621638

construct, the measurement error cannot be estimated but has to be set to a xed value. In these cases, we can only guess about the size of the measurement error. Probably, respondents can report with very little uncertainty whether or not they have a car. Hence, in this case it seems reasonable to x the error variance to zero. The other single-item construct included in the following analysisperceived controlis much more subjective, however. Therefore it is assumed that this measure contains about 20% error variance, which is typical for survey measures (Andrews, 1984) and not unlike what is found for the attitude and norm measures in this study. Hence, this error variance is xed to .2 times total item variance. The panel analysis partly utilizes a cross-lagged panel design (e.g., Bagozzi, 1980; Bentler and Speckart, 1981; Bynner, 1994; Engel and Meyer, 1996). Cross-lagged panel analysis is often considered a quasi-experimental procedure for ascertaining the direction of causality between two variables measured at two or more points in time (e.g., Bagozzi, 1980). The classical use of cross-lagged correlation analysis for this purpose has been heavily criticized, however (e.g., Rogosa, 1980). Acknowledging the criticism, modern cross-lagged panel analysis based on SEM is less mechanic and more theory-driven (e.g., Bentler and Speckart, 1981; Bynner, 1994; Deary, 1995; Engel and Meyer, 1996; Lorenz et al., 1995). An important issue in this relationship concerns the optimal length of the lag. The lag between waves should mirror the time that it is assumed to take for the causal eect to evolve (Lorenz et al., 1995). The problem is, however, that often we can only guess how long time that is. It is generally assumed in attitude theory that the inuence of attitudinal variables on behavior is rather instantaneous (Ajzen, 1996). The same is true for situational contingencies on choices, such as the availability of a car. Hence, because a lag of one year is probably far too long to assess the causal impact of these variables on behavior, their causal inuence on behavior measured at the same point in time is retained in the panel model. However, other relationships between variables may reect processes that evolve much more slowly, such as the development of a habit or the learning from and, hence, adjustment of attitudes and perceived control based on experience.6 For assessing these relationships, a lag of one year seems much more appropriate. 3.1. Subjects A random sample of Danish consumers was contacted in NovemberDecember 1998, and again one and two years later, for a telephone interview.7 In contacted households the respondent was picked randomly among 18 years old or older individuals (next birthday method) for the rst interview. In 58% of those cases where a contact with the right person was established, the person accepted to participate, resulting in a sample of 1112 individuals. However, responses from 188 individuals could not be used for the present study because they never took one or both of the types of trips investigated here, reducing the eective sample to 924. The sex distribution of the participants was 45% men/55% women. The average age was 42 years and the age range 18 85 years. Seventy-four percent were living with a partner and 73% had children. The highest completed education for 28% of the participants was high-school and 32% had a college or university degree. Twenty-six percent lived in an apartment, the rest in some sort of detached house. Finally, 81% lived in a household with at least one car. When the rst and second interview ended, we asked for permission to call back a year later, should we need to do so.8 Ninety-eight percent accepted in the rst wave and 99% in the second wave. Hence, they were contacted again one year later. Since the interviews took about 30 min, on average, we decided to oer an incentive for participating second and third time: the possibility to win one of ten prices worth approximately
An example illustrates how slowly learning processes evolve in transport behavior. Thorough and repeated studies of the commuting mode of employees in large companies in the Copenhagen area has revealed that commuting mode depends heavily on where the company ` is located vis-a-vis stations on the S-train network (Hartoft-Nielsen, 2001). An unpublished analysis of this data material indicates that when a company moves from a location far from to one close by a main station, or vice versa, it can take up to eight years before the modal split of the commuting has adjusted to the pattern that is normal for companies with a similar location (verbal communication by Peter Hartoft-Nielsen at the Trakdage conference, Aalborg 2728 August 2001). 7 These interviews were carried out by Jysk Analyseinstitut, a professional marketing research company. 8 It is mandatory in Denmark to obtain permission if a market research company wants to save a persons personal information for recontact. In order to minimize possible interviewer eects from such a request, we gave the impression that only a minor, random sample of those contacted in the rst (and second) round would be contacted again.
6

J. Thgersen / Transportation Research Part A 40 (2006) 621638

627

100 Euro. Still, 11% in the second and another 11% in the third wave rejected and 15% in the second and 13% in the third wave could not be contacted after 35 trials. In each round, additional respondents were drawn randomly in order to replace the dropouts. In total, 1320 individuals, who fullled the requirements of the study, were interviewed at least one time. In some analyses in the following, the combined sample of additional subjects drawn in waves 2 and 3 and participating in wave 3 (n = 424) is used as an independent replication sample to cross-validate results based on the sample drawn for the rst wave. 3.2. Variables In all three waves, answers were obtained to questions about the use of public transport for various purposes, a number of psychological and background variables assumed to inuence the use of public transport, and other issues not pertinent to the present study.9 3.2.1. Behavior Public transportation behavior was measured by means of two items: How often do you take bus or train (1) to work, (2) to shopping, and using a ve-point scale with the labels never, rarely, half of the times, often, and always/every time. In the structural equation analyses, answers are coded so that a higher number indicates more frequent use of public transport. 3.2.2. Attitude The attitude towards taking the bus or train to work, shopping, or the like, was measured on a three-item ve-point semantic dierential scale with the end points meaningfulmeaningless, rightwrong, and benecial harmful. In the structural equation analyses, answers are coded so that a higher number indicates a more positive attitude. 3.2.3. Subjective norms The subjective norm as regards taking the bus or train to work, shopping, or the like, was measured by means of two items: I think that most of my acquaintances expect that I take the bus or train to work, shopping, or the like, if the choice is between the car and bus or train, and I think that most of my acquaintances take the bus or train to work, shopping, or the like, if the choice is between the car and bus or train. Answers were measured on a ve-point scale with the end points totally disagree and totally agree and in the structural equation analyses they are coded so that a higher number indicates stronger subjective norms. 3.2.4. Perceived control, opportunities for using public transport, and car ownership We attempted to anchor the persons subjective assessment of how realistic it is to use public transportation in considerations about current conditions. We did that by applying a two-step procedure for questioning about (1) getting to work and (2) shopping. In the rst step, we asked: How do you usually go to work/shopping? In the second step, we followed up by asking: Would it be possible for you, if you wanted to, to take the bus or train to work/shopping? Answers in the second step were registered on a two-point (yes-no) scale and the question was only asked if the reply in the rst step was not walking (in which case the bus or train is not deemed relevant) or taking the bus or train. In order to construct a measure of (lack of) perceived control/ opportunities as regards using public transport, we coded walking,10 bus, and train-answers in the rst step as yes in the second, resulting in a binary measure for each of the two destinations. These two items were nally summed to produce a three-point index. In the structural equation analyses, answers are coded so that a higher number indicates lower perceived barriers against using public transport.
9

Because a number of studies have focused on the relationship between background variables, such as gender, education, and job situation, and transport behavior it was controlled whether any of these variables have a direct inuence on the analyzed behavior after controlling for the variables mentioned above. None of them have. Hence, because the focus in this paper is on proximal determinants of transport choices, they are not included here. 10 Because what we are really interested in is perceived constraints on using public transportation when it is relevant, walking was coded in the same way as using public transportation.

628

J. Thgersen / Transportation Research Part A 40 (2006) 621638

Whether the respondents household had a car was registered on a binary scale (yes = 1, no = 0). No correction was made for the fact that an individuals car availability may be limited if he or she lives in a household with only one car and more than one adult having a drivers license (cf. Staats et al., 2004). 4. Results 4.1. Item-level means, temporal stability and change In order to convey an impression of the level of attitudes, constraints, and behavior revealed by the data, and of the quality of the data, item-level means, stabilities (i.e., testretest correlations) and change (i.e., paired samples t-tests) are presented in Table 1. I will only comment on a few general observations and leave most of the details to the readers own observation. Consider rst the pattern of missing values. The study suers from substantial panel attrition, which is revealed by the lower sample sizes for statistics comparing two waves than for individual waves and by the decrease in sample size when the lag is extended from one to two years. On top of this there are missing values due to item non-response for most items. Consistent with other sources (e.g., Jensen et al., 1998), the study nds a rather low propensity to use public transport in Denmark and that a large majority of the households have a car. Most people also perceive no or a low social pressure to use public transport and limited opportunities for using public transport for work or shopping. Still, most people hold rather positive attitudes towards using public transport. The analysis of item means reveals a pattern of stability rather than change at the aggregate level. Only 5 of 21 t-tests are signicant at the Bonferroni corrected signicance level of p < .002 (0.05 divided by 21). Notice in particular that the t-test reveals no changes in the average use of public transport over the two years spanned by the study.11 The testretest correlations reveal two important general patterns. As should be expected, the correlation is typically lower when the time-lag is two than one year. Further, many of the attitudinal items have rather low testretest reliabilities, indicating that the measured attitudes and norms are relatively unstable within a timeframe of one year. This nding is consistent with the general assumption in attitude theory that the inuence of attitudinal variables on behavior is rather instantaneous (e.g., Ajzen, 1996) and with the nding that the relationship between attitudes and behavior grows weaker when their temporal distance between the two measures increases (e.g., Schwartz, 1978). 4.2. The cross-sectional model A three-wave multivariate panel model easily gets cluttered and unmanageable. For this reason, I wanted to remove variables that do not contribute to the prediction of behavior anyway before constructing the nal model. Primarily for this reason, the relationships between the assumed antecedents and behavior were analyzed at the cross-sectional level rst. When performing a cross-sectional analysis of a single wave of data, I expected to nd that the use of public transport increases with favorable attitudes, norms, and perceived control/opportunities and that it is lower among car owners than among non-owners. The cross-sectional structural equation model based on wave 1 data and relating public transport behavior to the assumed antecedents is shown in Fig. 1 together with the most important t indices. The measurement model is left out in order not to clutter the picture unnecessarily.12

Although this nding apparently contradict work on panels, which conclude that there is a high degree of volatility in transport related behavior patterns, rather than stability (e.g., Dargay and Hanly, 2003; Goodwin, 1986, 1997), the contradiction is apparent rather than real. It is perfectly possible for stability and volatility to be present at the same time, for dierent individuals, dierent aspects of their behavior, dierent levels of aggregation, and depending on the length of the time. Especially, stability at the level of average behavior can cover a quite substantial volatility at the individual level. 12 This and other remaining parts of the AMOS output can be obtained from the author.

11

J. Thgersen / Transportation Research Part A 40 (2006) 621638 Table 1 Item-level means, stabilities, and mean changes n Use public transport to worka Use public transport to shoppinga Meaningfulmeaninglessb Rightwrongb Benecialharmfulb Others expect from mec I expect from othersc Perceived controld Car in householde 924 924 874 891 891 896 898 924 924 1998 1.79 1.39 3.42 4.12 4.16 2.03 1.94 1.29 0.81 n 925 925 891 897 896 896 900 927 927 1999 1.85 1.38 3.51 4.09 4.08 2.07 1.95 1.27 0.82 n 921 921 901 903 902 921 921 921 921 2000 1.79 1.35 3.77 4.12 4.16 2.14 2.07 1.25 0.82 n 630 629 579 594 597 599 599 630 630 r9899 0.71 0.64 0.46 0.29 0.28 0.44 0.36 0.57 0.79 t9899 1.04 0.17 2.34 0.53 0.95 0.42 0.56 0.22 0.97 n 650 650 614 620 622 632 636 652 652 r9900 0.74 0.67 0.34 0.29 0.30 0.37 0.28 0.54 0.84 t9900 1.31 1.28 5.66 0.91 2.25 3.08 2.79 0.31 0.60 n 468 468 436 443 444 453 453 468 468 r9800 0.64 0.49 0.35 0.32 0.26 0.41 0.25 0.53 0.71 t9800 0.56 1.20 6.87 1.92 0.55 3.73 3.83 0.73 2.27

629

Note: t-values printed in boldface are signicant at the Bonferroni corrected level of p < .002, two-tailed. a Scale from 1 (never) to 5 (always). b Positive pole = 5, negative pole = 1. c Scale from 1 (totally disagree) to 5 (totally agree). d Scale from 0 (public transport neither possible to work nor shopping) to 2 (public transport possible to work and shopping). e Yes = 1, no = 0.

Fig. 1. Determinants of public transport behavior, Denmark 1998, cross-sectional analysis, standardized solution. N = 924 v2(19) = 99.186, RMSEA = .068, CFI = .95. A path marked by (n.s.) is non-signicant, p > .05.

The composite reliability indexes (qn) surpass conventionally accepted thresholds (qAtt = .86, qSn = .70,qB = .76), and the highest correlation between exogeneous variables is far from 1.00,13 thus indicating acceptable construct and discriminant validities. The t indices suggest an acceptable overall t, thus conrming that the measurement model and the suggested structural model give a good representation of the data. It appears from Fig. 1 that the attitude towards using public transport, perceived control, and car ownership, but not subjective norms, have a direct inuence on the use of public transport in the present case. Hence, the results are consistent with the key proposition that travel mode choices are co-determined by motivational factors (the attitude towards using public transport) and actual and perceived facilitators and constraints (whether public transport covers ones needs and car ownership). More specically, it shows that the use of public transport depends on the favorability of the attitude towards this behavior and of the assessment of the feasibility of covering ones transport needs this way, and that car owners use public transport less than non-owners. On the other hand, normative considerations seem to have no direct inuence on the use of public

13

The highest pairwise correlation between latent variables in the dataset is .49 (behavior and car ownership).

630

J. Thgersen / Transportation Research Part A 40 (2006) 621638

Fig. 2. Determinants of public transport behavior, 3 wave panel model, standardized solution. N = 1320 (NextPage) v2 = 622.129,158 d.f. RMSEA = .047. CFI = .94. A path marked by (n.s.) is non-signicant, p > .05.

transport. Of the signicant determinants, constraint-related factors seem to have a stronger direct inuence on behavior than the attitude. When the model shown in Fig. 1 is applied to the data from the independent, supplementary sample (n = 424) collected two years later, the exact same pattern is replicated.14 Hence, the nding that public transport behavior is related to the three signicant antecedents in Fig. 1 has a high reliability. 4.3. The panel analysis What happens to these results if we study the issue in a longitudinal context? As mentioned above, I expect to nd that past behavior explains a substantial amount of variance in current behavior, over and above the variables included in the former analysis. I also expect that people adjust their attitudes and perceptions regarding public transport, as well as their car-ownership status, based on past behavioral experiences. In order to test these expectations, three waves of measures of the constructs represented in Fig. 1 (except subjective norms) are combined in a panel model. The basic modeling assumptions are as before, with a few exceptions. It is customary in panel analysis to allow unique (error) variances of items measured repeatedly to correlate (e.g., Joreskog and Sorbom, 1988). Further, structural equation residuals at the same point in time are allowed to correlate,15 provided that the variables, which they belong to, are not connected by a path. With no further restrictions applied, the model converged to an inadmissible solution, however.16 An admissible solution could be obtained by assuming the regression weights of latent on manifest variables (i.e., factor loadings) to be constant over time, which is not unreasonable. Hence, this restriction was added to the measurement model. In accordance with the assumptions outlined in the introduction to the methods section, the structural model, which is shown in Fig. 2,17 includes the paths from behavioral antecedents to behavior measured

In order to save space, these results are not shown, but they can be obtained from the author. However, in order to conserve degrees of freedom, residual correlations that are not signicantly dierent from 0 at the .05 level in any wave are set to 0. 16 The inadmissible result was a negative equation residual for behavior in wave 2. The very high temporal stability of behavior in this case, to be presented shortly, makes such a result particularly likely. 17 In order not to clutter the picture unnecessarily, the measurement model is omitted. However, it can be obtained from the author.
15

14

J. Thgersen / Transportation Research Part A 40 (2006) 621638

631

at the same point in time, all latent variable rst-order temporal stabilities, and cross-lagged paths from behavior at tn to its antecedents at tn+1. According to the test statistics, the model produces an acceptable t to the data. 4.3.1. Results The analysis conrms that there is an eect of past on current behavior, over and above the previously analyzed variables. In fact, the dictum that past behavior is the best predictor of future behavior (Mischel, 1968) is conrmed in this case. In the context of the TPB, behavioral stability over time is attributed to the stability of its antecedents (Bamberg et al., 2003). Behavioral antecedents seem denitely to be quite stable in this case, particularly antecedents reecting non-volitional inuences on behavior. The fact that controlling for past behavior means that the path coecients of other behavioral antecedents are seriously attenuated, for car ownership to a degree where it is no longer signicant, is another indication that the past behavior measure reects these antecedents to some degree. However, the TPB cannot account for the substantial increase in explained variance when past behavior is added to the equation. Still, as mentioned earlier, a residual eect (even a large one) of past on current behavior is not direct proof of an inuence of habit on behavior. But there is evidence that, when making travel-mode choices, people at least have a strong tendency to recycle a decision made in the past. When a decision is repeated several times a week in a stable context and with the same outcome every time, it is unlikely that much reasoning is involved. On the other hand, it seems highly likely that habitual processes are active in that decision (Wood et al., 2002). When autoregressive eects (i.e., the temporal stabilities) have been controlled, a signicant path from another variable indicates that the other variable accounts for some of the change18 in the explained variable (e.g., Schnabel, 1996). Hence, Fig. 2 shows that, to the degree that public transport behavior changes, it adjusts to the individuals (current) attitude towards using public transport and to his or her (current) perceptions as regards constraints on using public transport. In this way, the use of public transport still depends on attitudes and perceptions about constraints, over and above the inuence of past behavior. The cross-lagged eects conrm the expectation that the assumed behavioral antecedents also adjust to behavior. The signicant paths from behavior to car ownership show, as predicted, that the more car owners use public transport the more likely they are to get rid of their cars and/or the less non-owners use public transport the more likely they are to get a car. Hence, it seems that car ownership can be reduced if car owners can be persuaded to tryand continue usingpublic transport; the latter of which undoubtedly depends on experiencing that public transport is an attractive alternative to driving. The signicant path from behavior to the attitude in wave three indicates that not only does the use of public transport depend on the attitude towards doing so. The reverse is also (sometimes) true. The latter eect could be the product of positive experiences from using public transportation (e.g., Fishbein, 1967), but it might also just reect that the attitude to some degree is inferred from behavior, as suggested by Bem (1972). The signicant paths from behavior to perceived control indicate that some people adjust their perceptions about whether their transport needs can be covered by public transport due to experience. In sum, all hypotheses about the inuence of past behavior on current behavior as well as on attitudes towards and perceptions about public transport and on car ownership are conrmed by this analysis. Fig. 2 also reveals a relationship about which no expectations were formulated on beforehand: an instance of correlated equation residuals at the same point in time. When auto-regressive eects have been controlled, such a correlation indicates that changes in the two variables are correlated. Hence, it appears from Fig. 2 that, at least in the last period, changes in car-ownership status are related to changes in perceived opportunities for using public transportation to work and shopping. What this means is most easily understood through an example. Say that the perceived opportunities change, for example as the result of moving residence or job. The correlated residuals indicate that such a change increases the likelihood of a change in car-ownership status as well. The sign of the correlation shows that a positive change in perceived opportunities for using public transport is related to a negative change in car-ownership status, as one would expect.

18

Adjusted for a possible common trend in the variables.

632

J. Thgersen / Transportation Research Part A 40 (2006) 621638

4.4. Does car ownership moderate the attitudebehavior relationship regarding public transport? Finally the proposition that car ownership moderates the attitudebehavior relationship with regard to public transport is analyzed by means of multiple group SEM, comparing car owners and non-owners. As before, cross-sectional as well as the panel analyses are performed. The results of a simple multiple group CFA analysis are presented in Table 2, based only on attitude and behavior items and including two independent, cross-sectional replications (i.e., the rst wave and the combined supplementary sample from wave 2 and 3, interviewed at wave 3). Table 2 shows that there is a stronger attitudebehavior covariance (and correlation) regarding public transport among non-owners than among car owners in both cases. The chi-square dierence tests show that the model t is signicantly reduced if the attitudebehavior covariance is restricted to be equal across groups. This result is consistent with the hypothesis that car ownership weakens the attitudebehavior relationship regarding travel mode choice, which is what we will expect if car ownership facilitates the formation of a transport habit, and thereby reduces the amount of deliberation spent on travel mode choices, or if it makes the attitude towards public transport unimportant. Next, the moderation hypothesis is tested in a 3-wave panel model setting, as above minus the car ownership variable. The restrictions applied in the earlier model were retained in this analysis and in addition equation residuals for behavior in wave 2 in both groups and in wave 3 for car owners were xed to 0 in order to obtain an admissible solution. Finally, the model t could be improved by xing behavioral stabilities to be equal over time in both groups. Comparisons were made between a model where only the mentioned restrictions were applied, and a model where, in addition, the attitudebehavior paths were restricted to be equal

Table 2 Multiple group analysis of the attitudebehavior covariance/correlation regarding the use of public transport for car owners and nonowners in Denmark, 1998 and 2000 Year Covariances/correlations Final model Dierence from model with covariance xed to be equal between groups RMSEA .061 .061 v2 57.523 27.972 d.f. 13 11 Dv2 130.795 36.999 d.f. 1 1

Car owners 1998 2000 .23/.36 .21/.34

Non-owners .50/.42 .57/.48

CFI .95 .97

Note: In order to obtain an acceptable model t, a number of equality constraints were imposed on the measurement models. In both cases, the regression weights of the latent on the observed variables (i.e., factor loadings) were set equal between groups. In addition, in the model for 1998 one attitude items unique variance and the variance of the latent attitude variable were set equal between groups.

Table 3 Multigroup analysis of the paths from attitude to public transport behavior, 3 wave panel model, unstandardized solution. N = 1320 Wave Selected regression weights Final model Dierence from model with paths xed to be equal between groups RMSEA .027 v2 434.970 d.f. 225 Dv2 140.600 d.f. 3

Car owners 1. 2. 2. 3. 3. AB AB BB AB BB .16 n.s. .94 n.s. .94

Non-owners .34 .28 .74 .31 .74

CFI .96

Note: AB: The path from attitude to behavior. BB: The path from past behavior to behavior. Restrictions on the model identical to the model in Fig. 2, with the following exceptions: In order to obtain an admissible solution, the equation residuals for behavior at wave two for both groups and at wave three for car owners only were xed to 0. Further, the model t was improved by xing stabilities for behavior to be equal over time in both groups.

J. Thgersen / Transportation Research Part A 40 (2006) 621638

633

between groups. The relevant t measures and the unstandardized19 attitudebehavior and behaviorbehavior path coecients for the two groups in the model without equality constraints on attitudebehavior paths are shown in Table 3. Consistent with the ndings presented in Tables 2, 3 shows (1) that a signicantly better t is achieved if the attitudebehavior paths are allowed to dier between car owners and non-owners, than if they are restricted to be equal, and (2) that the attitudebehavior relationship is weaker among car owners than among non-owners. Hence, the conclusion derived from the simple analysis presented in Table 2 is robust enough to persist when all the other relationships in the panel model are taken into consideration. Further, Table 3 shows that the tendency of behavior to adjust to the attitude, which was identied in the panel analysis above, is actually limited to non-owners. When past behavior is controlled for, current attitudes are apparently inconsequential for car owners use of public transport.20 In order to better understand why, the behaviorbehavior path coecients are also reported in Table 3. They show that behavior has a higher temporal stability among car owners than among non-owners. Hence, a weaker inuence of the attitude (the attitudebehavior relationship) is matched by a stronger inuence of past behavior (the behaviorbehavior relationship). 5. Summary and conclusions I argue in this paper that in order to properly understand travel mode choices, their repetitive nature should be considered as well as their dependence on both motivational factors and personal and external constraints. In the empirical study, I concentrate on the use of public transport and antecedents with a fairly direct inuence on travel mode choices. Future research should extend the analysis to other means of transport and more distal antecedents. The item-level analysis revealed that, as is known from many other studies, public transport has a rather low market share in Denmark and that most households have at least one car at their disposal. Hence, as in most other countries, there is a need for interventions to reduce the amount of car driving. The good news are that attitudes toward the use of public transport are quite favorable, however. The cross-sectional structural equation analysis based on a single wave of interviews identied three significant determinants of public transport behavior: the attitude towards using public transport, the perceived ability of public transport to cover ones transport needs, and car ownership. Normative considerations seem not to inuence this behavior, at least not in a direct way. According to this analysis, factors which at least partly reect non-volitional forces explain more of the variation in the use of public transport than the attitude. When three waves of interviews are combined into a panel analysis, we get a fuller picture of public transport behavior, including its temporal dynamics. A very high temporal stability is revealed for public transport behavior, and we learn that past behavior is a stronger predictor of current behavior than any of the previously mentioned antecedents. Further, the inuence of other antecedents is severely attenuated, and car ownership is no longer signicant, when past behavior is taken into account. From these observations, several important conclusions can be drawn. As suggested by Ajzen (1991, 2002), the behavioral stabilityand thereby the eect of past on current behaviorcan to some extent be explained by the temporal stability of the antecedents identied in the cross-sectional analysis. The panel analysis shows that especially car ownership and (perceptions regarding) public transports ability to cover ones transport needs are very stable. However, the panel analysis also reveals that past behavior explains a large amount of residual variance, which cannot be attributed to the temporal stability of the other included antecedents. Hence, past behavior reects important predictors of behavior that were not included in the cross-sectional analysis.

19 In multiple group comparisons, where variances may dier between groupsas they indeed do in this casethe unstandardized solution is most informative. In order to save space, only the most central information from the analysis is reported here. The rest of the AMOS output can be obtained from the author. 20 The fact that a similar nding was reported in an earlier panel study in a completely dierent setting (Golob, 2001)car commuters in South California and with a dierent set of travel mode options than those available in this studylends credibility to this nding.

634

J. Thgersen / Transportation Research Part A 40 (2006) 621638

There are good reasons for assuming that habitual processes are at least partly responsible for the eect of past on current behavior in this case. Habit produces temporal stability in behavior. Because transport behavior is performed frequently, in a stable context, and presumably with rewarding outcomes, the conditions strongly facilitate the formation of a habit for this type of behavior (e.g., Ouellette and Wood, 1998). And empirical studies have found a strong and positive correlation between self-reported measures of past transport behavior (i.e., how frequently the behavior has been performed in the past) and other measures of habit (e.g., Verplanken et al., 1994; Verplanken et al., 1998; Verplanken and Orbell, 2003). Vaguer and more inferential support for the assumption that habitual processes are partly responsible for the eect of past on current behavior can be drawn from the ndings regarding the eect of car ownership on using public transport. One of the mentioned reasons for expecting an eect of this kind was that car-owners transport habits are likely to be stronger than non-owners. From this followed the expectation that the eect would disappear when controlling for habit. Hence, the extinction of the signicant relationship between car ownership and behavior when controlling for past behavior, which is found in the panel analysis, is consistent with the assumption that the inuence of past on current behavior reects habitual processes. What distinguishes habitual processes from the conscious recycling of a previously formed intention is that, in the former case, behavioral inertia is due to some sort of automaticity (e.g., Aarts et al., 1998; Bargh and Barndollar, 1996) while, in the latter case, it is a deliberately made choice. In many cases, it probably makes no dierence whether behavioral inertia is the product of one or the other. However, the important point made by habit researchers is that sometimes, due to the force of habit, decisions may be repeated even though important conditions have changed and made a non-chosen alternative more preferable (e.g., Betsch et al., 2001; Wood et al., 2002). The nding in this study that attitudes towards using public transportand thereby probably also travel-mode preferencesare more volatile than behavior indicates that this is indeed the case here. To some extent, public transport behavior adjusts to the attitude towards using public transport, but only among non-owners, not among car owners, according to the multiple group analysis. Hence, at least among car owners it seems that past travel mode choices are repeated in spite of changes in conditions, or how they are perceived, that are suciently important to produce a change in the attitude towards using public transport. As a caveat, remember that two possible explanations for the weaker attitudebehavior relationship for car owners than for non-owners were suggested earlier: (1) that car owners hold stronger transport habits and (2) that the attitude towards using public transport is less important for car owners. The two explanations are not mutually exclusive. It is perfectly possible that they both contributed to the dierence between car owners and non-owners in this respect. The practical implications of the two explanations are also quite similar. If (especially car owners) travel mode choices are either strongly inuenced by habitual processes, or their attitudes towards means of transportation other than the one(s) they actually use are weak, they should be expected to pay little attention to information about alternative means of transportation (Aarts et al., 1997; Verplanken et al., 1997), including highly relevant information such as information about changes in relative prices. What can be done to promote public transport in this situation depends on the quality of the service. Informational and promotional means, such as those discussed later, are only useful if public transport is actually an attractive alternative to the private car. If this is not the case, should one succeed in persuading car drivers to try public transport, the experience will only conrm the individuals prior conviction that car transport is better. Hence, in this case structural conditions need to be changed in order to increase the attractiveness of public transport, or reduce the attractiveness of car transport. In the extreme case, such changes could force people to change behavior. However, as mentioned in the introduction, the use of force is likely to provoke individual reactance and political resistance, and it is therefore usually avoided. Where more moderate structural changes have been implemented and made public transport an attractive alternative to the car, they need to be accompanied by the types of informational and promotional means discussed in the following (see, e.g., Stern, 1999). If public transport is a suciently attractive alternative to the private car to make more people use it, if only they knew it better, a possibility is to adopt a hierarchy of eects approach, not unlike the one used in the marketing of detergents and other low-involvement consumer products (see, e.g., Peter et al., 1999). First, car drivers need to be persuaded to attend to information about the available public transport services. Next,

J. Thgersen / Transportation Research Part A 40 (2006) 621638

635

when they are aware of the options, they need to be persuaded to give public transport a try. Finally, when they have tried public transport on an experimental basis, their decision to use public transport needs to be reinforced and they need to be persuaded to repeat using public transport long enough for it to become a habit. Some of the specic elements in such an approach could be (a) use a targeted approach, (b) entertain, (c) use sales promotion, and (d) provide feedback. Targeting means that, rather than trying to reach everyone all the time, the information should be targeted at the individuals that are most motivated to attend to it, at the exact time and place where they are most motivated to attend to it (usually when they need it). Further, by making the information entertaining21 or in other ways gratifying in itself22 it is possible to catch peoples attention even though they are not interested in the message that the communicator wants to convey. Other proven ways to catch attention are to link the information to dramatic events (e.g., Greenberg et al., 1989) or to threats against things that the audience values (e.g., Dahlstrand and Biel, 1997). Various types of sales promotion, such as trips for free23 or at a reduced fare, can be used to induce trial and if free or reduced fares are oered for a longer period it may even induce repeated use. Feedback can be used to reinforce the decision, for instance by informing about the many others that are also using public transport (aiming at reinforcing a social norm about doing so) and about positive outcomes of using public transport, which are not immediately perceptible (aiming at reinforcing a positive attitude). The panel study showed that attitudes towards using public transport and perceptions about its ability to fulll ones transport needs are inuenced positively by the use of public transport, and that the more people use public transport the more likely it is that they will sell their car. These results indicate that public transport in Denmark is actually experienced as an attractive alternative by those who use it. It also illustrates the importance of getting people to try public transport. Not only is experience an eective way to correct unfavorable misconceptions. People may also change their evaluations of known attributes in a favorable direction due to practical experience. If, for example, a person experiences that the time spent in public transport can be used for valued purposes (e.g., working, reading, and sleeping), the belief that the travel time by public transport is longer may carry less negative weight than before. References
Aarts, H., Verplanken, B., Knippenberg, A.v., 1997. Habit and information use in travel mode choices. Acta Psychologica 96, 114. Aarts, H., Verplanken, B., Knippenberg, A.v., 1998. Predicting behavior from actions in the past: repeated decision making or a matter of habit. Journal of Applied Social Psychology 28, 13551374. Ajzen, I., 1985. From intention to actions: a theory of planned behavior. In: Kuhl, J., Beckmann, J. (Eds.), Action-Control: From Cognition to Behavior. Springer, Heidelberg, pp. 1139. Ajzen, I., 1988. Attitudes, Personality, and Behavior. Open University Press, Milton Keynes. Ajzen, I., 1991. The theory of planned behavior. Organizational Behavior and Decision Processes 50, 179211. Ajzen, I., 1996. The directive inuence of attitudes on behavior. In: Bargh, J.A., Golwitzer, P. (Eds.), Psychology of Action. Guilford, New York, pp. 385403. Ajzen, I., 2002. Residual eects of past on later behavior: habituation and reasoned action perspectives. Personality and Social Psychology Review 6, 107122. Andrews, F.M., 1984. Construct validity and error components of survey measures: a structural modeling approach. Public Opinion Quarterly 48, 409442. Arbuckle, J.L., 1996. Full information estimation in the presence of incomplete data. In: Marcoulides, G.A., Schumacker, R.E. (Eds.), Advanced Structural Equation Modeling: Issues and Techniques. Lawrence Erlbaum Associates, Mahwah, NJ, pp. 243277. Arbuckle, J.L., Wothke, W., 1999. Amos 4.0 Users Guide. Marketing Division SPSS Inc.: SmallWaters Corporation, Chicago, Ill. Bagozzi, R.P., 1980. Causal Models in Marketing. John Wiley & Sons, Inc, Pittsburgh. Bamberg, S., Ajzen, I., Schmidt, P., 2003. Choice of travel mode in the theory of planned behavior: the roles of past behavior, habit, and reasoned action. Basic and Applied Social Psychology 13, 175188. Bamberg, S., Schmidt, P., 1999. Regulating transport: behavioral changes in the eld. Journal of Consumer Policy 22, 479509.

21 The former Danish national rail company, DSB, used this approach in a series of TV spots starring a well-known Danish actor and a Muppet-Show type doll called Harry. 22 For instance by means of contests or lotteries. 23 This type of promotion was, e.g., used by DSB, who, after a period where construction work resulted in frequent delays and cancellations, oered its customers a free month trial of the new Svenborg-Odense line in January 2004.

636

J. Thgersen / Transportation Research Part A 40 (2006) 621638

Bargh, J.A., Barndollar, K., 1996. Automaticity in action: the unconscious as repository of chronic goals and motives. In: Bargh, J.A., Golwitzer, P. (Eds.), Psychology of Action. Guilford, New York, pp. 457481. Bem, D.J., 1972. Self-perception theory. In: Berkowitz, L. (Ed.), Advances in Experimental Social Psychology, Vol. 6. Academic Press, New York, pp. 162. Bentler, P.M., Speckart, G., 1981. Attitudes cause behaviors: a structural equation analysis. Journal of Personality and Social Psychology 40, 226238. Betsch, T., Haberstroh, S., Andreas, G., Haar, T., Fiedler, K., 2001. The eects of routine strength on adaptation and information search in recurrent decision making. Organizational Behavior and Human Decision Processes 84, 2353. Bizer, G.Y., Krosnick, J.A., 2001. Exploring the structure of strength-related attitude features: the relation between attitude importance and attitude accessibility. Journal of Personality and Social Psychology 81, 566586. Bjrner, T.B., 1994. Persontransport med bil (Person transport by car). AKF Forlaget, Kbenhavn. Bonsall, P., 2000. Legislating for modal shift: background to the UKs new transport act. Transport Policy 7, 179184. Bresson, G., Dargay, J., Madre, J.-L., Pirotte, A., 2003. The main determinants of the demand for public transport: a comparative analysis of England and France using shrinkage estimators. Transportation Research Part A: Policy and Practice 37, 605627. Bresson, G., Dargay, J., Madre, J.-L., Pirotte, A., 2004. Economic and structural determinants of the demand for public transport: an analysis on a panel of French urban areas using shrinkage estimators. Transportation Research Part A: Policy and Practice 38, 269 285. Bynner, J., 1994. Analyzing change over time using LISREL: social life and delinguency. In: Dale, A., Davies, R.B. (Eds.), Analyzing Social and Political Change. Sage, London, pp. 99117. Cameron, I., Kenworthy, J.R., Lyons, T.J., 2003. Understanding and predicting private motorised urban mobility. Transportation Research Part D: Transport and Environment 8, 267283. Cameron, I., Lyons, T.J., Kenworthy, J.R., 2004. Trends in vehicle kilometres of travel in world cities, 19601990: underlying drivers and policy responses. Transport Policy 11, 287298. Claxton, J.D., Ritchie, J.R.B., McDougall, G.H.G., 1983. Evaluating acceptability and eectiveness of consumer energy conservation programs. Journal of Economic Psychology 4, 7183. Conner, M., Armitage, J.C., 1998. Extending the theory of planned behavior: a review and arenas for further research. Journal of Applied Social Psychology 28, 14291464. Dahlstrand, U., Biel, A., 1997. Pro-environmental habits: propensity levels in behavioral change. Journal of Applied Social Psychology 27, 588602. Dargay, J.M., Hanly, M., 2003. Travel to work: an investigation based on the british household panel survey. Paper presented at the NECTAR Conference No. 7, Umea, Sweden. Deary, I.J., 1995. Auditory inspection time and intelligence: what is the direction of causation?. Developmental Psychology 31 237250. Ellaway, A., Macintyre, S., Hiscock, R., Kearns, A., 2003. In the driving seat: psychosocial benets from private motor vehicle transport compared to public transport. Transportation Research Part F: Trac Psychology and Behaviour 6, 217231. Emmerink, R.H.M., Nijkamp, P., Rietveld, P., 1995. Is congestion pricing a rst-best strategy in transport policy? A Critical Review of Arguments. Environment and Planning B 22, 581602. Enders, C.K., 2001. The impact of non-normality on full information maximum-likelihood estimation for structural equation models with missing data. Psychological Methods 6, 352370. Engel, U., Meyer, W., 1996. Structural analysis in the study of structural change. In: Engel, U., Reinecke, J. (Eds.), Analysis of Change. Advanced Techniques in Panel Data Analysis. Walter de Gruyter, Berlin, pp. 221252. Fishbein, M., 1967. Attitude and the prediction of behavior. In: Fishbein, M. (Ed.), Readings in Attitude Theory and Measurement. John Wiley, New York, pp. 477492. Fujii, S., Garling, T., 2003. Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis. Transportation Research Part A: Policy and Practice 37, 389402. Garon, P., 2003. The implementation of walking and cycling policies in British local authorities. Transport Policy 10, 235244. Garling, T., Boe, O., Fujii, S., 2001. Empirical tests of a model of determinants of script based driving choice. Transportation Research Part F: Trac Psychology and Behaviour 4, 89102. Garvill, J., Marell, A., Nordlund, A., 2003. Eects of increased awareness on choice of travel mode. Transportation 30, 6379. Geller, E.S., Winett, R.A., Everett, P.R., 1982. Preserving the Environment: New Strategies for Behavior Change. Pergamon, Elmsford, NY. Golob, T.F., 2001. Joint models of attitudes and behavior in evaluation of the San Diego I-15 congestion pricing project. Transportation Research Part A: Policy and Practice 35, 495514. Golob, T.F., 2003. Structural equation modeling for travel behavior research. Transportation Research Part B: Methodological 37, 125. Golob, T.F., Hensher, D.A., 1998. Greenhouse gas emissions and Australian commuters attitudes and behavior concerning abatement policies and personal involvement. Transportation Research Part D: Transport and Environment 3, 118. Golob, T.F., Kitamura, R., Long, L. (Eds.), 1997. Panels for Transportation Planning. Methods and Applications. Kluwer, Boston. Goodwin, P.B., 1986. A panel analysis of changes in car ownership and bus use. Trac Engineering and Control 27, 519525. Goodwin, P.B., 1997. Have panel surveys told us anything new? In: Golob, T.F., Kitamura, R., Long, L. (Eds.), Panels for Transportation Planning. Methods and Applications. Kluwer, Boston, pp. 7996. Greenberg, M.R., Sachsman, D.B., Sandman, P.M., Salomone, K.L., 1989. Risk, drama, and geography in coverage of environmental risk by network TV. Journalism Quarterly (Summer), 267276.

J. Thgersen / Transportation Research Part A 40 (2006) 621638

637

Harland, P., Staats, H., Wilke, H.A.M., 1999. Explaining pro-environmental intention and behavior by personal norms and the theory of planned behavior. Journal of Applied Social Psychology 29, 25052528. Hartoft-Nielsen, P., 2001. Hvad betyder nye boligers og arbejdspladsers lokalisering i bystrukturen for persontransporten? (What are the implications of the localization of new homes and workplaces in the urban structure for person transport?). In: Lohmann-Hansen, A., Nielsen, J. (Eds.), Trakdage pa Aalborg Universitet 2001. Transportradet og Trakforskningsgruppen, Vol. 2. Aalborg Universitet, Aalborg, pp. 455468. Holland, R.W., Meertens, R.M., van Vugt, M., 2002. Self-justication among car-drivers: processes that undermine policy measures. In: Nelissen, G.C.B.a.W.J.A. (Ed.), Marketing for Sustainability. Towards Transactional Policy-making. IOS Press, Amsterdam. Hoyer, W.D., MacInnis, D.J., 1997. Consumer Behavior. Houghton Miin, Boston. Jacobsson, C., Fujii, S., Garling, T., 2000. Determinants of private car users acceptance of road pricing. Transport Policy 7, 153158. James, W., 1890. The Principles of Psychology. Holt, New York. Jensen, M., Gudmundsson, H., Fenger, J., Christensen, L. (1998). Bilisme og miljen svr balance (Car-driving and environment a dicult balance) (Tema-raport fra DMU No. 18). Arhus: Miljog energiministeriet, Danmarks miljundersgelser. Jones, P.M., 1995. Road pricing: the public viewpoint. In: Johansson, B., Mattsson, L.-G. (Eds.), Road Pricing: Theory, Empirical Assessment, and Policy. Kluwer, Dordrecht, pp. 159179. Joreskog, K.G., Sorbom, D., 1988. LISREL7: A Guide to the Program and Applications, 2nd ed. SPSS, Chicago. Jovicic, G., Hansen, C.O., 2003. A passenger travel demand model for Copenhagen. Transportation Research Part A: Policy and Practice 37, 333349. Kim, Y., Kim, T.-Y., Heo, E., 2003. Bayesian estimation of multinomial probit models of work trip choice. Transportation 30, 351365. Lorenz, F.O., Conger, R.D., Simons, R.L., Whitbeck, L.B., 1995. The eects of unequal covariances and reliabilities on contemporaneous inference: the case of hostility and marital happiness. Journal of Marriage and the Family 57, 10491064. Mackenzie, J.J., 1997. Driving the road to sustainable ground transportation. In: Dower, R., Ditz, D., Faeth, P., Johnson, N., Kozlo, K., Mackenzie, J.J. (Eds.), Frontiers of Sustainability. Island Press, D.C. Washington, pp. 121190. Magelund, L. (1997). Valg af transportmiddel i storbyenbil og kollektiv transport i bolig-arbejdsrejsen (Choice of travel mode in the citycar and public transportation for commuting) (notat No. 97-03). Kbenhavn: Transportradet og HT. Milj-og Energiministeriet. (1999). Natur-og miljpolitisk redegrelse 1999. Kbenhavn: Milj-og Energiministeriet. Mischel, W., 1968. Personality and Assessment. Wiley, New York. Mller, B.T., Thgersen, J., in press. Car-use habits: an obstacle to the use of public transportation? In: Jensen-Butler, C., Madsen, B., Nielsen, O.A., Sloth, B. (Eds.), Road Pricing, The Economy, and the Environment. Amsterdam: Elsevier. Naturvardsverket, 1996. Att miljoanpassa Sveriges transportsystemen scenariostudie (Adapting Swedens transport system to the environment a scenario study) (rapport). MaTs, Stockholm. OECD. 1996. Towards sustainable transportation. The Vancouver conference. Paris: OECD. OFallon, C., Sullivan, C., Hensher, D.A., 2004. Constraints aecting mode choices by morning car commuters. Transport Policy 11, 17 29. Olander, F., Thgersen, J., 1995. Understanding of consumer behaviour as a prerequisite for environmental protection. Journal of Consumer Policy 18, 317357. Olsen, M.E., 1983. Public acceptance of consumer energy conservation strategies. Journal of Economic Psychology 4, 183196. Ouellette, J.A., Wood, W., 1998. Habit and intention in everyday life: the multiple processes by which past behavior predicts future behavior. Psychological Bulletin 124, 5474. Peter, J.P., Olson, J.C., Grunert, K.G., 1999. Consumer Behavior and Marketing Strategy, European edition. McGraw Hill, London. Rogosa, D., 1980. A critique of cross-lagged correlation. Psychological Bulletin 88, 245258. Ronis, D.L., Yates, J.F., Kirscht, J.P., 1989. Attitudes, decisions, and habits as determinants of repeated behavior. In: Pratkanis, A.R., Breckler, S.J., Greenwald, A.G. (Eds.), Attitude Structure and Function. Lawrence Erlbaum, Hillsdale, N.J., pp. 213239. Schlag, B., Teubel, U., 1997. Public acceptability of transport pricing. IATSS Research 21 (2), 134142. Schnabel, K., 1996. Latent dierence models as alternatives to modeling residuals of cross-lagged eects. In: Engel, U., Reinecke, J. (Eds.), Analysis of Change. Advanced Techniques in Panel Data Analysis. Walter de Gruyter, Berlin, pp. 253278. Schwartz, S.H., 1978. Temporal instability as a moderator of the attitudebehavior relationship. Journal of Personality and Social Psychology 36, 715724. Staats, H., Harland, P., Wilke, H.A.M., 2004. Eecting durable change. A team approach to improve environmental behavior in the household. Environment and Behavior 36, 341367. Steg, L., Geurs, K., Ras, M., 2001. The eects of motivational factors on car use: a multidisciplinary modelling approach. Transportation Research Part A: Policy and Practice 35, 789806. Steg, L., Vlek, C., 1997. Getting out of your car: the role of problem awareness in willingness-to-change car use and in evaluating relevant policy measures. In: Rothengatter, J.A., Vaya, E.C. (Eds.), Trac & Transport Psychology. Elsevier science publisher, Amsterdam, pp. 111. Steg, L., Vlek, C., Rooijers, T., 1995. Private car mobility. Problem awareness, willingness to change, and policy evaluation: a national interview study among Dutch car users. In: Zwerver, S., van Rompaey, R.S.A.R., Kok, M.T.J., Berk, M.M. (Eds.), Climate Change Research: Evaluation and Policy Implications, Vol. B. Elsevier, Amsterdam, pp. 11731176. Stern, P.C., 1999. Information, incentives, and proenvironmental consumer behavior. Journal of Consumer Policy 22, 461478. Stern, P.C., Oskamp, S., 1987. Managing scarce environmental resources. In: Stokols, D., Altman, I. (Eds.), Handbook of Environmental Psychology, Vol. 2. Wiley, New York, pp. 10431088.

638

J. Thgersen / Transportation Research Part A 40 (2006) 621638

Tertoolen, G., Kreveld, D.V., Verstraten, B., 1998. Psychological resistance against attempts to reduce private car use. Transportation Research Part A: Policy and Practice 32, 171181. Tolba, M.K., El-Kholy, O.A. (Eds.), 1992. The World Environment 19721992. Two Decades of Challenge. Chapman & Hall on behalf of UNEP, London. Van Raaij, W.F., 2002. Stages of behavioural change: motivation, ability and opportunity. In: Bartels, G., Nelissen, W. (Eds.), Marketing for Sustainability. Towards Transactional Policy-making. IOS Press, Amsterdam, pp. 321333. Van Raaij, W.F., Verhallen, T.M.M., 1983. A behavioural model of residential energy use. Journal of Economic Psychology 3, 3963. Verplanken, B., Aarts, H., 1999. Habit, attitude, and planned behaviour: is habit an empty construct or an interesting case of goal-directed automaticity? European Review of Social Psychology 10, 101134. Verplanken, B., Aarts, H., Knippenberg, A.v., Knippenberg, C.v., 1994. Attitude versus general habit: antecedents of travel mode choice. Journal of Applied Social Psychology 24, 285300. Verplanken, B., Aarts, H., Knippenberg, A.v., Moonen, A., 1998. Habit versus planned behavior: a eld experiment. British Journal of Social Psychology 37, 111128. Verplanken, B., Aarts, H., Van Knippenberg, A., 1997. Habit, information acquisition, and the process of making travel mode choices. European Journal of Social Psychology 27, 539560. Verplanken, B., Orbell, S., 2003. Reections on past behaviour: a self-report index of habit strength. Journal of Applied Social Psychology 33, 13131330. Wood, W., Quinn, J.M., Kashy, D.A., 2002. Habit in everyday life: thought, emotion, and action. Journal of Personality and Social Psychology 83, 12811297.

You might also like