Changing commuters’ behavior using rewards: A study of rush-hour avoidance

Eran Ben-Elia*, Centre for Transport and Society

Faculty of Environment and TechnologyUniversity of the West of England

Frenchay Campus, Bristol, BS16 1QY, United Kingdomeran.ben-elia@u we.ac.uk

Dick EttemaUrban and Regional research centre Utrecht

Faculty of GeosciencesUtrecht University

P.O. Box 801153508 TC, Utrecht, The Netherlands

d.ettema@geo.uu.nl

* (corresponding author)

Key WordsAttitudes, behavior-change, congestion, habitual behavior, information, motivation, reward.

AbstractIn a 13-week field study conducted in The Netherlands, participants were provided with daily rewards – monetary and non-monetary, in order to encourage them to avoid driving during the morning rush-hour. Participants could earn a reward (money or credits to keep a Smartphone handset), by driving to work earlier or later, by switching to another mode or by teleworking. The collected data, complemented with pre and post measurement surveys, were analyzed using longitudinal techniques and mixed logistic regression. The results assert that the reward is the main extrinsic motivation for discouraging rush-hour driving. The monetary reward exhibits diminishing sensitivity, whereas the Smartphone has endowment qualities. Although the reward influences the motivation to avoid the rush-hour, the choice how to change behavior is influenced by additional factors including gender and education, scheduling considerations, habitual behavior, and cognitive factors regarding attitudes and perceptions, as well as travel information availability factors.

1

1. IntroductionCongestion on urban roads throughout the European Union is increasing and is expected to worsen as the demand for trip making increases and supply of road infrastructure remains limited (European Commission, 2006a, 2006b). Loading of excess demand on the transportation system has considerable external costs such as pollution, noise and road user safety (Mayeres et al., 1996). Road overloading disrupts vehicle flow, increases the frequency of incidents and magnifies the uncertainty of travel schedules (Lomax & Schrank, 2003). Congestion is a collective, synchronic phenomenon: massive commuting at a more or less common time-frame (e.g. the morning rush-hour). Thus, shifting of commuters’ departure times to less congested times, before or after the rush-hour, change of transport mode (from car to public transport) or change of work mode (working from home), should, in theory, lead to considerable time savings, greater travel certainty and lower external costs of congestion.

Transportation demand-based solutions (e.g. road pricing, promoting modal alternatives, parking policy and land use planning policy) have been suggested to reduce congestion (Shiftan & Golani, 2005). In this respect, transport economists have been arguing for the implementation of road pricing as a first-best solution to efficiently alleviate congestion externalities (Nijkamp & Shefer, 1998; Rouwendal & Verhoef, 2006; Small & Verhoef, 2007). However, road pricing is controversial and its behavioral implications are not well understood. As suggested initially by Vickrey (1969), optimal pricing requires the design of variable tolls, making them quite complex for drivers’ comprehension (Bonsall et al., 2007; Verhoef, 2008). In addition, road pricing raises questions regarding social equity (Giuliano, 1994), fairness and public acceptability (Eriksson et al., 2006) as well as economic efficiency (Banister, 1994; Viegas, 2001).

Second-best schemes have been suggested to circumvent the difficulties in implementing first-best pricing solutions (Small & Verhoef, 2007). In The Netherlands the notion of using rewards to achieve desired outcomes in travelers’ behavior has been recently implemented in the context of the Spitsmijden1 program (Ettema et al., 2010; Knockaert et al., 2007), thus far, the largest systematic effort to analyze the potential of rewards in the field as a policy mean for changing commuter behavior. A pilot study (see section 3 for further details), involving 340 participants and lasting over 13 weeks, was organized in the second half of 2006. Its objective was to investigate, in an empirical field study, the potential impacts of rewards on commuters’ behavior during the morning rush-hour. Participants could earn a reward (money or credits to keep a Smartphone handset which also provided real-time traffic information), by driving to work earlier or later, by switching to another travel mode or by teleworking. Initial results provided evidence of substantial behavior change in response to the rewards, with commuters shifting to earlier and later departure times and more use of public transport and alternative modes or working from home (Ettema et al., 2010).

The effectiveness of rewards to reinforce a desirable behavior (e.g. identification and loyalty, work effort) is supported by a large volume of empirical evidence (Kreps, 1997; Berridge, 2001). However, in the context of travel and traffic behavior, rewards are poorly represented. Punishments and enforcement (such as policing, felony detectors, fines etc.), have been more widely documented than rewards (e.g. Rothengatter, 1992; Perry et al., 2002; Schuitema, 2003). The relative salience of negative motivational means reflects, to a large extent, a disciplinary bias. Given that travel behavior has been to the most part subjected and influenced by microeconomic theories (McFadden, 2007), it is not surprising that the behavioral rationale of many demand based strategies to manage traffic congestion is based on negative incentives that associate, through learning, the act of driving with punishments (such as tolls or increased parking costs).

1 translated literally as peak avoidance

2

The few examples where rewards have been applied in a travel context are short term studies involving the use of a temporary free bus ticket as an incentive to reduce car driving. To most parts, the results of these studies are inconclusive. For example, (Fujii et al., 2001; Fujii & Kitamura, 2003) found that an incentive did encourage a change towards reducing car driving; however the level of car driving returned to previous levels once the incentive was stopped. In contrast, (Bamberg et al., 2002; Bamberg et al., 2003), found that habitual behavior prevented substantial reductions in car use. It is not the scope of this paper to debate which policy (pricing or rewards) is more effective. However there is substantial evidence that people respond more favorably and are more motivated when rewarded rather than punished (Kahneman & Tversky, 1984; Geller, 1989). Thus, the potential of rewards as a base for traffic management policy is well worth considering if based on robust behavioral foundations.

The main aim of this paper is to comprehensively analyze and explore the changes in behavior during the course of the aforementioned pilot study and identify key factors that influenced the response to the rewards. The rest of the paper is organized as follows: Section 2 sets a number of theoretically driven research questions and hypotheses. Section 3 describes the experimental setup and methods. Results, based on a mixed logistic regression analysis are presented in section 4. A discussion is presented in section 5, followed by summary and conclusions in section 6.

2. Research questions & hypothesesSeveral key questions are postulated: First, how effective are rewards as a means for motivating travel behavior change? The literature does not provide a clear indication. One view suggests that satisfying rewards contribute to higher rates of motivation (Cameron et al., 2001; 1994). The other view propounds that rewards interfere and undermine intrinsic motivation, deflecting motivation from internal to external causes and reducing the amount of effort devoted to participate in activities (Deci, 1971; 1975; Lepper & Green, 1978). Theory of Cognitive Evaluation (TCE) further asserts that the effect of reward will depend on how it affects perceived self-determination and competence (Deci & Ryan, 1985).

Second, does the nature of the reward (monetary, in-kind) affect the willingness to change travel behavior and its tenacity? People seem more receptive to large monetary rewards compared to small ones (Gneezy & Rustichini, 2000; Gneezy, 2003). Moreover, a monetary reward might be framed as a prospective gain. According to Prospect Theory (Kahneman & Tversky, 1979), diminishing sensitivity to money can affect the perseverance of change. Participants’ apparently have greater satisfaction and motivation is higher with gifts compared to monetary rewards; however when asked, most people prefer receiving money (Shaffer & Arkes, in press). In-kind rewards may therefore encourage behavior change through a different cognitive path: the endowment effect. A Smartphone handset granted to some participants may be regarded as an uncertain endowment. An endowment is not easily relinquished, once given (Kahneman et al., 1991). The endowment effect may well motivate to change behavior just in order to avoid the loss associated with the possibility to give up a valued object. In this respect, the in-kind reward, unlike the monetary one may have affective as well as motivational properties.

Third, to what extent do personal and social characteristics (e.g. gender, education level, personal income, or household composition) sustain or diminish the potential impact of rewards? The connection between socio-economic characteristics and travel choices is well documented (e.g. Harris & Tanner, 1974; Ben-Akiva & Lerman, 1985; Axhausen & Gärling, 1992) In this respect income may well affect motivation in the case of the monetary reward. Diminishing sensitivity could suggest that participants with higher incomes might be less motivated to change behavior for a rather marginal monetary gain.

3

Fourth, do participants’ beliefs attitudes and norms influence their responsiveness to change behavior? Several studies (e.g. Gärling et al., 1998; Gärling et al., 2001) suggest attitudes towards travel alternatives, affect the choice of travel modes. The Theory of Planned Behavior (TPB) (Fishbein & Ajzen, 1975; Ajzen, 1991) suggests a positive attitude towards a certain behavior will influence a person’s intention to consciously engage in it. Rewards which create a positive attitude with a certain behavior, will contribute to this behavior being repeated. Another issue is that of personal norms that are self expectations or specific actions in specific situations (Schwartz, 1977). They refer to feelings of moral obligations to behave in a certain way (e.g. environmental friendly behavior). If a reward scheme is regarded as congruent with the personal norms and expectations, it is more likely to encourage behavior-change.

Fifth, are there situational factors (home and work-related) that affect the relative salience of rewards as means for travel behavior change (here, rush-hour avoidance)? TBP stresses the role of others’ attitudes, and the perceived situational control on influencing intentional behavior-change. If a person perceives behavior changes as difficult, the probability of repeating this action is relatively low. Scheduling constraints such as household obligations (e.g. child care, children chauffeuring) and work organization have been found to influence individuals’ responses to pricing schemes and limit their perceived effectiveness (Gärling & Fujii, 2006). Participants with child care or children chauffeuring responsibilities on one hand, or participants with inflexible working times, on the other hand, might have a limited ability to change behavior even when motivated by the reward. Conversely, the support a person gets from the household, workplace and from colleagues or friends that are also participating in a reward based scheme may well contribute to one’s own participation.

Sixth, to what extent options chosen to avoid the peak are determined by habits? In the long run habitual travel behavior, as asserted by Gärling et al. (2001) and Gärling & Axhausen (2003), is quite relevant for promoting or discouraging a behavior change different from the usual travel behavior. Theory of Interpersonal Behavior (TIB) (Triandis, 1977, 1980) stresses the role of habit in behavior. With habitual behavior, decisions are made with a lesser degree of consciousness which decreases the likelihood behavior will change in response to a change in context. Habitual behavior is less intentional more automated and script based (Ronis et al., 1989; Gärling & Garvill, 1993). Travel decisions (e.g. the drive to work) are an example of habitual behavior as repeated decisions which loose intention and become gradually routinized (Verplanken et al., 1997; Gärling et al., 1998).

Last, what is the role travel information plays in changing commuters’ behavior? Several studies point out that availability of information has significant effects on travelers’ behavior in the lab (Avineri & Prashker, 2006; Ben-Elia et al., 2008). For example in the case of route-choice, Ben-Elia & Shiftan, (2010) found real-time travel information expedites learning in unfamiliar environments and reduces initial exploration. At the same time, exposure to information is also associated with more heterogeneity in choice behavior and in risk attitudes. In this respect the Smartphone reward could well have instrumental value as it also provides access to real-time traffic information. Information might motivate change of behavior by facilitating the travel decision process and by reducing subjective effort and difficulty increasing the perceived situational control.

3. Method3.1 Participants Using license plate recognition cameras, 2,300 cars, both privately owned and leased company vehicles and traveling at least three times a week during the morning rush hour on the busy stretch of the A12 motorway (about 15 km connecting Zoetermeer to The Hague). The Dutch Department of Road Transport provided the names and addresses of the car owners and they were approached by mail with an invitation to participate in the experiment.

4

A total of 341 commuters - 221 men and 120 women – chose to participate in the experiment. Upon registration, the participants self selected one out of two types of reward. The first type of reward was an amount of money (3-7 Euros and see next subsection) for each day that the participant avoided driving during the morning rush-hour. In this case, participants were provided with a realistic estimate of how much they could earn in the course of the study. The second type comprised credits towards ultimately keeping a Smartphone (called Yeti) at the end of the experiment. 232 participants (60% men), selected a monetary reward (‘money’) and 109 (74% men) the Yeti reward. The Yeti’s market value was around € 500 at the time. All the participants were inhabitants of the town of Zoetermeer and the vast majority was working at the time in The Hague or its vicinities. They are characterized by relatively high percentage of higher education, moderate to high incomes and mostly families with children. Table 1 presents the descriptives of the participants by group.

***Table 1 about here***

3.2 ProcedureThe task and rules were communicated to the participants through the project's back office: Participation had to be voluntary. The participants were to commute at least three times a week from home to work. They had to have access to e-mail and the Internet. They were requested to complete surveys completely and timely. They were made aware that their movements by car would be recorded and had to agree to the installation of an on-board transponder in their car. In addition it was explained that only the car in which a transponder had been previously installed could be eligible for the reward. A travel log (i.e. logbook) was to be filled in daily on a personal webpage on the projects’ internet site. Participants that opted for the Yeti reward were also instructed to switch on the Smartphone during every car trip, in order to get full and easy access to real-time travel information. All communication was to be conducted via the project’s back-office which dealt with complaints or operational problems. A weekly newsletter was also sent to participants’ homes providing further information and clarifications. Participants’ earnings were shown on their personal webpage. The earnings were updated once a week according to the relevant treatment schemes. The monetary rewards were directly paid to participants’ bank accounts at the end of the working week by bank transfer.

3.3 Design Participants were instructed that they could avoid commuting during the morning rush-hour (defined between 7:30-9:30 AM) either by shifting their departure times to earlier or later times of travel, or by choosing other modes of travel (cycling, carpool, public transport), or by working from home (teleworking). The experiment ran for a period of 13 weeks. The first two weeks were without reward (pre-test). The data collected during the pre-test was used to determine participants’ reference travel behavior and subsequent assignment to reward classes. The final week (post-test) was also without rewards.

Those participants who opted for money were the subject of three consecutive reward treatments lasting 10 weeks in total: a reward of 3€ (lasting three weeks), a reward of 7€ (lasting four weeks) and a mixed reward (lasting three weeks) of up to 7€ - of which 3€ for avoiding the high peak (8:00-9:00) and an additional 4€ for avoiding also the lower peak shoulders (7:30-8:00, 9:00-9:30). A counterbalanced (blocked randomization) design was used to allocate participants randomly to 6 (that is 3! blocks) possible treatment orders (referred to as scheme). A few exceptions were applied to couples using the same vehicle. The scheme of treatments was communicated to the participants through their personal web

5

pages. Participants in possession of the Yeti could acquire credit during a period of five consecutive weeks. If they earned enough credit relative to a known threshold they could keep the Smartphone. This threshold was determined by their reward class (see below). The other five weeks were without credits but participants could still have access to traffic information. Participants were randomly divided between two schemes in relation to which of the first or second set of 5 weeks credits could be awarded. They were also made aware of their respective schemes.

Participants in possession of a Yeti also had 24 hour access to travel information via the handset during 11 weeks: the credit treatment, the no-credit treatment as well as the post-test. This information consisted of real-time travel times on the A12 motorway on the Zoetermeer – The Hague corridor and an online map showing congestion levels on other roads in the area. Information availability was not dependent on the reward itself. In contrast, participants in the money group had access to information available to all other drivers: pre-trip through internet and media and en-route from variable message signs along the motorway.

In addition to the treatments, each participant was also assigned to a reward class which determined his/her maximum eligible reward. In essence, a participant could only earn the reward as often as he/she was observed to drive in the morning peak during the pre-test. Thus, a participant who would drive in the peak three times per week in the pre-test, could only receive a reward for the third, fourth and fifth day in a week he/she avoided the peak, whereas one who drove in the peak five times per week was eligible for any working day he/she avoided the peak. This reward could be either the daily monetary reward or the threshold number of credits needed to keep the Yeti. It should be noted that retrospectively very few participants failed to meet their threshold. In order to avoid regret, it was also decided at the end of the study to allow all the participants to keep their Yeti’s. Accordingly, each participant was allocated into one of four possible reward classes. Once determined these classes were fixed throughout the rest of the experiment. The majority of participants belonged to classes A and B and the minority to classes C and D. Table 2 presents the number of participants (by gender) in each class. In both groups women are more prevalent in the classes with lower traveling frequencies. For a more detailed description of the experiment’s design see the report (in English) of Knockaert et al., (2007) also available from the authors by request.

Self selection of reward types by participants suggests by definition a quasi-experimental design. Like random experiments, quasi-experiments share the same basic principles of manipulation (cause precedes effect) and measurable associations (covariation). In contrast, causation requires more effort as compared to random assignment there are more threats to internal validity. In this we will follow the recommendations of Shadish et al., (2002) noting possible threats. An analysis of threats to internal validity is described in section 4. Lack of a control group also can contribute to validity problems. Several features in the design allow improved control and reduce possible threats. First, the pre-test / post-test design is fostered by additional measurements of stated behavior through the two surveys. Specifically threats resulting from history and novelty can be assessed by comparing between the preliminary survey and the pre-test. In addition, the measured factors from the surveys such as usual behavior, constraints and support measures, can provide relevant mediators to the observed behavior and verify if selection is a problematic issue. This is dealt in detail in Section 4. Second, norm comparisons with traffic counts on the main A12 trajectory suggest other drivers did not change behavior during this period. The sample is small enough not to have any real impact on traffic flow. Since no significant change in traffic occurred during the 13 week observed period we can assert that any difference between observed behaviors with treatments and without is likely to be related to the intervention. In retrospect, it is acknowledged that random assignment and group control would have been the preferred solution.

6

***Table 2 – about here***

3.4 Measurements Data was collected during the study in several stages. In the first stage, after volunteering (April-August, 2006), participants completed a web-based preliminary survey. This survey gathered data regarding several important pre-test factors including home to work daily travel routines, individual and household characteristics (gender, age, education level, income, family composition); work schedules (i.e. flexibility in departure from home and in starting work early/late, or ability to telework), family obligations (e.g. childcare or child chauffeuring duties), availability and use of alternative means of transport, attitudes towards alternative travel modes and regular use of travel information. The survey results can be requested from the authors.

The second stage was the actual experiment, lasting 13 weeks (of which weeks 3-12 were with rewards). It consisted of tracking participant’s observed behavior. Detection equipment using in-vehicle installed transponders and electronic vehicle identification (EVI) as well as backup road-side cameras was installed at the exits from Zoetermeer to the A12 motorway and on other routes leaving the city. This equipment allowed detecting each and every car passage during the course of the day, minimizing the ability of participants to cheat by trying to access alternative routes. In addition, participants were instructed to fill in their daily web-based logbook. They recorded whether or not they had commuted to work (and if not, why not), which means of transport they used and at what slot time they made their trip. This information was used to gain insight into situations in which the participant was not detected by the EVI.

In this paper we decided to focus on the logbook data. The main reasons were the completeness of the data which included not only car travel but also non-car travel. In addition, the logs provide a unique description of each days travel choice whereas detections could appear several times a day. Furthermore, the logbooks and detections were checked by the project’s back-office for consistency to avoid complaints and disagreements with participants regarding their eligibility for a reward. The logbook contained several entries: normal entries on working days about the choice of travel and abnormal entries (including situations like use of another car, holiday, illness, problems with the equipment etc). Only normal entries relating to working days were included in the analysis. Detection data is left for future research on dynamics of departure time choice.The third stage of the study was a posterior evaluation survey. In this survey questions were asked about the participant’s subjective experience during the course of the experiment. This dealt with their retrospective assessment of behavior adjustment (was it easy / difficult to adjust travel behavior and how much effort was involved in changing one's behavior). Other questions focused on support measures such as discussions with one's employer, colleagues and household members about flexible working times and household routines, practicing with behavior-change during the pre-test and purchasing of certain items. Questions were also asked regarding the use of travel information enabling a pre/post-test comparison that indicated a significant increase in usage of both traffic and public transport information.. Retrospective motivations to participate in the program were also inquired. One fact to be noted is that during the experiment disruptions occurred with the regional rail service and bus service replacements were not always adequately provided. In retrospect this was mentioned as causing participants some difficulty for using the public transport.

At the same time that data was collected about the participants, a survey of non-participants was also carried out. It was based on a representative sample of Zoetermeer residents,

7

regularly commuting to The Hague during the morning rush-hour, who did not participate in the experiment. The purpose was to determine whether the participants in the trial were representative of the total population of rush-hour drivers. Similar questions were put to these respondents. This analysis (see Ben-Elia & Ettema, 2009) demonstrated that although the reward is the main motivation in potentially choosing to participate in a similar reward-based scheme, lack of flexibility in daily schedules was the main reason to reject the scheme.

4. ResultsResponses appearing in the logbook were sorted into four distinctive and exclusive categories: rush-hour driving (RD), driving earlier (DE) or driving later (DL) than the rush-hour, and non-driving (ND) which included all non-auto modes of travel (public transport, cycling, car pool) as well as teleworking. Since the rewards were provided on a weekly basis, the number of rush-hour avoidances within the week, could well be correlated. Daily responses were therefore assembled to weekly average shares (i.e. proportions). Weekly averages were further aggregated to treatment averages for statistical testing purposes in the following way: in the money group five repeated measurements (pre-test, three treatment levels, post-test); and in the Yeti group four repeated measurements (pre-test, credit, non-credit, post-test). The data analysis itself consisted of two stages. In the first stage (available from the authors by request), each of the four response categories was analyzed separately using GLM-repeated measures. In the second stage a mixed logistic regression (MLR) model was estimated based on the significant factors found in the first stage.

The rationale behind using MLR was that the four response categories (RD, DE, DL, ND) attributed to each participant are in a sense a closed set of discrete choice alternatives and therefore correlated. The probability of choosing a discrete response (i.e. an alternative) is specified as the dependent variable and the independent factors explain this probability. Usually, the relationship between alternatives and explanatory factors is specified with an outcome function referred to as 'utility'. The greater the utility of an alternative is, the higher is the probability of a participant choosing it (Train, 2002). Simple logistic regression is unsuitable for analysis of repeated measurements (McFadden & Train, 2000). However the MLR model can accommodate this by specifying a panel data model (Revelt & Train 1998; Bhat, 1999;). We estimated the MLR model using the estimation program of NLOGIT 4.0 (Econometric Software Inc.,) and using share-based data with 1,000 random draws (see Train, 2000), for further details regarding drawing methods.

Formally, the utility of person n of alternative i in response t and the probability (P) of person n choosing alternative i in response t are (eq. 1, 2):

(1)

(2)

where P is the conditional probability that person n chooses alternative i out of a set of J alternatives, Y, is an indicator that i is chosen at response t, X is a vector of explanatory factors, , is a vector of fixed coefficients (including a constant), is a vector of random parameters with a distribution f (0 mean and a variance parameter ) and is a vector of independently, identically distributed (iid) Gumbel (or extreme-value type 1) error terms.

8

The MLR model's main purpose is to estimate the composite effects on all four (correlated) response categories accounting for the sequential structure of the data. Each category has its own utility specification which is linear. Factors were entered into the utilities in a sequential manner whereby, non significant factors are dropped out and significant ones remain. Random effects (i.e ‘s) are specified (for statistical restrictions only for three out of four categories) as normally distributed error terms (with zero mean and unknown variance

) to better capture differences between respondents (i.e. heterogeneity) across the observations. In addition we allowed covariation between the random effects to account for inherent correlations between the unobserved factors in the model. This is also due to the nature of the similarity between the three driving alternatives (RD, DE, DL) relative to not driving (ND).

Table 3 presents the treatments’ average measurements (also illustrated in Figure 1) and between group pre/post-test differences. Table 4, presents the coefficient estimates for the MLR model. As noted in Table 3, pre-measurement levels of RD were substantially higher than the pre-test and this difference is significant for both groups. Thus any significant change between the pre-test and other treatments is also expected to be significant relative to the pre-measurement level. Since around one third of the participants stated in the posterior survey that practicing with rush-hour avoidance during the pre-test assisted them to change their behavior, exploration with alternatives to rush-hour driving could be one way to explain the difference between stated and observed pre-test behaviors. However, since the pre-measurement is based on stated rather than observed behavior, to remain conservative we did not include it in the MLR analysis. In addition although the between-group analysis of RD, suggests that post-test differences are significant this is not confirmed in the more robust MLR. Consequently only the reward treatments were specified in the model whereby the coefficients reflect their effects relative to the pre-test.

In terms of goodness of fit the model has a final log likelihood of -1,648.12 and the rho-square is 0.22. A simple multinomial logistic regression model (without random effects) had a log likelihood of -1,678.24. The log likelihood ratio test shows this difference is significant (2 = 60.2, df=6, p<.05). Therefore, specifying the random effects structure is justifiable. The estimates of the standard deviations of the random effects, as well as their correlations are all significant.

The coefficients of the reward treatment were only found significant when specified for RD but not for the other response categories. The main effect of the reward is a decrease in RD and an increase in overall avoidance shares. Therefore we can assert that rewards have no apparent influence on the choice how to avoid the rush-hour. The model shows that all monetary treatments are significant and the sign of the coefficient is negative. The main decrease in RD is attributed to the 3€ reward whereas larger rewards have only a relatively marginal effect on the response (see Figure 1A). Thus, in the case of money, the difference in the average shares per treatment can be described as a diminishing sensitivity effect (initial GLM analysis confirms this with contrasts being not significant). The parameter size of Yeti credit is similar to that of the 3€ level. The no-credit also has a negative effect on RD however the parameter estimate is not significant. Although interactions (i.e. moderators) of reward levels and mediators were analyzed as well, no significant results were found.

Regarding the other response categories Figures 1B through 1D show the rewards increase the shares of driving at other times and not driving compared to both pre and post test levels. Diminishing sensitivity to money is also evident. The main noticeable differences are the relatively higher shares of DE and lower shares of DL for the money group compared to the Yeti group (Figure 1B, 1C). The latter is already noticeable at the pre-test levels. Yeti users reported in the posterior survey higher shares for arrangements with employers about flexible working hours as a support measure compared to the money group (Table 3). Therefore, it is possible that this allowed them greater flexibility in their behavior during the pre-test and the rest of the experiment. As noted exploration also seems to be an important factor in pre-test behavior.

9

Mediators (between-subjects factors) included the design related factors (reward class and treatment scheme), and factors relating to the participants’ stated behavior derived from the two surveys. First, as neither the treatment scheme nor any of its interactions are significant we can conclude that the order of treatments had no effect on behavior. Therefore the order effect is discarded from the final model in Table 4. Second, among socio-demographic characteristics gender has a marginally significant effect on RD (p≈0.1) suggesting men tend to change behavior more than women. In the case of money, higher education has a significant and negative effect on DE. A possible explanation is that education as a proxy for income could well be masking an income effect. However testing of moderation by income is not possible due to the small groups involved and consequent loss of statistical power.

Third, we find that factors relating to habitual behavior have significant results. The reward class, which relates to pre-test levels of driving at the rush-hour, has a negative association with behavior change. It was found that moderating the class effect by group proved significant. Participants, in both groups, associated with classes A, B (2.5 - 5 rush-hour trips at pre-test) were more likely to continue driving during the rush-hour compared to classes C and D (0-2.5 trips). In addition, the class coefficient for money is slightly larger than that of Yeti. The usual departure time has a negative association with DE: i.e. the earlier is the usual departure time - the more probable is a change of behavior by driving earlier. One may argue that similar factors that affect driving early in the non rewarded situation (such as household obligations) will still be at play during the rewarded period. The preferred start of work time, a likely proxy for the preferred arrival time, has a similar negative effect on DE but also a positive effect on DL. That means that participants driving later are those that are more accustomed to depart later in usual circumstances. Finally, the use of other modes for commuting has a positive effect on not driving. Fourth, concerning scheduling flexibility and constraints, a number of factors have been found to affect change of behavior. Child chauffeuring is positively associated with RD. Other constraints on early departure, such as childcare responsibilities, were not found significant. Conversely, participants who stated they had support from their employers with arranging flexible working times are less likely to drive during the rush-hour. These results demonstrate the relevancy of constraints and support measures as important factors that determine the probability to change behavior. The number of days (per week) that starting work late is possible has a positive effect on DL, a finding that suggests that participants with more flexible working schedules are more likely to drive later. Similarly but with a marginally significant positive effect (p<0.1), the ability to telework encourages to drive later.

Fifth, several stated experiences during the course of the experiment were found significant. In the case of money, the parameter for ‘practicing with behavior change’ is marginally significant (p<0.1); this suggests, participants in the money group who reported practicing with avoidance behavior during the pre-test were somewhat more likely to have changed behavior. In contrast, participants who stated that they incurred difficulties with the regional rail service (a main alternative to driving) were less likely to have changed behavior. Similarly participants in the money group, who reported in retrospect a higher level of effort in changing behavior, were also less likely to change behavior. These results indicate that positive or negative perceptions regarding experiences can have an influence on the likelihood to change behavior.

Sixth, attitudes in relation to public transport and cycling as realistic alternatives to driving are also important. Participants with a positive attitude towards public transport are less likely to change behavior by driving at other times (the parameters for both DE and DL are negative). In contrast, participants with a positive attitude to cycling are more likely to change behavior by not driving (the coefficient for ND is positive). This result indicates the significance of attitudes towards driving alternatives in influencing change of behavior.

Finally, there are significant effects of information usage. Participants with frequenter use of traffic information are more likely to drive later. The coefficient for DL is positive. In addition,

10

participants with frequenter use of public transport information or stating they had searched for public transport alternatives to support their behavior are more likely to change behavior by not driving ( the coefficients for both these factors are positive).

*** Table 3 – about here ***

***Table 4 – about here ***

As noted in section 3, this study is compromised of a quasi-experimental design. Based on the recommendation of Shadish et al. (2002), we describe here the plausible threats to the validity of the results. Threats to statistical inferences are not discussed here as we contend that these are likely to be low given the conservative nature of the analysis method applied which guarantees proper statistical identification of the measurable effects.

The two most plausible threats in our study are selection and history. Attrition is not an issue since no dropouts occurred. Maturation is also not relevant given the short period of time that the experiment was running. The issue of selection relates to a priori differences in the money and Yeti groups which could compromise the results. In Table 5 we present a statistical comparison of the differences between the self-selected groups by the factors that are associated with the response in the MLR model. Most factors have no significant difference, but gender, chauffeuring children and arrangements with employers regarding flexible working times do. The latter which has the most significant difference was measured in the posterior survey whereas the first two factors relate to pre-measurement. To contend with the threats we estimated the effects moderated by group, specifying the MLR model with group-specific coefficients for these three factors.

Regarding gender, it was barely significant in the model (p≈0.1). We tested moderating effects for money and Yeti but this was found not to be significant. We can therefore conclude that gender (woman) has a weak negative association with avoidance behavior. Chauffeuring children is significantly and negatively associated with avoidance behavior (p=0.02). We investigated if this might be moderated by gender but given the small group involved of both men and women who have this constraint we could not identify with confidence any significant moderating effect. Moderating by group also does not reveal significant group differences in the MLR model. Therefore we can suggest that the negative effect identified for chauffeuring on peak avoidance is probable.

The threat attributed to pre-arrangements regarding flexible work times requires more attention (p<0.001). It is clear that participants in the Yeti group reported in retrospect a greater share of prearrangements (55%) compared to the money group (34%). This threat relates to both history and to contending with novelty and disruptions (i.e. construct validity) in the daily schedule. It is reasonable that participants felt the need to prepare for what some may regard as a major disruption in their routine which would carry on for several weeks. It is plausible that these arrangements which were more apparent in the Yeti group had an effect on the direction of response i.e. increasing the share of driving later. Specifying the factor for DL as well as moderating by group resulted in insignificant coefficients. Therefore it seems likely that the differences in behavior between the two groups is related to the treatments i.e. traffic information rather than prearrangements. However the confounding of the Yeti effect with prearrangements makes both explanations seem plausible.

Although not significant in terms of group differences, practicing with behavior change during the pre-test is another threat attributed to history. About a third of participants in both groups stated in retrospect that they practiced with change of behavior even without rewarding during the pre-test measurements - a plausible explanation for the lower rush-hour frequencies compared to the usual stated behavior. However, since pre-test levels also

11

determine reward eligibility (reward class), the comparison between treatment and no-treatment (pre/post test) remains relatively valid. Practicing also has a weak negative effect on rush-hour driving (p=0.07) which was only found relevant for the money group.

***Table 5 – about here ***

5. DiscussionEffectiveness of rewarding The results demonstrate, that rewarding, at least in the short run, is effective as within a short period of time of several weeks, the share of rush-hour avoidance substantially increased. Thus in concordance with motivation theories (e.g. Cameron et al., 2001) rewards do influence the motivation to avoid the rush-hour. Moreover, we also found that the decision how to exercise this change of behavior, whether by driving at other times or by changing transport or work modes seem to be determined by other factors unrelated to the type or level of reward.

Nonetheless, it is difficult to conclude from a relatively short longitudinal study about the impacts of rewards in the long run. Motivation theories suggest that if intrinsic motivation kicks in, the change of behavior is more likely to be sustained. However, we observed in the post-test, once rewards ceased, avoidance shares had dropped and participants had returned more or less to their usual behavior of rush-hour driving (as observed in the pre-test). In this respect the results are similar to those obtained by Fujii & Kitamura (2003) regarding free bus tickets. Therefore at first glance it seems the change was not sustained for most of the participants. Notwithstanding, in the posterior survey less than 15% of participants stated they had returned to their previous behavior. Unfortunately, we do not have observations to corroborate this subjective evaluation. Further research is being carried out in this respect (see section 7). We also do not posses sufficient (post-test) data to conclude about the affective qualities of the rewards apart for the fact that the vast majority of participants (in both groups) answered affirmatively to the question 'did you like the reward' in the posterior survey.

Reward type and levelsWe found that both types of reward (monetary and in-kind) have a significant and negative effect on rush-hour driving. In the case of a monetary reward, diminishing sensitivity was clearly noted. The 7€ treatment has the largest overall effect on RD; however the largest marginal effect (the derivative) is associated with the 3€ treatment. Therefore, for practical purposes, a moderate monetary reward seems to be sufficient to encourage a relatively substantial change of behavior. In the case of the Yeti reward, the main effect is the credits which had an effect similar to the 7€ reward. In this sense an in-kind reward, likely perceived as an endowment, can be just as useful as the monetary reward. However, for practical reasons, there may be difficulty in implementing an in-kind reward over a long period of time.

Though not statistically significant in comparison to pre-test levels, avoidance behavior was also apparent without valid credits. This treatment had no extrinsic reward but travel information was still accessible to Yeti users. Furthermore, it was evident that Yeti users were more likely to drive later compared to participants in the money group. Two possible explanations are possible for this different behavior. On one hand, Yeti users had higher shares regarding support provided from employers. Thus, it is possible that pre-adjustments were involved in choosing to depart later (especially during the pre-test). On the other hand, the main advantage Yeti users had over the other group was 24 hour access to travel

12

information. This leads us to suggest that the decision how to change behavior is also influenced by travel information availability (discussed later on).

Socio-demographic characteristics As noted, this is hardly a novel assumption in travel behavior studies. Gender (marginally) and education, were found to have an impact on the response to the rewards. It seems that men (mainly in the money group) are more likely to avoid the rush-hour compared to women. The lower motivation of women to avoid the rush-hour can be associated with many issues. One idea that has been suggested in social mobility studies (Palma et al., 2009) is that women are more constrained in time compared to men for various reasons, mainly household tasks and child raising obligations. Dutch women quite often leave work early in the afternoon to pick up children from nurseries (Schwanen, 2007). This limits their ability to change their schedule - e.g. to start work later even when extrinsically motivated by a reward. However, a larger sample is needed to clearly mark the causation between gender and time-use behavior.

Education had a negative effect on behavior change (driving later). Participants (in the money group) with higher education were less likely to drive later. Education is a known proxy for latent income effects. Income is regarded as a key issue determining willingness to pay for travel purposes as well as the value of travel time savings (Ben-Akiva & Lerman, 1985; Axhausen & Gärling, 1992). In the context of the money group, the significance of higher education strengthens the notion of diminishing sensitivity in relation to the monetary reward: participants with higher real income are likely to be less sensitive to a marginal monetary gain compared to participants with lower incomes. As a result motivation to avoid the rush-hour would be negatively associated with real income. Education did not appear to be a relevant factor on the behavior of Yeti users, possibly because it is cognitively and affectively appreciated as an endowment, rather than in monetary (how much it’s worth) value.

Scheduling This is a new territory of travel behavior research, lately identified by Gärling and Fujii (2006). The results suggest that behavior change and more so the choice of behavior change is associated with the ability or disability to change daily schedules. Both home related and work related flexibilities are relevant. Family obligations, such as children chauffeuring - a constraint associate positively with rush-hour driving - make it more difficult for parents to change travel behavior. The ability to accommodate a flexible schedule and the support provided by others are also significant factors. Participants that could start working later or could telework were more likely to drive later. Participants reporting to have received support from their employer with arranging flexible working times were also less likely to drive in the rush-hour. We see these results as supporting evidence that flexibility, especially at the work place is a key issue in promoting changes in travel behavior. Contrary to our expectations, home-related support measures such as household arrangements did not have a significant effect on behavior-change. A possible explanation to the effects of scheduling is the extent of control over one’s actions and their outcomes. The Theory of Planned Behavior (Ajzen, 1991) suggests perceived situational control is a key factor in encouraging a conscience behavior-change. Thus flexibility in time-use promotes a sense of self confidence and ability to contend with the schedule’s change.

Habitual behavior, experience and attitudes As suggested by Theory of Interpersonal Behavior (Triandis, 1977, 1980) we found that factors relating to habitual behavior play an important role in the choice how to change behavior. This corroborates findings from other studies (e.g. Gärling et al., 2001; Gärling & Axhausen, 2003). The effect of habitual behavior is well manifested in the significance of the reward class, usual departure time, the preferred start of work time (in the case of shifting

13

driving times) as well as the use of other modes for commuting purposes (in the case of switching mode). Participants with higher rush-hour commute frequencies during the pre-test (reward class A, B) were relatively less likely to avoid the rush-hour compared to participants with lower rush-hour frequencies (class C, D). Two potential explanations are put forward. First, in terms of effort, one could argue that a similar relative response demands more rush-hour avoidances from frequent rush-hour drivers than from less frequent rush-hour drivers. Hence, the effort involved is higher for high frequency drivers. This is in line with Garling et al., (2004) and Cao and Mokhtarian (2005), who found that travelers prefer low effort responses over high effort responses. A second explanation is that the added value of additional rewards depends on the amount already gained, in the sense that the marginal utility of reward decreases (i.e. diminishing sensitivity). Thus, the extra rewards gained by high frequency drivers will have a lower impact on behavior. This is in line with the idea of satisficing behavior described by Simon (1987). In the case of Yeti users, the effect of reward class is weaker. This might be related to the affective qualities of the Smartphone endowment i.e. avoiding the displeasure of having to give back the handy Smartphone encouraged avoidance. In addition, real-time travel information may have been useful in reducing perceived effort and promoting self confidence in the ability to manage with rush-hour avoidance.

It is also evident that a relation exists between the usual schedules (usual departure and arrival times) and choice of behavior-change. The usual departure time was a decisive factor affecting the choice to depart earlier whereas the preferred start of work time, a likely proxy for the preferred arrival time, had a significant influence on both driving earlier and later. Furthermore, previous experience using other transport modes was an important contributor to the choice not to drive. That is, familiarity with an alternative seems to increase intrinsic motivation. It appears that the choice of behavior-change is closely related to the perceived gap between the usual behavior and the required change – the smaller the gap the more likely is that the change will be exercised. One may argue that alternatives that are more similar to the current behavior, will better meet the travelers’ preferences with respect to characteristics of the travel mode and timing.

Contrary to the usual behavior, a behavior-change requires gaining of knowledge through exploration and reinforced learning about the new situation. It is suggested that exploration had an important role during the pre-test. Practicing avoidance behavior during the pre-test was reported by almost a third of the participants( in both groups) and it is one explanation for the dramatic drop in the pre-test shares of rush-hour driving compared to the usual (stated) behavior recorded in the preliminary survey. This factor was also found (albeit weakly) to increase the likelihood of decreasing rush-hour driving for the money group. Recent findings in the context of route-choice behavior suggest information expedites learning in the short-run whereas lack of information requires greater effort devoted to exploration and learning (Ben-Elia et al., 2008; Ben-Elia & Shiftan, 2010). It is plausible that the pre-test was devoted by participants for information acquisition.

Attitudes have been recently gaining attention in travel behavior studies (Gärling & Axhausen, 2003). Moreover, perceptions and attitudes have been the focus of invigorating attempts to improve choice modeling (Walker, 2001; Cherchi, 2009). As suggested by TPB, attitudes and personal norms are significant factors in encouraging or discouraging a conscious decision to change behavior. Our results support this assertion in that participants’ attitudes regarding alternative modes were a key factor in determining the choice of avoidance behavior. Positive attitudes (defined as regarding a travel mode as a realistic alternative), regarding public transport and cycling, discouraged driving (including at off-peak periods) and encouraged mode switch away from the car. Conversely, perceptions regarding the (high) effort involved in changing behavior decreased the likelihood of changing behavior and were positively associated with rush-hour driving. We could not find real support for the relevance of personal norms in the decision to avoid the rush-hour.

14

Information availabilitySeveral studies confirm the key role that availability of travel information has on promoting sensible travel choices (Mahmassani & Liu, 1999; Srinivasan & Mahamassani, 2003). Our results sustain this association in three ways. First, there appear to be significant between-group differences in the behavior change. Yeti users had relatively higher shares of driving later compared to higher shares of driving earlier in the money group. Yeti users’ main advantage was the real-time access to travel information whereas the participants in the money group had to search for the same information i.e. it involved more effort. However, as noted this result could also be confounded by prior arrangements with employers over flexible working times which were more dominant with the Yeti group. Hence, we cannot be certain if the different response is attributed to the Yeti treatment or related to prearrangements which facilitated driving later. However, the information explanation is strengthened by the fact that moderating this effect by group was not significant.

Second, access of travel information, mainly traffic but also public transport information, intensified during the course of the experiment (pre/post-test comparison). Thus, decision-making in a changed environment apparently increased the need for information about the outcomes of alternatives. Third, information availability is positively associated with not driving or driving later. Participants who frequently accessed public transport information and who were actively perusing information over public transport connections were more likely to avoid driving altogether. In addition, participants with higher frequency of accessing traffic information where more likely to choose driving later. It seems therefore that information acquisition and choice of avoidance behavior are clearly related. However causality here is uncertain as participants could also increase information acquisition for the alternative they found is best.

6. Summary and conclusions The main conclusion regarding the use of rewards in encouraging commuters to change behavior is that it actually works. Rewards are effective extrinsic motivators for travel behavior change - here rush-hour avoidance. The monetary reward was likely perceived as a gain with diminishing sensitivity, whereas the Yeti should be regarded as an in-kind reward which had added endowment and instrumental qualities. The rewards were able to sustain the behavioral change throughout the experiment. Nonetheless, it is still an open question whether the change would be sustained in the long run and without rewards. We do not have enough post-test observations to provide an answer apart from subjective assessments by the participants. A second conclusion that can be drawn from this research is that the reward influences the magnitude of change – an increase or decrease in rush-hour avoidance. However the choice how to avoid – driving at other times, switching to another mode of transport or working from home, is determined by external factors relating to participants’ personal and social characteristics, scheduling flexibility, history and information availability, Although already of some interest to the travel behavior research community these issues deserve further attention in future research.

As a closing remark, following the success of the current study, application of reward-based schemes is now taking place across The Netherlands. Although some concern, based on traffic simulation models, indicated that too many people might start changing their schedules to gain a reward (Bliemer & van Amelsfort, 2008), the evidence in the field does not support this claim. Their effectiveness in mitigating congestion, especially in situations involving temporary road maintenance or lane closures has been verified (Bliemer et al., 2009). A recent survey of firms also has shown positive attitude amongst employers towards the reward scheme (Vonk Noordegraaf & Annema, 2009). So far, the majority of the Dutch public (apart for the public transport users who are ineligible and consequently grumbling) and the government are quite content with the results. However as recently published in the

15

media the (last) government also wanted to advance a punishment policy through universal kilometer road charging – a decision that stresses the importance of well-informed, evidence-based, as well as behaviorally-sound public policy.

AcknowledgmentsThis study was undertaken as part of the Spitsmijden project, which was funded by Transumo, the Ministry of Transport in the Netherlands, Bereik, RDW, NS, Rabobank, ARS T&TT, OC Mobility Coaching, Vrije Universiteit Amsterdam, TU Delft, Universiteit Utrecht. The modeling framework was discussed in the 5th Discrete Choice Modeling Workshop organized at EPFL (Lausanne, Swizterland) in August, 2009. The comprehensive comments and suggestions of two anonymous reviewers are very highly appreciated.

ReferencesAjzen, I. (1991), "The theory of planned behavior." Organizational Behavior and Human Decision Processes,

50, 179-211.

Avineri, E., Prashker, J., N. (2006), "The impact of travel time information on travelers’ learning under uncertainty." Transportation, 33 (4), 393-408.

Axhausen K & Gärling (1992) Activity-based approaches to travel analysis: conceptual frameworks, models and research problems. Transport Reviews 12: 324–341.

Bamberg, S., Ajzen, I., Schmidt, P. (2002), "Choice of travel mode in the Theory of Planned Behavior: The Roles of past behavior, habit, and reasoned action." Basic and Applied Social Psychology, 25, 175 - 187.

Bamberg, S., Rolle, D., Weber, C. (2003), "Does habitual car use not lead to more resistance to change of travel mode." Transportation, 30, 97-108.

Banister, D. (1994), Equity and Acceptability Questions in Internalising the Social Costs of Transport, Internalising the Social Costs of Transport, OECD, Paris.

Ben-Akiva, M. Lerman, S. R (1985), Discrete Choice Analysis: Theory and Application to Travel Demand, Cambridge, MA: MIT Press.

Ben-Elia, E., Erev, I., Shiftan, Y. (2008), " The combined effect of information and experience on drivers’ route-choice behavior." Transportation, 35 (2), 165-177.

Ben-Elia, E., Ettema, D. (2009), "Carrots versus sticks: Rewarding commuters for avoiding the rush-hour - a study of willingness to participate." Transport Policy, 16, 68-76.

Ben-Elia, E., Shiftan, Y. (2010), "Which road do I take? A learning based model of route-choice and real-time information." Transportation Research A (44), 249-264.

Berridge, K. C. (2001), "Rewarding learning: reinforcement, incentives and expectations." The Psychology of Learning and Motivation, D. L. Medin, ed., Academic Press.

Bhat, C. (1999), "An analysis of evening commute stop-making behavior using repeated choice observation from a multi-day survey." Transportation Research B., 33, 495–510.

Bliemer, M. C. J., van Amelsfort, D. H. "Rewarding instead of charging road users: a model case study investigating effects on traffic conditions." paper presented at 3rd Kuhmo-Nectar Summer School and Conference, Amsterdam, The Netherlands.

Bonsall, P., Shires, J., Maule, J., Matthews, B., Beale, J. (2007), "Responses to complex pricing signals: Theory, evidence and implications for road pricing." Transportation Research A, 41 (7), 672-683.

Cameron, J., Banko, K. M., Pierce, W. D. (2001), "Pervasive negative effects of rewards on intrinsic motivation: Te myth continues." The Behavior Analyst, 24, 1-44.

Cameron, J., Pierce, W. D. (1994), "Reinforcement, reward and intrinsic motivation: A meta-analysis." Review of Educational Research, 64, 363-423.

16

Cao, X., Mokhtarian, P.L. (2005), How do individuals adapt their personal travel? Objective and subjective influences on the consideration of travel-related strategies for San Francisco Bay Area commuters, Transport Policy, 12 (4), 291-302.

Cherchi, E. "Modelling individual preferences: State of the art, recent advances and future directions." paper presented at 12th International Conference on Travel Behaviour and Research, Jaipur.

Deci, E. L. (1971), "Effects of externally mediated rewards on intrinsic motivation." Journa of Personality and Social Psychology, 18, 105-115.

Deci, E. L. (1975), Intrinsic motivation, Plenum Press, New York.

Deci, E. L., Ryan, R. (1985), Intrinsic motivation and self determination in human behavior, Plenum Press, New York.

Eriksson, L., Garvill, J., Nordlund, A. M. (2006), "Acceptability of travel demand management measures: The importance of problem awareness, personal norm, freedom, and fairness." Journal of Environmental Psychology, 26, 15-26.

Ettema, D., Knockaert, J., Verhoef, E. "Using Incentives as Traffic Management Tool: Empirical Results of the “Peak Avoidance” Experiment.", Transportation Letters, 2, 39-51.

European Commission. (2006a), Keep Europe moving – Sustainable mobility for our continent. (Mid-term review of the European Commission’s 2001 Transport White Paper), Communication from the Commission, European Commission, COM(2006)314.

European Commission. (2006b), Raising Awareness of ICT for Smarter, Safer and Cleaner Vehicles, Communication from the Commission, European Commission, COM(2006)59.

Fishbein, M., Ajzen, I. (1975), Belief, Attitude, Intention and Behaviour: An Introduction to Theory and Research, Addison-Wesley, Reading, MA.

Fujii, S., Gärling, T., Kitamura, R. (2001), "Changes in drivers' perceptions and use of public transport during a freeway closure: Effects of temporary structural change on cooperation in a real-life social dilemma." Environment and Behaviour, 33, 796-808.

Fujii, S., Kitamura, R. (2003), "What does a one-month free bus ticket do to habitual? An experimental analysis of habit and attitude change ", Transportation, 30, 81-95.

Gärling, T., Axhausen, K. W. (2003), "Introduction: Habitual travel choice." Transportation, 30, 1-11.

Gärling, T., Fujii, S. (2006), "Travel behavior modification: Theories, methods and programs." paper presented at 11th IATBR conference, Kyoto, Japan.

Gärling , T., Fujii, S., Boe, O. (2001), "Empirical tests of a model of determinants of script-based driving choices." Tranportation Research F, 4, 89-102.

Gärling, T., Garvill, J. (1993), "Psychological explanations of participants in everyday activities." Behaviour and Environment: Psychological and Geographical Approaches, T. Gärling and R. G. Golledge, eds., Elsevier, Amsterdam.

Gärling , T., Gillholm, R., Gärling , A. (1998), "Reintroducing attitude theory in travel behaviour research: The validity of an interactive interview procedure to predict car use." Transportation, 25, 147-167.

Gärling, T., Jakobsson, C., Loukopoulos, P., Fujii, S. (2004), Adaptation of Private Car Use in Response to Travel Demand Management Measures: Potential Roles of Intelligent Transportation Systems, Journal of Intelligent Transportation Systems Technology, Planning, and Operations, 8 (4), 189-194.

Geller, E. (1989), "Applied behavior analysis and social marketing: An integration for environmental preservation." Journal of Social Issues, 45, 17-36.

Giuliano, G. (1994), Equity and fairness considerations of congestion pricing, in Special Report 242: Curbing gridlock: Peak-period fees to relieve traffic congestion. vol 2., Transportation Research Board, Washington DC.

Gneezy, U. (2003), The W effect of incentives, working paper, University of Chicago.

17

Gneezy, U., Rustichini, A. (2000), "par enough or don't pay at all." Quarterly Journal of Economics, 115, 791-810.

Harris, A. J., Tanner, J. C. (1974). Transport demand models based on personal characteristics, in D. J. Buckley (ed.), Transportation and Traffic Theory, Elsevier, Amsterdam.

Kahneman, D., Tversky, A. (1979), "Prospect theory – an analysis of decision under risk." Econometrica, 47 (2), 263-291.

Kahneman, D., Tversky, A. (1984), "Choices values and frames." American Psychologist, 39 (4), 341-350.

D. Kahneman, J. Knetsch and R. H. Thaler (1991), The Endowment Effect, Loss Aversion, and Status Quo Bias: Anomalies, Journal of Economic Perspectives, 5, 193-206.

Knockaert, J., Bliemer, M., Ettema, D., Joksimovic, D., Mulder, A., Rouwendal, J., van Amelsfort, D. (2007), Experimental design and modelling Spitsmijden, Bureau Spitsmijden, The Hague.

Kreps, D. (1997), "Intrinsic motivation and extrinsic incentives." American Economic Review Papers and Proceedings, 87, 359-364.

Lepper, M. R., Green, D. (1978), The hidden costs of reward: New perspectives in the psychology of human motivation, John Wiley and Sons, NJ.

Lomax, T., Schrank, D. (2003), 2002 Urban Mobility Study, Texas Transportation Institute.

Mahmassani, H. S., Liu, Y.-H. (1999), "Dynamics of commuting decisions behavior under advanced traveler information systems." Transportation Research. C, 7 (2), 91-107.

Mayeres, I., Ochelen, S., Proost, S. (1996), "Marginal and external costs of urban transport." Transportation Research D, 1 (2), 111-130.

McFadden, D. (2007), "The behavioral science of transportation ", Transport Policy, 14, 269-274.

McFadden, D., Train, K. (2000), "Mixed MNL models of discrete response." Journal of Applied Econometrics, 15, 447-470.

Nijkamp, P., Shefer, D. (1998), Urban Transport Externalities and Pigouvian Taxes: A Network Approach, Edward Elgar Publishing, Cheltenham.

Palma, P. A., González , C. D., Carrasco, S. L. "Gender differences in time use and mobility: Time poverty and dual consumption." paper presented at Time Use Observatory, Santiago, Chile.

Perry, O., I. Erev, E. Haruvy. (2002), "Frequent probabilistic punishment in law enforcement ", Economics of Governance 3,71-86.

Revelt, D., Train , K. (1998), "Mixed MNL models for discrete response." Review of Economics and Statstics, 80, 647-657.

Ronis, D. L., Yates, J. F., Kirscht, J. P. (1989), "Attitudes, decisions and habits as determinants of repeated behaviour." Attitude Structure and Function, A. R. Pratkanis, S. J. Breckler, and A. G. Greenwald, eds., Erlbaum, Hillsdale, NJ.

Rothengatter, T. (1992), "The effects of police surveillance and law enforcement on driver behaviour ", Current Psychological Reviews, 2, 349-358.

Rouwendal, J., Verhoef, E. T. (2006), "Basic economic principles of road pricing: From theory to applications." Transport Policy, 13 (2), 106-114.

Schuitema, G. (2003), "Pricing policies in transport." Human decision making and environmental perception: Understanding and assisting human decision making in real-life settings L. Hendrickx, W. Jager, and L. Steg, eds., University of Groningen, Groningen.

Schwanen, T. (2007), "Gender Differences in Chauffeuring Children among Dual-Earner Families." The Professional Geographer, 59, 447-462.

Schwartz S. (1977), Normative influences on altruism, in : Berkowitz L. (ed.), Advances in Experimental Social Psychology, vol. 10, Academic Press, London.

18

Shadish, W. R., Cook, T. D., Campbell D.T., (2002), Experimental and Quasi-Experimental Designs for Generalized Causal Inference, Wadsworth, Belmont.

Shaffer, V. A., Arkes, H. R. (in press), "Preference reversals in evaluations of cash versus non-cash incentives." Journal of Economic Psychology.

Shiftan, Y., Golani, A. (2005), "Effect of Auto Restraint on Travel Behavior ", Transportation Research Record, 1932, 156-163.

Simon H. A. (1987). Satisficing. In The New Palgrave: A Dictionary of Economics, Eatwell J, Miligate M, Newman P (eds.). Vol. 4: Stockton Press: New York; 243-245.

Small, K. A., Verhoef, E. T. (2007), The Economics of Urban Transportation, Routledge, London.

Srinivasan, K. K., Mahamassani, H. S. (2003), "Analyzing heterogeneity and unobserved structural effects in route-switching behavior under ATIS: a dynamic kernel logit formulation." Transportation Research B., 37, 793-814.

Train, K. (2002), Discrete Choice Methods with Simulation, Cambridge University Press, Cambridge, MA.

Triandis, H. C. (1977). Interpersonal behavior. Monterey, CA: Brooks/Cole.

Triandis, H. C. (1980). Values, attitudes, and interpersonal behavior. In H. E. Howe & M. M. Page (Eds.), Nebraska Symposium on Motivation 1979 (pp. 195-259). Lincoln: University of Nebraska Press.

Verhoef, E. (2008), "Road Transport Pricing: Motivation, Objectives and Design from an Economic Perspective." Pricing In Road Transport: A Multi-Disciplinary Perspective, E. Verhoef, M. Bliemer, L. Steg, and B. van Wee, eds., Edward Elgar Publishing, London.

Verplanken, B., Aarts, H., van Knippenberg, A. (1997), "Attitude versus general habit: Antecedents of travel mode choice." Journal of Applied Social Psychology, 24, 285-300.

Vickrey, W. S. (1969), "Congestion theory and transport investment." American Economic Review, 59 (251-261).

Viegas, J. M. (2001), "Making urban road pricing acceptable and effective: searching for quality and equity in urban mobility." Transport Policy, 8 (4), 289-294.

Vonk Noordegraaf, D. M., Annema, J. A. "Employer attitude to rush-hour avoidance." proceedings of the European Transport Conference, Noordwijkerhout

Walker, J. L. (2001), "Extended Discrete Choice Models: Integrated Framework, Flexible Error Structures, and Latent Variables," Ph. Dthesis, Massachusetts Institute of Technology.

19

Table 1: Participants’ characteristicsMoney Yeti

N % N %

Genderman 140 60.3 81 74.3woman 92 39.7 28 25.7

Education level

Secondary 24 10.4 9 8.3Low vocational 9 3.9 5 4.6Middle vocational 64 27.7 36 33.3Higher education 134 58.0 58 53.7

Income €(net person/month)

<1500 12 5.2 6 5.61500-3000 98 42.4 38 35.23000-4500 57 24.7 40 37.0>4500 11 4.8 3 2.8didn't answer 53 22.9 21 19.4

Household composition

single 35 15.2 10 9.3partner no kids 61 26.4 20 18.5partner + kids 118 51.1 73 67.6single parent 13 5.6 3 2.8other 4 1.7 2 1.9

Cars / Household1 120 51.9 45 41.72 103 44.6 59 54.63+ 8 3.5 4 3.7

Age (years)

Mean 41.3   44.8  Median 42.5   45  Per.25 34   37  Per.75 49   51  

Table 1: Reward classes* by gender and reward type (group)

Money Yeti

A B C D A B C D Thresholds** 5 4 2 1 15 20 23 25N

Men 83 33 13 11 34 27 13 7 221 62% 54% 57% 79% 72% 87% 59% 78% 65%Women 51 28 10 3 13 4 9 2 120

38% 46% 44% 21% 28% 13% 41% 22% 35%Total 134 61 23 14 47 31 22 9 341

* A: 3.5-5, B:2/5-3.5, C: 1-2.5, D: 0-1 trips/week. ** Money: maximum number of eligible rewards per week; Yeti: number of credits at the end of 5 weeks required to keep the phone.

20

Table 1: Mean values of response and between-group differences

Resp. Measurement

Money Yeti between-group difference

N Mean s.d N Mean s.d t-stat Mann-Whitney

URush Preliminary Survey - S 232 89.9 15.44 108 89.8 13.25 0.03 12,153Hour Pre test - R1 219 48.7 38.17 107 44.0 36.69 1.06 10,395

3€ - R2 229 22.4 28.667€ - R3 231 17.7 26.703-7€ - R4 230 17.8 26.88No Credit - R5 109 31.0 31.26Credit - R6 107 15.4 21.95Post test - R7 225 47.3 38.91 101 37.6 37.64 2.22 9,808

Driving Preliminary Survey - SEarly Pre test - R1 219 22.7 33.03 107 22.0 33.06 0.20 11,489

3€ - R2 229 37.7 37.877€ - R3 231 41.8 38.053-7€ - R4 230 42.4 38.49No Credit - R5 109 24.8 33.24Credit - R6 107 33.8 37.12Post test - R7 225 24.3 35.47 101 27.7 37.57 -0.78 10,768

Driving Preliminary Survey - SLate Pre test - R1 219 10.1 22.29 107 20.3 31.32 -3.37 9,740

3€ - R2 229 17.7 27.647€ - R3 231 15.9 26.203-7€ - R4 230 15.9 27.15No Credit - R5 109 24.1 32.19Credit - R6 107 25.6 31.76Post test - R7 230 15.9 27.15 101 19.1 31.87 -1.87 10,327

Not Preliminary Survey - SDriving Pre test - R1 219 18.5 30.25 107 13.8 26.00 1.39 10,816

3€ - R2 229 22.2 30.707€ - R3 231 24.6 30.723-7€ - R4 230 23.9 32.73No Credit - R5 109 20.1 28.50Credit - R6 107 25.1 31.79Post test - R7 225 15.6 29.02 101 15.6 30.28 0.02 11,086

21

Table 1: Results of mixed logistic regression model

Alt.* Parameter Est. S.E t-test pRD Constant rush-hour driving 0.750 0.31 2.44 0.010  3€ reward -1.440 0.46 -3.12 <.001  7€ reward -1.804 0.50 -3.58 <.001  3-7€ reward -1.780 0.50 -3.53 <.001  Yeti without credit reward -0.370 0.66 -0.56 0.560  Yeti with credit reward -1.426 0.66 -2.15 0.031  Class A,B for money 1.430 0.26 5.38 <.001  Class A,B for Yeti 1.180 0.28 4.29 <.001  Gender (woman) 0.250 0.15 1.57 0.110  High effort for behavior change (for money) 1.039 0.31 3.3 <.001  Practice with avoidance during pre-test (for money) 0.36 0.20 -1.79 0.074  Arrangements with employer over flexible working time -0.325 0.15 -2.17 0.029  Constraint chauffeuring children 0.470 0.21 2.25 0.024  Problems with regional rail – would use public transport more. 0.500 0.20 2.57 0.010DE Constant driving early 15.58 1.70 9.13 <.001  Usual departure time (min.) -0.022 0.003 -6.95 <.001

Preferred start of work time (min.) -0.007 0.001 -4.49 <.001  Public transport is realistic alternative -0.670 0.23 -2.92 0.003DL Constant driving late -2.050 0.92 -2.22 0.026  Higher education for money group -0.767 0.24 -3.17 0.010  Weekly frequency of accessing traffic information 0.070 0.03 2.18 0.030  Public transport is realistic alternative -0.570 0.26 -2.39 0.030  Preferred start of work time (min.) 0.004 0.001 2.52 0.011

Number of days teleworking is possible 0.348 0.21 1.69 0.090  Number of days starting work late possible 0.220 0.05 4.08 <.001ND Weekly frequency of accessing public transport information 0.281 0.08 3.32 0.009  Cycling is realistic alternative for commuting 0.668 0.21 3.16 0.001  Seek information on Public Transport connections 0.803 0.31 2.54 0.010  Use of other modes 0.884 0.23 3.78 0.002          

s.d – driving early 1.306 0.36 3.58 <.001s.d – driving late 0.863 0.27 3.19 <.001s.d – rush-hour driving 0.744 0.18 4.10 <.001corr (RD,DE) -0.834      corr (RD,DL) -0.629      corr (DE,DL) 0.450      L(0) -2123.800      L()

-1648.120      2

0.224      Adj 2

0.207      * RD – rush hour driving, DE – driving earlier, DL – driving later, ND – not driving.

22

Table 1: Between-group differences on Response-associated factors

FactorMoney (N=231

)

Yeti (N=108

)p*

Socio-demographics Gender (woman) 40% 26%

0.01

 High education 58% 53%

0.43

Alternative modes

Other modes used for commuting 21% 19% 0.58

Public transport is realistic alternative 35% 32% 0.63

Cycling is realistic alternative 20% 14% 0.21

Schedules

Usual departure time (hour.min) 7.52 7.57 0.27

Preferred start of work time (hour.min) 8.24 8.35 0.20

Start work later (days) 3.5 3.64 0.52

Telework (days) 0.46 0.58 0.37

 Chauffeuring children duties 16% 27%

0.02

DifficultiesHigh effort perceived with changing behaviour 6% 9%

0.29

  Problems with regional rain   33% 29%

0.72

Support measures

Arrangements with employer 34% 55% <.001

Practice during pre-test 30% 25% 0.34

Search for public transport connections 13% 13% 0.97

travel information Use of traffic information (days/week) 1.35 1.83 0.08

Use of public transport information (days/week) 0.13 0.01 0.15

* chi-square test for nominal factors, t-test for interval factors

23

Figure 1: Average response shares by group by treatment

24