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

Transportation Research Part F 14 (2011) 354–368

Contents lists available at ScienceDirect

Transportation Research Part F

journal homepage: www.elsevier .com/locate / t r f

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

Eran Ben-Elia a,⇑, Dick Ettema b

a Centre for Transport and Society, Dept. of Planning and Architecture, Faculty of Environment and Technology, University of the West of England,Frenchay Campus, Bristol BS16 1QY, United Kingdomb Urban and Regional Research Centre Utrecht, Dept. of Human Geography and Planning, Faculty of Geosciences, Utrecht University, P.O. Box 80115,3508 TC Utrecht, The Netherlands

a r t i c l e i n f o

Article history:Received 27 April 2009Received in revised form 28 December 2010Accepted 8 April 2011

Keywords:AttitudesBehavior changeCongestionHabitual behaviorInformationMotivationReward

1369-8478/$ - see front matter � 2011 Elsevier Ltddoi:10.1016/j.trf.2011.04.003

⇑ Corresponding author. Tel.: +44 117 3281430.E-mail addresses: [email protected] (E. Be

a b s t r a c t

In a 13-week field study conducted in The Netherlands, participants were provided withdaily rewards – monetary and in-kind, in order to encourage them to avoid driving duringthe morning rush-hour. Participants could earn a reward (money or credits to keep aSmartphone handset), by driving to work earlier or later, by switching to another modeor by teleworking. The collected data, complemented with pre and post measurement sur-veys, were analyzed using longitudinal techniques and mixed logistic regression. Theresults assert that the reward is the main extrinsic motivation for discouraging rush-hourdriving. The monetary reward exhibits diminishing sensitivity, whereas the Smartphonehas endowment qualities. Although the reward influences the motivation to avoid therush-hour, the choice how to change behavior is influenced by additional factors includingeducation, scheduling, habitual behavior, attitudes, and travel information availability.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Congestion on urban roads throughout the European Union is increasing and is expected to worsen as the demand for tripmaking increases and supply of road infrastructure remains limited (European Commission, 2006a, 2006b). Loading of excessdemand on the transportation system has considerable external costs such as pollution, noise and road user safety (Mayeres,Ochelen, & Proost, 1996). Road overloading disrupts vehicle flow, increases the frequency of incidents and magnifies the uncer-tainty of travel schedules (Lomax & Schrank, 2003). Congestion is a collective, synchronic phenomenon: massive commuting ata more or less common time-frame (e.g. the morning rush-hour). Thus, shifting of commuters’ departure times to less congestedtimes, before or after the rush-hour, change of transport mode (from car to public transport) or change of work mode (workingfrom 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 useplanning policy) have been suggested to reduce congestion (Shiftan & Golani, 2005). In this respect, transport economistshave been arguing for the implementation of road pricing as a first-best solution to efficiently alleviate congestion external-ities (Nijkamp & Shefer, 1998; Rouwendal & Verhoef, 2006; Small & Verhoef, 2007). However, road pricing is controversialand its behavioral implications are not well understood. As suggested initially by Vickrey (1969), optimal pricing requires thedesign of variable tolls, making them quite complex for drivers’ comprehension (Bonsall, Shires, Maule, Matthews, & Beale,2007; Verhoef, 2008). In addition, road pricing raises questions regarding social equity (Giuliano, 1994), fairness and publicacceptability (Eriksson, Garvill, & Nordlund, 2006) as well as economic efficiency (Banister, 1994; Viegas, 2001).

. All rights reserved.

n-Elia), [email protected] (D. Ettema).

http://dx.doi.org/10.1016/j.trf.2011.04.003

mailto:[email protected]

mailto:[email protected]

http://dx.doi.org/10.1016/j.trf.2011.04.003

http://www.sciencedirect.com/science/journal/13698478

http://www.elsevier.com/locate/trf

E. Ben-Elia, D. Ettema / Transportation Research Part F 14 (2011) 354–368 355

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’ behaviorhas been recently implemented in the context of the Spitsmijden1 program (Ettema, Knockaert, & Verhoef, 2010; Knockaertet al., 2007), thus far, the largest systematic effort to analyze the potential of rewards in the field as a policy mean for changingcommuter behavior. A pilot study (see Section 3 for further details), involving 340 participants and lasting over 13 weeks, wasorganized in the second half of 2006. Its objective was to investigate, in an empirical field study, the potential impacts of re-wards on commuters’ behavior during the morning rush-hour. Participants could earn a reward (money or credits to keep aSmartphone handset which also provided real-time traffic information), by driving to work earlier or later, by switching to an-other 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 workingfrom home (Ettema et al., 2010).

The effectiveness of rewards to reinforce a desirable behavior (e.g. identification and loyalty, work effort) is supported bya 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, and fines), have beenmore widely documented than rewards (e.g. Rothengatter, 1992; Perry, Erev, & Haruvy, 2002; Schuitema, 2003). The relativesalience of negative motivational means reflects, to a large extent, a disciplinary bias. Given that travel behavior has been tothe most part subjected and influenced by microeconomic theories (McFadden, 2007), it is not surprising that the behavioralrationale 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).

The few examples where rewards have been applied in a travel context are short term studies involving the use of a tem-porary free bus ticket as an incentive to reduce car driving. To most parts, the results of these studies are inconclusive. Forexample, Fujii, Gärling, and Kitamura (2001) and Fujii and Kitamura (2003) found that an incentive did encourage a changetowards reducing car driving; however the level of car driving returned to previous levels once the incentive was stopped. Incontrast, Bamberg, Ajzen, and Schmidt (2002) and Bamberg, Rolle, and Weber (2003), found that habitual behavior pre-vented substantial reductions in car use. It is not the scope of this paper to debate which policy (pricing or rewards) is moreeffective. However there is substantial evidence that people respond more favorably and are more motivated when rewardedrather than punished (Kahneman & Tversky, 1984; Geller, 1989). Thus, the potential of rewards as a base for traffic manage-ment 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 theaforementione pilot study and identify key factors that influence the response to the rewards. The rest of the paper is orga-nized as follows: Section 2 sets a number of theoretically driven research questions and hypotheses. Section 3 describes theexperimental setup and methods. Results, based on a mixed logistic regression analysis are presented in Section 4. A discus-sion is presented in Section 5, followed by summary and conclusions in Section 6.

2. Research questions and hypotheses

Several 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 ofmotivation (Cameron, Banko, & Pierce, 2001; Cameron & Pierce, 1994). The other view propounds that rewards interfereand undermine intrinsic motivation, deflecting motivation from internal to external causes and reducing the amount of ef-fort devoted to participate in activities (Deci, 1971, 1975; Lepper & Green, 1978). Theory of Cognitive Evaluation (TCE) furtherasserts 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 satisfac-tion and motivation is higher with gifts compared to monetary rewards; however when asked, most people prefer receivingmoney (Shaffer & Arkes, (2009)). In-kind rewards may therefore encourage behavior change through a different cognitivepath: the endowment effect. A Smartphone handset granted to some participants may be regarded as an uncertain endow-ment. An endowment is not easily relinquished, once given (Kahneman, Knetsch, & Thaler, 1991). The endowment effectmay well motivate to change behavior just in order to avoid the emotional loss associated with the possibility to give up a val-ued 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 householdcomposition) sustain or diminish the potential impact of rewards? The connection between socio-economic characteristicsand travel choices is well documented (e.g. Harris & Tanner, 1974; Ben-Akiva & Lerman, 1985; Axhausen & Gärling, 1992). Inthis respect income may well affect motivation in the case of the monetary reward. Diminishing sensitivity could suggestthat participants with higher incomes might be less motivated to change behavior for a rather marginal monetary gain.

1 Translated literally as peak avoidance.

356 E. Ben-Elia, D. Ettema / Transportation Research Part F 14 (2011) 354–368

Fourth, do participants’ beliefs attitudes and norms influence their responsiveness to change behavior? Several studies(e.g. Gärling, Gillholm, & Gärling, 1998; Gärling, Fujii, & Boe, 2001) suggest attitudes towards travel alternatives, affectthe choice of travel modes. The Theory of Planned Behavior (TPB) (Fishbein & Ajzen, 1975; Ajzen, 1991) suggests a positiveattitude towards a certain behavior will influence a person’s intention to consciously engage in it. Rewards which createa positive attitude with a certain behavior, will contribute to this behavior being repeated. Another issue is that of personalnorms that are self expectations or specific actions in specific situations (Schwartz, 1977). They refer to feelings of moralobligations to behave in a certain way (e.g. environmental friendly behavior). If a reward scheme is regarded as congruentwith 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 travelbehavior change (here, rush-hour avoidance)? TBP stresses the role of others’ attitudes, and the perceived situational controlon influencing intentional behavior change. If a person perceives behavior changes as difficult, the probability of repeatingthis action is relatively low. Scheduling constraints such as household obligations (e.g. child care, children chauffeuring) andwork organization have been found to influence individuals’ responses to pricing schemes and limit their perceived effective-ness (Gärling & Fujii, 2006). Participants with child care or children chauffeuring responsibilities on one hand, or participantswith inflexible working times, on the other hand, might have a limited ability to change behavior even when motivated bythe reward. Conversely, the support a person gets from the household, workplace and from colleagues or friends that are alsoparticipating 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 and Axhausen (2003), is quite relevant for promoting or discouraging abehavior change different from the usual travel behavior. Theory of Interpersonal Behavior (TIB) (Triandis, 1977, 1980) stres-ses the role of habit in behavior. With habitual behavior, decisions are made with a lesser degree of consciousness whichdecreases the likelihood behavior will change in response to a change in context. Habitual behavior is less intentional moreautomated and script based (Ronis, Yates, & Kirscht, 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 (Verplan-ken, Aarts, & van Knippenberg, 1997; Gärling et al., 1998).

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

3. Method

3.1. Participants

Using license plate recognition cameras, 2300 cars, both privately owned and leased company vehicles were identifiedtraveling at least three times a week during the morning rush-hour on the busy stretch of the A12 motorway (about15 km connecting Zoetermeer to The Hague). The Dutch Department of Road Transport provided the names and addressesof the car owners and they were approached by mail with an invitation to participate in the experiment. A total of 341 com-muters – 221 men and 120 women – chose to participate in the experiment. Upon registration, the participants self selectedone out of two types of reward. The first type of reward was an amount of money (3–7 Euros and see next subsection) foreach day that the participant avoided driving during the morning rush-hour. In this case, participants were provided with arealistic estimate of how much they could earn in the course of the study. The second type comprised credits towards ulti-mately keeping a Smartphone (called Yeti) at the end of the experiment. The Yeti’s market value was around € 500 at thetime. 232 participants (60% men), selected a monetary reward (‘money’) and 109 (74% men) the Yeti reward. All the partic-ipants were inhabitants of the town of Zoetermeer (a satellite city of The Hague) and the vast majority was working at thetime in The Hague or its vicinities. They are characterized by relatively high percentage of higher education, moderate tohigh incomes and mostly families with children. Table 1 presents the descriptives of the participants by group.

The task and rules were clearly communicated to the participants: Participation had to be voluntary. The participantswere to commute at least three times a week from home to work. They had to have access to e-mail and the Internet. Theywere requested to complete surveys completely and timely. They were made aware that their movements by car would berecorded and had to agree to the installation of an on-board transponder in their car. In addition it was explained that onlythe car in which a transponder had been previously installed could be eligible for the reward. A travel log (i.e. logbook) wasto be filled in daily on a personal webpage on the projects’ internet site. Participants that opted for the Yeti reward were alsoinstructed to switch on the Smartphone during every car trip, in order to get full and easy access to real-time travel infor-mation. All communication was to be conducted via the project’s back-office which dealt with complaints or operationalproblems. A weekly newsletter was also sent to participants’ homes providing further information and clarifications.

Table 1Participants’ characteristics.

Money Yeti

N % N %

Gender Man 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 2.8Did not 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/household 1 120 51.9 45 41.72 103 44.6 59 54.63+ 8 3.5 4 3.7

Age (years) Mean 41.3 44.8Median 42.5 45Per.25 34 37Per.75 49 51


Participants’ earnings were shown on their personal webpage. The earnings were updated once a week according to the rel-evant treatment schemes. The monetary rewards were directly paid to participants’ bank accounts at the end of the workingweek by bank transfer.

3.2. Design and procedure

Participants were instructed that they could avoid commuting during the morning rush-hour (defined between 7:30 and9:30 AM) either by shifting their departure times to earlier or later times of travel, or by choosing other modes of travel (cy-cling, carpool, public transport), or by working from home (teleworking). The experiment ran for a period of 13 weeks. Thefirst 2 weeks were without reward (pre-test). The data collected during the pre-test was used to determine participants’ ref-erence 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 intotal: a reward of 3€ (lasting 3 weeks), a reward of 7€ (lasting 4 weeks) and a mixed reward (lasting 3 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 to6 (that is 3! blocks) possible treatment orders (referred to as scheme). A few exceptions were applied to couples usingthe same vehicle. The scheme of treatments was communicated to the participants through their personal web pages. Par-ticipants in possession of the Yeti could acquire credit during a period of 5 consecutive weeks. If they earned enough creditrelative to a known threshold they could keep the Smartphone. This threshold was determined by their reward class (seebelow). The other 5 weeks were without credits but participants could still have access to traffic information. Participantswere randomly divided between two schemes in relation to which of the first or second set of 5 weeks credits could beawarded. They were also made aware of their respective schemes.

Participants in possession of a Yeti also had 24 h access to travel information via the handset during 11 weeks: the credittreatment, the no-credit treatment as well as the post-test. This information consisted of real-time travel times on the A12motorway 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 toinformation available to all other drivers: pre-trip through internet and media and en-route from variable message signsalong the motorway.

In addition to the treatments, each participant was also assigned to a reward class which determined his/her maximumeligible reward. In essence, a participant could only earn the reward as often as he/she was observed to drive in the morningpeak during the pre-test. Thus, a participant who would drive in the peak three times per week in the pre-test, could onlyreceive a reward for the third, fourth and fifth day in a week he/she avoided the peak, whereas one who drove in the peak fivetimes per week was eligible for any working day he/she avoided the peak. It should be noted that retrospectively very fewparticipants failed to meet their threshold. In order to avoid regret, it was also decided at the end of the study to allow all theparticipants 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 classesA and B and the minority to classes C and D. Table 2 presents the number of participants (by gender) in each class. In bothgroups women are more prevalent in the classes with lower traveling frequencies. For a more detailed description of theexperiment’s design see the report (in English) of Knockaert et al. (2007) also available from the authors by request.

3.3. Measurements

Data was collected during the study in several stages. In the first stage, after volunteering (April–August, 2006), partic-ipants completed a web-based preliminary survey. This survey gathered data regarding several important pre-test factorsincluding 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 tele-work), 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 fromthe authors and see Ben-Elia and Ettema (2011).

The second stage was the actual experiment, lasting 13 weeks (of which weeks 3–12 were with rewards). It consisted oftracking participant’s observed behavior. Detection equipment using in-vehicle installed transponders and electronic vehicleidentification (EVI) as well as backup road-side cameras was installed at the exits from Zoetermeer to the A12 motorway andon 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 instructedto 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 insightinto 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 in-cluded not only car travel but also non-car travel. In addition, the logs provide a unique description of each days travel choicewhereas detections could appear several times a day. Furthermore, the logbooks and detections were checked by the pro-ject’s back-office for consistency to avoid complaints and disagreements with participants regarding their eligibility for a re-ward. 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 relatingto working days were included in the analysis. Detection data is left for future research on dynamics of departure timechoice.

The third stage of the study was a posterior evaluation survey. In this survey questions were asked about the participant’ssubjective experience during the course of the experiment. This dealt with their retrospective assessment of behavior adjust-ment (was it easy/difficult to adjust travel behavior and how much effort was involved in changing one’s behavior). Otherquestions focused on support measures such as discussions with one’s employer, colleagues and household members aboutflexible working times and household routines, practicing with behavior change during the pre-test and purchasing of cer-tain items. Questions were also asked regarding the use of travel information enabling a pre/post-test comparison that indi-cated a significant increase in usage of both traffic and public transport information. Retrospective motivations to participatein the program were also inquired. One fact to be noted is that during the experiment disruptions occurred with the regionalrail service and bus service replacements were not always adequately provided. In retrospect this was mentioned as causingparticipants some difficulty for using the public transport. Main findings of the survey are reported in Ben-Elia and Ettema(2011).

At the same time that data was collected about the participants, a survey of non-participants was also carried out. It wasbased on a representative sample of Zoetermeer residents, 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 repre-sentative of the total population of rush-hour drivers. Similar questions were put to these respondents. This analysis (seeBen-Elia & Ettema, 2009) demonstrated that although the reward is the main motivation in potentially choosing to partic-ipate in a similar reward-based scheme, lack of flexibility in daily schedules was the main reason to reject the scheme.

Table 2Reward classesa by gender and reward type (group).

Money Yeti

A B C D A B C D

Thresholdsb 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 12038% 46% 44% 21% 28% 13% 41% 22% 35%

Total 134 61 23 14 47 31 22 9 341

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


4. Results

Responses 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 ofrush-hour avoidances within the week, could well be correlated. Daily responses were therefore assembled to weekly averageshares (i.e. proportions). Weekly averages were further aggregated to treatment averages for statistical testing purposes in thefollowing way: in the money group five repeated measurements (pre-test, three treatment levels, post-test); and in the Yetigroup four repeated measurements (pre-test, credit, non-credit, post-test). The data analysis itself consisted of two stages. Inthe first stage (available from the authors by request), each of the four response categories was analyzed separately usingGLM-repeated measures. In the second stage a mixed logistic regression (MLR) model was estimated.

The rationale behind using MLR was that the four response categories (RD, DE, DL, ND) attributed to each participant arein 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, therelationship between alternatives and explanatory factors is specified with an outcome function referred to as ‘utility’. Thegreater the utility of an alternative is, the higher is the probability of a participant choosing it (Train, 2002). Simple logisticregression is unsuitable for analysis of repeated measurements (McFadden & Train, 2000). However the MLR model canaccommodate this by specifying a panel data model (Revelt & Train, 1998; Bhat, 1999). We estimated the MLR model usingthe estimation program of NLOGIT 4.0 (Econometric Software Inc.,) and using share-based data with 1000 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 inresponse t are:

Unit ¼ ani þ bniXnit þ enit; ani � Nð0;r2niÞ; eni � iid EV1 ð1Þ

PnðYt ¼ iÞ ¼ eaniþbniXnit

PJj¼1;i2Je

anjþbnXnjtð2Þ

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 ischosen at response t, X is a vector of explanatory factors, b, is a vector of fixed coefficients (including a constant), a is a vectorof normally distributed random coefficients with 0 mean and a variance parameter r and e is a vector of independently, iden-tically distributed (iid) Gumbel (or extreme-value type 1) error terms.

The MLR model’s main purpose is to estimate the composite effects on all four (correlated) response categories account-ing for the sequential structure of the data. Each category has its own utility specification which is linear. Factors were en-tered into the utilities in a sequential manner whereby, non significant factors are dropped out and significant ones remain.Random coefficients (i.e. a‘s) are specified (for statistical restrictions only for three out of four categories) error terms to bet-ter capture differences between respondents (i.e. heterogeneity) across the observations. In addition we allowed covariationbetween the random coefficients to account for inherent correlations between the unobserved factors in the model. This isalso 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 Fig. 1) and between-group pre/post-test dif-ferences. Table 4, presents the coefficient estimates for the MLR model. As noted in Table 3, pre-measurement levels (relatingto the preliminary survey) 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-houravoidance during the pre-test assisted them to change their behavior, exploration with alternatives to rush-hour drivingcould be one way to explain this difference. However, since the pre-measurement is based on stated rather than observedbehavior, to remain conservative we did not include it in the MLR analysis. In addition although the between-group analysisof RD, suggests that post-test differences are significant this is not confirmed in the more robust MLR. Consequently only thereward treatments were specified in the model. Here the coefficients reflect the treatments’ effects relative to the pre-test.

In terms of goodness of fit the model has a final log likelihood of �1648.12 and the rho-square is 0.22. A simple multi-nomial logistic regression model (without random effects) had a log likelihood of �1678.24. The log likelihood ratio testshows this difference is significant (v2 = 60.2, df = 6, p < .05). Therefore, specifying the random coefficients is justifiable.The estimates of the standard deviations of the random coefficients, 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 responsecategories. The main effect of the reward is a decrease in RD and an increase in overall avoidance shares. Therefore we canassert that rewards have no apparent influence on the choice how to avoid the rush-hour. The model shows that all mon-etary treatments are significant and the sign of the coefficient is negative. The main decrease in RD is attributed to the 3€reward (p < .001) whereas larger rewards have only a relatively marginal effect on the response (see Fig. 1A). Thus, in thecase of money, the difference in the average shares per treatment can be described as a diminishing sensitivity effect (initialGLM analysis confirms this with contrasts being not significant). The parameter size of Yeti credit is similar to that of the 3€level (p < .05). The no-credit also has a negative effect on RD however the parameter estimate is not significant (p < .1).

Table 3Mean 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 U

Rush-hour Preliminary Survey – S 232 89.9 15.44 108 89.8 13.25 0.03 12,153Pre-test – R1 219 48.7 38.17 107 44.0 36.69 1.06 10,3953€ – 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* 9808*

Driving early Preliminary survey – SPre-test – R1 219 22.7 33.03 107 22.0 33.06 0.20 11,4893€ – 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 late Preliminary Survey – SPre-test – R1 219 10.1 22.29 107 20.3 31.32 �3.37* 9740*

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 driving Preliminary survey – SPre-test – R1 219 18.5 30.25 107 13.8 26.00 1.39 10,8163€ – 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

* p < .05.


Although interactions (i.e. moderators) of reward levels and mediators were analyzed as well, no significant results werefound.

Regarding the other response categories Fig. 1B–D show the rewards increase the shares of driving at other times and notdriving compared to both pre and post-test levels. Diminishing sensitivity to money is also evident. The main noticeable dif-ferences are the relatively higher shares of DE and lower shares of DL for the money group compared to the Yeti group(Fig. 1B and C). The latter is already noticeable at the pre-test levels. Yeti users reported in the posterior survey higher sharesfor arrangements with employers about flexible working hours as a support measure compared to the money group (Table 5).Therefore, it is possible that this allowed them greater flexibility in their behavior during the pre-test and the rest of theexperiment. As noted exploration also seems to be an important factor in pre-test behavior.

Mediators (between-subjects factors) included the design related factors (reward class and treatment scheme), and fac-tors relating to the participants’ stated behavior derived from the two surveys. First, as neither the treatment scheme nor anyof its interactions are significant we can conclude that the order of treatments had no effect on behavior. Therefore the ordereffect is discarded from the final model in Table 4. Second, among socio-demographic characteristics gender has a weak ef-fect on RD suggesting men tend to change behavior more than women; however the gender effect is not statistically signif-icant, p > 0.1. In the case of money, higher education has a significant and negative effect on DL (p < .05). A possibleexplanation is that education as a proxy for income could well be masking an income effect. However testing of moderationby 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 are significant. The reward class, which relates to pre-test levels ofdriving at the rush-hour, has a negative association with behavior change. It was found that moderating class by group issignificant (p < .001). Participants, in both groups, associated with classes A, B (2.5–5 rush-hour trips at pre-test) were morelikely to continue driving during the rush-hour compared to classes C and D (0–2.5 trips). In addition, the class coefficient formoney is slightly larger than that of Yeti. The usual departure time has a negative association with DE (p < .001): i.e. the ear-lier is the usual departure time – the more probable is a change of behavior by driving earlier. One may argue that similarfactors that affect driving early in the non rewarded situation (such as household obligations) will still be at play during therewarded period. The preferred start of work time, a likely proxy for the preferred arrival time, has a similar negative effect

Fig. 1. Average response shares by group by treatment.


on DE (p < .001) but also a positive effect on DL (p < .05). That means that participants driving later are those that are moreaccustomed to depart later in usual circumstances. Finally, the experience with using other modes for commuting has a po-sitive effect on not driving (p < .01). Fourth, concerning scheduling flexibility and constraints, a number of factors have beenfound to affect change of behavior. Child chauffeuring is positively associated with RD (p < .05). Other constraints on earlydeparture, such as childcare responsibilities, were not found significant. Conversely, participants who stated they had sup-port from their employers with arranging flexible working times are less likely to drive during the rush-hour (p < .05). Theseresults demonstrate the relevancy of constraints and support measures as important factors that determine the probabilityto change behavior. The number of days (per week) that starting work late is possible has a positive effect on DL (p < .05), afinding that suggests that participants with more flexible working schedules are more likely to drive later. Similarly, the abil-ity to telework encourages to drive later (p < .1).

Fifth, several stated experiences during the course of the experiment were found significant. In the case of money, theestimate of the parameter for ‘practicing with behavior change’ suggests, participants in the money group who reportedpracticing with avoidance behavior during the pre-test were somewhat more likely to have changed behavior (p < .1). In con-trast, participants who stated that they incurred difficulties with the regional rail service (a main alternative to driving) wereless likely to have changed behavior (p < .05). Similarly participants in the money group, who reported in retrospect a higherlevel of effort in changing behavior, were also less likely to change behavior (p < .001). These results indicate that positive ornegative 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. Partici-pants with a positive attitude towards public transport are less likely to change behavior by driving at other times. Theparameters for both DE (p < .01) and DL (p < .05) are negative. In contrast, participants with a positive attitude to cyclingare more likely to change behavior by not driving (the coefficient for ND is positive; p < .01). This result indicates the signif-icance of attitudes towards driving alternatives in influencing change of behavior.

Finally, there are significant effects relating to information usage. Participants with frequenter use of traffic informationare more likely to drive later (p < .05). The coefficient for DL is positive. In addition, participants with frequenter use of publictransport information (p < .01) or stating they had sought for public transport alternatives to support their behavior (p < .05)are more likely to change behavior by not driving; the coefficients for both these factors are positive.

Table 4Results of mixed logistic regression model.

Alt.a Parameter Est. SE t-test p

RD Constant rush-hour driving 0.750 0.31 2.44 0.0103€ reward �1.440 0.46 �3.12 <.0017€ reward �1.804 0.50 �3.58 <.0013–7€ reward �1.780 0.50 �3.53 <.001Yeti without credit reward �0.370 0.66 �0.56 0.560Yeti with credit reward �1.426 0.66 �2.15 0.031Class A and B for money 1.430 0.26 5.38 <.001Class A and B for Yeti 1.180 0.28 4.29 <.001Gender (woman) 0.250 0.15 1.57 0.110High effort for behavior change (for money) 1.039 0.31 3.3 <.001Practice with avoidance during pre-test (for money) 0.36 0.20 �1.79 0.074Arrangements with employer over flexible working time �0.325 0.15 �2.17 0.029Constraint chauffeuring children 0.470 0.21 2.25 0.024Problems with regional rail – would use public transport more 0.500 0.20 2.57 0.010

DE Constant driving early 15.58 1.70 9.13 <.001Usual departure time (min.) �0.022 0.003 �6.95 <.001Preferred start of work time (min.) �0.007 0.001 �4.49 <.001Public transport is realistic alternative �0.670 0.23 �2.92 0.003

DL Constant driving late �2.050 0.92 �2.22 0.026Higher education for money group �0.767 0.24 �3.17 0.010Weekly frequency of accessing traffic information 0.070 0.03 2.18 0.030Public transport is realistic alternative �0.570 0.26 �2.39 0.030Preferred start of work time (min.) 0.004 0.001 2.52 0.011Number of days teleworking is possible 0.348 0.21 1.69 0.090Number of days starting work late possible 0.220 0.05 4.08 <.001

ND Weekly frequency of accessing public transport information 0.281 0.08 3.32 0.009Cycling is realistic alternative for commuting 0.668 0.21 3.16 0.001Seek information on Public Transport connections 0.803 0.31 2.54 0.010Use 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) �0.834Corr. (RD, DL) �0.629Corr. (DE, DL) 0.450L(0) �2123.800L(b) �1648.120q2 0.224Adj. q2 0.207

a RD – rush-hour driving, DE – driving earlier, DL – driving later, ND – not driving.

Table 5Between-group differences on response-associated factors.

Factor Money (N = 231) Yeti (N = 108) pa

Socio-demographics Gender (woman) 40% 26% 0.01High education 58% 53% 0.43

Alternative modes Other modes used for commuting 21% 19% 0.58Public transport is realistic alternative 35% 32% 0.63Cycling is realistic alternative 20% 14% 0.21

Schedules Usual departure time (h.min) 7.52 7.57 0.27Preferred start of work time (h.min) 8.24 8.35 0.20Start work later (days) 3.5 3.64 0.52Telework (days) 0.46 0.58 0.37

Difficulties Chauffeuring children duties 16% 27% 0.02High effort perceived with changing behavior 6% 9% 0.29Problems with regional rain 33% 29% 0.72

Support measures Arrangements with employer 34% 55% <.001Practice during pre-test 30% 25% 0.34Search for public transport connections 13% 13% 0.97

Travel information Use of traffic information (days/week) 1.35 1.83 0.08Use of public transport information (days/week) 0.13 0.01 0.15

a Chi-square test for nominal factors, t-test for interval factors.



By definition this study which includes self selection follows a quasi-experimental design. Like random experiments, qua-si-experiments share the same basic principles of manipulation (cause precedes effect) and measurable associations (covari-ation). In contrast, causation requires more effort as compared to random assignment there are more threats to internalvalidity. Based on the recommendation of Shadish et al. (2002), we describe here the plausible threats to the internal validityof the results. Threats to statistical inferences are not discussed here as we contend that these are likely to be low given theconservative nature of the analysis method applied which guarantees proper statistical identification of the measurable ef-fects. Lack of a control group also can contribute to validity problems. However, some features in the design allow improvedcontrol and reduce possible threats. The pre-test / post-test design is fostered by additional measurements of stated behaviorthrough the preliminary and posterior surveys. Specifically threats resulting from history and novelty can be assessed bycomparing between the preliminary survey and the pre-test. In addition, the measured factors from the surveys such asusual behavior, constraints and support measures, can provide relevant mediators to the observed behavior and verify ifselection is a problematic issue. This is dealt in detail below. Norm comparisons with traffic counts on the main A12 trajec-tory suggest other drivers did not change behavior during this period. The sample is small enough not to have any real im-pact on traffic flow. Since no significant change in traffic occurred during the 13 week observed period we can assert that anydifference 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.

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 relatesto a priori differences in the money and Yeti groups which could compromise the results. In Table 5 we present a statisticalcomparison of the differences between the self-selected groups by the factors that are associated with the response in theMLR model. Most factors have no significant difference, but gender, chauffeuring children and arrangements with employersregarding flexible working times do. The latter which has the most significant difference was measured in the posterior sur-vey whereas the first two factors relate to measurement from the preliminary survey. To contend with the threats we esti-mated the effects moderated by group, specifying the MLR model with group-specific coefficients for these three factors.

Regarding gender, it was not significant in the model (p > 0.1). We tested moderating effects for money and Yeti but thiswas found not to be significant. We can therefore conclude that gender (woman) has a weak negative association with avoid-ance behavior. Chauffeuring children is significantly and negatively associated with avoidance behavior (p < .05). We inves-tigated if this might be moderated by gender but given the small group involved of both men and women who have thisconstraint we could not identify with confidence any significant moderating effect. Moderating by group also does not revealsignificant group differences in the MLR model. Therefore we can suggest that the negative effect identified for chauffeuringon peak avoidance is probable.

The threat attributed to prearrangements regarding flexible work times requires more attention (p < 0.001). It is clear thatparticipants 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 (affecting construct validity) in thedaily schedule. It is reasonable that participants felt the need to prepare for what some may regard as a major disruption intheir routine which would carry on for several weeks. It is plausible that these arrangements which were more apparent inthe Yeti group had an effect on the direction of response i.e. increasing the share of driving later. Specifying the factor for DLas well as moderating by group resulted in insignificant coefficients. Therefore it seems likely that the differences in behaviorbetween the two groups is related to the treatments i.e. traffic information rather than prearrangements. However the con-founding 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 anotherthreat attributed to history. About a third of participants in both groups stated in retrospect that they practiced with changeof behavior even without rewards during the pre-test measurements – a plausible explanation for the lower rush-hour fre-quencies compared to the usual stated behavior. However, since pre-test levels also determine reward eligibility (rewardclass), the comparison between treatment and no-treatment (pre/post-test) remains relatively valid. Practicing also has aweak negative effect on rush-hour driving (p < .1) which was only found relevant for the money group.

5. Discussion

5.1. Effectiveness of rewarding

The results demonstrate, that rewarding, at least in the short run, is effective as within a short period of time of severalweeks, the share of rush-hour avoidance substantially increased. Thus in concordance with motivation theories (e.g. Cam-eron et al., 2001) rewards do influence the motivation to avoid the rush-hour. Moreover, we also found that the decisionhow to exercise this change of behavior, whether by driving at other times or by changing transport or work modes seemto 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 longrun. 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 returnedmore 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 and Kitamura (2003) regarding free bus tickets. Therefore at first glance it seems the change was notsustained for most of the participants. Notwithstanding, in the posterior survey less than 15% of participants stated they hadreturned 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 Conclusions). We also do not posses sufficient (post-test) data toconclude 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.

5.2. Reward type and levels

We found that both types of reward (monetary and in-kind) have a significant and negative effect on rush-hour driving. Inthe case of a monetary reward, diminishing sensitivity was clearly noted. The 7€ treatment has the largest overall effect onRD; however the largest marginal effect (the derivative) is associated with the 3€ treatment. Therefore, for practical pur-poses, a moderate monetary reward seems to be sufficient to encourage a relatively substantial change of behavior. In thecase of the Yeti reward, the main effect is the credits which have an effect similar to the 7€ reward. In this sense an in-kindreward, 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 validcredits. This treatment has no extrinsic reward but travel information was still accessible to Yeti users. Furthermore, it wasevident that Yeti users were more likely to drive later compared to participants in the money group. Two possible explana-tions are possible for this different behavior. On one hand, Yeti users had higher shares regarding support provided fromemployers. Thus, it is possible that pre arrangements were involved in choosing to depart later (especially during thepre-test). On the other hand, the main advantage Yeti users had over the other group was 24 h access to travel information.This leads us to suggest that the decision how to change behavior is also influenced by travel information availability (dis-cussed later on).

5.3. Socio-demographic characteristics

As noted, this is hardly a novel assumption in travel behavior studies. Gender (weakly) and education, were found to havean impact on the response to the rewards. It seems that men (mainly in the money group) are somewhat more likely to avoidthe rush-hour compared to women. The lower motivation of women to avoid the rush-hour can be associated with manyissues. One idea that has been suggested in social mobility studies (Palma, González, & Carrasco, 2009) is that women aremore constrained in time compared to men for various reasons, mainly household tasks and child raising obligations. Dutchwomen quite often leave work early in the afternoon to pick up children from nurseries (Schwanen, 2007). This limits theirability to change their schedule - e.g. to start work later even when extrinsically motivated by a reward. However, a largersample is needed to clearly mark the causation between gender and time-use behavior given the lack of statistical signifi-cance of the results.

Education had a negative effect on behavior change (driving later). Participants (in the money group) with higher educa-tion were less likely to drive later. Education is a known proxy for latent income effects. Income is regarded as a key issuedetermining 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 notionof diminishing sensitivity in relation to the monetary reward: participants with higher real income are likely to be less sen-sitive 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 ofYeti users, possibly because it is cognitively and affectively appreciated as an endowment, rather than in monetary (howmuch it’s worth) value.

5.4. Scheduling

This is a new territory of travel behavior research, lately identified by Gärling and Fujii (2006). The results suggest thatbehavior change and more so the choice of behavior change is associated with the ability or disability to change daily sched-ules. Both home related and work related flexibilities are relevant. Family obligations, such as children chauffeuring – a con-straint associate positively with rush-hour driving – make it more difficult for parents to change travel behavior. The abilityto accommodate a flexible schedule and the support provided by others are also significant factors. Participants that couldstart working later or could telework were more likely to drive later. Participants reporting to have received support fromtheir employer with arranging flexible working times were also less likely to drive in the rush-hour. We see these resultsas 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 significanteffect on behavior change. A possible explanation to the effects of scheduling is the extent of control over one’s actionsand their outcomes. The Theory of Planned Behavior (Ajzen, 1991) suggests perceived situational control is a key factor inencouraging a conscience behavior change. Thus flexibility in time-use promotes a sense of self confidence and ability tocontend with the schedule’s change.


5.5. Habitual behavior, experience and attitudes

As suggested by Theory of Interpersonal Behavior (Triandis, 1977, 1980) we found that factors relating to habitual behav-ior play an important role in the choice how to change behavior. This corroborates findings from other studies (e.g. Gärlinget al., 2001; Gärling & Axhausen, 2003). The effect of habitual behavior is well manifested in the significance of the rewardclass, usual departure time, the preferred start of work time (in the case of shifting driving times) as well as the use of othermodes for commuting purposes (in the case of switching mode). Participants with higher rush-hour commute frequenciesduring the pre-test (reward class A, B) were relatively less likely to avoid the rush-hour compared to participants with lowerrush-hour frequencies (classes C and D). Two potential explanations are put forward. First, in terms of effort, one could arguethat a similar relative response demands more rush-hour avoidances from frequent rush-hour drivers than from less fre-quent rush-hour drivers. Hence, the effort involved is higher for high frequency drivers. This is in line with Gärling,Jakobsson, Loukopoulos, and Fujii (2004) and Cao and Mokhtarian (2005), who found that travelers prefer low effort re-sponses over high effort responses. A second explanation is that the added value of additional rewards depends on theamount already gained, in the sense that the marginal utility of reward decreases (i.e. diminishing sensitivity). Thus, the ex-tra rewards gained by high frequency drivers will have a lower impact on behavior. This is in line with the idea of satisficingbehavior described by Simon (1987). In the case of Yeti users, the effect of reward class is weaker. This might be related tothe affective qualities of the Smartphone endowment i.e. avoiding the displeasure of having to give back the handy Smart-phone encouraged avoidance. In addition, real-time travel information may have been useful in reducing perceived effortand 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 ofbehavior change. The usual departure time was a decisive factor affecting the choice to depart earlier whereas the preferredstart 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. Thatis, familiarity with an alternative seems to increase intrinsic motivation. It appears that the choice of behavior change is clo-sely related to the perceived gap between the usual behavior and the required change – the smaller the gap the more likely isthat the change will be exercised. One may argue that alternatives that are more similar to the current behavior, will bettermeet 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 reinforcedlearning about the new situation. It is suggested that exploration had an important role during the pre-test. Practicing avoid-ance behavior during the pre-test was reported by almost a third of the participants(in both groups) and it is one explanationfor the dramatic drop in the pre-test shares of rush-hour driving compared to the usual (stated) behavior recorded in thepreliminary survey. This factor was also found (albeit weakly) to increase the likelihood of decreasing rush-hour drivingfor the money group. Recent findings in the context of route-choice behavior suggest information expedites learning inthe 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, percep-tions 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 decisionto change behavior. Our results support this assertion in that participants’ attitudes regarding alternative modes influencedthe choice of avoidance behavior. Positive attitudes (defined as regarding a travel mode as a realistic commuting alternative),regarding public transport and cycling, discouraged driving (including at off-peak periods) and encouraged mode switchaway from the car. Conversely, perceptions regarding the (high) effort level involved in changing behavior decreased thelikelihood of changing behavior and were positively associated with rush-hour driving. We could not find real support forthe relevance of personal norms in the decision to avoid the rush-hour.

5.6. Information availability

Several studies confirm the key role that availability of travel information has on promoting sensible travel choices (Mah-massani & Liu, 1999; Srinivasan & Mahamassani, 2003). Our results sustain this association in three ways. First, there appearto be significant between-group differences in the behavior change. Yeti users had relatively higher shares of driving latercompared to higher shares of driving earlier in the money group. Yeti users’ main advantage was the real-time access to tra-vel information whereas the participants in the money group had to acquire the same information implying more effort.However, as noted this result could also be confounded by prior arrangements with employers over flexible working timeswhich were more dominant with the Yeti group. Hence, we cannot be certain if the different response is attributed to the Yetitreatment or related to prearrangements which facilitated driving later. However, the information explanation is strength-ened by the fact that moderating this effect by group had no significant effect.

Second, access to travel information, mainly traffic but also public transport information, intensified during the course ofthe experiment (pre/post-test comparison). Thus, decision-making in a changed environment apparently increases the needfor acquiring information about the outcomes of alternatives. Third, information availability is positively associated with notdriving or driving later. Participants who frequently accessed public transport information and who were actively seekinginformation 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 informa-tion acquisition and choice of avoidance behavior are clearly related. However causality here is uncertain as participantscould also increase information acquisition for the alternative they liked best.

6. 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 waslikely perceived as a gain with diminishing sensitivity, whereas the Yeti should be regarded as an in-kind reward which hadadded endowment and instrumental qualities. The rewards were able to sustain the behavioral change throughout theexperiment. Nonetheless, it is still an open question whether the change can be sustained in the long run and without re-wards. We do not have enough post-test observations to provide an answer apart from subjective assessments by the par-ticipants. 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 an-other mode of transport or working from home, is determined by external factors relating to participants’ personal and socialcharacteristics, scheduling flexibility, history and information availability, Although already of some interest to the travelbehavior 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 placeacross The Netherlands. Although some concern, based on traffic simulation models, indicated that too many people mightstart changing their schedules to gain a reward (Bliemer & van Amelsfort, 2008), the evidence in the field does not supportthis claim. Rewards’ effectiveness in mitigating congestion, especially in situations involving temporary road and bridgemaintenance or lane closures has been verified (Bliemer, Dicke-Ogenia, & Ettema, 2009). A recent survey of firms also hasshown positive attitude amongst employers towards the reward scheme (Vonk Noordegraaf & Annema, 2009). The majorityof the Dutch public (apart for the public transport users who are ineligible and consequently grumbling) and the governmentwere quite content with the results. However as recently published in the media the (last) government also wanted to ad-vance 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.

Acknowledgments

This study was undertaken as part of the Spitsmijden project, which was funded by Transumo (Transition to SustainableMobility), the Ministry of Transport in the Netherlands, Bereik, RDW, NS, Rabobank, ARS T&TT, OC Mobility Coaching, VrijeUniversiteit Amsterdam, TU Delft, Universiteit Utrecht. The modeling framework was discussed in the 5th Discrete ChoiceModeling Workshop organized at EPFL (Lausanne, Switzerland) in August, 2009. The comprehensive comments and sugges-tions of two anonymous reviewers are very highly appreciated. We would especially like to acknowledge the persistence andcommitment of Reviewer 2 to improving this manuscript. This has been a learning by doing process. The First author alsowarmly thanks the department of Human Geography and Planning at Utrecht University for their support during his stayas a postdoctoral researcher.

References

Ajzen, 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, T. (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., & Ettema, D. (2011). Rewarding rush-hour avoidance: A study of commuters’ travel behavior. Transportation Research Part A, in press.Ben-Elia, E., & Shiftan, Y. (2010). Which road do I take? A learning based model of route-choice and real-time information. Transportation Research Part A

(44), 249–264.Berridge, K. C. (2001). Rewarding learning: Reinforcement, incentives and expectations. In D. L. Medin (Ed.), The psychology of learning and motivation.

Academic Press.Bhat, C. (1999). An analysis of evening commute stop-making behavior using repeated choice observation from a multi-day survey. Transportation Research

Part B, 33, 495–510.Bliemer, M. C. J., van Amelsfort, D. H. (2008). 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.Bliemer, M., Dicke-Ogenia, M., & Ettema, D. (2009). Rewarding for avoiding the peak period: a synthesis of four studies in The Netherlands. In Paper

presented 617 at 12th international conference on travel behaviour research, Jaipur.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 Part A, 41(7), 672–683.


Cameron, J., Banko, K. M., & Pierce, W. D. (2001). Pervasive negative effects of rewards on intrinsic motivation: The 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.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. (2009). 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. Journal of Personality and Social Psychology, 18, 105–115.Deci, E. L. (1975). Intrinsic motivation. New York: Plenum Press.Deci, E. L., & Ryan, R. (1985). Intrinsic motivation and self determination in human behavior. New York: Plenum Press.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. (2010). 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. Reading, MA: Addison-Wesley.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. Transportation Research Part F, 4, 89–102.Gärling, T., & Garvill, J. (1993). Psychological explanations of participants in everyday activities. In T. Gärling & R. G. Golledge (Eds.), Behaviour and

environment: Psychological and geographical approaches. Amsterdam: Elsevier.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. Washington, DC: Transportation Research Boards.Gneezy, U. (2003), The W effect of incentives, working paper, University of Chicago.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.

Amsterdam: Elsevier.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.Kahneman, D., Knetsch, J., & Thaler, R. H. (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., et al (2007). Experimental design and modelling Spitsmijden. The Hague: Bureau

Spitsmijden.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. NJ: John Wiley and Sons.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

Part C, 7(2), 91–107.Mayeres, I., Ochelen, S., & Proost, S. (1996). Marginal and external costs of urban transport. Transportation Research Part 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. Cheltenham: Edward Elgar Publishing.Palma, P. A., González, C. D., Carrasco, S. L. (2009). Gender differences in time use and mobility: Time poverty and dual consumption. Paper presented at time

use observatory, Santiago, Chile.Perry, O., Erev, I., & Haruvy, E. (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 Statistics, 80, 647–657.Ronis, D. L., Yates, J. F., & Kirscht, J. P. (1989). Attitudes, decisions and habits as determinants of repeated behaviour. In A. R. Pratkanis, S. J. Breckler, & A. G.

Greenwald (Eds.), Attitude structure and function. Hillsdale, NJ: Erlbaum.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. In L. Hendrickx, W. Jager, & L. Steg (Eds.), Human decision making and environmental perception:

Understanding and assisting human decision making in real-life settings. Groningen: University of 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 L. Berkowitz (Ed.). Advances in experimental social psychology (Vol. 10). London: Academic Press.Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Belmont: Wadsworth.Shaffer, V. A., & Arkes, H. R. (2009). Preference reversals in evaluations of cash versus non-cash incentives. Journal of Economic Psychology, 30(6), 859–872.Shiftan, Y., & Golani, A. (2005). Effect of auto restraint on travel behavior. Transportation Research Record, 1932, 156–163.Simon, H. A. (1987). Satisficing. In J. Eatwell, M. Miligate, & P. Newman (Eds.). The new Palgrave: A dictionary of economics (Vol. 4, pp. 243–245). New York:

Stockton Press.Small, K. A., & Verhoef, E. T. (2007). The economics of urban transportation. London: Routledge.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 Part B, 37, 793–814.Train, K. (2002). Discrete choice methods with simulation. Cambridge, MA: Cambridge University Press.Train, K. (2000). Halton sequences for mixed logit. Working paper. Berkeley: Department of Economics, University of California.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. In E. Verhoef, M. Bliemer, L. Steg, & B. van Wee(Eds.), Pricing in road transport: A multi-disciplinary perspective. London: Edward Elgar Publishing.

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. (2009). Employer attitude to rush-hour avoidance. In Proceedings of the European transport conference,

Noordwijkerhout.Walker, J. L. (2001), Extended discrete choice models: Integrated framework, flexible error structures, and latent variables, Ph.D. thesis, Massachusetts

Institute of Technology.

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

Documents

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