Journal of Transport Geography 31 (2013) 258–266

Contents lists available at SciVerse ScienceDirect

Journal of Transport Geography

journal homepage: www.elsevier .com/locate / j t rangeo

Dynamic modelling of travellers’ social interactions and social learning

0966-6923/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.jtrangeo.2013.05.012

⇑ Corresponding author. Tel.: +972 3 7688777; fax: +972 3 7688692.E-mail address: avineri@afeka.ac.il (E. Avineri).

Yos Sunitiyoso a, Erel Avineri b,⇑, Kiron Chatterjee c

a School of Business and Management, Institut Teknologi Bandung, Ganesha 10, Bandung 40132, Indonesiab AFEKA – Tel Aviv Academic College of Engineering, 218 Bnei Efraim Road, Tel Aviv 69107, Israelc Centre for Transport & Society, Faculty of Enviornment and Technology, University of the West of England, Coldharbour Lane, Bristol BS16 1QY, United Kingdom

a r t i c l e i n f o a b s t r a c t

Keywords:Travel behaviourSocial interactionSocial learning

Social interaction and social learning are likely to be influential factors in the travel choices made by indi-viduals and the dynamics of these choices. This study aims to understand the influence these socialaspects have on travellers’ decision making and behaviour. Furthermore, this research seeks to find outthe possibility of utilising this understanding to enhance policies on behavioural change. Social interac-tions, which may due to an interdependent situation between travellers, social information about othertravellers’ behaviour and communication between travellers enable social learning and social influenceprocesses between travellers. Social psychology theories have been used to provide the underlyingframework for the study as well as the methods for analysing the data using the individual and sociallearning models. This study utilises a laboratory experiment to capture the role of social interactionsand social learning in the dynamics of travellers’ decision making over time. A major finding of the lab-oratory experiment is that social interaction and social learning influence individuals’ behaviour. How-ever providing more social information makes people behave in a less cooperative way and be moreunstable in making choices. It also influences more people to make contrarian than direct responses. Anal-yses reveal that people learn individually from their previous experience and socially from other people.It is revealed that confirmation (keeping previous behaviour when observed individuals also chose thesame choice) and conformity (following the choice of the majority) are exhibited whenever individualshave access to social information, and therefore could be incorporated into models of travel choice. Thesefindings elicit some behavioural, policy and methodological insights.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Travellers’ decision making and behaviour can be considered asdynamic processes, since individuals can and do change theirbehaviours over time. The understanding of how travel behaviourdevelops and changes over time is important in order to improvedynamic models of travel choice, and to identify possibilities forinfluencing behaviour. A traveller’s decision to change behaviourmay be due to new information gained from their own experienceand/or information and influence from the experience and behav-iour of others.

Kitamura et al. (1999) raised the issue of the need when analys-ing individual behaviour to assess the effects on them of othertravellers who also respond to demand management measures.Jones and Sloman (2003) stated that there is some evidence thatbehavioural changes may be very slow at first, but then accelerateas people observe their colleagues and neighbours changing their

travel behaviour. These and other studies indicate that socialinteraction and social learning may influence travellers’ changeof behaviour.

Social interaction is likely to influence individuals’ behaviourinside a group, or in a wider scope, a society. It also contributesto changes of the broader environment. In his theory of social pro-cess, Douglas (1974) stated that the social environment is con-stantly changing due to the contribution of individuals andgroups engaged in social interactions. Social interaction always ex-ists whenever an individual is in an interdependence situation thatinvolves other individuals where their actions affect each other.The scale of interactions may depend on the size of group (or soci-ety). In a group, actions of a group member achieve more influencethan that within a population, since inside a group there exists afeeling of belonging and responsibility as a group member. In apopulation, these feelings may not strongly exist.

An individual may expect that other individuals will contributeto a collective action so that she does not need to contribute any-thing. Olson’s (1965) experiment found two factors that can pro-mote cooperation: repeated social interactions and communication

Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266 259

among the participants. In a travel choice context, more peoplemight car share to work if they are aware of what others do andthis can take place only if some kind of interaction and communi-cation occur between them. A way of communicating throughword-of-mouth has been identified to allow an efficient sociallearning process. For example, Taniguchi and Fujii (2007), in theirstudy of promoting a community bus service, found that word-of-mouth advertising through recommendations to friends andfamily plays an important role in promoting bus use.

There are three levels of social interaction which may influencetravellers’ behaviour. The first level of social interaction is due toan interdependent situation where none of the individuals engagedin a collective action can be excluded from enjoying the benefits/costs of the sum of the individual decisions (e.g. a social dilemmaof public road users, where the decision of each user affects notonly herself but also the state of the system, hence affects otherusers). The second type of social interaction happens throughobservation by a traveller of others’ choices. This type of socialinteraction is more direct than the first type but it is one-wayand does not involve communication or exchange of informationwith other travellers. The third type of social interaction is themost direct interaction which happens through communication orexchange of information between travellers regarding their travelexperience (choices and their outcomes) and/or intentions. Thecommunication in this context covers all possible communicatingmedia, including face-to-face, telephone, video and text messaging.Both the second and third levels of social interaction may be due tothe fact that individuals are not indifferent to the outcomes re-ceived by others (Messick, 1985) since travellers sometimes takeinto account and are concerned about choices by other travellers(Van Lange et al., 2000). This study focuses on the first and secondlevels of social interaction (see Sunitiyoso et al., 2011a for a studyinvolving the third level of social interaction, communication).

Travel behaviour can be seen as a dynamic process that occursover time and may involve a learning process. Kimble (1961) de-fines learning as a relatively permanent change in behaviouralpotentiality that occurs as a result of reinforced practice. Learningis more likely to happen when there is a change in the situationalcontext (or behavioural goal), when deliberation is prompted byinformation or when the situation is uncertain due to its natureor due to interdependence between people.

The concept of individual learning suggests that an individuallearns from her past experience and utilises an adaptive decisionmaking process to cope with uncertainty. In another form of learn-ing, social learning, individuals learn from others’ experiences orobserved behaviours. In travel behaviour modelling, the individuallearning concept has often been studied (for a review, see Arentzeand Timmermans, 2005), while social learning has not been inves-tigated significantly although evidence from other disciplines (e.g.economics and behavioural sciences) have shown that this kind oflearning process is influential and important (e.g. Offerman andSonnemans, 1998).

The lack of understanding of the potential role of social interac-tion and social learning in influencing travellers’ behaviour encour-aged the authors to explore the social aspects using a laboratoryexperiment. This study also aims at understanding the dynamicsof behaviour at both aggregate and individual levels when individ-uals are provided with social information about other individuals’behaviour. The authors aim to answer the following hypotheses:(H1) whether providing more information will change individuals’behaviours in making choice; (H2) whether it will make individu-als choose a more cooperative choice; (H3) whether it will makeindividuals more decisive in making choice; and (H4) whether itwill make individuals make direct responses rather than contrarianresponses.

2. Laboratory experiment

To the authors’ knowledge, the effects of social interaction ontravellers’ intentions and behaviours have not been much exploreddirectly, i.e. by observations of choices in a laboratory or field envi-ronment. In the transport field, laboratory experiments have beenused to study travellers’ choice behaviours, particularly the dynam-ics of route choice making (e.g. Mahmassani and Jou, 2000; Seltenet al., 2004), departure time choice making (e.g. Ramadurai andUkkusuri, 2007) and the effects of traveller information servicesor ITS on travellers’ departure time and route choice making (e.g.Mahmassani and Liu, 1999). However, the effects of social interac-tion and social learning on individuals’ choice making and behav-iour have not been investigated in the transport context.

The laboratory experiment utilises a human–computer inter-face developed on Z-tree (Fischbacher, 2007), an experimental eco-nomics tool that allows the experimenter to design, develop andcarry out experiments with features, including communication be-tween computers, data saving, time display, profit calculation andtools for screen layout, as well as communication features. The lab-oratory experiment simulates a multi-player repeated decisionmaking environment. The experiment is a part of a major studyinvolving a behavioural survey and a series of laboratory and sim-ulation experiments which have been conducted to understand theinfluence of social aspects on travellers’ behaviour (Sunitiyosoet al., 2009, 2011a,b). The hypothetical choice situation used inthe experiment is based on the public goods (social) dilemmagame, where each participant is requested to contribute to collec-tive goods in order to obtain benefits for all participants, with lim-ited information (controlled by the researchers) on the actions ofother participants and their contribution to the social dilemmapayoffs. In this experiment, the social dilemma is put in the contextof car-sharing based on a real-life situation relating to car-sharingand car-parking in a university setting. Each participant is asked tochoose whether to drive alone (car-alone) or to share a car with an-other person (car-share) for a trip to the university. Each individualwho chooses to car-share is randomly partnered with another par-ticipant who also decides to car-share. If the number of car-sharersis not even then an extra traveller is generated by the computerserver. Whenever people travel to the university by car, regardlessof their choice of travel (car-alone or car share), they have to parktheir car at the university car park. The car park capacity is fixedand there may not be enough spaces for all cars. If the car park isfull, then they have to wait for a parking space. This costs a£3.00 penalty per traveller, reflecting the value of waiting time.There is no guarantee that participants will get a parking spaceeven if they choose to car-share but the chance is higher (since lesscars means less competition for finding a space). The participantsare given a briefing on the situation prior to the experiment.

The generalised costs of travel (without penalty) for each indi-vidual for a trip from the city centre to the university for eachmode are: (a) the total travel cost for driving car alone is £3.50,which consists of walking time cost (£1.40), in-vehicle time cost(£1.30) and vehicle operating cost (£0.80); and (b) the total costof car-sharing is £4.50, which consists of walking time cost(£1.40), in-vehicle time cost (£1.30), shared vehicle operating cost(£0.80/2 = £0.40), and picking up/waiting time cost (£1.40). Thesecosts are roughly estimated based on the DfT’s Transport AnalysisGuidance: Values of Time and Operating Costs (DfT, 2004) for a sin-gle trip by car from Bristol City Centre (UK) to the University of theWest of England (UWE) Frenchay Campus via M32. Participants areasked to select between car-alone and car-share. After all partici-pants have made decisions, they are provided with feedback onthe outcomes of their individual decisions (travel cost, penalty,and money left). The cost that a participant has to pay does not

260 Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266

only depend on her choice, but also on the output situation whichis uncertain since the availability of parking spaces depends on thechoices of other participants. The value of time was set to reflectthe penalty for waiting for a parking space, if not available; parkingspaces at the campus may get full at peak-hour and it might belong before a parking space become available, The penalty of notgetting a parking space is quite high (£3.00) which is almost thesame value as the cost of travelling alone (£3.50).

Three experimental treatments were developed based on theavailability of individual information regarding past experienceand availability of so-called ‘social information’ regarding otherparticipants’ choices. In Treatment 1, each participant is giveninformation about their previous choice and its outcome (cost,penalty, and money left), but no information is provided on thechoices of other participants. In Treatment 2, in addition to infor-mation about their previous experience, the participant is able toobserve another participant’s choice and corresponding outcomein the previous round(s). The target to be observed is anonymousand assigned randomly. He/she is continuously observed for thewhole rounds of the corresponding treatment. In Treatment 3 eachparticipant is able to observe the choices of all other participants inthe previous round. Each of the three treatments has 10 rounds. Ineach round of the experiment, participants make a choice simulta-neously. They do not know the outcomes of their choices until allparticipants have made a choice and the number of cars requiringparking spaces is calculated.

The experimental settings of the three treatments aim to inves-tigate differences in participants’ responses to travel information,and test the four hypotheses given in the introduction section.Three levels of social interactions were discussed: independent sit-uation (in which social information is absent), observation of oth-ers’ behaviours, and direct communication between travellers. Thiswork focuses on the effects of travel information provision on indi-vidual responses that might be associated with the first two levelsof social interactions: the independent situation is intuitively asso-ciated with the expected responses of participants to the setting ofTreatment 1; observation by a participant of others’ choices isassociated with the expected responses of participants to the set-ting of Treatment 2. This type of one-way social interaction doesnot involve direct communication or exchange of information withother participants; thus the third level of social interactions, verbalor face-to-face communication between participants, was not facil-itated in the experiment. While this level of social interaction is ofrelevance to the understanding of travel choices in some transportcontexts, its investigation calls for different methodological ap-proaches (see, for example, Sunitiyoso et al., 2011a; Bartle et al.,2013), and has been left out of the scope of this work.

Information on the system’s performance is provided in alltreatments and is fed back to all participants in the form of travelcost which depends on their choice (e.g. car-alone or car-share)and on the availability of parking space that depends on the de-mand. Treatment 1 becomes the control treatment. Feedback aboutthe choices made by another participant in Treatment 2 is used tocompare the effect between individuals receiving only system’sinformation (Treatment 1) and individuals receiving limited infor-mation based on the choices of another participant; these are alsocompared to the responses of participants who have been providedwith complete information on choices made by all other partici-pants (Treatment 3). In real life situation, somebody may observeand learn from his/her colleague(s) or peer(s) and may make achoice following or contrasting with the observed colleague/peeror a number of colleagues/peers, or even developing his/her strat-egies in response to other individuals behaviour.

Post-experiment questionnaire and interview survey is alsoconducted to gain more insight on the effect of providing limitedor complete information as well as the learning processes of

participants during the experiment. The participants are asked toexplain the way the make a decision, the reasons of the decisionand other questions regarding their experience in each treatmentof experiments. Both post-experiment questionnaire and intervieware also used to re-check whether participants really understoodthe experiment that they have done and whether they had anyproblems during the experiment.

At the end of the experiment, each participant is paid with realmoney which is a fraction of the accumulated earning of experi-mental money gained during all the rounds (up to £10) plus a showup fee (£5). The use of financial incentives may lead to a change ofbehaviour closer to the predictions of the normative models, whichare usually intended in experimental economics, for example totest decision-theoretic or game-theoretic models (Hertwig andOrtmann, 2001). Although participants can be assumed to beessentially cooperative and to be intrinsically motivated to partic-ipate in the experiment, there is no guarantee that self- motivatedresponses to an experiment will reflect real-world behaviour (Bon-sall, 2002). If incentives are decided to be given, then a decisionshould be made whether they are flat or performance-based incen-tives. The amount of incentives should also be just sufficient toneutralize the ‘good citizen’ sense of duty while not being so gen-erous as to generate a sense of obligation (Bonsall, 2002).

Participants in the laboratory experiment are undergraduateand postgraduate students at UWE, Bristol, recruited using adver-tisements via emails, notice boards, flyers and the UWE StudentsUnion online forum. The 40 participants were divided into 6 groupswith 4 groups of 7 participants and 2 groups of 6 participants. Eachgroup participated in all three treatments with a sequence of eitherTreatments 1-2-3 or Treatments 1-3-2.

To simulate the process of interactive decision making (involv-ing feedback) the experiments utilise a human–computer interfaceas the medium. Each participant faced a computer screen, madea choice based on the information presented, and received feed-back on the outcome of the choice and the choices of other partic-ipants depending on the treatment that is being applied. Eachexperimental treatment, which consisted of 10 rounds of decisionmaking, took around 15 min. After participants took part in allthree treatments a short interview was conducted to explore par-ticipants’ experiences and to provide additional qualitative under-standing of their behaviour during the experiment.

The difference in the size of the groups (N = 6 and N = 7) pro-duces a slightly different social dilemma situation. Groups ofN = 6 face less severe dilemma than that of groups of N = 7 sincesimilar size of public goods (4 car parking spaces) is available fora less number of people. To illustrate the dilemma situation, ex-pected costs of choosing car-alone (CA) and car-share (CS) in everypossible choice situation are calculated using the functions in thefollowing equation:

expðctxÞ ¼ Ppark � ctx þ ð1� PparkÞ � ðctx þ PnÞ

Ppark ¼NC

Cap; NC ¼ NCA þ NCS

2

totalðctÞ ¼X

8x2fCA;CSgexpðctxÞ � Nx

avgðctÞ ¼ totalðctÞP8x2fCA;CSgN

x

ð1Þ

where exp (ctx) is the expected cost of choice x (x e {CA, CS}); Ppark isthe probability of getting a parking space, ctx is the cost of choice x;Pn is the amount of penalty for not getting a parking space; NC is thenumber of cars (both solo cars and shared cars); NCA and NCS are thenumber of solo drivers and the number of cars sharers respectively;Cap is the capacity of car park; total(ct) is the total cost of the alldrivers and car sharers; and avg(ct) is the average cost of a choice.

Table 1Expected, system, and average costs of choices for N = 7 (all costs in £).

NCA NCS NC exp(ctCA) exp(ctCS) Total(ct) avg(ct)

1 6 4 3.50 4.50 30.5 4.363 4 5 4.10 5.10 32.7 4.675 2 6 4.50 5.50 33.5 4.797 0 7 4.80 33.6 4.80

Table 2Expected, system, and average costs of choices for N = 6 (all costs in £).

NCA NCS NC exp(ctCA) exp(ctCS) Total(ct) avg(ct)

0 6 3 4.50 27 4.502 4 4 3.50 4.50 25 4.174 2 5 4.10 5.10 26.6 4.436 0 6 4.50 27 4.50

Table 3Average proportion of car-sharers in each group.

Group N Treatment Treatment 1 Treatment 2 Treatment 3

Order Mean Var. Mean Var. Mean Var.

1 7 1-2-3 0.416 0.019 0.444 0.019 0.486 0.0182 7 1-2-3 0.386 0.023 0.330 0.027 0.314 0.0313 7 1-3-2 0.444 0.019 0.485 0.036 0.472 0.0264 6 1-3-2 0.450 0.019 0.517 0.034 0.416 0.0145 7 1-2-3 0.458 0.017 0.486 0.018 0.416 0.0116 6 1-2-3 0.450 0.025 0.550 0.019 0.483 0.034

Total 40 All 0.433 0.020 0.468 0.029 0.431 0.024

Table 4ANOVA test with one between-subjects variable and two within-subjects variables.

Source Sums ofsquares

df Meansquare

F-ratio

Sig.

Between-subjects effectsGroup size 0.098 1.000 0.098 1.271 0.323Error 0.309 4.000 0.077

Within-subjects effectsTreatment 0.082 2 0.041 3.495b 0.081Treatment � group size 0.046 2 0.023 1.939 0.206Error (treatment) 0.094 8 0.012Round 0.382 9 0.042 2.477a 0.026Round � group size 0.333 9 0.037 2.158a 0.049Error (round) 0.617 36 0.017Treatment � round 0.377 18 0.021 0.973 0.499Treatment � round � group

size0.385 18 0.021 0.994 0.477

Error (treatment � round) 1.549 72 0.022

Note:a Significant at level of significance a = 0.05.b Significant at a = 0.10.

Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266 261

Tables 1 and 2 show that the expected cost of choosing car-alone (exp(ctCA)) is always cheaper than the expected cost ofchoosing car-share (exp(ctCS)) in every situation for both N = 6and N = 7. The social dilemma situation resembles the ‘prisoners’dilemma where the expected cost of choosing a ‘non-cooperative’choice (in this case, car-alone) is always cheaper than the cost ofchoosing the ‘cooperative’ one (car-share).

For N = 7, the social dilemma means that theoretically the user-equilibrium is when each traveller tries to maximise her individualutility, thus everyone chooses car-alone (NCA = 7, avg(ct) = 4.80 inTable 1). The system cost reaches its optimum point when thenumber of car-sharers is 6 persons or only one person drives caralone (NCA = 1, avg(ct) = 4.36). This is not an ideal situation sincethere will be a temptation for becoming the person who drivescar-alone and pays less cost than the other 6 persons, which willdrive the system into a higher system cost. On the other hand, ifall 7 persons choose car-share, this will also drive the system toa higher cost. One may predicts that if the user-equilibrium or sys-tem optimum can be reached, they will not be stable consideringthese situations. Table 2 shows the costs for N = 6. Similar type ofsocial dilemma situation as the case of N = 7 presents in this case.However, the situation is less severe since the system optimum canbe reached when 4 persons choose car-share leaving two personsenjoying less cost by choosing car-alone (compared with 6 personsrequired in N = 7). The average cost at the system optimum is£4.17, which is also cheaper than the case of N = 7 (£4.36). It meansthat the number of participants in the group with N = 6 who choosecar-share might be expected to be less than that of N = 7 (assumingperfect knowledge and utility maximisation by all participants).This effect of group size is tested on the experimental data theexperiments as presented in the next section.

3. Analyses of results

To make the results comparable and to ease aggregation, theproportion rather than the number of car-sharers in each of the 6groups is presented in the analyses. This proportion is defined asthe ratio between the number of car-sharers and the group size(N). The mean of contribution for each group is provided in Table 3.The analysis indicates that there are always a number of car-shar-ers in each round of the experiment. Hence the system is never atits user-equilibrium point where all group members choose car-alone. The system optimum, where the system cost is at a mini-mum, is reached in some (mostly non-consecutive) rounds. How-ever, it is an unstable situation as the system continues tofluctuate even after reaching the system optimum. One reason

could be that the temptation to ‘free-ride’ by choosing car-alonewhen other people choose car-share has made the situation unsta-ble even after a high level of cooperation (large number of car-sharers) is reached. It may also be due to the cognitive limitationsof the participants; they may not fully understand either the ‘coop-erative’ or ‘non-cooperative’ strategy in a social dilemma situation,as their choices might be mainly based on own experience and/orinformation about other participants’ choices.

3.1. The effects of between- and within-subject variables

This section addresses Hypothesis 1.

3.1.1. (H1) Providing more information will change individuals’behaviours in making choice

An ANOVA test is conducted (Table 4) to test the effects of be-tween-subjects variable and within-subjects variables. As dis-cussed in the last part of Section 2, group size affects the severityof the social dilemma situation, thus it may influence individuals’behaviour and is an important variable that needs to be tested inthe ANOVA test as a between-subjects variables. The within-sub-jects variables are the type of treatment and round, which are existwithin each individual as they experience all three treatments and10 rounds in each treatment.

While the effect of the sequences of individuals’ experience un-der different type of treatments (the sequence of treatments iseither Treatments 1-2-3 or Treatments 1-3-2) may also have an ef-fect on the results. To test the effect of the order of treatment, t-Tests was conducted to directly test the difference of individuals’behaviour in Treatments 2 and 3, between whether it was given

43% 43% 38%

35% 30% 45%

23% 28% 18%

0%

25%

50%

75%

100%

1 2 3

Treatment

% o

f par

ticip

ants

non-cooperative indifference cooperative

Fig. 1. Distribution of participants’ cooperativeness levels.

Table 5aTurnover in participants’ cooperativeness level from Treatment 1 to 2.

From To Treatment 2 Total

Low Medium High

Treatment 1 Low 12 5 0 17Medium 4 6 4 14High 1 1 7 9Total 17 12 11 40

Note: Positive changes = 9; negative changes = 6; asymmetric churn.

Table 5bTurnover in participants’ cooperativeness level from Treatment 2 to 3.

From To Treatment 3 Total

Low Medium High

Treatment 2 Low 12 5 0 17Medium 3 9 0 12High 0 4 7 11Total 15 18 7 40

Note: Positive changes = 5; negative changes = 7; asymmetric churn.

262 Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266

before or after Treatments 3 and 2 respectively. Treatment 1 wasnot used in this analysis as all groups received Treatment 1 asthe first treatment. The tests revealed that there is no significantdifference resulted from the order of treatments, therefore this var-iable is not included in the ANOVA test.

The ANOVA test (see Table 4) found that group size has no sig-nificant effect which means than the difference in severity of socialdilemma situation between group size of 6 and 7 did not affect thebehaviour of participants. For the within-subjects variables the ef-fect of the variable treatment is significant at a = 0.10. This givessome indication that providing participants with more informationabout other participants’ behaviour may influence individuals’behaviour (with confidence level of 90%). The variable round is sig-nificant at a = 0.05, which indicates the existence of a learning pro-cess. However, the proportion of car-sharers decreases over timewhich means that the more information regarding other partici-pants the participants received over the round, the more they areinclined to choose car-alone. The post-experiment survey revealedthat participants understood that if more people choose car-sharethen the less likely it is they will get a penalty. This means that ifall participants are bandwagon individuals then more participantswill choose car-share when the critical mass of cooperation (choos-ing car-share) is reached. However the existence of opportunisticindividuals, who ‘free-ride’ by choosing car-alone whenever mostof the others choose car-share, might have encouraged those band-wagon individuals not to choose car-share thus producing lowcooperation (fewer car-sharers).

The survey shows that own experience is used by most partici-pants before making a decision in Treatment 1. The strength ofinfluence of past-experiences varies between individuals. ForTreatment 2, there is no clear evidence that can be found onwhether participants were influenced by (or learning from) an-other participant’s behaviour before making a choice in Treatment2. The results from the post-experiment questionnaire also showthat the degree of influence varies evenly across participants. ForTreatment 3, the findings from the interview and questionnairesupport the existence of social learning mechanism where individ-uals follow other individuals’ choices.

3.2. Frequency of choosing car-share

This section addresses Hypothesis 2.

3.2.1. H2: Providing more information will make individuals choose amore cooperative choice

The frequency of car-share choices among participants was ap-plied as an indicator of the changes in their level of ‘cooperative-ness’. A Chi-Square test on the frequencies of choosing car-sharein the three treatments found that they do not belong to the samedistribution (v2 = 8.1, df = 2, a < 0.05), which means that the differ-ences between the frequency data in the three treatments are sig-nificant. Three levels based on car-share frequency within each 10-round treatment are used: low (0–3); medium (4–6); high (7–10).Fig. 1 shows the distribution of participants by their level of coop-erativeness in the three treatments. Treatments 1 and 2 have sim-ilar proportions of members in all three groups, while Treatment 3has the highest number of participants with 45% ‘medium’ level ofcooperativeness. This could be an indication that the situation inTreatment 3, participants’ decisions are more deliberate and arebased on greater consideration of information from previousrounds so that they are less fixed in choosing car-alone or car-share. A further explanation is that due to cognitive limitations,the relatively large amount of information provided to them andthe variability of information, participants failed to observe andlearn the differences between choices.

Providing more information was expected to increase the levelof cooperation more than in the situation in which only limitedinformation was provided (in Treatment 2), as it was hypothesisedthat when provided with more information people would tend tomake more cooperative choices. However the experiment showedthat this has not been the case, as having more information hasmade individuals less cooperative, less stable with their choices,and made contrarian responses to information (when others coop-erate, he/she will defect). Unfortunately we have limited under-standing of this behaviour; a possible explanation of it might berelated to the so-called ‘payoff variability effect’, a phenomenonobserved in repeated-choice situations: decision-makers’ choicesbecome uniformly distributed as their variability increase and itseems to move individuals’ choice behaviour toward randomchoice (see Myers and Sadler, 1960; Erev and Barron, 2005; andtravel choice applications in Avineri and Prashker, 2005, 2006,and Ben-Elia et al., 2008). Hence the provision of additional infor-mation is expected to encourage participants to make more ‘coop-erative’ choice (car-share) actually made them less decisive ineither choosing car-share or car-alone.

Changes in cooperativeness levels between treatments are pre-sented in Tables 5a–c. A positive change means a change from alower to a higher level of cooperativeness (e.g. from ‘‘low’’ to‘‘medium’’ or from ‘‘medium’’ to ‘‘high’’), while a negative changeis a change in the opposite direction. It is found that comparedwith a situation where no social information is provided (as inTreatment 1), limited social information provided (Treatment 2)produces more positive than negative changes (9 positive and 6negative). However, more complete information provided in

Table 5cTurnover in participants’ cooperativeness level from Treatment 1 to 3.

From To Treatment 3 Total

Low Medium High

Treatment 1 Low 11 6 0 17Medium 4 9 1 14High 0 3 6 9Total 15 18 7 40

Note: Positive changes = 7; negative changes = 7; symmetric churn.

40% 43% 35%

38% 43% 55%

23% 15% 10%

0%

25%

50%

75%

100%

1 2 3

Treatment

% o

f par

ticip

ants

stable unstable highly unstable

N=40

Fig. 2. Distribution of participants’ frequencies of switching between choices (0–2:stable; 3–5: unstable; 6–9: highly unstable).

Table 6t-Test rate of switching within-treatment.

Pair Rounds Mean Var. t Sig.

Treatment 1 1st block 0.425 0.090 2.980 0.005a

2nd block 0.290 0.070

Treatment 2 1st block 0.338 0.082 0.948 0.3492nd block 0.295 0.084

Treatment 3 1st block 0.288 0.095 -1.733 0.091b

2nd block 0.390 0.061

Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266 263

Treatment 3 compared to limited information provided in Treat-ment 2 produces more negative than positive changes (5 positiveand 7 negative). Between Treatments 3 and 1 there is a similaramount of changes in both directions (7 positive and 7 negative).These changes may be small or unobserved in the analysis ofaggregated behaviour but actually involve a number individualsbeing persuaded to change their behaviour by changes to theirenvironment. This phenomenon is known as the ‘churn’ effect(Chatterjee, 2001), which in travel behaviour context means thatmany individuals were being tempted and persuaded to changetheir behaviour by changes to their environment and circum-stances but only slow and small apparent changes could be ob-served in the overall picture of behaviour.

Note:a Significant at level of significance a = 0.05.b Significant at level of significance a = 0.1.

3.3. Frequency of switching between choices

This section addresses Hypothesis 3.

3.3.1. H3: Providing more information will make individuals moredecisive in making choice

To study the nature of the learning process in the experimentparticipants’ frequency of switching between choices was ana-lysed. A switching is recorded when a choice made by a participantin current round is different from the choice made in the previousround. The maximum number of switches in each treatment is 9times. Each individual’s frequency of switching between choicesis measured by accumulating the number of changes betweentwo consecutive rounds. In a Chi-Square test on the frequenciesof switching between choices in the three treatments it was foundthat the frequency data in the three treatments did not come fromthe same distribution (v2 = 8.086, df = 2, a < 0.05), which meansthat the differences between the frequency data in the three treat-ments are significant. We categorised the frequencies of switchingbetween choices into three categories: stable (freq. 0–2), unstable(freq. 3–5), and highly unstable (freq. 6–9).

Fig. 2 shows the distribution of the frequency of switching amongparticipants. The number of ‘unstable’ group members increasesfrom Treatment 1 to Treatment 2, and further increases in Treatment3. In Treatment 2, the increase may be compensated by the reductionin the number of members of the ‘highly unstable’ group. However,in Treatment 3 both the ‘stable’ and ‘highly unstable’ groups havesome members who become ‘unstable’. These results indicate thatproviding social information to participants regarding the choicesof all other participants may have influenced them and made themless decisive. Thus, it has made them switch between choices moreoften than they have done in the other two treatments.

A significant reduction of the rate of switching between choicesat the final state compared to that of the initial state can be an evi-dence of learning, since individuals stick to a particular choice as aresult of reinforced experience of choosing the choice (e.g. Kimble,1961). This situation happened when individuals only use theirown experience as the only source of information in their learningprocess (as in Treatment 1). However, when social information isprovided, individuals’ stability of choices may be affected. Individ-

uals may switch between choices more often as they try to antici-pate the choices of other individuals. Table 6 shows the changes inthe rate of switching between the first block (Rounds 1–5) and sec-ond block (Rounds 6–10) of each treatment, as a measure of thestability of choices.

In Treatment 1, the rate of switching in the first block is signif-icantly higher than the second block. This is an indication of anindividual learning process within this treatment which resultsin a more stable situation (less switching) in the second block ofthe treatment. While no significant change is found in Treatments2 and 3 the rate of switching in the second block is significantlyhigher than that of the first block. The social information providedin this treatment might have an effect on the choices of partici-pants; this information might influence participants to switch be-tween choices after observing the choices of other participants inthe previous round(s).

The widely perceived belief is that providing travellers withmore information helps them to make better choice decisions(and higher payoff for them). However, in this study, it is indicatedthat providing more (social) information increases uncertainty atthe individual level. The reason could be that participants knowwhat other participants are doing and they might react accordinglyby changing their choices, thus producing a more unstable situa-tion (higher rate of switching between choices).

The ‘payoff variability effect’ may again explain the instabilityof choices since the uncertainty produces high payoff variability.This may also be an indication of liability in attitudes (Beale andBonsall, 2007), which can be expected if individuals lack commit-ment to either a given attitude profile or are subject to conflictinginformation. The latter is more likely to happen in this experiment,thus resulting in instability of choices. The reason may due to thesocial information provided to them which may be in conflict withtheir experience or previously-received social information.

3.4. Response to outcomes

This section addresses Hypothesis 4.

Table 7Variables for calculating Yule coefficient.

Previous round’s outcome Num. of changes Num. of stays

High cost (bad payoff) C� S�Low cost (good payoff) C+ S+

264 Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266

3.4.1. H4: Providing more information will make individuals makedirect responses rather than contrarian responses

A response mode analysis using Yule coefficients is used to ana-lyse participants’ responses to the outcomes of their decisions (e.g.Selten et al., 2004). Participants with direct responses expect thatthey will pay similar ‘high’ cost if they choose similar choice, sothey change their choice in the next period (round). While thosewho have contrarian responses expect that other participants willchange their choices into the current ‘low’ cost choice, so theychange their choices in the next period. In this experiment it is as-sumed that costs perceived as high (‘bad’ payoff) are costs of mak-ing a choice and getting a penalty (£6.5 and £7.5 for car-alone andcar-share respectively) and costs perceived as low (‘good’ payoff)are costs of making a choice and not getting a penalty (£3.5 and£4.5 for car-alone and car-share respectively). To calculate a Yulecoefficient information needs to be collected for every participant,as shown in Table 7.

A Yule coefficient (Q) is then calculated as follows in Eq. (2). TheYule coefficient has a range from �1 to +1. A high Yule coefficientreflects a tendency towards direct responses and a low one reflectsa tendency toward contrarian responses. In each of the three treat-ments there are some participants for whom no Yule coefficientcould be determined since the participants have not paid highcosts in the treatment. Fig. 3 shows the distribution of the experi-mental Yule coefficients for Treatments 1, 2 and 3.

Q ¼ c�:sþ � cþ:s�c�:sþ þ cþ:s�

ð2Þ

One can interpret these results in the context of the social infor-mation schemes participants were provided with. When there is noinformation about other participant’s choices (as in Treatment 1;Fig. 3a), most of the participants have strong direct responses bychanging their choice when they experience high cost (MeanQ = 0.57). When participants receive social information regardingthe choices of another participant (Treatment 2, Fig. 3b) some ofthem make strong contrarian responses. However there are still alarger number of participants who have strong direct responses(Mean Q = 0.31). In these two treatments only a small number ofparticipants cannot be classified into one of the two types. In Treat-ment 3 (Fig. 3c) (when participants receive social informationregarding the choices of all other participants) more participantshave strong contrarian responses, thus creating more balance be-tween direct and contrarian responses (Mean Q = 0.16).

Fig. 3. Yule coefficients for (a) Treatment 1

The number of participants who cannot be classified into eitherof the response types is also larger in Treatment 3 than the othertwo treatments. These findings indicate that participants may haveutilised the social information for developing a strategy in mini-mizing their cost. Thus the more information they have the morepeople make contrarian responses.

4. Limitations

The laboratory experiment conducted in this study has somelimitations, including the small sample size (n = 40) and the hypo-thetical experimental setting presented in the experiment. A groupof 40 experiment participants might be seen as small size of sam-ple. However, such a sample size is not uncommon is psychologicalexperiments. For example, Marszalek et al. (2011) observed thatthe median sample size reported in four leading APA (The Ameri-can Psychological Association) journals at the year 2006 was 40.Studies in behavioural sciences cited in this work have usually ap-plied small sample sizes; for example, the sample size reported instudies of dynamic choice making reviewed by Erev and Barron(2005) range from 10 to 24. It is not uncommon in travel behaviourliterature to have experiments designed over relatively small sam-ples – for example 47 participants at Avineri and Prashker (2006),49 participants at Ben-Elia et al. (2008), and 23 participants atRamadurai and Ukkusuri (2007). It is important to mention thatthe unit of analysis in these and similar experimental studies isnot the individual participant but decisions made by participants,and therefore sets a reasonably large database. For example, theamount of data collected in this work offers a reasonably largenumber of decision observations (40 participants � 10 rounds � 3treatments = 1200 observations) as the study takes an approachof understanding the decision process of each individual, then find-ing a commonality of behaviour across all the rounds in the threeexperimental treatments for the respective individuals. Somemight argue that the issue of relatively small sample size in psy-chological research (and some of the travel behaviour research in-spired by it) might be inadequate, as it is closely tied to powereffect (the measure of the strength of a phenomenon) and effectsize. Ideally, future work in this area will explore the behaviouralresponses within larger-scale sample sizes, and might contributeto validation and generalisation of results presented in this work.

There are diverse decision-making styles and behaviours ofindividuals in the experiment, and the diversity is also likely tobe the case among the population in real-world situations. In orderto generalise the findings of the laboratory experiment to a widerpopulation further experimental research should be carried outwith different group sizes and also in other experimental settings.

In the laboratory experiment, verbal or face-to-face communi-cation between participants was not facilitated although face-to-face communication might be expected to have a greater influenceon individuals’ behaviour than observation of other individuals

, (b) Treatment 2, and (c) Treatment 3.

Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266 265

choices or even ‘anonymous’ communication (e.g. through ‘chat’services, such as in Sunitiyoso et al., 2011a). Effects might be differ-ent if participants already know one another, in which case sharedsocial norms and concerns about reputation might influencebehaviour within the group. Cooperation in social dilemmas mightimprove when participants communicate with one another face-to-face. Thus, face-to-face communication has a potential to be ex-plored in the study of social interaction and may have importanteffect on participants’ behaviour during an experiment (see, forexample, Bartle et al., 2013).

Experiments provide a useful approach to the study of complexhuman decision systems, but they are primarily intended to devel-op underlying theoretical constructs or behavioural mechanisms ofindividuals (Mahmassani and Jou, 2000). They are based primarilyon simulated situations which may not necessarily correspond toactual setting, as the factors considered may be much simpler thana real life situation. Laboratory experiments are favoured due tothe ability to control the environment and confounding variableshowever, this may cause a lack of external validity that may makeindividuals who participate in the experiment behave differentlyfrom their behaviour in real life. To gain better insights on thebehaviour of participants during the experiment as well as to as-sess whether the validity problem affect their behaviour, post-experiment questionnaire and interview survey were conducted.

5. Conclusions

The study utilised an innovative methodology involving labora-tory experiments to capture the role of social interaction and sociallearning in the dynamics of traveller decision making over time.Small scale experiments with six groups of participants were con-ducted using a hypothetical scenario of car-sharing in a universitycontext.

The laboratory experiments reveal that social interaction andsocial learning influenced individuals’ behaviour. It is also foundthat social interaction affects individuals’ level of cooperativeness,their stability of choices and their responses to the outcomes re-ceived. Providing more social information makes people less coop-erative and more unstable in making choices. It is also found thatthe more information they have, the more people make contrarianresponses rather than direct responses.

It was observed that in Treatment 2 the cooperation level hasincreased; however it has decreased again during the Treatment3 until it has reached the same level as in Treatment 1. The reasonsfor that may be explained by the analysis of (a) level of coopera-tion, (b) stability of choices, and (c) response mode analysis. Look-ing at the level of cooperation, it has decreased in Treatment 3. Thechoices also became unstable in Treatment 3 and the participantswere expecting others to change their choices rather than havingtheir own choices changed (‘contrarian responses’). These showedthat when full information is provided and people know whateveryone else is doing, then it created a temptation to become anopportunist and the social dilemma arise again in similar way aswhen there is no information about other participants’ choices(in Treatment 1).

This provides an insight on the information acquisition processas more social information does not necessarily improve travellers’abilities in making decisions. Moreover, it may also unexpectedlycreate difficulties for travellers due to their cognitive limitationsin processing large amounts of information and/or in dealing withthe uncertainty created by the social information. The experimentalso shows a process of behavioural change within each individual,which involves a gradual process of adaptation where individualsmake decisions to initiate changes. The change may be small sincepeople in general want to preserve the status quo. This may be due

to internal inhibitors (e.g. habit) or external influence (e.g. expect-ing others to change before they change themselves).

The study highlights benefits of travel behaviour studies using alaboratory experiment. It is demonstrated in the study that a labo-ratory experiment can be used as a tool to capture the dynamics oftravellers’ decision making over time under various informationschemes, both in terms of insights into overall system or aggregatebehaviour and into the underlying individual behavioural process.This shows the potential of the research method to test various set-tings or treatments which resemble scenarios of a potential trans-port policy measure in order to gain informed insights of howtravellers would respond to each scenario. The insights could laterbe useful in designing transport policy measures that are poten-tially effective in influencing travellers’ behaviour (e.g. behaviouralchange programmes, user communication features of AdvancedTraveller Information Systems). Another potential applicationwould be in incorporating social interactions and social learningin traditional models of choice behaviour that assume individualsdo not have (or do not value) social information, for example bytaking the choices of other individuals with whom an individualinteract into account as one of attributes of the individual in themodels.

Despite its potential benefits, a laboratory experiment shouldnot be considered as a substitute to a field study, since in an exper-iment it would not be possible to represent precisely travellers’real decision situation and all factors that influence their decisions.An experiment is generally based on a simulated situation whichmay not necessarily correspond to actual setting, as the factorsconsidered may be much simpler than a real life situation. If a fieldstudy is possible, then real world observations of behaviour couldbe used to confirm the substantive conclusions resulting from suchexperiment. The findings of the laboratory experiment can also beexplored further in a simulation model (e.g. Kameda and Nakani-shi, 2002; Sunitiyoso et al., 2011b). A simulation model could rep-resent large-scale environment of more complex nature (numberof individuals and interactions between them) and over a largetime period. This enables the researcher to observe whether andwhen individuals’ choices are likely to be converged to an equilib-rium point, how they converge, and the dynamics before conver-gence. Further empirical investigation of more realistic situationsand observation of actual behaviour in a field study are requiredto validate the experimental findings of this study.

References

Arentze, T., Timmermans, H., 2005. Modelling learning and adaptation intransportation context. Transportmetrica 1 (1), 13–22.

Avineri, E., Prashker, J.N., 2005. Sensitivity to travel time variability: travellers’learning perspective. Transportation Research C 13 (2), 157–183.

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

Bartle, C., Avineri, E., Chatterjee, K., 2013. Online information-sharing: a qualitativeanalysis of community, trust and social influence amongst commuter cyclists inthe UK. Transportation Research Part F: Traffic Psychology and Behaviour 16 (1),60–72.

Beale, J.R., Bonsall, P.W., 2007. Marketing in the bus industry: A psychologicalinterpretation of some attitudinal and behavioural outcomes. TransportationResearch Part F: Traffic Psychology and Behaviour 10, 271–287.

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

Bonsall, P., 2002. Motivating the respondent: how far should you go? In:Mahmassani, H.S. (Ed.), Perpetual Motion: Travel Behavior ResearchOpportunities and Application Challenges. Elsevier.

Chatterjee, K., 2001. Asymmetric Churn – Academic Jargon or a Serious Issue forTransport Planning? <www.tps.org.uk/files/Main/Library/2001/0001chatterjee.pdf>.

DfT, 2004. Transport analysis guidance: values of time and operating costs. <http://www.dft.gov.uk/webtag/documents/expert/unit3.5.6.php>.

Douglas, J.D., 1974. Understanding Everyday Life. London, Routledge & Kegan Paul.Erev, I., Barron, G., 2005. On adaptation, maximization and reinforcement learning

among cognitive strategies. Psychological Review 112 (4), 912–931.

266 Y. Sunitiyoso et al. / Journal of Transport Geography 31 (2013) 258–266

Fischbacher, U., 2007. Z-Tree: Zurich toolbox for readymade economic experiments.Experimental Economics 10 (2), 171–178.

Hertwig, R., Ortmann, A., 2001. Experimental practices in economics: amethodological challenge for psychologists? Behavioural and Brain Sciences24, 383–451.

Jones, P., Sloman, L., 2003. Encouraging behavioural change through marketing andmanagement: what can be achieved? In: 10th International Conference onTravel Behavior Research. Lucerne, Switzerland.

Kameda, T., Nakanishi, D., 2002. Cost/benefit analysis of social/cultural learning in anonstationary uncertain environment: an evolutionary simulation and anexperiment with human subjects. Evolution and Human Behavior 23, 373–393.

Kimble, G.A., 1961. Interpersonal Relation: A Theory of Interdependence, second ed.Prentice Hall, Englewood Cliffs.

Kitamura, R., Nakayama, S., Yamamoto, T., 1999. Self-reinforcing motorization: cantravel demand management take us out of the social trap? Transport Policy 6,135–145.

Mahmassani, H.S., Jou, R.C., 2000. Transferring insights into commuter behaviourdynamics from laboratory experiments to field surveys. TransportationResearch A 34, 243–260.

Mahmassani, H.S., Liu, Y.H., 1999. Dynamics of commuting decision behaviourunder advanced traveller information systems. Transportation Research C 7,91–107.

Marszalek, J.M., Barber, C., Kohlhart, J., 2011. Sample size in psychological researchover the past 30 years. Perceptual and Motor Skills 112 (2), 331–348.

Messick, D.M., 1985. Social interdependence and decision making. In: Wright, G.(Ed.), Behavioral Decision Making. Plenum Press, New York.

Myers, J.L., Sadler, E., 1960. Effects of range of payoffs as a variable in risk taking.Journal of Experimental Psychology 60, 306–309.

Offerman, T., Sonnemans, J., 1998. Learning by experience and learning by imitatingsuccessful others. Journal of Economic Behavior and Organization 34, 559–575.

Olson, M., 1965. The Logic of Collective Action: Public Goods and the Theory ofGroups, second ed. Harvard University Press.

Ramadurai, G., Ukkusuri, S.V., 2007. Dynamic traffic equilibrium: theoretical andexperimental network game results in single-bottleneck model. TransportationResearch Record 2029, 1–13.

Selten, R., Schreckenberg, M., Pitz, T., Kube, S., Hafstein, S., Chrobok, R., Pottmeier, A.,Wahle, J., 2004. Experimental investigation of day-to-day route choice behaviorand network simulations of Autobahn traffic in North Rhine Westphalia. In:Schreckenberg, M., Selten, R. (Eds.), Human Behavior and Traffic Networks.Springer, Berlin, pp. 1–21.

Sunitiyoso, Y., Avineri, E., Chatterjee, K., 2009. The role of minority influence on thediffusion of compliance with a demand management measure. In: Kitamura, R.,Yoshii, T., Yamamoto, T. (Eds.), The Expanding Sphere of Travel BehaviourResearch. Emerald, UK, pp. 643–672.

Sunitiyoso, Y., Avineri, E., Chatterjee, K., 2011a. The effect of social interactions ontravel behaviour: an exploratory study using a laboratory experiment.Transportation Research A 45 (4), 332–344.

Sunitiyoso, Y., Avineri, E., Chatterjee, K., 2011b. On the potential of social interactionand social learning in modelling travellers’ change of behaviour underuncertainty. Transportmetrica 7 (1), 5–30.

Taniguchi, A., Fujii, S., 2007. Promoting public transport using marketing techniquesin mobility management and verifying their quantitative effects. Transportation34, 37–49.

Van Lange, P., Van Vugt, M., De Cremer, D., 2000. Choosing between personalcomfort and the environment: solutions to the transportation dilemma. In:Vugt, M.V., Snyder, M., Tyler, T.R., Biel, A. (Eds.), Cooperation in Modern Society:Promoting the Welfare of Communities, States and Organizations. Routledge,London.