Smart Mobility : influencing transportation mode choice with behavioural economics

Smart Mobility

Influencing transportation mode choice with behavioural economics

Milo Boers, 10529047 August, 2018

University of Amsterdam

Statement of Originality

This document is written by Student Milo Boers who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Ethical statement

This research was conducted during several concert nights in the ArenaPoort area of Amsterdam, in collaboration with the Amsterdam ArenA. The concert nights are specified as event 1, event 2, etc, instead of by exact name and date of the concerts.

During the research departure addresses and IP addresses from subjects were gathered. According to the May 25th, 2018 GDPR European data protection law, these are counted as personal information and are only allowed to be registered upon consent. During this research care was taken to comply to these regulations. The logging of this data is mentioned in the online privacy statement. Furthermore, the logging of IP address was specifically mentioned during conducted surveys. During the survey people could give consent for the logging of their IP address and only IP addresses were logged for people who gave consent.

IP addresses were only used for the purpose of trying to match survey data with data logged on the online portal. From departure addresses only the zip-4 code was used to match neighbourhood level statistics based on open data from the Dutch Central Bureau for Statistics.

All data is anonymised by default, was not used for individual profiling during the research and was removed afterwards. No subject was negatively influenced by unknowingly partaking in this research. In essence, the research did no more than trying to nudge people from taking the car to taking public transport, by either informing them that other people do the same or that public transport is better for the environment.

Summary

To test whether people’s transport mode choice can be influenced with behavioural economics, a natural field experiment was conducted. During a real life travel choice under information design subjects were confronted with randomly assigned persuasive texts, to see to what extent people can be nudged to come to a concert by public transport instead of by car. The strongest effect was caused by providing a social norm related to the specific event, with a reported effect of 8.2% in reducing interest to come by car. Although not all significant, the more specific the social identity, the stronger the effect, with 1.0%, 1.9% and 8.2% in order of specificness. However, environmental concerns directly followed the most specific social norm, increasing interest in public transport with 4.7%. In addition, a heterogeneity analysis was performed, which showed that there are differences between how people react on different treatments, based on their economic background.

Table of content

1. Introduction

In this research it was tested to what extent behavioural economics can be applied to

influence the transportation choice of people. The focus was to reduce transportation by car in favour of transportation by public transport since this reduces carbon emissions, local traffic jams and air and noise pollution.

1.1 Motivation

It becomes increasingly important for companies to reduce their direct and indirect carbon footprint, this can, amongst others, be done by incentivizing both their employees and their customers to behave more environmental friendly. For instance, make agents not come by car, but alternative modes of transportation instead.

Making agents come less by car can also have all sorts of other benefits, such as reduced traffic jams, reduced local air pollution and noise disturbance and over time may even lead to decreased need for infrastructure such as additional roads and parking spaces. (Minett, 2015) Companies, governments and other organizations can make use of behavioural economics to influence agents according to their strategy, instead of incentivizing via financial measures, as according to traditional economics. (Kondyli, 2017)

The field of transportation economics has not seen much application of behavioural economics yet. Especially in the form of real life application (Garcia-Sierra et al., 2015). Large hubs such as stadiums, airports, business districts and shopping malls make an interesting case for this, as they see a lot of event related traffic. Since these trips are less habitual than daily commutes it was hypothesised that they would make an easier case to see if people can be influenced in their transportation choice with behavioural economic insights.

1.2 Research question

The main goal of this research is to show whether behavioural economic insights can be used to persuade concert visitors to come by other modes of transportation. More specifically by using descriptive norms. This is tested in the setting of a field experiment. During several concert nights, visitors are persuaded to come by public transport instead of by car. Therefore, the following research question was postulated:

Main question

To what extent can descriptive norms be used to persuade people to come to an event by public transportation instead of by car?

Sub questions

- What kind of descriptive norms are most effective?

- How does this relate to factors such as travel distance, time and economic background? - How does this relate to people’s stated preferences?

1.3 Contribution

A field experiment was conducted to test to what extent descriptive norms can be used to persuade people to come to an event by public transport instead of by car. It was shown that nudging visitors with descriptive norms significantly decreased their interest in coming by car, in favour of public transport. The strongest effect is caused by providing a social norm related to the specific event, followed by environmental concerns. Furthermore, it was shown that the extent to which people react on different norms, is related to their social-economic background. Not much literature is yet available where behavioural economics is tested in the field of transportation. Most related literature with respect to this is psychological research, where people are asked for their hypothetical choice preferences in surveys. This experiment shows both the applicability of norm provision as a possible way to influence people’s transportation behaviour in a practical transportation context and adds to the broader set of literature on economic research to nudge people’s behaviour in different contexts.

1.4 Structure

Literature with respect to a classical economic approach to incentivising people to change their transportation behaviour is combined with insights from behavioural economics. Next the experimental method and experimental data handling is explained in the methodology. In the results section, first the summary statistics and main graphs are presented for the experiment, followed by a regression and heterogeneity analysis. Next the survey results are presented to add more background information with respect to the demographic context of the experiment. The results are interpreted and their internal and external validity discussed in the discussion session. The report concludes with summarising the main answers to research question and hypotheses and recommendations for further research.

2. Literature review

This research builds upon the economic framework of travel choice modelling, however instead of classical incentives people are nudged. The experiment is a real life travel options choice under information design, comparable with the lab experiment of Gaker et al. (2010), combined with the persuasive message design of Goldstein et al. (2008).

2.1 Transport economics

Governments can discourage overall car usage and can specifically discourage high fuel consuming vehicles by raising Pigouvian taxes and/or by subsidising more fuel efficient cars and alternative modes of transportation. With respect to congestion management, congestion pricing can be applied by regional traffic managers such as municipalities or local governments. First best congestion pricing would mean real time tolls that match with the degree of congestion at the location, given the type of vehicle and weather conditions and other current circumstances such as accidents. While first best pricing would maximize efficiency, it is difficult to practically apply. (Lindsey & Verhoef, 2001).

The final transportation choice that agents make is modelled as a utility optimisation problem. An agent has several choice options and certain preferences. Some of these preferences are comparable with agents who belong to the same social-economic subgroups, for instance educational level, income, age, are dependent on access to facilities such as nearest train station and roads. Finally, with relation to the actual quality of these facilities. These factors are then often modelled in a multinomial logit or probit model, to define the effect of the situation specific parameters on a certain choice. (Koppelman & Bhat, 2006)

2.2 Psychology in transportation

Financial incentives to discourage car usage in general and at certain times at certain locations are often highly unpopular and can crowd out intrinsic motivation.. With regard to car use reduction popular psychological frameworks are Theory of Planned Behaviour (TPB) and the Norm Activation Model (NAM), which are more psychologically oriented. Both models are based on the idea that behaviour is a result of behavioural intention, which is affected by perceived behavioural control, attitude and especially personal norms. These are directly affected by feelings of guilt and social norms and indirectly by awareness and perceived responsibility (Bamberg et al. 2011). TPB and NAM are relatable to traditional economic analysis, in which the actor is a rational agent that behaves in a way that optimizes utility according to his or her preferences. To what extent the different components of the

model explain the variation in behaviour, can be analysed with path analysis and can be empirically tested (Gardner & Abraham, 2008). Multiple empirical researches on reducing car usage have partially or mainly focused on social norms within the TPB and NAM framework and have shown that these can be very effective measures to influence intended behaviour (Harland et al., 1999; Nordlund & Garvill, 2003; Cialdini, 2003; Matthies et al., 2006; Bamberg, Hunecke, & Blöbaum, 2007; Zhang, Schmöcker, Fujii, & Yang, 2016; Olsson, 2018).

2.3 Behavioural economics

The above mentioned researches used surveying to analyse to what extent people agreed they would be influenced by certain phrases. However, meta-analytic research shows that intentions or stated preferences have rather low predictive power on actual behaviour change. Only a small percentage of the variance in causal studies can be explained by intentions (Metcalfe & Dolan, 2012). A very often cited article in that sense is: “A Room with a Viewpoint “. In the research several descriptive norms are used to request hotel guests to use their towel more often. The research successfully showed that people actually don’t behave as they state in surveys. While people stated in surveys that they would be strongly persuaded by environmental reasons and least by social norms related to the location, actually the opposite was shown to be true (Goldstein et al., 2008).

Standard economic theory and some psychological theories (such as used in the field of transportation or market research) are based on the assumptions that humans behave fully rational according to their preferences. Although these assumptions make it relatively easier to calculate and optimize, behavioural economics challenges these underlying assumptions by showing human biases, combining economics with psychological principles. People’s behaviour is not consistent, but context related. MINDSPACE is a mnemonic to summarize this contextual impact. It stands for Messenger, Incentives, Norms, Defaults, Salience, Priming, Affect, Commitments and Ego. Above described problems of the causal relationship between incentives and actual behaviour are caused by biases related to those behavioural impacts: People overly dislike losses, focus on change from their status quo, overweigh small chances, think in discrete bundles, overvalue right now versus the future, care about other people and can be negatively impacted by their own incentives. The 2017 Nobel price’s laureate R.H. Thaler received the price in economics for his work in the field of behavioural economics, amongst others on Nudging theory. Nudging tries to provoke non-forced

compliance to behaviour that is both beneficial for the individual and the group, by making smart usage of behavioural biases. (The Committee for the Prize in Economic Sciences in Memory of Alfred Nobel, 2017). However, only few nudges have yet been applied and tested in the field of transportation (Metcalfe & Dolan, 2012).

2.3 Synthesis

Despite using behavioural insights to decrease cost, the example of (Prabhakar, 2013) in essence uses financial incentives and could therefore at least partially be regarded as classical economic persuasion. In contrary, the research of Goldstein et al. (2008) only reframed the text on cards already used to persuade hotel visitors to reuse their towel. But while the decision to throw a towel on the floor or not is small and in the moment, choosing another mode of transportation than a default option, seems like a bigger decision. Doing this on a regular basis seems like a large behavioural change. Doing this only in relation to one event seems like an acceptable change for one individual, while it can have a high impact if it shows to influence several hundreds or even thousands coming to an event.

A lab experiment on information and social norms in the field of transportation shows high transferability of behavioural economic insights (Gaker et al., 2010). There have been some natural field experiments undertaken in the related field of environmental economics (Garcia-Sierra et al., 2015). However, a comparable natural field experiment as “A room with a viewpoint” is not yet conducted within the field of transportation (Metcalfe & Dolan, 2012). This may be explained by the difficulty to both have a platform to persuasively inform people and be able to collect data about their actual transport behaviour at the same time, without them knowing they are part of an experiment.

This research aims to build upon the economic analysis framework of travel choice modelling, however instead of classical incentives people are nudged with persuasive messages.

3. Methodology

To test whether people’s transport mode choice can be influenced by providing them with descriptive norms, a natural field experiment was conducted, using the Amsterdam ArenAPoort area as a test location. The online platform ‘Mobility Portal’ was used to influence the transportation mode choice for concert visitors of four different concert nights. Their behaviour was logged via their click data. In addition, visitors were surveyed to reveal the concert demographics and people’s stated preferences with regards to their transportation mode choice. Furthermore, neighbourhood level economic statistics from the Dutch Central Bureau of Statistics were added based on zip code, to test for balance of the treatment groups and for heterogeneity in behaviour based on economic background.

3.1 Experimental design

Visitors of the concerts were informed about possible modes of transportation and via randomly assigned treatments were nudged to come by public transport instead of by car. 3.1.1 Test location

The test location, Amsterdam’s ArenAPoort area, is a city region with high rates of event related visitors coming to the area. The area functions as an entertainment, business and living area with a soccer and concert stadium, two concert halls, a large cinema, several shopping malls, Amsterdam’s second largest business district and growing inhabitant numbers. Visitors of the area can make use of an online portal, called Mobility Portal.

The Mobility Portal works very much like the Google maps route planner, but with added functionality for the ArenAPoort area. After clicking “Plan trip” the fastest routes for public transport, car and park & ride are shown, including possible information on road or rail construction works. Furthermore, people have the possibility to buy train, touring car and parking tickets or book a taxi or even hotel. The website is hosted both in Dutch and in English.

3.1.2 Intervention

After visitors have clicked “plan trip” it takes some time before the travel advice is loaded, similar to a comparison website for flights or hotels. In the meantime, a neutral banner pops up in the screen, with a loading sign and the text “Please wait while your personal travel advice is loaded.” As soon as the travel advice is loaded, the popup disappears.

During the experiment this ‘Neutral’ popup was randomly assigned to visitors of the portal with a chance of 1/5, functioning as a control group ( {T0} ). The treatment group was

randomly assigned a text meant to persuade them to come by public transport instead of by car. Four different persuasive texts were used ( {T1, .., T4} ), randomly assigned with a chance

of 1/5 each. The used messages are an adaptation to the messages used in the towel reuse experiment of Goldstein et al. (2008):

Neutral (T0): “Please wait while your personal travel advice is loaded.”

Environmental (T1): “Travel environmentally conscious.”

Regional (T2): “Join many concert visitors from your region.”

General Local (T3): “Join many fellow visitors of Amsterdam Arena”

Specific Local (T4): “Join many fellow visitors of this event”

Furthermore, all non-neutral banners showed a text asking visitors to consider to come by public transport instead of by car, to reduce carbon emissions, reduce possible traffic jams and save on fuel and parking cost, since Gaker et al. (2010) showed the persuasive importance of providing this information in their lab experiment. In addition, people are given the possibility to immediately click on buttons to either immediately get directed to ‘Public Transport’ or ‘Organised Transport’. As Metcalf and Dolan (2012) explain in their literature review, the car is the default choice for many travellers, without first considering their different options every time. Providing the greener travel options first instead, is supposed to divert the agent from their default choice and make them consider other options as well. This is a method often used in food placement, where the more expensive products in the supermarket are placed on eye sight or where the healthier options in a company restaurant are placed first. For a picture of a popup as it was shown on the website, see appendix.

3.2 Experimental data

The provided information and click data is stored per unique visitor session. Both first and last viewed transport modes are used as proxy for transportation mode choice. Information such as travel distance and travel time per transport mode is stored as independent variables. Only visitors departing from addresses in the Netherlands were considered for four different concert nights, resulting in a total of 4282 observations.

3.2.1 Stored data

Broadly speaking the data from the portal can be divided in three main categories. What information does the visitor fill out, what information is provided to the visitor and where does the visitor click on. Some data will always be gathered (all), other data is situation dependant (optional):

Table 1: Overview of information logged via Mobility Portal

Information filled out Information provided Click behaviour Departure address (all) Popup type (all) Popup: click on either

buttons or no click (all) Departure or Arrival time

(opt)

Public Transport: Travel duration (all)

Transport modes (opt) What event (opt) Car: Travel duration +

parking price (all)

Book (opt) Park & Ride: Travel duration

+ free parking (all) Organised Transport: minimum price (all) Hotel: minimum price (all) Taxi: price (all)

Bike and motor: parking spots (all)

Disabled parking: info on how to contact beforehand 3.2.2 Data gathering process

The visitor fills out his or her departure address and either a preferred departure or arrival date and time or, as most visitors, clicks on the specific event they would want to visit. Then the randomly assigned popup appears while the travel advice is being loaded. The visitor can read the text and wait until the travel advice is loaded and the popup disappears or click on either the direct to “Public Transport” or “Organised Transport” button.

In case of the Neutral popup or if the visitor does not click on either direct buttons, after loading, all travel advice appears as stated above in the “Information provided” column. From here the visitor can choose to click on a transportation mode of interest to be provided with the exact travel details. When the visitor clicked either one of the two direct buttons, the

visitor will directly be provided with the travel information for that specific travel mode. The visitor can keep clicking on different transportation modes to be provided with more detailed information. Eventually the visitor either may choose to book a train ticket, parking ticket, organised transport seat or hotel in advance and leave the portal by getting directed to an external website or leave the webpage without booking.

Since many visitors don’t directly buy their parking or train ticket online, a proxy has to be used for transport choice. It can be assumed that either the first or the last viewed transport mode has the highest chance to be chosen eventually. The visitor may either first view the mode of transport he or she is most interested in and then look at the other options out of interest or may look at different options and end with the option of highest interest. Both ‘firstviewed’ and ‘lastviewed’ transportation modes are therefore considered as variable of interst.

3.2.3 Data processing

As soon as a new visitor conducts a first action on the portal, the Mobility Portal backend data storage creates a unique, anonymous code as a visitor session identity. Every action and what information is provided to that visitor during that session will be stored as a record with that unique code and a date and time stamp. As soon as a visitor leaves the page, reloads or closes the browser, the session is over. If the same visitor would come to the portal again at a later moment in time, a new code is generated, independently from the previous visit. A visitor is therefore stored as unique visitor, not as unique individual.

During the data preparation process millions of records are traversed, grouped by the same session ids and ordered by time stamp, to reveal in what order individual visitors navigated trough the portal. Next all the records are parsed and the information of interest is stored in a common column row structure, with one row accounting for a single visitor session and its retrieved variables of interest stored in the columns.

3.2.4 Data cleaning

During the period of measurement 5074 visitors used the Mobility Portal. However, 451 left or reloaded the page resulting in 4623 actual behavioural observations. As a case study four consecutive concerts were chosen with around 1.000 visitors making use of the Mobility Portal per concert night. Searches for which no concert was specified, were taken into account as well. Since by far most searches during the measurement period can be attributed to either one of the four concerts, it is assumed that these are mainly related to either one of

those and can therefore serve as a base scenario for concert fixed effects. Other events are filtered out, resulting in 4438 observations. Next, since most concert visitors are from the Netherlands and the small sample of visitors from abroad can cause very strong outlier distortion, the 156 non-Netherlands departure locations were filtered out, resulting in 4282 observations used for the analysis.

3.3 Survey data

To gather additional information on demographics of the concert visitors, a survey was conducted. An attempt was made to connect survey data with the experimental data based on IP address, however, the overlapping sample was too small.

3.3.1 Data gathering process

The dense amount of audience packed during a concert makes it very difficult to have a good mobile 3G or 4G connection. Visitors who want to make use of the free wifi network during a concert, have to fill out their email address and agree to receive a survey. The survey is sent several days after the concert and includes questions mainly related to customer satisfaction. During the events that are part of the experiment, several questions were included in relation to the transportation mode and decision factors of respondents. They were asked what transportation mode they used, their gender, to what age group they belong, their group size, travel time back and forth, whether they used the Mobility Portal and to rank to what extent they consider several factors important in their travel choice on a one to five Likert scale (see Appendix for question form).

3.3.2 Matching

Preferentially respondents of the survey can be matched with users of the Mobility Portal. In that case chosen mode of transportation and stated preferences can be compared with which popup they have seen and their online behaviour. In order for this to have enough statistical power, the overlapping group that both used the Mobility Portal ànd filled out the survey has to be large enough (see figure). Furthermore, a way of matching is needed. The most obvious way of matching would be to ask respondents to fill out their address, as this is already logged in the Mobility Portal data. However, for a survey this was deemed as too intrusive to specifically ask for with the only goal to connect the data sets.

The logging of IP addresses is a very common way for online analytics to track users. The use of an IP address is semi-anonymously. Without information from the internet provider it is not directly traceable to the person. It was therefore decided to use IP logging instead. Mobility Portal visitor’s IP addresses are logged, which is mentioned and explained in the privacy statement. The IP addresses of the internet connection from which respondents filled out the survey was logged on respondent’s consent. Respondents were asked if they agreed that their survey results would be connected with their possible usage of the Mobility Portal. Below a yes/no option, was a dropdown menu with the text: “Why do I get this question?”. By clicking on this dropdown a text appeared explaining which data would be connected, how and for what reason, see appendix.

Despite the effort, from 2054 total respondents of which 154 said they had also used Mobility Portal, only 56 had a matching IP address. The most important reason for this is that users have to both use Mobility Portal and fill out the survey in the same home internet environment. Using a mobile device for either one, making use of another wifi network or making use of a business wifi network results in different IP addresses.

Figure 1: Schematic overview of relation between different datasets

All concert visitors

Portal users

Survey respondents IP match

3.4 Neighbourhood level economic data

Neighbourhood level economic statistics from the Dutch Central Bureau of Statistics were added based on zip code, to test for balance of the treatment groups and for heterogeneity in behaviour based on economic background. The data used is based on zip-4 code. For average age, cars per household and number of addresses per km2 the most recent data was from 2017 (CBS, 2017), for distance to train station and distance to highway the data was from 2015 (CBS, 2015) For average neighbourhood income the data was from 2015 (CBS(2), 2015) as well, but a different dataset.

3.5 Hypotheses

The main goal of the research is to test whether people can be persuaded by descriptive norms to come by public transport instead of by car. Thus, whether the chance that an individual will be interested to take public transport increases as a result of being informed with a descriptive norm instead of a neutral text. Furthermore, Goldstein et al. (2008) showed that social norms were more persuading than a text related to being environmentally conscious. The more specific to the local situation, the stronger the effect, while surveys revealed people think they would be most influenced by environmental concerns and least by local social situations. This research tests these outcomes in the context of travel mode choice, leading to the following hypotheses:

1. Informing visitors with descriptive norms increases the chance that they will choose public transport instead of the car.

2. Social norms are more persuading than environmental concerns.

3. Social norms are more persuading when they are made more situation specific. 4. In stated preferences, environmental concerns will be overvalued and social norms

4. Descriptive Norms

While the control variables are well balanced over treatments, the outcomes are not, indicating the treatment had effect. To test this a regression analysis was performed. It was shown that using descriptive norms does decrease the interest in taking the car in favour of coming by public transport, although results were not significant in all cases. In addition, a heterogeneity analysis was performed, which showed that there are differences between how people react on different treatments, based on their economic background.

4.1 Summary statistics

The summary statistics are divided into two tables. The first (table 2) contains the number of observations per event and the mean and standard deviation for the control variables. The second table (table 3) contains the different outcome variables. Both tables give these numbers for the Neutral control group, the four treatments and all groups combined in Total. The control variables should be balanced and the treatments not distinguishable, while the outcome variables should be unbalanced. From eyeballing the control variables in table 2 seem equally distributed over the treatments, while the outcome variables in table 3 seem less equally distributed. All variables are formally tested for balance with an F-test, testing the null hypothesis that a row variable does not differ for the different popups, formally described as:

!"#$%&',) = ,-+ ,/0/,)+ ,101,) + ,202,) + ,303,)+ 4) Fo& 5 = 1, … . , 5 test:

9_-: ,_/ = ,₁ = ,₂ = ,₃ = 0 vs 9_/: ,_< ≠ 0 for at least one >, > = 1, … ,4

When the p-value, in both tables reported in the right column as “Prob > F”, is more than 0.05, this would indicate that row variable is balanced over the treatments. In addition, the treatments are tested for joint orthogonality. where it is tested if a treatment can be recognised based on all row variables, formally described as:

0<,) = ,-+ ,/!"#$%&/,) + ⋯ + ,'!"#$%&',)+ 4) For > = 1, … ,4 test :

Outcomes are reported in the bottom row. Indeed, in table X no p-value below 0.05 is reported, while multiple are for table Y. While the treatments are not distinguishable based on the control variables, the Neutral popup is distinguishable from the other popups, based on the outcome variables. The different treatments and especially the specific local identity report lower interest in car and higher interest in public transport. These differences are significant for the last viewed and especially first viewed modalities.

Table 2: Mobility Portal control variables per treatment

Treatments F(4, N)

Neutral Environmental Regional General Local Specific Local Total Prob > F Concerts Eventless Search count 48 53 57 48 54 260 0.9521 Event 1 count 126 119 122 123 119 609 0.8457 Event 2 count 104 102 105 87 94 492 0.6864 Event 3 count 279 303 276 301 294 1453 0.4167 Event 4 count 273 313 311 274 297 1468 0.6649 Total observations count 830 890 871 833 858 4282 Travel info Distance [km] mean 93 92 89 90 90 91 0.4792 std 51 52 50 51 51 51 Duration Car [min.] mean 62 61 60 60 60 60 0.4434 std 29 29 28 28 29 29 Duration Public Transport [min.] mean 121 122 123 121 123 122 0.9629 std 84 87 93 87 93 89 Duration Park & Ride [min.] mean 59 59 57 58 57 58 0.2671 std 28 29 27 28 28 28 Neighbourhood Income [€] mean 35473 35640 35855 35387 35561 35586 0.6695 std 6488 7315 7214 6341 6914 6873 Age [years] mean 41 41 41 41 41 41 0.3887 std 4 4 5 4 4 4 Urban density [adresses/km2_{] mean 2090 2188 2171 2218 2217 2177 0.6471} std 1768 1928 1915 1946 1932 1899 Cars per household mean 0.98 0.97 0.97 0.96 0.94 0.96 0.2190 std 0.33 0.36 0.34 0.35 0.36 0.35 Distance to highway [km] mean 1.8 1.7 1.7 1.7 1.9 1.8 0.0315* std 1.2 1.0 1.0 1.1 1.2 1.1 Distance to trainstation [km] mean 4.2 4.4 4.3 4.3 4.6 4.4 0.4025 std 4.9 5.0 4.3 4.4 5.1 4.7 F(14, 3679) Prob > F 0.1305 0.5532 0.8918 0.7950 0.1036

Table 3: Mobility Portal outcome variables per treatment

Treatments

Neutral Environmental Regional General Local Specific Local Total F(4, 4273)

% % % % % % Prob > F Plan direct No click 75.5 77.4 75.0 74.7 75.7 0.5553 Public Transport 22.4 20.9 21.8 24.0 22.3 0.2543 Touring Car 2.1 1.7 3.1 1.3 2.1 0.0515 Total 100.0 100.0 100.0 100.0 100.0 First viewed No view 6.1 4.4 4.4 5.0 5.5 4.8 0.4851 Car 60.4 58.1 59.1 56.7 51.6 56.4 0.0036** Public Transport 22.3 28.9 28.0 27.6 30.7 28.8 0.0023** Park & Ride 7.0 5.2 4.4 5.0 7.6 5.5 0.0175* Touring Car 1.2 1.6 2.0 2.9 1.9 2.1 0.1350 Custom content 2.0 1.2 1.4 2.3 1.7 1.7 0.4175 Taxi 0.2 0.2 0.3 0.4 0.6 0.4 0.7311 Hotel 0.7 0.4 0.5 0.1 0.5 0.4 0.4849 Total 100.0 100.0 100.0 100.0 100.0 100.0 Last viewed No view 6.1 4.4 4.4 5.0 5.5 4.8 0.4851 Car 60.1 59.0 59.5 58.1 52.0 57.1 0.0049** Public Transport 18.6 23.3 21.7 21.2 24.2 22.6 0.0515 Park & Ride 9.2 7.0 8.0 7.9 9.6 8.1 0.2923 Touring Car 2.2 3.6 3.3 3.6 4.4 3.7 0.1490 Custom content 2.4 1.5 2.4 2.4 2.7 2.2 0.4746 Taxi 0.8 0.3 0.5 1.0 0.8 0.6 0.4429 Hotel 0.6 1.0 0.2 0.7 0.8 0.7 0.3537 Total 100.0 100.0 100.0 100.0 100.0 100.0 Booked No Booking 73.6 72.7 72.6 74.1 74.4 73.4 0.8973 Parking 24.7 24.7 24.9 23.3 22.4 23.8 0.6887 Train 1.1 2.1 2.3 1.9 2.8 2.3 0.1587 Hotel 0.6 0.4 0.2 0.7 0.5 0.5 0.6674 Total 100.0 100.0 100.0 100.0 100.0 100.0 F(16, 4261) Prob > F 0.0164* 0.5811 0.4296 0.2975 0.0884

4.2 Main results

The graphs visualise per treatment the distribution for first and last viewed modality pages.

Figure 2: Modal share per treatment for first viewed modalities

Figure 2 and 3 seem to indicate a decrease for car for both first and last viewed modality, although the effect seems slightly stronger for first viewed. Furthermore, the interest in public transport increases for all treatments, compared to Neutral, while other slightly decreases. Finally, a difference between first viewed and last viewed seems to be that public transport slightly decreases again in favour of park&ride. Since park & ride combines car and public transport, using the car for the largest share of the route, it has the advantage for visitors to be quick, free parking and decreases the chance of possible traffic jams. However, from a carbon emission perspective it is only slightly more environmental friendly than making the full trip by car.

4.3 Regression analysis

Since figure 1 and 2 seem to indicate some treatment effects, the effects were tested for dependant variables car, public transport and park & ride. This was both done with and without adding control variables. Not all variables reported in table X were used together, since correlation analysis revealed certain strong relationships between controls.

4.3.1 Correlation analysis

Correlation analysis revealed that travel distance has a correlation of 0.99 with travel time for car, 0.95 for P&R and 0.57 for public transport. However, the travel distance not only influences the travel time, but also the travel cost and carbon emissions. Especially for longer distance the marginal cost of the car decrease, because of the fixed parking cost. Travel time however on itself is also assumed to be an important control variable, however, not the absolute travel time is assumed to be a decision factor, but the relative travel time between transport modes. Thus, the factor of travel time of public transport versus car and of P&R versus car. Now correlations decrease to -0.10 and -0.13 for the public transport and P&R ratios to car, respectively.

Furthermore, urban density was not used as a control because it has a correlation with neighbourhood income of -0.38, age of -0.36, cars per household of -0.75, distance to a main road of 0.26 and distance to train station of -0.38. Except for a correlation between cars per household with income of 0.50 and with distance to train of 0.33 all other correlations are mainly in the around or sub 0.1 range.

4.3.2 Regression model

The different treatments are regressed on the last viewed modality as dependant variable. A binary linear probability model is used to estimate the effect of a popup on the chance of a portal visitor to view the specified modality (1) or view another (0). The analysis was done for dummies of car, public transport and P&R, individually. Since these have a relatively high share of occurrence, the linear probability model shows very comparable results with a logit or probit model, but is easier to directly interpret and was therefore preferred. The analysis as reported in table 4 was carried out both with and without control variables. As part of the controls, different concert nights are added as fixed effects, with Eventless search as left out dummy. This results in the following model:

No controls

Pr #$%&'()* = , -_., . . -₁ = 3₄+ 3_.-_.+. . +3₁-₁ With controls

Pr #$%&'()* = , -_., . . -₁, 6_., . . 6₇, 89_., . . 89₁

= 3₄+ 3_.-_.+. . +3₁-₁+6_.+ . . +3_7:;6₇+ 3_7:<89_.+. . +3_7:=89₁ With Car, Public Transport, P&R, Other for m=1,2,3,4 respectively 4.3.3 Regression results

The regression results show that interest in car decreases relatively to the neutral message for all popups, while public transport increases compared to the popup for all messages. For car, the general and specific local identity seem to have a stronger effect than environmental. Although, only specific local identity has a significant effect, decreasing the interest in car with 8.2%. Interest in public transport increases with up to 6.6% with control variables included. However, the environmental message also significantly increases the interest in public transport up to 4.7% without control. P&R shows a small, not significant decrease of less than 2.6% and a just above positive, not significant effect for Specific Local Identity. The last thing to notice is that in most cases including controls does not much influence the reported treatment effects. An exception is General Local Identity on Public Transport, that shifted a full standard deviation and shows a significant effect of 5.0% with controls. Since there are almost 600 or about 14% less observations with all controls included, this may probably be caused by some random imbalance.

Table 4:Treatment effects for Car, Public Transport and P&R, with and without controls

4.3.4 Heterogeneity analysis

To see the effect of the different treatments on groups of different economic backgrounds, a heterogeneity analysis was performed. The analysis is performed on the three control variables that seem most relevant from an economic perspective. Travel distance influences travel time, cost and carbon emissions, income was shown to be strongly correlated with car ownership and serves as a proxy for education, while urban density makes a division between urban and rural and thereby ease of access to facilities such as train stations, parking spots and roads.

The results are reported in table 5 for car (1), public transport (2) and P&R (3). The groups are equally split in below median (low) and above median (high). People with higher travel distance, lower income and from more urban areas are more interested to consider not coming by car. The same groups have increased interest to come by public transport, except for travel distance, where interest for public transport is actually higher for the short distance group. Environmental concerns specifically strongly influence the short distance, high income, urban concert visitors to come by public transport, while for these same groups interest in P&R decreases. P&R generally sees decreased interest from short distance travellers from more urban areas and increased interest from longer distance travellers from less urban areas.

(1) (2) (3) (4) (5) (6) Environmental -0.0164 -0.00965 0.0470* 0.0409* -0.0219 -0.0263 (0.0238) (0.0244) (0.0199) (0.0199) (0.0133) (0.0144) Regional Identity -0.0103 -0.00197 0.0314 0.0284 -0.0112 -0.0139 (0.0239) (0.0244) (0.0200) (0.0199) (0.0134) (0.0143) General Local Identity -0.0190 -0.0265 0.0269 0.0500* -0.0123 -0.0187 (0.0241) (0.0248) (0.0203) (0.0203) (0.0135) (0.0146) Specific Local Identity -0.0817*** -0.0814*** 0.0569** 0.0662*** 0.00400 0.00321 (0.0240) (0.0246) (0.0201) (0.0201) (0.0134) (0.0145) Constant 0.610*** 0.417*** 0.186*** 0.469*** 0.0916*** 0.0706 (0.0171) (0.114) (0.0143) (0.0928) (0.00958) (0.0669)

With controls No Yes No Yes No Yes

N 4282 3694 4282 3694 4282 3694

Standard errors in parentheses * p<0.05 ** p<0.01 *** p<0.001

Table 5: Heterogeneity analysis of treatment effects for different travel distance, income and urbanisation groups, for Car (1), Public Transport (2) and P&R (3), top to bottom

Low High Low High Low High

Environmental -0.0372 0.00583 -0.00823 -0.0238 -0.0161 -0.0172 (0.0340) (0.0324) (0.0340) (0.0332) (0.0329) (0.0336) Regional Identity -0.00284 -0.00873 0.0345 -0.0522 0.00170 -0.0241 (0.0337) (0.0331) (0.0344) (0.0332) (0.0330) (0.0339) General Local Identity -0.00578 -0.0239 -0.0288 -0.0106 -0.0117 -0.0218 (0.0340) (0.0335) (0.0349) (0.0335) (0.0337) (0.0339) Specific Local Identity -0.0644 -0.0967** -0.0967** -0.0674* -0.0809* -0.0847* (0.0342) (0.0328) (0.0344) (0.0334) (0.0331) (0.0340) Constant 0.537*** 0.683*** 0.596*** 0.623*** 0.677*** 0.547*** (0.0245) (0.0233) (0.0246) (0.0238) (0.0237) (0.0241) N 2219 2063 2077 2205 2078 2204 Standard errors in parentheses * p<0.05 ** p<0.01 *** p<0.001

Travel distance Income Urban density

Low High Low High Low High

Environmental 0.100** -0.00817 0.0377 0.0559* 0.0304 0.0634* (0.0309) (0.0237) (0.0290) (0.0273) (0.0240) (0.0306) Regional Identity 0.0523 0.000161 0.00932 0.0521 0.00616 0.0579 (0.0306) (0.0242) (0.0294) (0.0273) (0.0240) (0.0308) General Local Identity 0.0429 0.00120 0.0508 0.00550 0.0188 0.0308 (0.0309) (0.0245) (0.0298) (0.0275) (0.0246) (0.0308) Specific Local Identity 0.0740* 0.0367 0.0942** 0.0216 0.0175 0.0972** (0.0310) (0.0240) (0.0294) (0.0274) (0.0241) (0.0309) Constant 0.237*** 0.133*** 0.194*** 0.178*** 0.124*** 0.243*** (0.0222) (0.0170) (0.0210) (0.0195) (0.0173) (0.0220) N 2219 2063 2077 2205 2078 2204 Standard errors in parentheses * p<0.05 ** p<0.01 *** p<0.001

Travel distance Income Urban density

Low High Low High Low High

Environmental -0.0450** 0.00162 -0.0183 -0.0261 0.00354 -0.0459** (0.0172) (0.0205) (0.0190) (0.0186) (0.0201) (0.0175) Regional Identity -0.0361* 0.0165 -0.0303 0.00679 0.00198 -0.0236 (0.0170) (0.0209) (0.0193) (0.0186) (0.0202) (0.0177) General Local Identity -0.0378* 0.0160 -0.0337 0.00758 0.0201 -0.0417* (0.0172) (0.0211) (0.0195) (0.0187) (0.0207) (0.0177) Specific Local Identity -0.0279 0.0373 -0.0348 0.0406* 0.0433* -0.0338 (0.0173) (0.0207) (0.0193) (0.0187) (0.0203) (0.0178) Constant 0.0983*** 0.0847*** 0.107*** 0.0773*** 0.0796*** 0.103*** (0.0124) (0.0147) (0.0138) (0.0133) (0.0145) (0.0126) N 2219 2063 2077 2205 2078 2204 Standard errors in parentheses * p<0.05 ** p<0.01 *** p<0.001

Travel distance Income Urban density

1

2

5. Demographics and stated preferences

5.1 Summary statistics

A survey was conducted to gain more information on the concert demographics, how the transportation mode choice was distributed and what influenced this choice. Furthermore, people were asked if they had used the portal. As table 6 shows, there are differences between the portal users and the other concert visitors. Most importantly, much more portal users actually came by car. Furthermore, the sample of portal users who ended up coming by car, with 70% is even higher than the 60% of people who first viewed and/or last viewed car. However, there also seem reasons why this sample might have been more inclined to come by car, for instance their travel time is longer and they are relatively older.

Table 6: Summary statistics of concert demographics for survey respondents who said they did or did not use the online portal

Yes No All % % % Travelmode Car 70.1 53.4 54.6 Public Transport + P&R 29.3 43.9 42.8 Other 0.7 2.7 2.6 Total 100.0 100.0 100.0 Gender Male 13.6 15.9 15.8 Female 86.4 84.1 84.2 Total 100.0 100.0 100.0 Age group Below 18 15.0 20.2 19.9 18 to 25 29.3 35.7 35.2 26 to 35 11.6 12.7 12.6 36 to 45 16.3 12.0 12.3 46 to 55 19.7 16.7 16.9 Above 55 8.2 2.7 3.1 Total 100.0 100.0 100.0 Groupsize One person 7.5 7.7 7.7 Two people 45.6 45.4 45.4 Three people 20.4 15.5 15.9 Four people 20.4 23.4 23.2 Five people 2.7 4.0 3.9 More than six 3.4 4.0 3.9 Total 100.0 100.0 100.0 Traveltime avg Less than 30 min. 4.8 10.1 9.7 Between 30 and 60 min. 16.3 20.1 19.8 Between 60 and 90 min. 25.5 29.0 28.7 Between 90 and 120 min. 28.9 21.8 22.3 More than 120 min. 24.5 19.0 19.4 Total 100 100 100 Observations 147 1907 2054 Used the online portal

5.2 Main results

In addition to aspects such as gender and age, survey respondents were asked to rate how important different factors were to them in deciding what mode of transportation to use. Figure 3 shows the average responses for people who came by car and people who came by public transport. Answers followed a normal distribution around the mean, with a standard deviation of just below 1 for most factors. With flexibility for car having the lowest standard deviation with 0.81 and preferences of other event visitors for public transport users having the highest standard deviation, with 1.04. Distinction was made between portal and non-portal users, however, this showed to make not much of a difference in stated preferences. As this additional subdivision would double the amount of bars, this is not reported in figure 3.

Figure 4: Stated preferences for people who came by car and people who came by train

The graph shows that public transport users rank most decision factors slightly higher than respondents who came by car. Especially traffic updates are ranked higher by public transport users. Flexibility on the other hand is ranked more important by car users as is travel time. Furthermore, in line with expectation, people rank carbon emissions and social norms quite low. They valued this even lower than the weather. However, the outcomes in chapter 4 show that people actually are influenced by information on carbon emissions and what other people do, while they reacted not very strong on differences in travel time. Cost were not even reported on the portal and probably not investigated in detail by most via a different source.

5.3 Regression analysis

Women are more inclined to come by public transport then men and an increase in age increases the chance to come by car instead of public transport. Visual inspection showed that above 60 this effect slightly reduces again and car users are mainly in the age group 35 to 60, which is also an age that people often have a lease car with business tank card. For both car and public transport group size has a decreasing effect. This is because larger groups more often come by public transport. The effect is much stronger for public transport, because especially smaller groups use public transport, since the cost of coming by car is split with multiple people. Furthermore, in line with portal results, longer travel time strongly increases the chance of coming by car. Finally, all preferences are in line with discussion in section 5.2.

Table 7: Regression results for factors influencing the actual transport choice to come by car or by public transport

(1) (2) (3) (4) (5) (6) Female -0.0421 -0.00792 0.0632** 0.0287 (0.0238) (0.0232) (0.0226) (0.0220) Agegroup 0.0611*** 0.0475*** -0.0590*** -0.0478*** (0.00573) (0.00564) (0.00544) (0.00537) Groupsize -0.00689 -0.00680 -0.0444*** -0.0430*** (0.00416) (0.00405) (0.00395) (0.00385) Actual Traveltime 0.0400*** 0.0409*** -0.0390*** -0.0407*** (0.00737) (0.00719) (0.00700) (0.00685) Traveltime info 0.0113 0.0110 -0.00470 -0.00920 (0.0107) (0.0106) (0.0104) (0.0101) Cost -0.0483*** -0.0464*** 0.0365*** 0.0371*** (0.0106) (0.0106) (0.0104) (0.0101) CO2emission 0.0293** 0.0175 -0.00874 0.000282 (0.00976) (0.00973) (0.00952) (0.00926) Trafficupdates -0.0865*** -0.0778*** 0.110*** 0.0927*** (0.00992) (0.00986) (0.00968) (0.00939) Groupprefs 0.00956 0.0112 -0.0247* -0.0129 (0.0106) (0.0105) (0.0103) (0.0100) Regionalprefs 0.00402 0.000647 -0.0142 -0.000653 (0.0114) (0.0112) (0.0111) (0.0107) Concertprefs -0.0509*** -0.0424*** 0.0492*** 0.0347** (0.0113) (0.0112) (0.0110) (0.0106) Weather 0.00179 0.00597 -0.0260** -0.0287*** (0.00926) (0.00914) (0.00903) (0.00870) Flexibility 0.139*** 0.128*** -0.101*** -0.0956*** (0.0107) (0.0107) (0.0105) (0.0101) Constant 0.350*** 0.489*** 0.259** 0.702*** 0.327*** 0.695*** (0.0562) (0.0635) (0.0802) (0.0533) (0.0619) (0.0763) N 2970 2970 2970 2970 2970 2970 Standard errors in parentheses * p<0.05 ** p<0.01 *** p<0.001 Car Public Transport

6. Discussion

Care was taken to randomly assign treatments and it was shown control variables are equally balanced over the different groups. The most important risk for internal validity is that instead of actual final transport choice a proxy was used. The most important risk for external validity is that visitors of a specific concert do not reflect a more general population and using concerts as a case study might make the results broader applicable to event related transportation choices, but not to more habitual transport choices such as commute to work.

6.1 Discussion of outcomes

In line with hypothesis 1, descriptive norms do have a significant effect on several outcome variables. Especially the interest in car for both first and last viewed modality decrease significantly. Interest in public transport increases, however only significantly for the first viewed modality. This may suggest that while the descriptive norms initially interest the visitor to first look at public transport, other factors, such as longer travel time or the amount of transits may divest that interest again.

In line with hypothesis 3, the strongest effect is caused by the specific local identity. However, in contrast with hypothesis 2, based on the outcomes of Goldstein et al. (2008), this is not followed by less location specific social norms, but by environmental concerns instead. The heterogeneity analysis showed that environmental concerns especially have a strong effect on the groups with higher income, from more urban areas and with shorter travel distance.

Since income actually had a high positive correlation with car ownership, higher income group more often choosing the train may presumably be caused by educational level. Furthermore, while urban density strongly positively correlates with access to a train station and negatively with access to a main road, inhabitants of more urban, higher income neighbourhoods, thus presumably higher educated, may be quicker persuaded to come by public transport by environmental concerns. People from less urban areas, may see the car more as a default option, but are more persuaded by social norms, especially when specific to the event they are visiting. As a result, they may consider coming by public transport. However, the difficulty of reaching the first train station may direct them to then consider Park & Ride instead.

6.2 Internal validity

Since the popups in the experiment were automatically, randomly assigned, there is no risk of reverse causality. The popups a subject viewed was not related to what choices he or she made. To show the experiment was randomized, the control variables were checked for balance. Only for one control variable, distance to highway, the null hypothesis that the treatments do not affect control variables, could not be rejected. However, from eyeballing the differences for this variable seem very small, furthermore, by random assignment a small share of control variables can end up not perfectly well balanced. Therefore, in addition a test of joint orthogonality was performed, in which none of the treatments could be recognised from the others based on distribution of control variables.

Since the experiment was a fully automated natural experiment there was no risk of failure to follow a treatment protocol. There might however be a slight risk that some subjects may have understood that they were partaking in an experiment. Some visitors may make use of the portal more than once. Since popups are randomly assigned, chances are that some subject may have seen more than one popup. While to many people this may not come across as uncommon, for some individuals it may be. However, even if so, it is questionable if this may have influenced choice behaviour. More importantly in that sense is that when individuals have seen multiple popups, it becomes less clear to what extent they were affected by which popup. This might have resulted in a light averaging of treatment effect across treatments. Thus the effects of different treatments might slightly more differentiate from one another.

Furthermore, there is no problem with attrition. Some visitors did leave the portal without looking at any modality, this however is included in the affected variable table as no view. It actually seems that the popups decrease the share of people leaving the portal without looking at their travel advise. This effect however was not significant.

With more than 4200 observations to test five different popups, the sample size is sufficiently large. However, when including controls, this reduces with about 15%. This is mainly as a result of missing data for distance to train station and main roads. On closer inspection, the data was often missing for zip codes with low population density, if this data was available. This may have caused some bias in the controlled regressions as a result of sample selection.

A chance does exist that in some cases wrong measurements have been recorded for some of the control variables, since the control variables retrieved from the portal are parsed out of the data log, risks of misinterpretation in exception cases may occur. Some of the most common reasons for this to happen are special symbols that are not recognised by the data logging program. An example of this is a French c-cedilla (ç). Special care was taken to manually filter and change as many of these exceptions as possible, based on visual inspection. While the author cannot guarantee that every exception was included in the parsing progress, the overall effect is expected to be very small due to it affecting control variables and due to the relatively small occurrence of exceptions on total sample size.

The research makes use of the last viewed mobility page as a proxy for the choice people make. While the research does show significant changes in interest of visitors to view public transportation instead of the car mobility page, this interest does not reflect the final choice that people made. In comparison with the towel experiment of Goldsteind et al. (2008), there is much more of a delay between the influencing message and the final choice. Preferably buying of a parking or train ticket would have been used. However, in the Netherlands many people ow a personal public transport card and therefore will not buy an online ticket. For this reason, both the number of observations of people buying an online ticket stays small and as the people who did buy an online ticket do not own a personal public transport card, using this data instead would have been affected by a strong selection bias.

In order to test the strength of last viewed mobility page as a proxy for final transportation choice, it was proposed to match usage of the portal with survey answers based on IP address. However, only 65 matches were found. Based on these matches 75% of visitors took the same transportation as the last mobility they viewed. While this seems to indicate that the final treatment effect may be slightly different than reported based on the used proxy, not enough data was available to test this.

A missing variable in the overall analysis is the actual cost of the different transportation modes. While this was reported as one of the most important decision factors in the stated preferences in the survey, this is not fully included in the portal. The portal does report on certain cost, such as price of a parking place, but does not compare the different modes on price. An important reason not to do this is that this can only be based on averages. The actual fuel price for the car an individual takes is unknown, unless specifically asked for this data. However, it is unclear to what extent people are aware of and do think about the cost of

the different transportation modes. Since no one had this additional information from the portal, it is assumed that the amount of people who take this more and the amount of people who take this less into account are equally balanced over the treatment groups, as was shown for the other control variables.

One may question whether a linear probability model is the right model and whether the right control variables were included. Linear probability model was chosen because of ease of interpretation. Comparison with a logit an probit model showed that results are very comparable as a result of the relative high share of occurrences of the dependent variables within the total number of observations. The control variables seem reasonable from an economic perspective and most were significant or close to significant, while the reported treatment effects did not change much with or without including these controls. Finally, it is known that LP models are prone to heteroskedastic errors. Several model were tested with and without robust standard errors. Although the differences were rather small; the model was therefore regressed using robust.

6.2 External validity

As the survey summary statistics revealed, there is a difference in population between the users of Mobiity Portal and the other visitors of these specific concerts. However, more importantly, the largest share of the audience of these concerts were below 35 and around 80% were woman. It may be assumed that the survey itself had some problems with selection bias, as it is a voluntary survey and is only send to people who make use of the wifi network. However, several thousands of people make use of the wifi network and thus receive a survey. Furthermore, the gender and age distribution is very much in line with the distribution as visually observed during the concert nights. Furthermore, since both gender and age were shown to be not very strongly correlated with the revealed preferences for environmental and social concerns it may be expected that this would not make much of a difference on the average treatment effects. However, as it was shown that the effect of especially social norms was lower in the stated preferences than the revealed effect size, it might be assumed that different genders and age groups actually more or less underrate this and the effect on external validity may actually be larger.

A population effect that may be of more concern for the external validity is the economic background of the visitors. Environmental concerns showed to be more powerful than was

hypothesised based on the literature. The main article on which this was based, conducted a field research in the US with a specific hotel chain. It may be expected that strongly affected the subject group, since people in the US may have different values with regarding to social and environmental concerns and one hotel attracts different people than another. In the same way, a certain artist may attract a certain audience. As the heterogeneity analysis showed, there are differences between people from different economic backgrounds in the way they responded on the different norms that were used. Extending the research to more events, with different audiences, would make the results more generalizable.

Except for the environmental concerns having a stronger effect than expected, the effects were in line with expectations based on the literature from experiments in different contexts. However, while this experiment shows that descriptive norms could be used to influence people’s transportation choice, this is now only tested for events, not to chance people’s habits. It is uncertain if getting confronted with a message that reminds people of the higher environmental impact of the car than public transport increases or decreases their response to this over time.

7. Conclusion

It was shown that using descriptive norms does have an effect and this differs per norm used. Furthermore, people respond differently, based on economic background.

7.1 Man findings

Nudging visitors with descriptive norms significantly decreased their interest in coming by car, in favour of coming by public transport. The strongest effect is caused by providing a social norm related to the specific event, with a reported effect of 8.2% in reducing interest to come by car. It was hypothesised that the more specific the social norm, the stronger the effect. Although not all significant, indeed the more specific the social identity, the stronger the effect, with 1.0%, 1.9% and 8.2% for reducing interest in car in order of local situation specificness. However, it was also expected that the social norms would all have a stronger effect than environmental concerns, but environmental concerns directly followed the most specific social norm, increasing interest in public transport with 4.7%.. The heterogeneity analysis shows that different groups react differently on either environmental concerns or social norms. It is therefore assumed that environmental concerns are more prevalent among certain groups in the Netherlands, namely people from higher income more urban areas. To more successfully nudge people, it could actually be useful to show people from higher income, urban zipcodes an environmental message, while other groups get to see a location specific social norm.

7.2 Recommendations for further research

Within the direct experimental setup of the research a step for further research could be to actually make a distinction in assigning popups to people and see if this increases the effect for both the environmental message and the social message. Furthermore, the research itself could be extended to both gather data on a broader spectrum of audience and to gather more IP address matches. Furthermore, additional information could be added, such as expected cost or carbon emissions of the trip per traveller and other types of messages could be tested. In addition, other insights from behavioural economics could be tried.

In a broader sense the use of these descriptive norms can either be tested in a different field or in the same field but different setting. It could for instance be used in a business district to test whether people’s habitual choice of commute can be influenced or in food consumption in the office or university, to promote people to either eat healthier or more environmental friendly.

Appendix

Survey questions (in Dutch)

De volgende additionele vragen opnemen in survey. Algemene vragen

- Persoonlijke gegevens, zoals gender en leeftijdsgroep worden al gevraagd - Type vervoer wordt al gevraagd

Vragen over Mobility Portal

V1 Van welke informatiebron heeft u gebruik gemaakt voor uw reisadvies richting de ArenA?

Meerkeuze, niet verplicht

o Heb ik niet zelf opgezocht, ik reed met iemand mee

o Ik ben gewoon gaan rijden en/of heb alleen de autonavigatie gebruikt o Ik heb geen reisplanner gebruikt, want ik weet de weg

o Google Maps o OV9292 o NS-Reisplanner

o Bereikbaarheidsinformatie op de website van de ArenA = Mobility Portal

o Bereikbaarheidsinformatie op de website van de Organisator o Anders, namelijk: <open>

V2 Hoe lang moest u ongeveer reizen tijdens de HEENREIS naar de ArenA?

• Minder dan 30 minuten • Tussen 30 minuten en een uur • Tussen een uur en anderhalf uur • Tussen anderhalf en twee uur • Meer dan twee uur

V3 Hoe lang moest u ongeveer reizen tijdens de TERUGREIS vanaf de ArenA?

• Minder dan 30 minuten • Tussen 30 minuten en een uur • Tussen een uur en anderhalf uur • Tussen anderhalf en twee uur • Meer dan twee uur

Motivatie voor vervoerskeuze

V4 In hoeverre vindt u de volgende factoren belangrijk in uw vervoerskeuzekeuze?:

Zeer

onbelangrijk Onbelangrijk Neutraal Belangrijk belangrijk Zeer Informatie over reistijd verschillende vervoersopties

o

o

o

o

o

o

o

o

o

o

Informatie over CO2 uitstoot verschillende vervoersopties

o

o

o

o

o

Actuele informatie zoals werkzaamheden aan spoor en weg

o

o

o

o

o

De voorkeuren van

mijn reisgezelschap

o

o

o

o

o

andere bezoekers uit mijn regio

o

o

o

o

o

De vervoerskeuze van andere bezoekers van het concert

o

o

o

o

o

Weersomstandigheden

_o

_o

_o

_o

_o

Flexibiliteit te vertrekken wanneer ik dat zelf wil