“Dear diary, I took the bus today!” – Cost-efficient travel behavior influencers, June 15th, 2020 By Saumya Jain
“Soft” transportation policy measures can influence a significant reduction in personal car use, according to a new research published in Transportation Research Part D.
The Behavioral Toolkit identifies six psychological variables that can affect travel behavior: attitudes; emotions; habits; social, cultural, and moral norms; knowledge and awareness; and capability and self-efficacy. To compare travel behavior measures, this study categorized each intervention under the six variables. The paper reviewed literature and analyzed experimental studies on transportation behavioral measures from the past 30 years.
The results show that interventions that focus on social, cultural, and moral norms have the most significant effect on travel behavior. The most effective interventions were travel feedback programs that made participants aware of their environmental impact and suggested alternate transportation modes and routes. Measures that were accompanied by incentives like transit vouchers were more effective than just education.
Ineffective interventions were those focused on changing established habits and attitudes.
While policies and infrastructure that create barriers to car trips or alternative transportation options (congestion pricing, parking fees, blocking roads, bike-lanes, etc.) are impactful, they require substantial resources. Policymakers can use cost-effective and more easily-implementable soft interventions discussed in the study in conjunction with policies and infrastructure to change traveler behavior and mode.
Saumya Jain is a Senior Associate at SSTI.
Smarter Choices: Assessing the Potential to Achieve Traffic Reduction Using ‘Soft Measures’
S. Cairns ,L. Sloman,C. Newson,J. Anable,A. Kirkbride &P. GoodwinPages 593-618 | Received 03 May 2007, Accepted 03 Jan 2008, Published online: 21 Aug 2008 Download citation https://doi.org/10.1080/01441640801892504 Full Article
In recent years, there has been a growing interest in a range of transport policy initiatives which are designed to influence people’s travel behaviour away from single‐occupancy car use and towards more benign and efficient options, through a combination of marketing, information, incentives and tailored new services. In transport policy discussions, these are now widely described as ‘soft’ factor interventions or ‘smarter choice’ measures or ‘mobility management’ tools. In 2004, the UK Department for Transport commissioned a major study to examine whether large‐scale programmes of these measures could potentially deliver substantial cuts in car use. The purpose of this article is to clarify the approach taken in the study, the types of evidence reviewed and the overall conclusions reached. In summary, the results suggested that, within approximately ten years, smarter choice measures have the potential to reduce national traffic levels by about 11%, with reductions of up to 21% of peak period urban traffic. Moreover, they represent relatively good value for money, with schemes potentially generating benefit:cost ratios which are in excess of 10:1. The central conclusion of the study was that such measures could play a very significant role in addressing traffic, given the right support and policy context.
Transportation Research Part D: Transport and Environment, Volume 85, August 2020, 102397
30 Years of soft interventions to reduce car use – A systematic review and meta-analysis
Author links open overlay panelAlinSemenescu AlinGavreliuc PaulSârbescu w morehttps://doi.org/10.1016/j.trd.2020.102397 Get rights and content Under a Creative Commons license open access
Highlights
Soft transport interventions are effective measures for reducing personal car use.
Effectiveness of soft interventions does not significantly decrease over time.
Effectiveness was moderated by psychological variable targeted.
Soft travel interventions are generally regarded as effective measures for reducing personal car use. However, doubts about the validity of such claims are raised, primarily fueled by the low methodological quality of evaluation studies upon which such conclusions are based and by the fact that the literature is, for the most part, narratively synthesized. The present systematic review addresses these critiques by investigating the effect of soft interventions on car use through a meta-analysis, which includes only experimental and well-controlled quasi-experimental studies. Results revealed that interventions (k = 41) lead to a significant reduction of 7% in car modal split share (Hedges’ g = 0.163). Moderators of interventions’ effectiveness were investigated in a meta-regression. Effectiveness was moderated by the type of intervention and by the main psychological variable targeted by the interventions, whereas the other studied moderators (i.e. residential relocation of participants, study design, percentage of females in the study, the presence of incentives, passed time to follow-up, interventions’ measurement instrument, city size in which interventions were applied and the setting where they were conducted) were non-significant. Limitations of the present findings, together with implications for policy and practice are discussed.
Keywords
Soft interventionsCar useSustainable transportationSystematic reviewMeta-analysis
1. Introduction
Road transportation accounts for about 20% of all carbon dioxide (CO2) emitted annually into the atmosphere (Bamberg and Rees, 2017), which makes it a strategic sector for policymakers interested in mitigating climate change. Technological innovations such as electric vehicles could potentially reduce transportation’s impact on the environment. However, it seems unlikely that these technologies will scale quickly enough to have a significant impact on CO2 emissions in the following decades. For example, the market share of electric vehicles in the EU in 2017 reached only 1.48%, with only a modest increase compared to 2016 (European Environment Agency, 2018). In most other countries outside of the EU, this percentage is considerably lower. Therefore, it is likely that, to reduce greenhouse gas emissions from transportation, governments and local authorities will continue to rely mostly on interventions aimed at motivating individuals to reduce environmentally harmful behaviors such as driving, and adopt more environmentally friendly means of transportation.
The last decades have witnessed a variety of policy measures aimed at promoting sustainable transportation. Such measures are commonly known as travel demand management (TDM; Loukopoulos, 2007), which have been defined by Litman (2003, p. 245) as “strategies and programs that encourage more efficient use of transport resources.” Even though a plethora of TDM measures has been proposed, scholars generally classify them into two categories (see Bamberg et al., 2011, Graham-Rowe et al., 2011, Steg, 2003). Strategies aimed at modifying social conditions and structures are the so-called structural interventions or “hard” interventions (Steg, 2003). These measures aim to change transportation behavior by altering the physical environment (e.g. closing roads, building bicycle lanes, etc.) and through legal or economic policies (e.g. prohibiting car traffic in city centers, congestion pricing, introducing parking fees, etc.). Economic incentives typically require substantial resources, while coercing behavior through modifications of the physical environment or legal policies may involve political costs, as such measures can be met with opposition from the public. Hard measures have also been criticized as being insufficient for reducing personal car use (Stopher, 2004). As a result, policymakers have increasingly orientated towards the so-called “soft” or psychological interventions, which are defined by Steg (2003, p. 190) as “strategies aimed at influencing people’s perceptions, beliefs, attitudes, values, and norms.” They rely on persuasion and motivation of car users to switch to sustainable travel modes, instead of using coercive means (Möser and Bamberg, 2008) and therefore are sometimes referred to also as voluntary change measures (Bamberg et al., 2011). This approach involves providing people with information, education, behavioral examples, prompts, feedback and other strategies aimed at changing people’s travel behavior by increasing their knowledge, awareness, self-efficacy, changing their attitudes, habits, norms or by influencing their emotions.
The effectiveness of soft interventions was investigated in large-scale projects in countries such as Australia (e.g. Rose and Ampt, 2001), the UK (e.g. Parker et al., 2007) or Germany (see Richter et al., 2009) and, on a smaller scale, in Japan (see Fujii et al., 2009), Austria (e.g. Steininger et al., 1996), Netherlands (see Cairns et al., 2008) and other European countries (see Richter et al., 2009). Based on this accumulated evidence, a report from Halcrow Group (2002) estimated that soft transport measures could reduce traffic in the UK with around 5%. Similarly, Cairns et al. (2004) estimated a decrease in UK’s traffic levels by 4–5% nationally in a low-intensity scenario, in which soft policies are applied inconsistently, and a 10–15% decrease in a high-intensity scenario, in which soft policies are consistently applied and are complemented by hard measures. A similar conclusion was reached by Ogilvie et al. (2004), who estimated that soft interventions could decrease the car modal split share (the proportion of car trips from the total number of trips) with about 5% in motivated subgroups.
Even though the evidence seems to concur, existing reviews were mostly based on intervention studies conducted by private consultancy firms (e.g. Socialdata in Australia, Steer Davies Gleave in the UK), which often employed weak designs in their evaluations. This significant limitation was identified also by other researchers (e.g. O’Fallon and Sullivan, 2004, Richardson, 2003), who have cautioned about the overoptimistic picture that such studies may have painted. For example, Möser and Bamberg (2008), who reviewed 141 studies assessing the effectiveness of soft interventions, concluded that all 141 studies included in their meta-analysis used a one group pre-test-post-test design.
A second important criticism is related to the way results from available studies have been synthesized. Even though most reviews used narrative techniques (e.g. Cairns et al., 2004, Ogilvie et al., 2004), there is increased concern that this type of analysis does not provide a reliable mechanism for drawing valid conclusions from quantitative data (Borenstein et al., 2011, Lipsey and Wilson, 2001). As an alternative, numerous researchers recommend using meta-analysis as the more appropriate method for synthesizing research results. It is thus surprising that, despite recent methodological advances, no meta-analyses have been conducted in the field until 2007 and that, to our knowledge, only five have been conducted up to June 2018.
2. Previous meta-analyses
Bamberg and Möser, 2007, Möser and Bamberg, 2008 were the first to employ meta-analytical techniques in the context of transportation research. They synthesized the results of 141 soft transport measures from different narrative reviews published on the topic and have concluded that soft interventions increased by 7% the proportion of trips not conducted by car (from 39% to 46%), an effect which corresponded to a Cohen’s h = 0.15 (Möser and Bamberg, 2008). However, the validity of their conclusion is seriously threatened by the weak methodological quality of the included studies. Two later meta-analyses have addressed this substantial limitation. Fujii et al. (2009) assessed the effect of 15 soft interventions with experimental designs implemented in Japan and found an effect size of Cohen’s d = 0.165 as a result of such programs. Later, Bamberg and Rees (2017) meta-analyzed the results from eleven soft transport interventions with quasi-experimental and experimental designs, ten of them conducted in the UK and one in Germany. Across the analyzed studies, an effect size of Cohen’s h = 0.12 was found, which corresponded to a reduction of approximately 5% in car modal split share. The validity of these findings is nevertheless restricted by the limited cultural contexts in which the evaluation studies were conducted (i.e. Japan, UK and Germany) and by the fact that conclusions were not substantiated by a systematic review of the literature.
The study conducted by Arnott et al. (2014) addressed such limitations by conducting a systematic review from which only Randomized Controlled Trials, Cluster Randomized Controlled Trials and Controlled Before and After studies were meta-analyzed. The authors reached the surprising conclusion that “there is no evidence for the efficacy of behavioral interventions […] to change transport behaviors.” They suggested that their findings can be reconciled with previous research by taking into account the methodological superiority of the studies included in their analysis. However, considering that their meta-analysis was conducted on only four studies, the authors’ claim must be regarded with a sense of apprehension. Such a limited number of studies is likely to result in biased estimates and raises serious questions about the validity of their conclusion. Nevertheless, their study highlighted an important question: When considering only methodologically sound evaluation studies, is there any evidence for the effectiveness of soft interventions on car use reduction?
3. The present study
Even though soft interventions were implemented on a large scale and considerable evidence has accumulated throughout the years, a debate exists concerning their effectiveness. Previously conducted meta-analyses have not offered a satisfactory answer to this debate, as they suffer from several important limitations that raise doubts about the validity of their conclusions. The present study aims to contribute to the discussion by a twofold purpose. First, we will inspect whether studies with robust methodological designs show an effect of soft interventions on car use behavior by reviewing the literature from the past 30 years and meta-analyzing only studies with experimental and strong quasi-experimental designs. We decided to conduct our analysis only on studies with strong methodological designs, as these types of studies are the only ones that warrant causal conclusions. Our decision to focus on studies from the past 30 years was informed by the fact that in the late ‘80s (and early ‘90s) large scale soft interventions began to be implemented and evaluation studies started being conducted. Our second focus is to investigate what type of soft interventions are most effective and under which circumstances they produce the largest effects. To this purpose, we planned three tests (for which we devised two hypotheses), to investigate the moderating effects of the type of intervention, the relocation status of participants and the study design used. We will also conduct several exploratory analyses.
Adopting the classification of soft interventions proposed by Cairns et al., 2004, Möser and Bamberg, 2008 were the first to investigate the moderating effect of the type of intervention in a meta-analytical context. They found that soft interventions such as work travel plans, school travel plans and travel planning/travel awareness/PT marketing, significantly differed in their efficacy in decreasing car use. While work travel plans had an average effect size of Cohen’s h = 0.24, school travel plans proved to be less effective, with an average effect of only 0.08. Aside from the classification proposed by Cairns et al. (2004), soft interventions can also be classified based on the main psychological variable that they attempt to influence. The Behavioural Insights Toolkit (Savage et al., 2011), which draws on several theoretical models, identifies six such variables that are particularly relevant for transportation behavior. These include attitudes (defined as the degree of subjectively favorable or unfavorable rational evaluations of a perceived object, person or behavior), emotions (the degree of subjectively favorable or unfavorable feelings towards an object, person or behavior), social, cultural and moral norms (the perceived prevailing attitudes, behaviors or moral standards in a given context), habits (behaviors triggered by events or environmental cues that are performed automatically), knowledge and awareness and capability and self-efficacy (the subjective belief that one can succeed in a certain task or perform well in a specific situation). Because meta-analytical evidence shows that such variables are correlated in varying degrees of strength with car use (see Gardner and Abraham, 2008) and because previous research (i.e. Möser and Bamberg, 2008) has already identified significant differences in effectiveness between different soft interventions, we hypothesized that:
Hypothesis 1
The type of soft intervention implemented acts as a moderator of interventions’ effectiveness.
Travel behaviors can be dependent on people’s habits (e.g. Aarts et al., 1998, Thøgersen and Møller, 2008), which are defined as automatic responses triggered by certain events or environmental cues (Verplanken et al., 1997). If this is the case, then individuals whose travel choices are habitual are less likely to be impacted by interventions aimed at changing their travel behavior. For instance, they may be less likely to notice and process new travel information in their environment, even if that information might offer them a better travel alternative then the one they are used to. A second mechanism through which habits can impede behavior change is through weakening the relation between intentions and actual behavior (e.g. De Bruijn et al., 2007, Verplanken et al., 1998). Bamberg (2006) argues that a residential relocation represents a significant life event, which can disrupt formed habits and force people to deliberately reorganize their lives. Consequently, the first weeks of a residential relocation may create a window of opportunity in which people are motivated to pay more attention to new information in relation to their travel options, which can enhance the impact of interventions conducted in this sensitive phase. This hypothesis, commonly known as the habit discontinuity hypothesis, has received some empirical support (e.g. Fujii et al., 2001, Verplanken and Roy, 2016), yet it has not been investigated in a meta-analytical context to present. We have therefore hypothesized that:
Hypothesis 2
Soft interventions conducted on recently relocated individuals are more effective in reducing car use than interventions conducted on individuals who did not relocate.
A debate still exists on the relation between the methodological quality of evaluation studies and their reported effect sizes. While some authors argue that quasi-experimental studies produce inconsistent and often unpredictable results when compared with Randomized Controlled Trials (e.g. Bloom et al., 2005, Glazerman et al., 2003), others conclude that results of well-designed quasi-experimental studies and those from Randomized Controlled Trials are remarkably similar (Abraham et al., 2010, Benson and Hartz, 2000, Concato et al., 2000). Though we agree that experimental designs remain the preferred choice, it is often the case in transportation research that experimentation is impossible or results in severe limitations to external validity, forcing researchers to resort to quasi-experimental designs. We therefore considered it a worthwhile endeavor to assess to what extent results from quasi-experimental studies converge towards “the gold standard” in research, namely experimental studies. Because the evidence is ambiguous, we treat this as an exploratory question.
In an explorative manner, we will also investigate the moderating effects of other variables, which we grouped into three categories: (1) Recipient characteristics (2) intervention characteristics and (3) contextual factors. Concerning recipient characteristics, some research suggests that males are more oriented than women towards the affective and symbolic benefits of driving (e.g. Steg, 2005) and derive their self-esteem from this activity (Ellaway et al., 2003). Consequently, males may be more reluctant to give up driving. If gender is a moderator of interventions’ effectiveness, then future studies will have to investigate how interventions need to be gender-tailored, to maximize their impact. Concerning intervention characteristics, we will explore the moderating effects of incentives offered for participating in the study, the time to follow-up and intervention’s measurement instrument. These variables represent important factors to consider when designing, implementing and measuring interventions’ effect. Incentives have been linked with increased motivation and performance (see Ariely et al., 2009, Cerasoli et al.,2014), while a longer time to follow-up may show, if effects are unstable over time, a dilution of intervention’s effect. An important issue for evaluators is whether self-reported and objective measures of car use concur, as this can have a considerable impact on the cost of evaluation studies. Finally, concerning contextual factors, we will explore two possible moderators, namely the city size in which interventions were applied and the setting where they were conducted. A smaller city size may be linked with a less effective intervention, as people in such places tend to have fewer transportation options available, while the setting may influence effectiveness through subtle processes such as social influences (e.g. interventions conducted in a work setting) or a deeper processing of the message (e.g. interventions conducted at home).
4. Method
The present systematic review was conducted in line with PRISMA recommendations (Moher et al., 2009). However, risk of bias in individual studies was not assessed because (a) most studies did not report relevant information and (b) we considered the stringent inclusion criteria as sufficiently restrictive to select only high-quality studies for this review.
4.1. Eligibility criteria
Studies were included in the present meta-analysis only if they fulfilled the following criteria:1.
Study design: To consider only the most reliable evidence available, only studies with an experimental or quasi-experimental design were included in the current analysis. If random assignment was not possible, then intervention and control groups had to be matched and a pre-test measure of the dependent variable was required, to correct for possible initial differences in car use. If matching was not possible, then relevant controlling variables had to be considered, together with a pre-test measure of the dependent variable.2.
Participants: Only studies conducted on adult participants (i.e. 18 years or older) were included. Studies in which participants had a clinical diagnosis or were younger than 18, together with studies on driving cessation programs for old age, were excluded.3.
Interventions: Only studies that included soft interventions in the context of repeated car journeys were included. Studies evaluating soft interventions that also provided minor incentives such as free public transport tickets were considered eligible. To be included, interventions needed to be implemented at the individual or group level. Population-level interventions were excluded because of the increased difficulty of controlling for confounding variables.4.
Comparison: Only studies using a control group as comparison, which received minimal or no intervention were included. A control group with minimal intervention was defined as a group of participants that had to perform a task for the study but did not receive the active components of the intervention (e.g. they received a leaflet with information, engaged in a one-on-one conversation with the intervention provider, etc.). Studies that included a control group receiving an alternative intervention were excluded, to avoid possible biases in effect size estimations.5.
Outcomes: Only studies that reported a quantitative outcome measure taken in the context of repeated car journeys (holiday travels were excluded), such as travelled distance, travelled time, number of trips, the proportion of trips or proportion of time in which the car was used, were included. Outcome measures could be both objectively measured or self-reported, as long as they measured actual travel behavior. Studies that measured intentions of traveling or which used imaginary/hypothetical scenarios were excluded.6.
Period: Only studies between 1988 and 2018 were considered.7.
Language: Only studies reported in English were included.
4.2. Sources of included studies
Comprehensive searches were conducted in June 2018 in the following databases: PsychINFO (568 results), Scopus (1620 results), Web of Science (974 results), GreenFILE (60 results), Science Direct (309 results), ROSAP (1 result), Transportation Research International Documentation (TRID; 79 results) and ProQuest Dissertations & Theses A& I (4 results). An example of a search strategy used for PsychINFO database can be found in supplementary material. Returned results from PsychINFO were limited to “quantitative studies”, to reduce the number of qualitative studies that were irrelevant for the purpose of this study. To eliminate irrelevant entries, results from Scopus and Web of Science were restricted solely to the subject areas of interest. Results from Scopus were restricted to the following “Subject areas”: Social Sciences, Environmental Science, Decision Sciences, Business Management & Accounting, Psychology, Earth and Planetary Sciences, Energy, Economics Econometrics and Finance, Multidisciplinary and Undefined. Results from Web of Science were restricted to the following “Research Areas”: Transportation, Business economics, Environmental Sciences Ecology, Public Environmental Occupational Health, Psychology, Urban Studies, Social Sciences, Other Topics, Sport Sciences, Physiology, Energy Fuels and Behavioral Sciences. Results from the other five databases were not filtered by any criteria. These search terms returned 3615 articles, book chapters, conference proceedings and dissertations. Additionally, 134 potentially relevant articles were identified through ancestry searches: 38 articles were identified from the reference lists of four meta-analyses already conducted on the topic and 96 were identified from the reference lists of eighteen reviews, resulting in a total of 3749 entries.
4.3. Study selection
Included studies in the meta-analysis came from two different sources: Previous reviews and our literature search. The 3749 entries were analyzed using a reference management software and, after duplicates were removed, the resulting 2920 entries were evaluated in a two-step process. First, all titles and abstracts were inspected against the inclusion criteria. This strategy eliminated 2751 irrelevant entries. Second, full texts of the remaining 169 publications that seemed to meet the inclusion criteria were evaluated. Of these, sixteen met our inclusion criteria and were included in the meta-analysis, while 153 were excluded for the following reasons: 50 studies did not have a control group, 21 studies did not measure car use as dependent variable, thirteen studies did not report enough information to calculate effect sizes, eleven were theoretical or review articles, ten were not intervention studies, nine studies could not be retrieved, six studies were about interventions which were not yet implemented, six were studies about “hard” measures, five studies had interventions implemented at population level, five had imaginary/fictional interventions, four studies had the same dataset as the one used in other studies that were already included in the analysis, four studies used a control group which received an alternative intervention, two studies measured car use with a categorical variable, two were not conducted on an adult sample, two did not use an adequate control group, one was a qualitative study, one was an intervention on a clinical sample and one was a study about a driving cessation program for older adults (see Fig. 1). The corresponding authors of the thirteen studies that did not report enough information for effect size calculation, were contacted by e-mail for supplementary information. Until the writing of this manuscript, only one of these authors positively responded to our inquiry, yet the study was not considered eligible because of the inability to control for pretest differences between groups. A detailed file with all excluded studies and the reason for their exclusion can be provided upon request to the main author.
Unfortunately, not all of the research meeting our inclusion criteria was publicly available. This is because local authorities often commissioned private companies (e.g. Socialdata in Australia or Steer Davies Gleave in the UK) with the implementation and evaluation of soft policy measures. In their evaluations, such companies often reported only a limited amount of information, while the rigor of scientific reporting was frequently not respected. This situation made it sometimes impossible to extract the necessary data for effect size calculation. Therefore, to also include such studies, we relied on data reported in two reviews (i.e. Fujii et al., 2009, Parker et al., 2007) under the assumption that the authors were more successful in obtaining the relevant information and that they reported it correctly. By adopting this strategy, fourteen additional studies were included in our analysis, which resulted in a total of 30 individual studies and 41 intervention arms. An asterisk symbol (*) marks the entries obtained directly from these reviews.
4.4. Study coding
All included studies were coded based on a coding manual, which can be provided, upon request, by the main author. The coded data included information about the study (year and type of publication), participants, interventions, outcomes and study design. Participant-related data encoded information about the sample size for each group, percentage of females in the sample, mean age of the sample, the type of sample used in the study (e.g. general population, students, etc.), participants’ nationality and whether they have relocated recently or not. Relocation status was coded as “relocated” if a specific mention about this aspect was made in the manuscript. If no mention existed, we assumed that the intervention was conducted on individuals who did not relocate recently. Intervention-related data encoded information about the type of intervention, whether participants received an incentive for participating in the study, information about the setting where the intervention took place (e.g. at home, at work, etc.), the size of the city in which it was implemented (measured in millions of inhabitants), the time to follow-up (measured as the number of weeks passed since the last component of the intervention ended until evaluation was conducted) and information about the main psychological variable targeted by the intervention. The main psychological variable targeted was coded according to the variables defined by the Behavioural Insights Toolkit and authors’ suggestions. Generally, Travel Feedback Programs (a particular type of soft intervention in which participants receive feedforward and feedback information) were coded as targeting capability and self-efficacy, as these programs involve close mentoring of intervention recipients by travel consultants, who conceive a weekly travel calendar by suggesting travel routes in such a way that greenhouse gas emissions are minimized. Interventions providing information were coded as targeting attitudes or knowledge and awareness (depending on the content of the information provided), interventions providing feedback about the environmental or financial consequences of car use were coded as targeting knowledge and awareness, while those that were based on the formation of action plans or implementation intentions were coded as targeting habits.
Data about outcomes encoded information about the type of outcome (e.g. travelled distance, travelled time, etc.), the way it was measured (objectively or self-reported) and the instrument of measurement (e.g. travel diary, odometer, etc.). Lastly, study design data encoded information about the design used (experimental or quasi-experimental).
Studied moderators were coded also by a second author. Inter-rater reliability was assessed by calculating Cohen’s kappa, for categorical moderators, and the intraclass correlation coefficient (ICC), for continuous moderators. Agreement was satisfactory on both statistics: Min K = 0.727, max K = 1; Min ICC = 0.980, max ICC = 1. Disagreements on coding were solved through discussion.
4.5. Strategies used for the meta-analysis
Throughout study evaluation, the following strategies were applied:(i)
In case interventions were measured with multiple outcomes, a combined effect size was calculated by taking into consideration all the outcomes reported in the study.(ii)
Where multiple time points were reported, intervention’s effect size was calculated based on the first measurement after the intervention ended.(iii)
In Controlled Before and After studies, the effect sizes were calculated by taking account also of the baseline measurement. In these situations the correlation between pre and post measurements needed to be known or estimated. Because the large majority of studies failed to report this statistic, we relied on Fujii et al. (2009), who estimated in their meta-analysis that the average correlation between pre and post measurement was r = 0.77.
5. Results
5.1. Study characteristics
A total of 11,206 participants were included in the present analysis, with a mean age of 38.8 years (mean range from 22 to 53), while the percentage of women was 52.4% (range from 7% to 72%). These figures were calculated only from studies that reported these statistics directly or from studies from which they could be approximated. If the mean age of participants was not reported, an approximation was calculated, using data reported in the study. If insufficient data was provided, the cell was left blank. Also, in some isolated cases, studies did not report sample sizes for each intervention group. In these cases, the total sample size was split equally between the control and intervention group(s).
Of all included interventions, seventeen were conducted in Japan, nine in England, four in Germany, four in the United States, three in Sweden, two in Taiwan and one in China and Australia each. Six interventions were conducted with student samples, while 35 were conducted on samples from the general population. Five were conducted on individuals who recently relocated, while 36 were conducted on individuals who did not relocate. Concerning study design, nine studies had quasi-experimental designs while the rest were experimental studies. Three interventions from a single study (i.e. Graham et al., 2011) used a control group that received minimal intervention. In this study, participants in the control group were required to monitor their traveling behavior, while those in the intervention groups additionally received different types of feedback. The other 38 interventions used a completely passive control group. Table 1 provides summary information about the characteristics of all interventions included in the meta-analysis.
Table 1. Intervention characteristics of included studies.
| Study/Intervention arm | % Women | Relocation | Incentives | Type of intervention | Psychological variable targeted | Setting | Time to follow-up (weeks) | Study design |
|---|---|---|---|---|---|---|---|---|
| Armitage et al., 2011 | 56 | not relocated | incentive | Implementation intentions | habit | at home | 4 | experimental |
| Bamberg and Rees, 2017 | 51.2 | relocated | no incentive | Information + free PT ticket | capability and self-efficacy | at home | 6 | experimental |
| Bamberg, 2006 | 47 | relocated | incentive | Information + free PT ticket | capability and self-efficacy | at home | 6 | experimental |
| Bamberg, 2013_information | 72 | not relocated | no incentive | Information | knowledge and awareness | at home | 6 | experimental |
| Bamberg, 2013_information + dialog marketing | 72 | not relocated | no incentive | Information + dialog marketing | knowledge and awareness | at home | 6 | experimental |
| Eriksson et al., 2008 | 48.8 | not relocated | incentive | Increased awareness + implementation intentions | habit | at home | 1 | experimental |
| Fujii and Kitamura, 2003 | 7 | not relocated | incentive | Information + free PT ticket | capability and self-efficacy | at university | 0 | experimental |
| * Fujii et al., 2009_Fukuoka 2005 | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_no change intention; feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_no change intention; no feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_non PT user; feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_non PT user; no feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_non PT user; ticket; feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_non PT user; ticket; no feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_PT user; feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Kawanishi 2003_PT user; no feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Ryugasaki 2005_feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Ryugasaki 2005_no feedback | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Ryugasaki 2005_Study 2 | relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Sapporo 2003_GIS-based | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Sapporo 2003_paper-based | not relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| * Fujii et al., 2009_Takasaki 2005 | relocated | Travel Feedback Program | capability and self-efficacy | at home | experimental | |||
| Gärling et al., 1998_Study 1 | 50 | not relocated | no incentive | Request for behavior change | social, cultural and moral norms | at home | 1 | experimental |
| Garvill et al., 2003 | 51 | not relocated | incentive | Increased awareness | knowledge and awareness | at home | 0 | experimental |
| Geng et al., 2016 | 45.07 | not relocated | incentive | Information | attitudes | mobile | 0 | experimental |
| Graham et al., 2011_ pollution feedback | 64.8 | not relocated | incentive | Pollution feedback | knowledge and awareness | mobile | 2 | experimental |
| Graham et al., 2011_financial feedback | 64.8 | not relocated | incentive | Financial feedback | knowledge and awareness | mobile | 2 | experimental |
| Graham et al., 2011_pollution + financial feedback | 64.8 | not relocated | incentive | Pollution + financial feedback | knowledge and awareness | mobile | 2 | experimental |
| Haq et al., 2008 | not relocated | incentive | Travel Feedback Program + free PT ticket | capability and self-efficacy | at home | 0 | quasi-experimental | |
| Hsieh et al., 2017_action plan | 42.1 | not relocated | incentive | Action plan | habit | mobile | 4 | experimental |
| Hsieh et al., 2017_action plan + coping plan | 42.1 | not relocated | incentive | Action plan + coping plan | habit | mobile | 4 | experimental |
| Ma et al., 2017 | 54.4 | not relocated | no incentive | Travel Feedback Program | capability and self-efficacy | at home | 0 | quasi-experimental |
| Nakayama and Takayama, 2005 | not relocated | no incentive | Ecotravel Coordinator Program | social, cultural and moral norms | various places | 8 | quasi-experimental | |
| * Parker et al., 2007_Bristol_Bishopston | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 12 | quasi-experimental | ||
| * Parker et al., 2007_Bristol_Bishopsworth | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 24 | quasi-experimental | ||
| * Parker et al., 2007_Bristol_Hartcliffe | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 24 | quasi-experimental | ||
| * Parker et al., 2007_Darlington_Phase 1 | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 24 | quasi-experimental | ||
| * Parker et al., 2007_Lancashire_South Ribble | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 32 | quasi-experimental | ||
| * Parker et al., 2007_Peterborough_Phase 1 | not relocated | Travel Feedback Program | capability and self-efficacy | at home | 24 | quasi-experimental | ||
| Rodriguez and Rogers, 2014 | relocated | no incentive | Information | attitudes | mobile | 24 | experimental | |
| Sargeant et al., 2004_Cambridgeshire | not relocated | no incentive | Travel Feedback Program | capability and self-efficacy | informal place | 11 | experimental |
Regarding outcomes, 20 interventions used the number of trips conducted by car as the main outcome, eleven used proportion of trips, five used travelled distance, one used travelled time and four interventions used multiple outcomes to measure car use. The largest part of these outcomes (39 interventions) were self-reported, using instruments such as travel diaries (34 interventions) or recollections from memory of the number of trips the car was used in the previous week (5 interventions). Only two interventions were evaluated through objectively measured outcomes: One used GPS measurements while the other used calculated distances on the map. A summary of participant and outcome characteristics for each intervention can be found in Table 1, in supplementary material.
5.2. Meta-analysis
Studies included in the meta-analysis reported various outcomes in the context of repeated car journeys, such as travelled distance, travelled time, number of trips, the proportion of car trips or the proportion of time travelled by car. We considered all these outcomes as reliable standardized measures of car use and therefore pulled them together in the same meta-analysis. This strategy provides a more reliable estimate of the summary effect by basing it on a larger pool of studies. To support our decision though, we inspected those studies that reported multiple outcomes of car use, to verify whether systematic differences in effect sizes existed between different operationalizations. In the three studies that reported multiple outcomes (i.e. Haq et al., 2008, Hsieh et al., 2017, Ma et al., 2017), no statistically significant differences were found. In a meta-regression, we also inspected whether the operationalization of car use predicted effect size. There was no relation between the type of operationalization and effect size, Q(4) = 6.585, p = .159 (see Table 2). Therefore, we proceeded further by including all operationalizations of car use in the same analysis.
Table 2. Studied moderators of interventions’ effectiveness.
| Variable | k | g | Min g | Max g | Z | p | Q | I2 |
|---|---|---|---|---|---|---|---|---|
| Outcome | Between-groups difference: Q(4) = 6.585, p = .159 | |||||||
| Multiple | 4 | 0.040 | −0.256 | 0.337 | 0.267 | 0.789 | Q = 9.179, p = .027* | 67.317 |
| Number of trips | 20 | 0.224 | 0.115 | 0.334 | 4.018 | <0.001*** | Q = 23.917, p = .199 | 20.557 |
| Proportion of trips | 11 | 0.117 | 0.070 | 0.163 | 4.916 | <0.001*** | Q = 16.889, p = .077 | 40.789 |
| Travelled distance | 5 | 0.389 | 0.093 | 0.684 | 2.574 | 0.010** | Q = 9.834, p = .043* | 59.324 |
| Travelled time | 1 | 0.089 | −0.109 | 0.287 | 0.881 | 0.378 | Q = 0.000, p = 1.000 | 0.000 |
| Type of intervention | Between-groups difference: Q(14) = 37.304, p < .001*** | |||||||
| Information | 3 | 0.131 | −0.001 | 0.263 | 1.948 | 0.051 | Q = 0.314, p = .855 | 0.000 |
| Information + free PT ticket | 3 | 0.112 | 0.099 | 0.125 | 17.003 | <0.001*** | Q = 1.605, p = .448 | 0.000 |
| Travel Feedback Program | 23 | 0.131 | 0.073 | 0.188 | 4.474 | <0.001*** | Q = 28.803, p = .151 | 23.620 |
| Psychological variable targeted (PVT) | Between-groups difference: Q(4) = 26.739, p < .001*** | |||||||
| Attitudes | 2 | 0.119 | −0.033 | 0.272 | 1.534 | 0.125 | Q = 0.225, p = .636 | 0.000 |
| Habit | 4 | −0.009 | −0.301 | 0.283 | 0.062 | 0.950 | Q = 7.100, p = .069 | 57.747 |
| Knowledge and awareness | 6 | 0.348 | 0.194 | 0.502 | 4.421 | <0.001*** | Q = 4.496, p = .480 | 0.000 |
| Capability and self-efficacy | 27 | 0.120 | 0.082 | 0.159 | 6.108 | <0.001*** | Q = 30.872, p = .233 | 15.781 |
| Social, cultural and moral norms | 2 | 0.738 | 0.467 | 0.1009 | 5.337 | <0.001*** | Q = 0.382, p = .537 | 0.000 |
| Relocation | Between-groups difference: Q(1) = 0.665, p = .415 | |||||||
| Relocated | 5 | 0.125 | 0.000 | 0.249 | 1.960 | 0.050* | Q = 5.821, p = .213 | 31.280 |
| Not relocated | 36 | 0.183 | 0.119 | 0.247 | 5.600 | <0.001*** | Q = 65.752, p = 001** | 46.770 |
| Type of design | Between-groups difference: Q(1) = 0.848, p = .357 | |||||||
| Experimental | 32 | 0.193 | 0.121 | 0.265 | 5.245 | <0.001*** | Q = 49.509, p = .019* | 37.386 |
| Quasi-experimental | 9 | 0.140 | 0.052 | 0.227 | 3.129 | 0.002** | Q = 23.326, p = .003** | 65.703 |
Notes. k = number of interventions included in the analysis; g = the average effect size; min/max g = the minimum and maximum limits of the 95% confidence interval for g; Z = the statistical test used for computing the significance of the average effect size; p = the significance level of the test; Q = the statistical test used for estimating the heterogeneity in effect sizes; I2 = the proportion of effect size variance that can be attributed to real differences between studies ; * p < .05; ** p < .01; *** p < .001
The analysis was conducted with the help of Comprehensive Meta-Analysis software, version 3. All effect sizes were transformed into Hedges’ g, to correct for the slight bias which estimates based on Cohen’s d have (Borenstein et al., 2011). For the ease of interpretation, we reported the summary effects and other statistics as positive values. A positive g-value signifies that soft interventions were effective in reducing car use, while a negative one signifies the opposite. Because interventions were very dissimilar and the contexts where they were applied were diverse, we specified a random-effects model to conduct the analysis.
Across the 41 interventions included in the meta-analysis, the random-effects model estimated a significant summary effect, g = 0.163, Z = 6.419, p < .001, 95% CI [0.113, 0.213], indicating that soft interventions were effective in reducing private car use. The effect was estimated with 95% confidence to be between 0.113 and 0.213, which corresponds to a small effect size, according to Cohen’s guidelines (Cohen, 1988).
5.3. Moderator analysis
The first step in any moderator analysis is to inspect heterogeneity, to assess to what extent observed variations in effect sizes represent real differences or just measurement errors around a common effect size value. The Q-test rejected the hypothesis that all studies shared a common effect size, Q(40) = 73.078, p < .001. The true variability between studies was estimated at about 45% (I2 = 45.264) which, according to Higgins et al. (2003), represents a medium value. The 95% confidence interval for I2 ran from 20.8% to 62.1%, indicating the true variability between studies could range from small to medium.
We continued our analysis by investigating the moderators that could explain these differences. To test our first hypothesis, we added the discrete variable type of intervention as a covariate in a meta-regression model. The type of intervention was a significant moderator, Q(14) = 37.304, p < .001, explaining about 64% of the true variability in effect sizes. This indicates significant differences between interventions existed, a result that supported our first hypothesis. However, the diversity in implemented interventions was so large that many categories ended up containing just one entry. We therefore reported the mean effects and other relevant statistics for only those categories that contained more than one intervention (see Table 2). Travel Feedback Programs had a summary effect, g = 0.131, Z = 4.474, p < .001, 95% CI [0.073, 0.188], interventions that provided information had a summary effect, g = 0.131, Z = 1.948, p = .051, 95% CI [−0.001, 0.263], while those that provided information and a free public transport ticket had a summary effect, g = 0.112, Z = 17.003, p < .001, 95% CI [0.099, 0.125].
Because this type of categorization was overly broad, we have categorized all interventions also based on the main psychological variable that they targeted, by using the Behavioural Insights Toolkit. The 41 interventions could be grouped into five categories, namely interventions targeting attitudes, social cultural and moral norms, habit, knowledge and awareness and capability and self-efficacy. The variable psychological variable targeted (PVT) proved to be a significant predictor of effect size, Q(4) = 26.739, p < .001, explaining about 76% of the true variability that existed between interventions.
Results showed that only interventions targeting knowledge and awareness, capability and self-efficacy and those targeting social, cultural and moral norms had significant summary effects. The most effective interventions were the ones targeting social cultural and moral norms, which had a mean effect size, g = 0.738, Z = 5.337, p < .001, 95% CI [0.467, 1.009], followed by the ones targeting knowledge and awareness, g = 0.348, Z = 4.421, p < .001, 95% CI [0.194, 0.502] and those targeting capability and self-efficacy, g = 0.120, Z = 6.108, p < .001, 95% CI [0.082, 0.159]. The least effective interventions were the ones targeting habit, g = −0.009, Z = 0.062, p = .950, 95% CI [−0.301, 0.283], yet there was substantial variability in effect sizes within this group, I2 = 57.747, p = .069. However, because there was a limited number of studies in each intervention category (except for interventions targeting capability and self-efficacy), we treat these conclusions as preliminary.
To test our second hypothesis, we investigated whether interventions targeting individuals who recently relocated were more effective than interventions targeting individuals who did not relocate. There were 36 interventions conducted on individuals who did not relocate, with a mean effect size, g = 0.183, Z = 5.600, p < .001 and five interventions on individuals who relocated to another city, g = 0.125, Z = 1.96, p = .050. Results suggest that both types of interventions are effective in reducing car use. However, the difference was in the opposite direction to what we expected, yet non-significant, Q(1) = 0.665, p = .415. The difference in effectiveness remained non-significant also when we controlled for differences in interventions, by introducing PVT into the regression model. Our second hypothesis was therefore not supported by the data.
To test whether experimental and quasi-experimental studies in transportation research differed in terms of their effect sizes, we added the dummy coded variable type of design as a covariate in a meta-regression model. There were nine studies with quasi-experimental designs, while the rest were experimental. The mean effect of quasi-experimental studies g = 0.140, Z = 3.129, p = .002, was slightly lower than the one from experimental designs, g = 0.193, Z = 5.245, p < .001, yet the difference was non-significant, Q(1) = 0.848, p = .357. A non-significant p-value does not warrant the conclusion that the two summary effects are the same, yet it also provides no evidence to conclude the contrary. This conclusion also held when we controlled for differences in interventions by introducing PVT into the model.
For our exploratory analyses, we investigated the moderating effects of the percentage of females in the study, of incentives for participation, of passed time to follow-up, of interventions’ measurement instrument, of the city size where interventions were conducted and of the setting where they were applied. For these analyses, two alternative models were created. In the first model (M1), all these moderators were assessed independently. In the second model (M2) the same moderators were assessed, but differences in interventions were controlled, by introducing PVT in the model. When PVT was in the model, none of the explored moderators reached significance, indicating that they did not significantly predict additional variance in effect sizes over the one already explained by PVT (see Table 3).
Table 3. Additional studied moderators of interventions’ effectiveness.
| M1 | M2 | |||||||
|---|---|---|---|---|---|---|---|---|
| Variable | k | Q | df | p | k | Q | df | p |
| Percentage of females | 16 | 5.083 | 1 | 0.024* | 16 | 0.213 | 1 | 0.643 |
| PVT | 16 | 6.226 | 4 | 0.182 | ||||
| Incentives | 20 | 2.424 | 1 | 0.119 | 20 | 0.009 | 1 | 0.923 |
| PVT | 20 | 19.353 | 4 | <0.001*** | ||||
| Time to follow-up | 26 | 2.452 | 1 | 0.117 | 26 | 0.552 | 1 | 0.457 |
| PVT | 26 | 25.958 | 4 | <0.001*** | ||||
| Measurement instrument | 41 | 0.174 | 1 | 0.675 | 41 | 1.098 | 1 | 0.294 |
| PVT | 41 | 27.376 | 4 | <0.001*** | ||||
| City size | 39 | 2.343 | 1 | 0.125 | 39 | 0.258 | 1 | 0.611 |
| PVT | 39 | 29.086 | 4 | <0.001*** | ||||
| Setting | 41 | 14.963 | 4 | 0.004** | 41 | 5.450 | 4 | 0.244 |
| PVT | 41 | 15.733 | 4 | 0.003** |
Notes. k = number of interventions included in the analysis; Q = the statistical test used for computing the significance of the predictors; df = degrees of freedom for the test; p = the significance level of the test; * p < .05; ** p < .01; *** p < .001.
5.4. Impact of bias
The representativeness of the studies included in the meta-analysis for the body of research conducted on the topic is a key aspect for the validity of the findings. We assessed the potential for reporting bias (Rosenthal, 1979) by visually inspecting the funnel plot (Borenstein et al., 2011). Funnel plots are bivariate graphs in which effect sizes are plotted against the number of participants in each study or (as it is the case here) against the standard error of the effect size. Plotting against the standard error has the advantage of spreading out the points in the bottom half of the graph, making it easier to identify asymmetry. If no bias is present, the shape of the plot should be like a funnel, with more spread on the bottom part and less spread at the top. A visual inspection of the funnel plot indicates that reporting bias seems to be present to a low extent (see Fig. 2).
Concerning retrieval bias, even though careful searches and rigorous evaluation of studies were employed, it is almost certain that not all of the research on the topic meeting our inclusion criteria was found through our searches. Also, all returned results were evaluated by a single author, which increased the possibility that some relevant studies were omitted. However, if the omitted studies represent a random selection of all published material on the topic (i.e. if no systematic bias is present) this omission should merely affect the power of our meta-analysis and not introduce bias in our results. As a last step in our analysis, we estimated the impact of bias on our findings by using the Trim and Fill method suggested by Duval and Tweedie (2000). The estimated summary effect across the 41 interventions was g = 0.163, 95% CI [0.113, 0.213], while the most accurate estimate for the unbiased summary effect under a random-effects model was g = 0.110, 95% CI [0.055, 0.165]. We therefore conclude that the impact of bias on our summary effect is probably modest. If all relevant studies would have been included in our meta-analysis, the summary effect might decrease a bit, but our conclusion regarding the effectiveness of soft transport interventions would probably remain unchanged.
6. Discussion
The primary focus of the current systematic review was to investigate whether evaluation studies with robust designs offer support for the pervasive assumption that soft transport policy measures are effective in reducing private car use. The second focus was to investigate what type of interventions work and under which conditions they are most effective.
The most important finding of the present study is that, across the 41 interventions meeting our inclusion criteria, a statistically significant summary effect of Hedges’ g = 0.163 was found. Even though this effect size is considered, according to Cohen’s (1988) guidelines, a small effect, it is nevertheless sufficiently large to be practically significant. Expressed in percentages, it corresponds to a decrease of approximately 7% in car use modal split share. The effect is similar in magnitude to the one reported by Fujii et al. (2009), who found an effect size of Cohen’s d = 0.165 in a sample of fifteen soft interventions with experimental designs conducted in Japan. It is also strikingly similar to Möser and Bamberg’s (2008) finding, who concluded that soft interventions increased by 7% the proportion of trips not conducted by car (Cohen’s h = 0.15). This might seem a surprising finding, as one would expect that superior research designs, which control for seasonal effects, testing, or social desirability, would yield smaller effect sizes. However, the discrepancy can be partially explained by the inclusion of school travel plans in Möser and Bamberg’s analysis, which had a considerably lower summary effect compared to other types of soft interventions, pulling therefore their overall estimate downwards.
As expected, there was significant variability in interventions’ effectiveness and our results indicated that
- the largest amount of variability was explained by the psychological variable targeted by the interventions. The most effective interventions were the ones targeting social, cultural and moral norms, with a summary effect estimated at Hedges’ g = 0.738 (approximately a 32% decrease in car modal split share),
- Followed by interventions targeting knowledge and awareness of own driving behavior, with a summary effect estimated at g = 0.348 (approximately a 14% decrease in car modal split share).
- Interventions targeting capability and self-efficacy, produced a lower summary effect, of g = 0.120 (approximately a 5% decrease in car modal split share), while our analyses suggest that
- Interventions targeting attitudes and habits may not be effective in reducing car use, at least when applied independently.
- Combining different soft interventions may however reveal a different picture. For instance, interventions targeting attitudes (e.g. providing tailored information) may be more effective if barriers to behavior change such as automatic behaviors are also addressed. Future studies will need to explore this possibility by including additional conditions that investigate interactions between different types of soft interventions.
We also hypothesized that relocation status acts as a moderator of interventions’ effectiveness, a hypothesis that was not supported by the data. This can indicate that, contrary to what the habit discontinuity theory postulates, individuals may have default modes of traveling that remain stable even when they change contexts. We have to address this conclusion with a sense of caution, for several reasons: (a) the relocation group is comprised of only a limited number of studies, (b) interventions that were coded as “not relocated” may (at least partly) contain relocated individuals, reducing therefore the differences between the two groups and (c) time since relocation was not considered in our analysis. Studies show that the “window of opportunity” may be limited to a defined period of time (e.g. Verplanken and Roy, 2016) and that car use increases with time since relocation (Thomas et al., 2016). Consequently, there is the possibility that the “relocation effect” may have decayed in evaluation studies that were conducted after longer periods of time.
Our investigation regarding the possible moderating effects of study design revealed the summary effect from quasi-experimental studies was not statistically different from the one resulted from experimental ones. This result might suggest that quasi-experimental designs may be a viable alternative in transportation research, especially when experiments are impossible to conduct. However, when experimentation is possible, we recommend this type of design, if this does not result in serious limitations to external validity.
Our exploratory analyses revealed that interventions’ effectiveness was not moderated by the gender of participants, the presence of incentives for participation, the time to follow-up, the measurement instrument of the interventions, the size of the city in which interventions were conducted or by the setting where they were applied. From these results several meaningful conclusions with implications for policy and practice can be drawn: First,
- a non-significant difference between genders indicates that both males and females respond similarity to soft transport interventions. This is contrary other authors’ suggestions that males are more motivated by symbolic and affective factors derived from driving (e.g. Steg, 2005) and thus less likely to change their driving behavior.
- Second, since time to follow-up was not a significant moderator of interventions’ effectiveness, it may indicate that soft interventions are not only short-term fixes, but can retain their effectiveness even after extended periods. This conclusion is supported by other researchers, who found that behavioral changes were maintained or even increased over time (e.g. Fujii and Taniguchi, 2006, Taylor and Ampt, 2003, Taylor, 2007). This is relevant, because it justifies the use of such interventions on a large scale, but also because it legitimizes the use of public funds for financing them.
- Third, a non-significant relationship between the measurement instrument and effect size may indicate that self-reported measures of car use may be just as valid as objective ones. If this is the case, then the cost of evaluations might be drastically reduced by using only self-reported measures. However, we once more have to caution that this analysis was based on only a limited number of studies that employed objective measures and, until further research is conducted, this conclusion has to be regarded as preliminary.
- Fourth, our results suggest that the effectiveness of soft interventions is not influenced by incentives, the setting where they are conducted or by the size of the city where they are applied. This is good news for intervention providers, as such results suggest that soft interventions are versatile enough to be effective even in markedly diverse contexts and circumstances and that their effectiveness cannot be attributed to temporary increases in motivation or social desirability as a result of incentives offered.
The practical implications of our findings are important, especially in the context of the current environmental crisis and the governments’ inability to mitigate ever-increasing levels of atmospheric CO2. Taking into consideration that road transportation accounts for about a fifth of the yearly CO2 emitted into the atmosphere (Bamberg and Rees, 2017), soft transport interventions might prove to be indispensable instruments for mitigating climate change in the coming decades. Nevertheless, a critical factor when deciding upon their feasibility is also their cost. If such interventions are to be implemented consistently, they should not only be effective but also be cost-efficient. Cost-benefit analyses were conducted in the UK (e.g. Cairns et al., 2008, Parker et al., 2007) as well as in Australia (e.g. Ker et al., 1999, Taylor, 2007) and reported greater benefits than costs for such interventions. For example, Ker et al. (1999) reported that for the IndiMark© project conducted in South Perth, they expect that, over a ten-year period, benefits would exceed costs by a factor between 11 and 13, thanks to decreases in greenhouse gasses, air pollution, travel time, road congestion and increases in public transportation. Additionally, Ker (2003) concludes that increases in public transportation trips, as a result of soft transport interventions conducted in Australia, will generate increased revenues that can offset the full costs of such programs in two to five years. For a more extensive discussion about this subject see Parker et al. (2007).
6.1. Limitations
One of the limitations of the present study derives from the search strategy used, which relied solely on published articles, book chapters, conference proceedings, dissertations and reports. Therefore, unpublished research was omitted. Also, even though our search strategy was sensitive enough to find all studies included in previously published meta-analyses, it is improbable that all relevant published material was identified. The search strategy was also limited to published material in English, which may have resulted in some eligible studies published in other languages to be overlooked. We therefore tried to compensate these downsides by also conducting searches in the reference lists of previously published reviews in the hope that relevant material that escaped our searches was found by other researchers. However, even though retrieval bias might be present, if this bias is non-systematic, it should merely result in a loss of power and not in biased estimates. The second source of potential bias is the small reporting bias present in our sample of studies. Nevertheless, our estimate of bias using Duval and Tweedie’s (2000) Trim and Fill method suggests that bias has only a modest impact on the summary estimate and does not affect our conclusion about the effectiveness of soft transport interventions.
A second limitation lies in the way study selection and coding of one of the studied moderators were done. Even though studies were selected in multiple steps and extensive care was given to this process, the fact that this was done by a single person, increased the chance of omissions. Also, when coding the variable PVT, we had to restrain our coding to a single target variable, to be able to conduct the analysis. However, complex interventions targeted, directly or indirectly, more than a single variable. This can diminish the differences between categories, which could be larger than the ones we reported in this review. This issue is further complicated by the often strong correlation that exists between psychological variables, such that interventions aimed at changing one variable are likely to have effects on other variables as well, making it difficult to identify the underlying mechanism through which such interventions work. Future longitudinal studies will have to disentangle this issue by identifying which targeted variables mediate the relation between type of intervention and effectiveness.
A third limitation is related to the inclusion of interventions that contained also minor incentives (i.e. provision of free public transport tickets), which are generally regarded as structural measures. However, such incentives were provided in only four interventions, whose average effect size was g = 0.112, which was somewhat lower than the average effect of the rest of the interventions (g = 0.180). Although it is impossible to isolate the influence of such incentives, their impact on the summary effect estimated in the present review is probably minimal.
A final limitation is related to the generalizability of our findings, as some of the studies included in the present meta-analysis were conducted on unrepresentative samples, such as student samples. Also, in some of the included studies, there was some element of participant selection, as these interventions were conducted on individuals who were already open to changing their travel patterns. Stopher and Bullock (2003) have already identified this problem for IndiMark© projects in Australia, suggesting the effects of such programs were somewhat overestimated. Even though the majority of the included studies have not used biased samples, it is probable that, if the summary effect would be estimated only from representative samples, the figure would be somewhat lower than the one estimated from our analysis.
6.2. Directions for future research
Even though the present systematic review shows that soft interventions cause a reduction in private car use, further questions still demand clarification: What type of soft interventions are most effective? Which are the factors enhancing or impeding their effectiveness? How do multiple interventions interact? How do soft interventions interact with hard ones? Future research should therefore focus on investigating relevant moderators and on comparing different soft interventions in terms of their effectiveness. At a minimum, future studies should carefully report all relevant information for assessing intervention’s effectiveness (such as sample sizes, means and standard deviations) and detailed information about the population, the characteristics of the intervention, outcome measurement and contextual factors. Ideally, reporting standards are developed to standardize this procedure and change publication practice, which will greatly help future research synthesis.
Future research should also investigate the comparative efficacy of hard and soft interventions yet, more importantly, it should assess to what extent they could work synergistically to reduce car use. Hard interventions are costly and do not always deliver the expected results (e.g. Sanjust et al., 2015), thus there is a stringent need to investigate to what extent soft interventions might help in enhancing their effectiveness.
A combination of hard and soft interventions might prove to be more effective than either one in isolation and might represent a powerful combination for shifting people’s transportation preferences towards more sustainable alternatives (see Gärling and Schuitema, 2007).
An essential change in evaluation practice, especially in a time of widespread technological diffusion in the population, is to include objective measurements of car use. The use of smartphones has become so pervasive in Western societies that they mark the use of technology as a viable, fast, cheap and accurate mechanism of collecting data. A significant limitation to current data collection practices based primarily on self-reported information is the possibility of bias, both from the intervention evaluator as well as from the intervention recipient. These biases could be eliminated by the process of collecting data with the use of smartphones or GPS devices. Such changes should be adopted not only by progressive researchers but should become standard practice for intervention providers, especially if financing organizations demand higher quality evaluations as an important part of the intervention program.
7. Conclusions
Soft transport policy measures determine a significant reduction in private car use. Expressed in percentages, this effect corresponds to approximately 7% reduction in car modal split share (Hedges’ g = 0.163). This figure might be a slight overestimation of the true population effect, as some included studies did not use representative samples and because a small bias might be present in our sample of studies. Soft interventions varied significantly in their effectiveness and this variance was largely predicted by the type of psychological variable targeted by the interventions. Interventions’ effectiveness was not predicted by the relocation status of participants, study design, the percentage of females in the study, the presence of incentives for participation, the city size in which interventions were applied, the setting where they were conducted, the passed time to follow-up or the interventions’ measurement instrument. However, because of the limited number of studies included in certain analyses, some of our conclusions about moderators of interventions’ effectiveness remain only preliminary.
Alin Semenescu: Conceptualization, Methodology, Formal analysis, Data curation, Writing – original draft, Funding acquisition. Alin Gavreliuc: Writing – review & editing, Supervision, Project administration. Paul Sârbescu: Data curation, Writing – review & editing.
We would like to thank Florin Sava and Tobias Arbogast for their valuable comments on the first draft of this manuscript.
This work has received funding from the Romanian grant BID (PN-III-P1-PFE-28), Ministry of Research and Innovation.
References
References marked with an asterisk indicate studies included in the review.
Aarts et al., 1998H. Aarts, B. Verplanken, A. Van KnippenbergPredicting behavior from actions in the past: Repeated decision making or a matter of habit?J. Appl. Soc. Psychol., 28 (15) (1998), pp. 1355-1374, 10.1111/j.1559-1816.1998.tb01681.xCrossRefView Record in ScopusGoogle ScholarAbraham et al., 2010N.S. Abraham, C.J. Byrne, J.M. Young, M.J. SolomonMeta-analysis of well-designed nonrandomized comparative studies of surgical procedures is as good as randomized controlled trialsJ. Clin. Epidemiol., 63 (3) (2010), pp. 238-245, 10.1016/j.jclinepi.2009.04.005ArticleDownload PDFView Record in ScopusGoogle ScholarAriely et al., 2009D. Ariely, A. Bracha, S. MeierDoing good or doing well? Image motivation and monetary incentives in behaving prosociallyAmerican Economic Review, 99 (1) (2009), pp. 544-555CrossRefView Record in ScopusGoogle ScholarArmitage et al., 2011C.J. *Armitage, J.C. Reid, C.P. SpencerEvidence that implementation intentions reduce single-occupancy car use in a rural population: moderating effects of compliance with instructionsTransportmetrica, 7 (2011), pp. 455-466, 10.1080/18128602.2010.505591CrossRefView Record in ScopusGoogle ScholarArnott et al., 2014B. Arnott, L. Rehackova, L. Errington, F.F. Sniehotta, J. Roberts, V. Araujo-SoaresEfficacy of behavioural interventions for transport behaviour change: systematic review, meta-analysis and intervention codingInt. J. Behav. Nutr. Phys. Activ., 11 (1) (2014), p. 133, 10.1186/s12966-014-0133-9View Record in ScopusGoogle ScholarBamberg, 2013S. *BambergApplying the stage model of self-regulated behavioral change in a car use reduction interventionJ. Environ. Psychol., 33 (2013), pp. 68-75, 10.1016/j.jenvp.2012.10.001ArticleDownload PDFView Record in ScopusGoogle ScholarBamberg, 2006S. *BambergIs a residential relocation a good opportunity to change people’s travel behavior? Results from a theory-driven intervention studyEnviron. Behav., 38 (6) (2006), pp. 820-840, 10.1177/0013916505285091CrossRefView Record in ScopusGoogle ScholarBamberg et al., 2011S. Bamberg, S. Fujii, M. Friman, T. GärlingBehaviour theory and soft transport policy measuresTransp. Policy, 18 (1) (2011), pp. 228-235, 10.1016/j.tranpol.2010.08.006ArticleDownload PDFView Record in ScopusGoogle ScholarBamberg and Möser, 2007S. Bamberg, G. MöserWhy are work travel plans effective? Comparing conclusions from narrative and meta-analytical research synthesisTransportation, 34 (6) (2007), pp. 647-666, 10.1007/s11116-007-9121-0CrossRefView Record in ScopusGoogle ScholarBamberg and Rees, 2017S. *Bamberg, J. ReesThe impact of voluntary travel behavior change measures–A meta-analytical comparison of quasi-experimental and experimental evidenceTransp. Res. Part A: Policy Pract., 100 (2017), pp. 16-26, 10.1016/j.tra.2017.04.004ArticleDownload PDFView Record in ScopusGoogle ScholarBenson and Hartz, 2000K. Benson, A.J. HartzA comparison of observational studies and randomized, controlled trialsN. Engl. J. Med., 342 (25) (2000), pp. 1878-1886, 10.1056/NEJM200006223422506View Record in ScopusGoogle ScholarBloom et al., 2005H.S. Bloom, C. Michalopoulos, C.J. HillUsing experiments to assess nonexperimental comparison-group methods for measuring program effectsH.S. Bloom (Ed.), Learning More From Social Experiments: Evolving Analytic Approaches, Russell Sage Foundation (2005), pp. 173-235Google ScholarBorenstein et al., 2011M. Borenstein, L.V. Hedges, J.P. Higgins, H.R. RothsteinIntroduction to meta-analysisJohn Wiley & Sons (2011)Google ScholarCairns et al., 2004Cairns, S., Sloman, L., Newson, C., Anable, J., Kirkbride, A., Goodwin, P., 2004. Smarter choices-changing the way we travel. Retrieved from https://discovery.ucl.ac.uk/id/eprint/1224/1/1224.pdf.Google ScholarCairns et al., 2008S. Cairns, L. Sloman, C. Newson, J. Anable, A. Kirkbride, P. GoodwinSmarter choices: assessing the potential to achieve traffic reduction using ‘soft measures’Transp. Rev., 28 (5) (2008), pp. 593-618, 10.1080/01441640801892504CrossRefView Record in ScopusGoogle ScholarCerasoli et al.,2014Cerasoli, C.P., Nicklin, J.M., Ford, M.T., 2014. Intrinsic motivation and extrinsic incentives jointly predict performance: A 40-year meta-analysis. Psychol. Bull. 140(4), 980–1008. https://doi.org/10.1037/a0035661.Google ScholarCohen, 1988Cohen, J., 1988. Statistical power analysis for the behavioral sciences. 1988, L. Lawrence Earlbaum Assoc. 2, Hillsdale, NJ.Google ScholarConcato et al., 2000J. Concato, N. Shah, R.I. HorwitzRandomized, controlled trials, observational studies, and the hierarchy of research designsN. Engl. J. Med., 342 (25) (2000), pp. 1887-1892, 10.1056/NEJM200006223422507View Record in ScopusGoogle ScholarDe Bruijn et al., 2007G.J. De Bruijn, S.P. Kremers, E. De Vet, J. De Nooijer, W. Van Mechelen, J. BrugDoes habit strength moderate the intention–behaviour relationship in the Theory of Planned Behaviour? The case of fruit consumptionPsychol. Health, 22 (8) (2007), pp. 899-916, 10.1080/14768320601176113CrossRefView Record in ScopusGoogle ScholarDuval and Tweedie, 2000S. Duval, R. TweedieTrim and fill: a simple funnel-plot–based method of testing and adjusting for publication bias in meta-analysisBiometrics, 56 (2) (2000), pp. 455-463, 10.1111/j.0006-341X.2000.00455.xCrossRefView Record in ScopusGoogle ScholarEllaway et al., 2003A. Ellaway, S. Macintyre, R. Hiscock, A. KearnsIn the driving seat: psychosocial benefits from private motor vehicle transport compared to public transportTransp. Res. Part F: Traff. Psychol. Behav., 6 (3) (2003), pp. 217-231, 10.1016/S1369-8478(03)00027-5ArticleDownload PDFView Record in ScopusGoogle ScholarEriksson et al., 2008L. *Eriksson, J. Garvill, A.M. NordlundInterrupting habitual car use: The importance of car habit strength and moral motivation for personal car use reductionTransp. Res. Part F: Traff. Psychol. Behav., 11 (1) (2008), pp. 10-23, 10.1016/j.trf.2007.05.004ArticleDownload PDFView Record in ScopusGoogle ScholarEuropean Environment Agency, 2018European Environment Agency, 2018, June 11. Electric vehicles as a proportion of the total fleet. Retrieved from https://www.eea.europa.eu/data-and-maps/indicators/proportion-of-vehicle-fleet-meeting-4/assessment-2 (last accessed, April 2019).Google ScholarFujii et al., 2009S. *Fujii, S. Bamberg, M. Friman, T. GärlingAre effects of travel feedback programs correctly assessed?Transportmetrica, 5 (2009), pp. 43-57, 10.1080/18128600802591277CrossRefView Record in ScopusGoogle ScholarFujii et al., 2001S. Fujii, T. Gärling, R. KitamuraChanges in drivers’ perceptions and use of public transport during a freeway closure: Effects of temporary structural change on cooperation in a real-life social dilemmaEnviron. Behav., 33 (6) (2001), pp. 796-808, 10.1177/00139160121973241View Record in ScopusGoogle ScholarFujii and Kitamura, 2003S. *Fujii, R. KitamuraWhat does a one-month free bus ticket do to habitual drivers? An experimental analysis of habit and attitude changeTransportation, 30 (1) (2003), pp. 81-95, 10.1023/A:1021234607980View Record in ScopusGoogle ScholarFujii and Taniguchi, 2006S. Fujii, A. TaniguchiDeterminants of the effectiveness of travel feedback programs–a review of communicative mobility management measures for changing travel behaviour in JapanTransp. Policy, 13 (5) (2006), pp. 339-348, 10.1016/j.tranpol.2005.12.007ArticleDownload PDFView Record in ScopusGoogle ScholarGardner and Abraham, 2008B. Gardner, C. AbrahamPsychological correlates of car use: A meta-analysisTransp. Res. Part F: Traff. Psychol. Behav., 11 (4) (2008), pp. 300-311, 10.1016/j.trf.2008.01.004ArticleDownload PDFView Record in ScopusGoogle ScholarGärling et al., 1998T. *Gärling, R. Gillholm, A. GärlingReintroducing attitude theory in travel behavior research: The validity of an interactive interview procedure to predict car useTransportation, 25 (2) (1998), pp. 129-146, 10.1023/A:1005004311776View Record in ScopusGoogle ScholarGärling and Schuitema, 2007T. Gärling, G. SchuitemaTravel demand management targeting reduced private car use: effectiveness, public acceptability and political feasibilityJ. Social Issues, 63 (1) (2007), pp. 139-153, 10.1111/j.1540-4560.2007.00500.xCrossRefView Record in ScopusGoogle ScholarGarvill et al., 2003J. *Garvill, A. Marell, A. NordlundEffects of increased awareness on choice of travel modeTransportation, 30 (1) (2003), pp. 63-79, 10.1023/A:1021286608889View Record in ScopusGoogle ScholarGeng et al., 2016J. *Geng, R. Long, H. ChenImpact of information intervention on travel mode choice of urban residents with different goal frames: A controlled trial in Xuzhou, ChinaTransp. Res. Part A: Policy Pract., 91 (2016), pp. 134-147, 10.1016/j.tra.2016.06.031ArticleDownload PDFView Record in ScopusGoogle ScholarGlazerman et al., 2003S. Glazerman, D.M. Levy, D. MyersNonexperimental versus experimental estimates of earnings impactsAnn. Am. Acad. Polit. Soc. Sci., 589 (1) (2003), pp. 63-93, 10.1177/0002716203254879View Record in ScopusGoogle ScholarGraham et al., 2011J. *Graham, M. Koo, T.D. WilsonConserving energy by inducing people to drive lessJ. Appl. Soc. Psychol., 41 (1) (2011), pp. 106-118, 10.1111/j.1559-1816.2010.00704.xCrossRefView Record in ScopusGoogle ScholarGraham-Rowe et al., 2011E. Graham-Rowe, S. Skippon, B. Gardner, C. AbrahamCan we reduce car use and if so, how? A review of available evidenceTransp. Res. Part A: Policy Pract., 45 (5) (2011), pp. 401-418, 10.1016/j.tra.2011.02.001ArticleDownload PDFView Record in ScopusGoogle ScholarHaq et al., 2008G. *Haq, J. Whitelegg, S. Cinderby, A. OwenThe use of personalised social marketing to foster voluntary behavioural change for sustainable travel and lifestylesLocal Environ., 13 (7) (2008), pp. 549-569, 10.1080/13549830802260092CrossRefView Record in ScopusGoogle ScholarHiggins et al., 2003J.P. Higgins, S.G. Thompson, J.J. Deeks, D.G. AltmanMeasuring inconsistency in meta-analysesBmj, 327 (2003), pp. 557-560, 10.1136/bmj.327.7414.557CrossRefView Record in ScopusGoogle ScholarHsieh et al., 2017H.S. *Hsieh, Y. Kanda, S. FujiiReducing car use by volitional strategy of action and coping planning enhancementTransp. Res. Part F: Traff. Psychol. Behav., 47 (2017), pp. 163-175, 10.1016/j.trf.2017.05.002ArticleDownload PDFView Record in ScopusGoogle ScholarKer, 2003Ker, I., 2003. Travel demand management: Public transport business case. Contract Report RC5051 for Department of Infrastructure, Victoria, Canada.Google ScholarKer et al., 1999Ker, I., James, B.G., Square, C., 1999. Evaluating behaviour change in transport: benefit cost analysis of individualised marketing for the city of South Perth. Western Australian Department of Transport.Google ScholarLipsey and Wilson, 2001M.W. Lipsey, D.B. WilsonPractical meta-analysisSAGE publications Inc. (2001)Google ScholarLitman, 2003T. LitmanThe Online TDM Encyclopedia: mobility management information gatewayTransp. Policy, 10 (3) (2003), pp. 245-249, 10.1016/S0967-070X(03)00025-8ArticleDownload PDFView Record in ScopusGoogle ScholarLoukopoulos, 2007Loukopoulos, P., 2007. A classification of travel demand management measures. Threats from Car Traffic to the Quality of Urban Life: problems, causes and solutions, 273–292. https://doi.org/10.1108/9780080481449-015.Google ScholarMa et al., 2017L. *Ma, C. Mulley, W. LiuSocial marketing and the built environment: What matters for travel behaviour change?Transportation, 44 (5) (2017), pp. 1147-1167, 10.1007/s11116-016-9698-2CrossRefView Record in ScopusGoogle ScholarMoher et al., 2009D. Moher, A. Liberati, J. Tetzlaff, D.G. Altman, T.P. GroupPreferred Reporting Items for Systematic Reviews and Meta-Analyses : The PRISMA StatementPLoS Med, 6 (2009), 10.1371/journal.pmed.1000097Google ScholarMöser and Bamberg, 2008G. Möser, S. BambergThe effectiveness of soft transport policy measures: A critical assessment and meta-analysis of empirical evidenceJ. Environ. Psychol., 28 (1) (2008), pp. 10-26, 10.1016/j.jenvp.2007.09.001ArticleDownload PDFView Record in ScopusGoogle ScholarNakayama and Takayama, 2005S. *Nakayama, J.I. TakayamaEcotravel coordinator program: Effects on travel behavior and environmental attitudeTransp. Res. Rec., 1924 (1) (2005), pp. 224-230, 10.3141/1924-28View Record in ScopusGoogle ScholarO’Fallon and Sullivan, 2004C. O’Fallon, C. SullivanPersonalised marketing-improving evaluationTransp Eng. Australia, 9 (2) (2004), p. 85Google ScholarOgilvie et al., 2004D. Ogilvie, M. Egan, V. Hamilton, M. PetticrewPromoting walking and cycling as an alternative to using cars: systematic reviewBMJ, 329 (2004), p. 763, 10.1136/bmj.38216.714560.55CrossRefView Record in ScopusGoogle ScholarParker et al., 2007*Parker, J., Harris, L., Chatterjee, K., Armitage, R., Cleary, J., Goodwin, P., 2007. Making Personal Travel Planning Work : Research Report.Google ScholarRichardson, 2003A.J. RichardsonTemporal variability of car use as an input to design of before and after surveysTransp. Res. Rec., 1855 (1) (2003), pp. 112-120, 10.3141/1855-14CrossRefView Record in ScopusGoogle ScholarRichter et al., 2009J. Richter, M. Friman, T. GärlingSoft transport policy measures 1: Results of implementationsKarlstads universitet (2009)Google ScholarRodriguez and Rogers, 2014D.A. *Rodriguez, J. RogersCan housing and accessibility information influence residential location choice and travel behavior? An experimental studyEnviron. Plan. B: Plan. Design, 41 (3) (2014), pp. 534-550, 10.1068/b120032pCrossRefView Record in ScopusGoogle ScholarRose and Ampt, 2001G. Rose, E. AmptTravel blending: an Australian travel awareness initiativeTransp. Res. Part D: Transp. Environ., 6 (2) (2001), pp. 95-110, 10.1016/S1361-9209(00)00015-8ArticleDownload PDFView Record in ScopusGoogle ScholarRosenthal, 1979R. RosenthalThe file drawer problem and tolerance for null resultsPsychol. Bull., 86 (3) (1979), p. 638, 10.1037/0033-2909.86.3.638ArticleDownload PDFCrossRefView Record in ScopusGoogle ScholarSanjust et al., 2015B. Sanjust, I. Meloni, E. SpissuAn impact assessment of a travel behavior change program: a case study of a light rail service in Cagliari, ItalyCase Stud. Transp. Policy, 3 (1) (2015), pp. 12-22, 10.1016/j.cstp.2014.04.002ArticleDownload PDFView Record in ScopusGoogle ScholarSavage et al., 2011B. Savage, T. Knight, J. Bacon, A. Millington, H. Bullock, J. BucklandBehavioural Insights ToolkitDepartment for Transport, London (2011)https://www.sustainabilityexchange.ac.uk/files/decc_behavioural_insights_toolkit.pdfGoogle ScholarSargeant et al., 2004J. *Sargeant, T. Carter, S. Mcsweeney, W. HughesCambridgeshire Travel Choice Project. Final ReportUK Department for Transport, Cambridgeshire County Council, Cambridge University Hospitals NHS Foundation Trust, Cambridgeshire Travel for Work Partnership, Cambridge, UK (2004)Google ScholarSteg, 2003L. StegFactors influencing the acceptability and effectiveness of transport pricingAcceptab. Transp. Pric. Strat., 27 (2003), pp. 187-202, 10.1108/9781786359506-012CrossRefView Record in ScopusGoogle ScholarSteg, 2005L. StegCar use: lust and must. Instrumental, symbolic and affective motives for car useTransp. Res. Part A: Policy Pract., 39 (2–3) (2005), pp. 147-162, 10.1016/j.tra.2004.07.001ArticleDownload PDFView Record in ScopusGoogle ScholarSteininger et al., 1996K. Steininger, C. Vogl, R. ZettlCar-sharing organizations: The size of the market segment and revealed change in mobility behaviorTransp. Policy, 3 (4) (1996), pp. 177-185, 10.1016/S0967-070X(96)00024-8ArticleDownload PDFView Record in ScopusGoogle ScholarStopher and Bullock, 2003Stopher, P., Bullock, P., 2003. Travel behavior modification: a critical appraisal. Paper Presented at the 26th Australasian Transport Research Forum, Wellington, New Zealand. Retrieved from https://pdfs.semanticscholar.org/b83f/e85f7221f34acf7f37ccf5a15eb32f6642b9.pdf?_ga=2.45181346.726671213.1590578065-1838460617.1590578065.Google ScholarStopher, 2004P.R. StopherReducing road congestion: a reality checkTransp. Policy, 11 (2) (2004), pp. 117-131, 10.1016/j.tranpol.2003.09.002ArticleDownload PDFView Record in ScopusGoogle ScholarTaylor, 2007M.A. TaylorVoluntary travel behavior change programs in Australia: The carrot rather than the stick in travel demand managementInt. J. Sustain. Transp., 1 (3) (2007), pp. 173-192, 10.1080/15568310601092005CrossRefView Record in ScopusGoogle ScholarTaylor and Ampt, 2003M.A. Taylor, E.S. AmptTravelling smarter down under: policies for voluntary travel behaviour change in AustraliaTransp. Policy, 10 (3) (2003), pp. 165-177, 10.1016/S0967-070X(03)00018-0ArticleDownload PDFView Record in ScopusGoogle ScholarThomas et al., 2016G.O. Thomas, W. Poortinga, E. SautkinaHabit discontinuity, self-activation, and the diminishing influence of context change: evidence from the UK understanding society surveyPLoS ONE, 11 (4) (2016), 10.1371/journal.pone.0153490Google ScholarThøgersen and Møller, 2008J. Thøgersen, B. MøllerBreaking car use habits: The effectiveness of a free one-month travelcardTransportation, 35 (3) (2008), pp. 329-345, 10.1007/s11116-008-9160-1CrossRefView Record in ScopusGoogle ScholarVerplanken et al., 1998B. Verplanken, H. Aarts, A.D. Van Knippenberg, A. MoonenHabit versus planned behaviour: A field experimentBr. J. Soc. Psychol., 37 (1) (1998), pp. 111-128, 10.1111/j.2044-8309.1998.tb01160.xCrossRefView Record in ScopusGoogle ScholarVerplanken et al., 1997B. Verplanken, H. Aarts, A. Van KnippenbergHabit, information acquisition, and the process of making travel mode choicesEur. J. Soc. Psychol., 27 (5) (1997), pp. 539-560, 10.1016/S0001-6918(97)00008-5View Record in ScopusGoogle ScholarVerplanken and Roy, 2016B. Verplanken, D. RoyEmpowering interventions to promote sustainable lifestyles: Testing the habit discontinuity hypothesis in a field experimentJ. Environ. Psychol., 45 (2016), pp. 127-134, 10.1016/j.jenvp.2015.11.008ArticleDownload PDFView Record in ScopusGoogle Scholar© 2020 The Authors. Published by Elsevier Ltd.
