330 likes | 517 Views
Promoting Eco-Driving Habits: A Randomised Controlled Trial. Dimitrios Xenias Lorraine Whitmarsh Paul Haggar Cardiff University Steve Skippon Shell. Habits. Much (most?) of what we do is habitual (contra. most soc. psych. models) 3 ingredients to habit: . . .frequency
E N D
Promoting Eco-Driving Habits: A Randomised Controlled Trial Dimitrios Xenias Lorraine Whitmarsh Paul HaggarCardiff University Steve SkipponShell
Habits • Much (most?) of what we do is habitual (contra. most soc. psych. models) • 3 ingredients to habit: • ...frequency • ...automaticity • ...cued by stable contexts (i.e. spatial, social and temporal environment) Verplanken& Wood (2006)
Habituation ...attenuates attention to new information(Verplanken et al., 1997) ...attenuates attention to changing conditions (Horeniet al., 2007; Neal et al., 2011)
Habit discontinuity • Information about alternative choices (e.g., bus travel) tends to be ignored when we have strong habits (e.g., to drive) • But when habits are disrupted by events/decisions (e.g., relocation, new job) behaviour-relevant information becomes more salient and influential • = Habit discontinuity hypothesis • (Verplanken & Wood, 2006) • Tailored public transport info and 1-day bus pass given 6-weeks post-relocation was significantly more effective (increase from 18% to 47%) than when given to those not relocating (18% to 25%, n.s.; Bamberg, 2006) Context Habit
Context change • Relocation • Family circumstances • Change of employment • Change of vehicle • … = Window of opportunity Thompson et al (2011); Schäfer et al. (2012)
Change of vehicle - interventions • 1) Information provision: • Shown to reduce energy use by up to 9% (Maibach, 2008) • More likely to be effective if situated where action occurs (Whitmarsh et al., 2011) • 2) Feedback provision: • Drivers save up to 10% fuel, esp. under little stress (Dogan et al., 2011) • Interventions more likely to work when real time (Stillwater & Kurani,2012) • 3) Social influence: • Talking to people we identify with (Ellmers et al., 2002) helps behaviour change towards a stated norm (Rabinovich et al., 2010)
Design & Hypotheses D,E,F > A,B,C - B,E > A,D - C,F > B,E,A,D
Interventions 1. Information 2. Feedback 3. Social influence
Interventions 1. Information 2. Feedback 3. Social influence
Measures • Eco-driving habit strength (e.g. Verplanken & Orbell, 2003) • Personality measures(e.g. TIPI: Gosling, Rentfrow & Swann, 2003) • Driving style (e.g. MDSI; Taubman-Ben-Ari et al., 2004) • Goals when travelling (Skippon et al., 2013) • Vehicle and personal information • Fuel consumption (receipts and mileage)
Sample size • Assuming medium effect size (e.g., 6% improvement in mpg = 0.25 f; Boriboonsomsin et al., 2010) required sub-group sizes of 50 (i.e., total N = 400). • Target N= 500 (400 after attrition)
Recruitment strategy • >700 members of the Cardiff Community Panel (personalised emails) • Advertisement in two local newspapers • Advertisement on the University Intranets (Cardiff and Bath) • Flyers in >20 garages in Cardiff and Bristol • TRL panel =substantial help, mainly for car changers (hardest to get!) • Google AdWords (an expensive idea!) • Responding participants were directed to a vetting survey
Attrition... • Recruitment strategies brought 670 participants to the vetting survey (bias: younger + female) • 383 passed vetting – 55 quit immediately after • 328 began study (cut off mid-July 2013) • 165 completed study ’000s reached 670 vetted 383 passed 328 started 165 completed
Fuel efficiency (car change) T-test for efficiency calculated as cost (t(1,163)=.48, p=.63) T-test for efficiency calculated as fuel volume (t(1,163)=.31, p=.76). (Error bars represent 95% Confidence Intervals.)
Fuel efficiency (intervention) T-test for efficiency calculated as cost (t(1,163)=1.13, p=.26) T-test for efficiency calculated as fuel volume (t(1,163)=.90, p=.33). (Error bars represent 95% Confidence Intervals.)
Fuel efficiency (intervention) F-test for efficiency calculated as cost (F(3,158)=.50, p=.68) F-test for efficiency calculated as fuel volume (F(3,158)=.34, p=.77). (Error bars represent 95% Confidence Intervals.)
Habit (SRHI) seems to increase, regardless of car change (Error bars represent 95% Confidence Intervals.) F(2,318)=28.093, p<.001
Habit (SRHI) seems to increase, regardless of car change (Error bars represent 95% Confidence Intervals.) All t(163)<.916, all p>.483
Habit (SRBAI) increases most in information condition • Although ANOVA showed overall trend was not sig. across conditions (F(3,158)=1.91, p=.13, partial η2=.04, observed power =.49), specific condition contrasts revealed sig. difference between information and control conditions (contrast estimate=.44, p=.02). • CIs suggest lack of an overall significant trend is likely due to issues with sample sizes.
Increase in careful driving for feedback condition • ‘Careful’ dimension of MDSI showed a marginal change after the study, compared to before (F(3,154)=2.39, p=.07, partial η2=.04, observed power=.59). • This change only sig. for in-car feedback intervention (condition 2), compared to control (Dunnett’s t=.25, p=.03)
Fuel efficiency does not really correlate with anything, except these trends • Meteorological conditions also did not affect fuel efficiency
Habit (SRHI) is related to driving style (MDSI) ** indicates p <.001
Some conclusions • Type of intervention did notlead to change in fuel consumption, when measured using means available to drivers in real world (mileage, fuel purchases) • Did find eco-driving habit strength increased over the duration of the study, particularly for condition 1 (information provision), whereas condition 2 (in-car feedback) was associated with increase in careful driving style • Our RCT design allows confidence in our findings and suggests real-world interventions to change driving style may be more problematic than previously thought • Thus, may be hard to make effective real-world eco-driving interventions • Working with real-world samples introduces issues with fuel data and mileage reporting accuracy, which may have added significant measurement error. Error could be mitigated in future studies by using in-car fuel monitors, this could compromise external validity: if an intervention does not lead to changes the drivers themselves can perceive/measure, it is rather unlikely to succeed
Thank you xeniasd@cf.ac.ukwhitmarshle@cf.ac.uk
Some considerations... • Measures 1: Fuel data (took a lot of debugging!) : • >1,300 fuel receipts, • 117 of which (8.4%) with cost only (quantity had to be estimated). • £57,156 represented in fuel receipts • 36,159 litres represented in fuel receipts • £7,206 (12.5%) does not correspond to fuel quantity, as 8.4% of receipts report cost only – therefore missing fuel had to be estimated. • Fuel efficiency calculated fortnightly: data miss 13% - 20% of mileage data • Fuel efficiency calculated 6-weekely: data miss around 3% of mileage data • 1-Week interval data cannot be computed (most 1 week windows don’t have fuel receipts artificial consumption data. The narrower the timeslots, the less fuel efficient participants appear to be. • 6-Weekly = much more accurate • Truly Unknown = fuel remaining in tank • Generally, a lot of missing fuel data
SRHI is unrelated to fuel efficiencyDriving style is unrelated to fuel efficiency This is very similar to MDSI driving style, too
Pre-post efficiency change? (Car change x intervention) Cost Volume F(3,165)=.340, p=.769 F(3,165)=.487, p=.692
Pre-post efficiency change? (Car change x intervention) Cost Volume F(7,165)=.756, p=.625 F(7,165)=.978, p=.449