Promoting Eco-Driving Habits: A Randomised Controlled Trial

Promoting Eco-Driving Habits: A Randomised Controlled Trial

Dimitrios Xenias Lorraine Whitmarsh Paul Haggar Cardiff UniversitySteve Skippon Shell

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 (Horeni et al., 2007; Neal et al., 2011)

Habit discontinuity

Habit

Context

• 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 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

Information Feedback Social influence

Control

Existing vehicle A B C

New vehicle D E F

D,E,F > A,B,C - B,E > A,D - C,F > B,E,A,D

Interventions1. Information 2. Feedback 3. Social influence

Interventions1. 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

Information Feedback Social influence Control

Existing vehicle 62 (50) 62 (50) 62 (50) 62 (50)

New vehicle 62 (50) 62 (50) 62 (50) 62 (50)

▫ 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

One-off info Feedback Social influence Control

Existing vehicle 62 (50) 62 (50) 62 (50) 62 (50)

New vehicle 62 (50) 62 (50) 62 (50) 62 (50)

▫ >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...

One-off info Feedback Social influence Control

Existing vehicle 61 (27) 85 (41) 64 (26) 66 (33)

New vehicle 03 (03) 36 (12) 07 (04) 29 (19)

▫ 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

Sample (♂= 84, ♀=81)Started Completed

Age Frequency % Cumulative % Frequency % Cumulative %18-19 6 1.7 1.7 - - -20-24 54 15.2 16.9 15 9.1 9.125-29 48 13.5 30.4 19 11.5 20.630-34 51 14.4 44.8 17 10.3 30.935-39 32 9.0 53.8 12 7.3 38.240-44 35 9.9 63.7 24 14.5 52.745-49 29 8.2 71.8 17 10.3 63.050-54 31 8.7 80.6 16 9.7 72.755-59 29 8.2 88.7 17 10.3 83.060-64 25 7.0 95.8 18 10.9 93.965-69 11 3.1 98.8 8 4.8 98.870-74 4 1.1 100.0 2 1.2 100.0Total 355 100.0 165 100.0

No differences between finishers and startersItem (examples) N Range Mean N Range Mean

When I am driving I try to save fuel (DrivingGoals5item_BL_1) 165 6 5.15 355 6 5.18

When I am driving I try to get to my destination as quickly as possible (DrivingGoals5item_BL_2) 165 6 4.30 355 6 4.34

If I drive in the next week, I intend to drive in a fuel-efficient way (Intention_ecodriving_BL_1) 165 6 5.25 355 6 5.20

At the moment, how easy would you find it to drive in a / fuel-efficient way? (PerceivedBehabiouralControl_BL_1) 165 3 1.95 355 3 1.97

Do any of your friends or family drive in a fuel-efficient way? (Social Norm (present)_BL_1) 165 2 1.42 355 2 1.36

Quicker than alternatives (ReasonsForDriving_BL_1) 164 9 2.30 354 9 2.31

Inadequate alternatives / no other option (ReasonsForDriving_BL_2) 164 9 3.46 354 9 3.66

Cheaper than alternatives (ReasonsForDriving_BL_4) 164 10 4.65 354 10 4.58

Gives me a sense of prestige (ReasonsForDriving_BL_10) 164 9 9.32 354 9 9.23

How much time pressure are you usually under / when you are driving to work / studies? (Time Pressure_BL) 128 2 .87 281 2 .92

Distribution of fuel efficiency

Fuel efficiency (car change)

-0.008

-0.006

-0.004

-0.002

0

0.002

0.004Efficiency (by cost)

NoChange

CarChange

£ /

mil

e

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03Efficiency (by fuel)

NoChange

CarChange

Lt /

mil

e

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.)

-0.012

-0.01

-0.008

-0.006

-0.004

-0.002

0

0.002

0.004

0.006Efficiency (by cost)

NoIntervention

Intervention

£ /

mil

e

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

Efficiency (by fuel)

NoIntervention

Intervention

Lt /

mil

e

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.)

-0.04

-0.02

0

0.02Efficiency (by cost)

Infor-mationFeedbackDiscussionControl

Intervention type

£ /

mil

e

-0.04

-0.02

0

0.02

0.04

0.06Efficiency (by fuel)

Infor-mation FeedbackDiscussionControl

Intervention type

Lt

/ m

ile

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Habit strength

SRHI_ Base-line

SRHI_Halfway

SRHI_Final

Time of measurement

Mean

habit

str

en

gth

(Error bars represent 95% Confidence Intervals.)

F(2,318)=28.093, p<.001

Habit (SRHI) seems to increase, regardless of car change

All t(163)<.916, all p>.483

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Habit strength by

Car Change

SRHI_BaselineSRHI_Halfway

Non car changers Car changers

Mea

n h

abit

str

ength

(Error bars represent 95% Confidence Intervals.)

Habit (SRHI) seems to increase, regardless of car change

• 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.

Habit (SRBAI) increases most in information condition

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8Habit (SRBAI)

Infor-mationFeedbackDiscussionControl

Intervention type

Mean

hab

it s

tren

gth

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)

-0.1

-0.05

0

0.05

0.1

0.15

0.2

MDSI 'Careful' change

Information

Feedback

Discussion

Control

Intervention type

Mean

pre

- po

st s

tudy

ch

an

ge

Fuel efficiency does not really correlate with anything, except these trends

Pearson Correlation

Sig. (2-tailed) N

Emotional Stability (baseline) -.146 .062 160

Agreeableness (halfway) -.138 .081 165

• Meteorological conditions also did not affect fuel efficiency

Habit (SRHI) is related to driving style (MDSI)

Measure StatisticSRHI

BaselineSRHI

HalfwaySRHIFinal

MDSI Reckless at Baseline Pearson’s r -.27** -.29** -.27**

MDSI Careful at Baseline Pearson’s r .35** .32** .27**

** indicates p <.001

Some conclusions• Type of intervention did not lead 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

[email protected] [email protected]

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

Concerns about efficiency data trustworthiness

Week 1-2 Week 3-4 Week 5-6 Week 7-8 Week 9-10 Week 11-12

-10

-8

-6

-4

-2

0

2

4

6

8

10

AverageMileage(%ofaveragelvalue) Averagefuel(%ofaveragevalue)mean-litres/mean-miles

% d

iffer

ence

from

res

pecti

ve m

ean

SRHI is unrelated to fuel efficiencyDriving style is unrelated to fuel efficiency

SRHI

BaselineSRHI

Halfway SRHI Final HabitChangeNet Cost per Distance

Pearson Correlation .038 .054 .093 .082

Sig. (2-tailed) .629 .501 .234 .297

N 165 160 165 165

Net Fuel per Distance

Pearson Correlation .018 .039 .088 .106

Sig. (2-tailed) .821 .627 .263 .177

N 165 160 165 165

This is very similar to MDSI driving style, too

Pre-post efficiency change? (Car change x intervention)

F(3,165)=.487, p=.692

Cost Volume

F(3,165)=.340, p=.769

Pre-post efficiency change? (Car change x intervention)

F(7,165)=.978, p=.449

Cost Volume

F(7,165)=.756, p=.625