Can commitment contracts boost participation in public health programmes?

Abstract

Commitment devices aim to help people make better choices in the face of their inherent biases: they are voluntary strategies aimed at changing behaviours by introducing costs to your current self, to bring about gains for your future self. Adherence to a structured health intervention is an important part of achieving health goals, and may be improved by commitment devices designed to keep people on track with their health goals. A field experiment set in a public weight management programme tests whether a personal commitment device in the form of a contract with oneself, which relies solely on self-reputation costs, can raise weekly participation and completion of the programme. Results suggest the commitment contract can significantly improve attendance (p = 0.05) and completion rates (p = 0.032), with some suggestive evidence that the contract works especially well for people with more myopic health attitudes. Findings also suggest the commitment contract can substitute for, but does not necessarily add to, wider commitment features in the health programme; raising new questions around threshold effects and the theory underlying commitment devices.

Full text

Contents lists available at ScienceDirect
Journal of Behavioral and Experimental Economics
journal homepage: www.elsevier.com/locate/jbee
Can commitment contracts boost participation in public health
programmes?
Manu Manthri Savani
1
Brunel University London, Department of Social and Political Sciences, United Kingdom
ARTICLE INFO
Keywords:
Commitment device
Health behaviour
Field experiment
Behavioural public policy
Dual-self theory
ABSTRACT
Commitment devices aim to help people make better choices in the face of their inherent biases: they are
voluntary strategies aimed at changing behaviours by introducing costs to your current self, to bring about gains
for your future self. Adherence to a structured health intervention is an important part of achieving health goals,
and may be improved by commitment devices designed to keep people on track with their health goals. A field
experiment set in a public weight management programme tests whether a personal commitment device in the
form of a contract with oneself, which relies solely on self-reputation costs, can raise weekly participation and
completion of the programme. Results suggest the commitment contract can significantly improve attendance
(p= 0.05) and completion rates (p= 0.032), with some suggestive evidence that the contract works especially
well for people with more myopic health attitudes. Findings also suggest the commitment contract can substitute
for, but does not necessarily add to, wider commitment features in the health programme; raising new questions
around threshold effects and the theory underlying commitment devices.
1. Introduction
The influence of present-bias on inter-temporal choices is well es-
tablished in the literature (Frederick, Loewenstein, & O'Donoghue,
2002; Loewenstein et al., 2012; O’ Donoghue & Rabin 1999), and has
been linked to a range of undesirable outcomes including weaker aca-
demic performance (Ariely & Wertenbroch, 2002), vulnerability to
natural disasters (Kunreuther, Meyer, & Michel-Kerjan, 2013), and in-
sufficient savings for retirement (Thaler & Benartzi, 2004). Many health
choices involve trade-offs: the benefits of a higher quality and longer
life may be felt years down the line, but rely on health investments,
including costly self-restraint, at the present time. Empirical evidence
associates present bias with under-investment in preventative health
(Dupas, 2011), including weight management (Liu et al., 2014). Fan
and Jin report that “a lack of self control capability is associated with
poor eating and exercise behaviours, as well as an increase in obesity
risk and BMI” (2013, p. 18). Individuals often identify an optimal
course of action – the right diet, the right physical activity – but when
the moment comes to put it into practice, they often delay or quit
(Rogers et al., 2015); they are time-inconsistent (Strotz, 1955). Getting
these intertemporal choices right is not easy, particularly in an
obesogenic environment (Costa-Font et al., 2013).
Obesity remains a “widespread threat to health and wellbeing” in
the UK (Department of Health, 2011, p. 5), despite a range of policy
efforts, including changes to food retailing, unprecedented access to
health information, long-running public health campaigns such as
‘Change4Life’, and the growth in digital health tools and wearable de-
vices. The Health Survey for England 2017 reports that 64% of adults
are now overweight or obese compared to 53% two decades ago, while
the proportion of people who have a normal body mass index (BMI) has
fallen from 45% to 34% during that time (Health Social Care
Information Centre, 2015; NHS Digital, 2018). Better access to health
information is not enough to improve individual decision-making
(Downs, Loewenstein, & Wisdom, 2009). Behavioural science is in-
creasingly being brought to bear on health policy (Perry et al., 2015),
and are seen as an opportunity to address the biases linked to poor
health choices.
Commitment devices are one amongst a menu of ‘nudges’ designed
to tackle behavioural biases (Oliver & Ubel, 2014; Thaler & Sunstein,
2008), and are widely employed in the weight loss sector. A commit-
ment device is any voluntary arrangement that restricts or binds future
choices, to “fulfil a plan for future behaviour that would otherwise be
https://doi.org/10.1016/j.socec.2019.101457
Received 13 February 2019; Received in revised form 23 August 2019; Accepted 27 August 2019
E-mail address: manu.savani@brunel.ac.uk.
1
This work was undertaken with the support of UCL's Department of Political Science. My thanks to Peter John and Roland Kappe for their support and advice on
this project, and to Albert Weale and David Torgerson. Participants at the LSE International Health Policy Conference, and the UK Faculty of Public Health Annual
Conference provided useful comments on the study. My thanks also to two anonymous reviewers. Any errors are my own. I owe a debt of gratitude to staff at Camden
Active Health team for hosting the fieldwork, in particular Verena Trend and Ian Reddington.
Journal of Behavioral and Experimental Economics 82 (2019) 101457
Available online 28 August 2019
2214-8043/ © 2019 Elsevier Inc. All rights reserved.
T

difficult owing to intra-personal conflict, stemming from, for example, a
lack of self-control” (Bryan, Karlan, & Nelson, 2010, p. 672). These
arrangements are taken up by the individual in the pursuit of some goal,
with no strategic motives relating to others. The commitment device
may take the form of an actual contract with a third party
(Halpern et al., 2015), or it may be a more ad hoc arrangement created
by individuals as a “promise to oneself” (Benabou & Tirole, 2004, p.
849).
Bryan et al. refer to ‘soft’ and ‘hard’ commitment devices, depending
on whether they involve primarily a psychological or financial cost
respectively (2010, p. 672), but note that this distinction is not strictly
binary, as financial commitments may also have psychological costs to
failure attached. Thaler and Shefrin's (1981) typology of commitment
devices differentiated between methods to alter incentives and to alter
opportunities. They also distinguish between external and internal
rules, depending on whether the individual relies on some external
assistance or applies personal, self-enforced rules. Informal commit-
ment devices for weight management include strategies such as signing
up to a running club, or sharing weight loss goals on social media.
Commercial weight loss programmes incorporate reputational and fi-
nancial commitment strategies in the form of publicising weekly weight
readings to the group, and offering membership fees back if weight loss
targets are met. Such features have been tested in public health pro-
grammes (Relton, Strong, & Li, 2011), and are advocated as valuable
behaviour change instruments (Rogers, Milkman, & Volpp, 2014).
Much of the empirical evidence focuses on financial commitment de-
vices, or deposit contracts, that stake money on achieving an outcome;
the individual earns their money back by meeting the goal. Such in-
terventions have been found to support weight loss and smoking ces-
sation, particularly in the short run (Gine, Karlan, & Zinman, 2010;
Halpern et al., 2015; John et al., 2011; Volpp et al., 2008); with some
evidence of both short and long term effects, for example on healthier
food shopping (Mochon et al., 2016; Schwartz et al., 2014).
Commitment contracts that rely purely on reputational costs are
relatively under-researched. The evidence base lacks conclusive an-
swers on whether they work, for whom they might be important, and
for what outcomes. For example, can these milder forms of commitment
device affect more straightforward behaviour change goals around
adherence and participation in weight loss programmes?
The prevalence of commitment devices in the weight management
sector, their potential for promoting desired health behaviours, and the
relatively weak evidence base, motivates this paper to ask: can re-
putational commitment devices promote behaviour change? The paper
contributes to the behavioural economics literature on commitment
devices, with novel experimental evidence on health behaviours and
biases. Specifically, I examine the effects of a commitment contract on
attendance and completion of a community weight management pro-
gramme in the UK, and weight loss outcomes. The next section provides
a more detailed review of the literature on reputational commitment
contracts and health behaviours. Section 3 details the programme
context and experimental design. Section 4 presents results. Section 5
discusses the findings that the commitment contract promoted atten-
dance, and explores the relationship with weight loss. Section 6 con-
cludes that self-reputational commitment contracts exert a positive in-
fluence on behaviour, and being cheap and straightforward to
administer may be readily incorporated into health services. It is not yet
clear they can effectively promote weight management outcomes, and
further research is needed to identify how commitment contracts can be
optimally designed and tailored to individual needs.
2. Theory
2.1. The planner–doer model
Maintaining sound health behaviours for long-term health requires
a strategic deferral of rewards, which is challenging in the context of
present bias and limited self-control. Cavaliere et al. find that the
probability of being “overweight or obese increases when consumers
are less future-concerned”, and conversely a healthy BMI is “associated
with a high orientation to the future” (2014, p. 135).
How might a commitment device change behaviour? Behavioural
economics theory explains this as a rebalancing of intertemporal costs,
bringing forward the incentive to invest in health now, rather than
relegating the health investment decision to a later time. In their
seminal article, "An Economic Theory of Self-Control", Thaler and
Shefrin (1981) arrive at the same conclusion as Strotz's earlier treat-
ment of the problem: a commitment strategy will address time incon-
sistency. Thaler and Shefrin however pose the issue as a principal-agent
problem, and delve deeper into the individual's self-control problems.
The individual is modelled as having two competing sub-selves, a
"planner" and a "doer". The framework successfully combines the con-
cepts of present bias (implicitly) and self-control (explicitly) to enrich
our understanding of why we may set a plan for the future, and when
the future arrives we delay, procrastinate, or quit.
Savani (2018) provides a more formal application of the planner-
doer framework to health behaviours. The key premise is the diver-
gence between the preferences of the myopic doer and the far-sighted
planner, which sets the scene for an internal tussle on inter-temporal
decisions. The myopia of the doer exaggerates the notion that humans
tend to be present-biased, focused solely on the rewards available now,
at the cost of any longer term thinking; there is no utility from delaying
gratification because the doer only considers the present period. The
planner, however, has a longer time horizon and a utility function that
takes account of longer-term payoffs, and therefore recognises that
costs now can pave the way for higher rewards later. Two important
implications of the planner-doer model are, firstly, that such time in-
consistency can be overcome through commitment strategies that
change the doer's choices (Thaler & Shefrin, 1981, p. 398); and sec-
ondly, the desire to maximise long-run utility prompts the planner to
demand such commitment strategies (Savani, 2018, p. 4). The way that
commitment strategies elicit change in individual behaviour is to exert
some costs to the doer sub-self continuing to act in a time-inconsistent
way. These costs might arise through financial penalties, as in the case
of deposit contracts, or they may take the form of reputational costs, as
discussed further below.
2.2. Heterogeneous effects of commitment devices
A possibility arising the planner-doer model, and one that is not
fully formalised in existing theory, is the potential for heterogeneity in
how commitment devices might affect different people (Savani, 2018).
Thaler and Shefrin discussed the plausibility of “individual rates of
impatience” (1981, p. 403). The intuition of varying rates of time
preference has a considerable history (Frederick et al., 2002). The field
experimental literature on commitment devices is beginning to address
the question of heterogeneous treatment effects directly. Nyer and
Dellande (2010) use a personality trait to identify heterogeneous ef-
fects, finding that a public pledge on weight loss is more effective
amongst those participants who were highly susceptible to normative
influence. Royer, Stehr, and Sydnor (2015) also test for sub-group ef-
fects, reporting that baseline variables including gym membership,
weight, gender and levels of exercise explain some of the variation in
treatment effects of a commitment contract on exercise behaviour.
These studies provide useful empirical pointers, but lack a theoretical
grounding in a dual-self model that also explains the original need for a
commitment device.
I argue that a key variable determining the effectiveness of a com-
mitment device is the extent to which the planner-doer scenario applies
within the individual. Empirical evidence demonstrates that people
have different rates of time preference (Andreoni, Kuhn, & Sprenger,
2015), and recent data from the Health Survey for England reports that
the degree of short-termism detected in health attitudes differs across
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
2

respondents (Robinson, 2012). People experience and exhibit degrees
of myopia or far-sightedness when it comes to their health outlook and
behaviours. Who are most likely to experience positive behaviour
change from a commitment device? Perhaps it is those who would
otherwise face the largest divergence between the doer's actions and the
planner's health goals, in other words those who behave more in line
with their doer sub-selves to begin with. People who are initially more
myopic in their health attitudes are therefore expected to experience
larger and positive effects on their behaviour from a commitment de-
vice.
2.3. Commitment devices for health behaviours
Commitment has been defined as the “pledging or binding of the
individual to behavioural acts” (Kiesler & Sakumura, 1966, p. 349). The
commitment makes an act less changeable. The magnitude of the
commitment is associated with how publicly it is stated, because of an
individual's desire to be consistent with what he has declared to others,
and to avoid the personal and social disapproval that accompanies in-
consistency. Nyer and Dellande (2010) find positive effects on weight
loss when individuals display their pledge on a public noticeboard at
the gym compared to those whose goals remain private.
Pledges do not have to be very public to be effective, indeed “in-
dividuals may not need the collaboration of other parties in order to
self-regulate their behaviour” (Brocas, Carillo, & Dewatripont, 2003, p.
54). Making a commitment even to oneself may be sufficient to dis-
cipline short-term health impulses. A pre-written grocery list can serve
as a guide to help the doer (walking around the shopping aisles) stay on
track with the planner's dietary regimen (Au et al., 2013). The most
informal type of reputational commitment device is a personal rule,
which can range from rules of thumb and one-off resolutions (‘no more
chocolate today’) to more active practices such as preparing written
plans for healthier behaviour. A simple contract signed to oneself is
another way of formalising a personal rule, taking a mental note and
making it something more tangible.
The working definition of a commitment contract in this paper is
any arrangement that relies solely on non-financial elements and has
some element of ingrained formality, perhaps by being written down or
through some verbal agreement with another person. This is in contrast
with Rogers et al. (2014) who refer to commitment devices, commit-
ment contracts and deposit contracts interchangeably. The argument
put forward below isolates and tests the effectiveness of one particular
type of commitment strategy: a commitment contract attached to self-
reputational costs.
Self-enforcement relies on there being some short-term cost if the
doer reneges on the planner's long-term goal; a ‘psychological tax’ that
helps align the doer's and planner's incentives (Miller & Prentice, 2013).
Such a cost can arise from the potential damage to individual's per-
ception of themselves, as stated by Benabou and Tirole: “[people] see
their own choices as indicative of ‘what kind of person’ they are”; such
that a fear “of losing faith in oneself then creates an incentive that helps
counter the bias toward instant gratification” (2004, p. 849), as ex-
emplified by the doer sub-self. Following this logic, I argue that a
commitment contract to oneself has the capacity to rein in momentary
impulses and improve health behaviours, but this proposition is rela-
tively untested.
2.4. Adherence to health programmes as an outcome itself
Existing work on commitment devices for health have largely fo-
cused on weight outcomes (John et al., 2011), smoking cessation
(Gine et al., 2010) or physical activity outcomes (Royer et al., 2015),
such as visiting the gym (Milkman, Minson, & Volpp, 2014). Studies
have not tested the role of commitment strategies in promoting ad-
herence to public health programmes. This is an important gap, because
evidence shows that attendance and completion rates in weight
management programmes are often lower than desired (Blane et al.,
2017; Brown et al., 2015); that adherence has the potential to enhance
weight management outcomes, for example through increased self-
monitoring (Butryn et al., 2007; Peterson et al., 2014); and there is a
need for more research on behavioural factors that affect dropout from
weight loss interventions (Moroshko, Brennan, & Brien, 2011).
This paper contributes to the scholarly literature by asking two
questions: can a personal commitment contract keep people on a weight
loss programme for longer? And can a commitment contract keep more
present-biased individuals on the programme for longer? Planner-doer
theory implies that a commitment device increases the costs of de-
viating from a set plan, and so will encourage greater adherence to a
structured weight loss programme. Individuals who would otherwise
face a greater tussle between planner and doer may benefit most from a
commitment contract that binds the doer's actions.
Hypothesis 1. A commitment contract will promote participation in a
weight management programme.
Hypothesis 2. A commitment contract will generate larger treatment
effects for individuals with myopic health attitudes
3. Study setting and experimental design
3.1. Population and programme context
The trial is set within a public weight loss service that was provided
by the Local Authority's public health team in the London borough of
Camden. Obese and overweight individuals resident in Camden were
eligible for the ‘Shape Up’ service: a free, 11-week, group-based pro-
gramme that sets a 5% weight loss target.
2
The target population for the
study was Camden's Shape Up client base, who had already been
screened for eligibility based on their body mass index (BMI) and home
postcode, and entered the programme through self-referral or referral
by a health professional. Groups met weekly at sessions facilitated by a
Shape Up tutor, to learn about various aspects of weight loss and life-
style such as portion sizes, taking account of food labelling, and phy-
sical activity. Tutors helped set weight loss goals (5% of initial body
weight) at the start of the programme, recorded participants’ atten-
dance and weight, facilitated group discussion, and coached individuals
on their food journals and weight readings. Under a normal Shape Up
programme the population made no financial commitment, but may
have been conscious of a mild reputational commitment to the tutor.
3
3.2. Sample size, randomisation and recruitment
A total of 197 participants were recruited from 27 Shape Up groups
over 2014–16, in line with ex ante sample size calculations. Clients
were recruited face-to-face in one of the early weeks of the Shape Up
programme they enrolled in (see Fig. 1).
4,5
The majority of participants
were randomised in advance of recruitment using administrative re-
cords of registration, but in cases where these records were not up to
date participants were randomly assigned to their experimental groups
2
The 11-week course was designed by the non-profit organization Weight
Concern.
3
Clients chose Shape Up over other similar service providers, and knew from
their first meeting that if they successfully completed the programme and
achieved their goals they would be awarded free gym sessions at local leisure
centres. The experiment focused on testing the effects of a commitment strategy
and not the role of incentives in encouraging people to lose weight, and these
incentives were common to all participants.
4
The study was approved by UCL Ethics Committee in December 2013,
project 4518/003. The study was registered at the AEA trial registry in
November 2015, after the trial had begun but before data collection concluded.
5
In line with CONSORT reporting standards (Schulz, Altman, & Moher,
2010).
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
3

at the time of recruitment using a randomisation rule with a 50/50
treatment allocation probability. This ensured that willing and eligible
participants were not excluded. Forty nine per cent were assigned to the
treatment group (n= 97). Tutors were blind to participant status.
Participants were recruited (and those in the treatment group offered
the commitment contract) either during the introductory class or the
next available class they attended.
6
A good balance was achieved in
baseline covariates between the experimental groups (see Table 1).
3.3. Participant profile
The average participant was female (83% of the sample), obese
(52%) and in their late 40 s. Starting BMI averaged 31.0 and ranged
from 24.5 to 47.1, with 98% of participants found to be carrying excess
weight (BMI > 25). A small proportion (6%) were severely obese (BMI
of 40 or over). Routes on to the Shape Up programme included referral
by a GP (32%), other health practitioner (38%), or self-referral (27%);
and 31% had participated in a weight loss programme previously.
There was a strong sense of motivation to lose weight – 83% of all
participants stated they intended to live a healthy lifestyle over the
following 12 months. Yet, dropouts and poor attendance were seen as
key barriers to Shape Up programme effectiveness, and Camden were
keen to increase their attendance and completion rates.
3.4. Intervention
The treatment group were offered a commitment contract (see
Fig. 2) that they were invited to personalise by writing in their names,
signature, and date. The wording of the contract was designed to make
two distinct goals salient: participation in the programme to its com-
pletion, and achievement of the 5% weight loss target set for all clients
in the programme. The contract was designed to have a degree of visual
gravitas and formality, and was printed on card of A5 size that could be
carried around in a handbag or satchel, or stuck on a fridge or wall.
Participants who received the contract were advised to keep it some-
where they would see it on most days, and to discuss it with friends and
family if they wished but not with the other group members. They
signed the contract at the time of recruitment. The control group
completed the baseline survey and, like the treated individuals, were
asked to continue with the Shape Up programme as they normally
would.
7
Recorded attendance and weight readings were gathered at the
end of the course.
3.5. Data and empirical strategy
Participation was measured using two outcome variables: atten-
dance rates and completion status of the programme. Completion status
(a binary variable coded 0 or 1) was based on whether the participant
attended until week 7 of the programme. This benchmark was derived
from Camden's monitoring and performance indicators, and is in line
with comparable benchmarks used for other weight management pro-
grammes (Logue et al., 2014). Shape Up group tutors routinely main-
tained a client register to monitor attendance, and collected end weight
each week. Attendance rates were calculated by the number of sessions
attended as a proportion of total sessions the individual could have
attended, depending on when they joined the Shape Up group – while
Camden aimed to have everyone join from the introductory week 0
class, some participants were signed up in weeks 1 and 2 depending on
personal circumstances. By design, attrition was not a threat to the
validity of the attendance and completion data.
A baseline survey collected data on individual characteristics such
as age, gender, health behaviours and attitudes. Group characteristics
controlled for tutor and proportion of group members who received the
Fig. 1. Field experiment flowchart.
6
Participants only became aware of the trial (and commitment contract, for
individuals assigned to the treatment group) once they attended a Shape Up
class. The commitment contract did not affect the decision to attend the in-
troductory session.
7
Participants who received the contract took it away with them in an en-
velope along with their participant information sheet and consent form. Control
group participants were given a copy of the consent form and information sheet
in an A4 brown envelope to take away with them. Each participant therefore
returned to the class with a brown envelope, whether they belonged to the
control or treatment group, and this was a deliberate design feature to minimise
the risk of contamination across groups and associated problems; such as the
John Henry Effect, where control group members may have felt resentful they
did not receive a contract (Glennerster & Takavarasha, 2013).
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
4

contract -because treatment was randomly assigned at the individual
level, groups varied randomly in the proportion of treated members.
Further control variables take account of recruitment wave and sea-
sonal effects. Camden collected data on initial behaviours around ex-
ercise sessions (per week) and diet (fruit and vegetable portions per
day). This data is valuable but somewhat limited due to missing ob-
servations where tutors failed to collect the Shape Up ‘starter surveys’.
Missing administrative data for baseline exercise levels was provided by
the trial's baseline survey, but missing data on baseline diet could not be
gathered. Other baseline variables identified whether participants had
previously taken part in any structured weight loss programmes, whe-
ther they had recently experienced major life changes, and whether
they were undertaking other activities alongside the Shape Up pro-
gramme in order to lose weight.
Initial health attitudes are a useful way of identifying individuals
who are more likely to be dominated by their doer sub-selves, therefore
exhibit time inconsistency, and be more likely to go off track with their
health goals in the absence of a commitment device. To operationalise
myopic health attitudes, the baseline survey incorporated a 19-item
questionnaire from the Healthy Foundations Segmentation model
(Williams et al., 2011), which was recently applied within the Health
Survey for England to identify how health attitudes are associated with
health outcomes (Robinson, 2012). The model identifies five health
motivation groups, of which three groups are particularly susceptible to
short-termist health attitudes. The analysis uses a binary variable to
capture myopic health attitudes, which describes 54% of the sample.
An important administrative variable that is included is attendance
at the introductory class that launched the 11-week programme. It fo-
cused on social introductions and building rapport with peers and the
tutor – potentially forging a sense of reputational commitment to the
tutors; explaining that Shape Up is about understanding themselves and
their habits (“it's a lifestyle, not a diet” was a common refrain from
tutors); reminding clients of the importance of regular attendance; and
highlighting the programme incentives for completing the programme
(including free gym passes). Overall 67% of participants attended the
introductory class, and there were no significant differences between
treatment and comparison group that might affect causal inference.
Group level dynamics might have played a role in participants’
willingness to complete the programme, so the models incorporate
variables on tutor and the proportion of participants in the group who
also received a contract. The latter variable is particularly useful for
investigating evidence of spillover effects – for example if the treatment
raises performance, this could generate a more positive atmosphere and
a virtuous circle of visible progress, fuelling motivation and further
progress for all group members, not just the treated individuals.
Average treatment effects were isolated using the model below:
= + + + +
=
Y C W S. . .
ic
j
J
jij i
1
(1)
Heterogeneous treatment effects were isolated using the model
below:
= + + + + +
=
Y C C Trait W S. . . . .
ic tr
j
J
jij i
1
(2)
Yis the outcome variable for attendance and completion rates in
Eq. (1). Treatment status is captured by variable C, where C= 1 if the
Table 1
Baseline variables by experimental group.
N Intervention group mean
(1)
Comparison group mean
(2)
Hypothesis test of (1) = (2), p-value
Starting weight 197 82.1 85.3 0.098
Body mass index (BMI) 190 30.7 31.3 0.276
Age 193 47.4 50.2 0.195
Female (%) 197 83.0 83.5 0.924
Myopic health attitudes (%) 197 58.8 49.0 0.169
Subjective wellbeing 190 6.4 6.3 0.805
Exercise (sessions per week) 197 1.63 1.27 0.088
Diet (fruit and veg portions 138 3.75 3.74 0.964
Prior weight loss programme (%) 197 30.0 33.0 0.652
Recent life change (%) 197 36.0 34.0 0.771
Takes part in other activities (%) 196 68.0 70.8 0.667
Self-referred to programme (%) 197 29.0 25.8 0.618
Attended introductory class (%) 197 68.0 65.0 0.653
Recruited in programme wk 1 (%) 197 52.0 50.5 0.835
Recruited in programme wk 2 (%) 197 34.0 34.0 0.998
Recruited in programme wk 3 (%) 197 14.0 15.5 0.772
Recruited in wave 1 (%) 197 28.0 29.9 0.769
Recruited in wave 2 (%) 197 32.0 29.9 0.750
Recruited in wave 3 (%) 197 40.0 40.2 0.976
Pre-randomised 197 73.0 71.1 0.770
Data is drawn from baseline surveys completed by all participants, except variable on diet which was taken from Shape Up starter survey administered by tutors.
Hypothesis testing on group means using t-tests for continuous variables two-group proportion tests for binary variables. For further detail on recruitment waves and
dates see Appendix Table A1. Pre-randomised participants were those who were registered and randomized in advance of recruitment. * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p
< 0.001.
Fig. 2. A commitment contract for health behaviour change.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
5

participant is in the treatment group. The OLS estimator for β
C
provides
the average treatment effect (intent-to-treat) for the commitment con-
tract on attendance. The probit estimator provides the average treat-
ment effect from achieving completer status. The combined linear effect
from OLS estimators on the treatment and treatment xtrait interaction
term is used to identify sub-group effects on attendance in Eq. (2).Wis
a series of individual covariates J, with coefficients γ
j
as described in
the summary of baseline data earlier. Sis a series of administrative
variables including tutor and recruitment wave (see Appendix for full
details). Three regression models are run for each outcome: model 1 is a
baseline model covering individual characteristics, initial health atti-
tudes and behaviours; model 2 adds a binary administrative control
variable for attendance at the introductory session; and model 3 adds a
variable for initial diet, trading off explanatory power from initial be-
haviour for sample size due to missing baseline observations on diet.
The hypotheses imply a positive and statistically significant value for β
C
from models fit for Eq. (1), and a positive and statistically significant
combined treatment and interaction term effect in Eq. (2).
8
4. Results
The data supports hypothesis 1, that the commitment contract can
raise adherence to the weight loss programme. Participants who were
offered the contract recorded average attendance of 71% compared to
67% amongst the comparison group, and higher completion rates of
77% compared to 69% amongst the comparison group. Regression
analysis indicates the commitment contract raises attendance rates by
7.5% (p= 0.050) in the baseline model. The average treatment effect
size is similar in model 2, incorporating the binary variable for at-
tending the introductory class, but not statistically significant
(p= 0.056). Model 3 finds a statistically significant effect of the con-
tract, raising attendance by 12% (p= 0.003), which is equivalent to
one additional session of the 11-week course. With the caveat that the
results are sensitive to model specification, Table 2 suggests a statisti-
cally significant causal impact from the commitment contract on par-
ticipation.
The commitment contract boosts completion rates, and this finding
is robust across model specifications (see Table 3). In the baseline
model, the commitment contract raises the probability of completing
the course (p= 0.032), with a marginal effect of 14% (rising to 19% in
model 3). Model 2 indicates the importance of having attended the
introductory class on subsequent participation in the programme.
Model 3 suggests that the treatment effect was stronger for both com-
pletion and attendance rates amongst the sample of participants who
provided dietary information; but the reasons for this are not clear.
These participants were different to the rest only in that they were
given the Shape Up starter survey by the tutors (the programme's own
baseline survey, distinct from the research project's baseline survey).
While this starter survey was intended for all Shape Up clients, orga-
nisational lapses meant that not everyone received a survey. With at-
tendance at the introductory session and differences in tutors already
controlled for, and the dietary variable itself not being statistically
significant, the most plausible remaining interpretation is that the
contract had a more positive effect when the programme elements were
administered most consistently. The potential interactions between the
programme elements and the commitment contract are discussed fur-
ther in Section 5.
Tables 2 and 3identify other variables correlated with participation:
older people are more likely to have higher attendance and complete
the programme, those who report higher levels of exercise at the
baseline are also more likely to complete. There are no significant
differences based on initial body mass index or gender. Self-referral
might be expected to capture initial motivation, but is not significantly
associated with participation rates. Myopic health attitudes at the
baseline are also not significantly associated with either measure of
participation.
Heterogeneous treatment effects are tested using the same three
model specifications (see Table 4), and the combined linear effect of the
treatment and interaction terms are reported below. The results provide
mixed support for hypothesis 2, and are sensitive to model specifica-
tion. In all three cases, the commitment contract improves attendance
amongst people who initially reported more myopic health attitudes,
but only in model 3 is this finding statistically significant (p= 0.001).
Models 1 and 2 suggest the size of the effect is in line with average
treatment effects reported in Table 2; however model 3 indicates a
larger, positive effect from the commitment contract on those in-
dividuals exhibiting myopic health attitudes. The mixed findings in-
dicate the need for further research on potential heterogeneous treat-
ment effects amongst those most likely to be affected by present bias.
5. Discussion
The commitment contract generates significant and positive parti-
cipation effects, enhancing adherence to the weight management pro-
gramme. There is limited evidence to confirm heterogeneous treatment
effects based on initial health attitudes, but the mixed results suggest
this is a promising line of enquiry in future research. The findings
prompt two further questions. The statistical analysis uncovered the
importance of attending the introductory class, which was argued to
have a potential commitment effect of its own. How did the commit-
ment contract interact with this administrative variable, and did it have
differential effects on participation amongst those who missed the in-
troductory class? Secondly, while it is valuable to improve adherence to
the weight loss programme, this is based on the assumption that ad-
herence promotes weight management goals. So, did the improved at-
tendance translate into greater weight loss?
5.1. The commitment contract as a substitute for other commitment
elements
Incorporating a binary variable for attendance at the initial Shape
Up class improves the explanatory fit of the statistical models for both
attendance and completion outcomes. In both cases, attendance at the
introductory class is positively associated with higher participation
throughout the weight management programme. As discussed in sec-
tion II, the first class was the opportunity for group members to meet,
get to know each other and the tutor, and the content of the class was
largely about motivating people to stick with the programme through to
the end for the best results. The tutor's introduction would have been
repeated in brief in the following weeks for newcomers, but more time
was given to social introductions and discussing personal motivations
for weight loss goals in the first class. Tutors would also have had more
time to share their own background and personal journeys. For all these
reasons, the introductory class was different to the rest, and this is
borne out in the data.
If the first class aimed to generate a sense of togetherness, perhaps
this was a form of commitment creation towards the tutor, the group,
and to the programme itself; albeit not in the form of a commitment
device as defined in this article. Participants who attended may have
experienced an initial sense of commitment nevertheless; conversely,
those who missed the meeting may have started the programme with a
slight disadvantage as far as commitment goes. To investigate further, I
examine whether the participation rates vary based on their having
attended the introductory class.
Descriptive statistics suggest there is an interaction between the
contract and attendance at the first class (Table 5). When a person
missed the introductory class, and the encouragement it delivered to
stick with the programme, they reported low attendance rates of just
over 50% - unless they were randomly assigned a contract. Having a
8
All statistical analysis was undertaken with Stata (v 12).
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
6

contract boosts participation up to 71% for those who missed the in-
troductory class (column 1). Regression analysis incorporating a binary
variable ‘missed initial class’ as an interaction term with treatment
status further indicates a statistically significant, combined linear effect
of the contract and interaction term of 18% (p= 0.018).
There are limits to how far commitment contracts can increase at-
tendance. Where a participant attended the class, and experienced the
initial efforts to create a sense of commitment to the programme, the
contract has little impact on their subsequent participation (column 2).
Amongst those who did not receive a contract, attending the in-
troductory class offsets any commitment gap and is associated with
higher attendance of 75%. But, when the individual has received a
contract, there is a marginal difference on their participation rates
based on whether they attended the introductory class or not.
The results raise the possibility that the contract is able to plug a
commitment gap (from not attending the initial class), but that it does
not continue to enhance participation amongst those who have at-
tended the initial class. In other words, the contract appears to work
effectively as a substitute for other commitment elements in the weight
loss programme, but not additively with these wider elements. These
findings are exploratory in nature and rely on small samples, so cannot
be used for reliable causal inference and extrapolation, but point to-
wards fruitful future research enquiry; they also highlights gaps in our
understanding of commitment devices, and points to the presence of
threshold effects and substitution effects that are not well formalized in
the available theoretical frameworks (Savani, 2018).
Table 2
Can a commitment contract boost attendance rates?.
(1) (2) (3)
Received a commitment contract 0.075* (0.038) 0.071 (0.037) 0.123
⁎⁎
(0.041)
Female 0.081 (0.052) 0.102* (0.051) 0.123 (0.068)
Age 0.004
⁎⁎
(0.001) 0.005
⁎⁎⁎
(0.001) 0.005
⁎⁎⁎
(0.001)
Overweight −0.044 (0.098) −0.030 (0.108) 0.014 (0.141)
Obese −0.079 (0.100) −0.067 (0.110) −0.012 (0.143)
Severely obese −0.062 (0.124) −0.041 (0.133) 0.011 (0.171)
Myopic health attitudes 0.016 (0.039) 0.027 (0.036) 0.011 (0.042)
Exercise 0.025 (0.014) 0.024 (0.014) 0.012 (0.016)
Prior weight loss programme −0.004 (0.045) −0.022 (0.041) −0.034 (0.045)
Recent life change 0.053 (0.038) 0.039 (0.037) 0.021 (0.043)
Takes part in other activities −0.065 (0.039) −0.064 (0.039) −0.065 (0.042)
Self referred to class 0.020 (0.044) 0.027 (0.041) 0.018 (0.042)
Group treatment intensity −0.141 (0.189) −0.182 (0.182) −0.180 (0.205)
Controls for tutor Y Y Y
Controls for wave Y Y Y
Attended introductory class 0.135
⁎⁎
(0.043) 0.127* (0.054)
Diet 0.007 (0.015)
Observations 192 192 136
Adjusted R
2
0.073 0.131 0.157
Group treatment intensity measures the proportion of group members who were randomly assigned a commitment contract. Controls applied for each of the
eight tutors and three waves of participant recruitment. All results are OLS estimates. Column 1 reports a parsimonious model, column 2 includes admin-
istrative variable for attending the introductory class, and column 3 adds baseline diet. Further robustness checks indicate model specification 2 run on the
sub-sample for column 3 generates similar results as column 3. Robust standard errors clustered at the level of the individual are in parentheses, * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p< 0.001.
Table 3
Can a commitment contract boost completion rates?.
(1) (2) (3)
Received a commitment contract 0.473* (0.220) 0.467* (0.222) 0.844
⁎⁎
(0.293)
Female 0.316 (0.295) 0.428 (0.301) 0.626 (0.400)
Age 0.015 (0.008) 0.017* (0.008) 0.031
⁎⁎
(0.010)
Overweight 0.153 (0.645) 0.328 (0.655) 1.226 (0.772)
Obese −0.254 (0.647) −0.095 (0.654) 0.721 (0.736)
Severely obese −0.519 (0.768) −0.371 (0.797) 0.305 (0.954)
Myopic health attitudes 0.171 (0.228) 0.228 (0.224) 0.379 (0.316)
Exercise 0.181* (0.089) 0.184* (0.093) 0.072 (0.107)
Prior weight loss programme −0.009 (0.255) −0.116 (0.256) −0.098 (0.325)
Recent life change 0.448 (0.245) 0.400 (0.250) 0.258 (0.325)
Takes part in other activities −0.403 (0.238) −0.422 (0.241) −0.593 (0.337)
Self referred to class −0.103 (0.257) −0.046 (0.263) 0.006 (0.338)
Group treatment intensity −0.206 (1.031) −0.298 (1.037) −0.131 (1.303)
Controls for tutor Y Y Y
Controls for wave Y Y Y
Attended introductory class . 0.689
⁎⁎
(0.230) 0.855
⁎⁎
(0.311)
Diet . . 0.120 (0.097)
Observations 192 192 136
Pseudo R
2
0.141 0.179 0.242
Statistical analysis as in Table 2, with Probit analysis on completion status. Column 1 reports a parsimonious model, column 2 includes the administrative
variable for attending the introductory class, and column 3 adds baseline diet. Further robustness checks indicate model specification 2 run on the sub-
sample for column 3 generates similar results as column 3. Robust standard errors clustered at the level of the individual are in parentheses, * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p< 0.001.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
7

5.2. Commitment, adherence, and weight outcomes
While the contract has demonstrable effects on boosting participa-
tion in the weight management programme, adherence is valued largely
as a stepping stone towards better health outcomes. Participation for
the full course of the programme matters because it is expected to
promote weight loss (Moroshko et al., 2011). The discussion below
briefly examines the distribution of weight outcomes across partici-
pants; and what evidence there is for a positive relationship between
attendance and weight loss.
The Shape Up programme routinely collects weight data. Given the
practice for tutors to treat attendance to week seven as the benchmark
to identify completers, final weight loss outcomes were taken from the
latest reading available during the final four weeks of the programme.
By this measure, 18% of observations lacked an outcome measure,
which is in line with other weight loss studies (Elobeid et al., 2009).
9
Attrition is higher in the comparison group than the contract group
(22% compared to 14%, p= 0.169). These attrition patterns create
challenges for inferring a causal effect of the contract on weight out-
comes (see Appendix Table A2). But the data does allow for descriptive
analysis to explore the potential relationships between commitment,
adherence and weight loss.
The comparison group reported average weight loss of 2.6% (of
initial weight) against the treatment group's 2.8%, and the difference is
not statistically significant (p= 0.677). The boxplot in Fig. 3 visualises
this finding. This is puzzling, as we would reasonably expect that more
commitment to the course would reflect in stronger weight manage-
ment results. One explanation is that the design of this particular
commitment device is strong enough to change tightly-specified beha-
viours (for example: ‘I must go to the class this evening’), but too mild
to bring about more complex changes that require multiple, com-
plementary and sustained actions (such as: ‘I must avoid high fat foods
for three months’).
Another explanation is that the contract is effective in promoting
adherence to the programme, but the programme is not effective in
delivering weight loss. This might be because the informational content
did not influence dietary and exercise behaviours outside the
Table 4
Treatment effects on attendance for individuals with myopic health attitudes.
(1) (2) (3)
Received a commitment contract 0.073 (0.054) 0.075 (0.051) 0.037 (0.054)
Contract x myopic health attitude 0.004 (0.072) −0.007 (0.070) 0.147 (0.077)
Female 0.081 (0.052) 0.102* (0.051) 0.127 (0.067)
Age 0.004
⁎⁎
(0.001) 0.005
⁎⁎⁎
(0.001) 0.006
⁎⁎⁎
(0.001)
Overweight −0.043 (0.098) −0.031 (0.110) 0.034 (0.127)
Obese −0.079 (0.100) −0.067 (0.111) 0.008 (0.129)
Severely obese −0.063 (0.124) −0.040 (0.134) 0.018 (0.157)
Myopic health attitudes 0.014 (0.055) 0.030 (0.050) −0.056 (0.060)
Exercise 0.025 (0.014) 0.024 (0.014) 0.011 (0.016)
Prior weight loss programme −0.004 (0.045) −0.022 (0.041) −0.043 (0.045)
Recently experienced life change 0.053 (0.039) 0.039 (0.038) 0.024 (0.043)
Takes part in other activities −0.065 (0.039) −0.064 (0.039) −0.055 (0.042)
Self referred to class 0.020 (0.044) 0.027 (0.041) 0.006 (0.040)
Group treatment intensity −0.141 (0.190) −0.182 (0.183) −0.144 (0.204)
Controls for tutor Y Y Y
Controls for wave Y Y Y
Attended introductory class – 0.135
⁎⁎
(0.043) 0.125* (0.053)
Diet – – 0.008 (0.015)
Linear combined effect of contract + interaction term 0.077 (0.051) 0.068 (0.051) 0.185
⁎⁎⁎
(0.057)
Observations 192 192 136
Adjusted R
2
0.067 0.126 0.174
Heterogeneous treatment effects drawn from OLS analysis. Column 1 reports a parsimonious model, column 2 includes the administrative variable for attending the
introductory class, and column 3 adds baseline diet. Further robustness checks indicate that model specification 2 run on the sub-sample for column 3 generates
similar results as column 3. Robust standard errors clustered at the level of the individual are reported in parentheses, * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p< 0.001.
Table 5
The commitment contract and the introductory Shape Up class.
Missed first class Attended first class
Mean attendance rates (1) (2)
No contract 53% (n= 35) 75% (n= 65)
Contract 71% (n= 31) 72% (n= 66)
Difference +18% −3%
Table 6
Is the commitment contract a substitute for other commitment elements?.
Received a commitment contract 0.010 (0.040)
Contract xmissed introductory Shape Up class 0.169* (0.086)
Female 0.104* (0.046)
Age 0.004
⁎⁎
(0.001)
Overweight −0.029 (0.095)
Obese −0.069 (0.097)
Severely obese −0.044 (0.118)
Myopic health attitudes 0.029 (0.035)
Exercise 0.022 (0.014)
Took part in a weight loss programme before −0.012 (0.041)
Recently experienced a major life change 0.046 (0.037)
Takes part in other activities to stay healthy −0.057 (0.039)
Self referred to class 0.022 (0.042)
Missed introductory Shape Up class −0.214
⁎⁎⁎
(0.058)
Group treatment intensity −0.168 (0.181)
Controls for tutor Y
Controls for wave Y
Linear combined effect of contract + interaction term 0.179* (0.075)
Observations 192
Adjusted R
2
0.153
Statistical analysis as in Table 2 column 1, with OLS regression on attendance
rates, incorporating binary variable on whether participant attended the in-
troductory Shape Up class. Robust standard errors clustered at the level of the
individual are in parentheses, * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p< 0.001.
9
Two alternative weight outcome measures were considered. Last observa-
tion or baseline observation carried forward measures were discarded due to
the likely bias they would introduce, and the unjustified assumptions they
would impose on weight loss performance in the absence of the Shape Up
course. Secondly, weight loss outcomes from a narrower window in the final
two weeks of the programme was considered, but attrition was significantly
associated with this outcome measure (p = 0.012) as reported in Appendix
Table A2.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
8

classroom. Fig. 4 offers some evidence against this explanation, as it
suggests a modest but positive association between attendance and
reported weight outcomes. Moral license effects might also affect
weight performance overall: if by attending the class participants felt
they had earned “moral credits” (Mullen & Monin, 2016, p. 367),
whereby “engaging in positive behaviour may increase the likelihood of
engaging in indulgent behaviour” (Chang & Chiou, 2014, p. 9), the net
effect on health goals might have been undermined. Further statistical
evidence is not available to assess these explanations, but these are
promising questions for future research.
Fig. 3 raises two other important points. Firstly, it demonstrates the
diverse range of weight loss outcomes across the participant sample:
this is explicit evidence of heterogeneous experiences in weight man-
agement. Secondly, the contrast between the two box plots highlights a
distinct likelihood of truncated outcome data, with fewer observations
of weight gain in the comparison group. Plausibly, the commitment
contract encourages people to return to the class week after week, and
greater adherence therefore leads to a fuller spectrum of performance
data at both ends. Administrative data is then better able to capture the
true range of participant experiences, and this is why the intervention
group boxplot reflects that some people lost a lot of weight, some lost a
little, and some actually gained weight while on the programme (to the
Fig. 3. Weight loss outcomes across experimental groups.
Notes: plot demonstrates median weight loss as a% of initial body weight, with dots representing outliers and whiskers representing 95% CI.
Fig. 4. Attendance and weight management outcomes.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
9

left of the ‘zero’ reference line in Fig. 3). Participants in the treatment
group returned to class, even when they were under-performing, and
had their weight readings recorded; equally, they may have been more
likely to return when performing well in order to have their progress
recorded. Without the contract, however, comparison group partici-
pants may have been more likely to drop out of the programme alto-
gether if they were under-performing against their weight loss target
(hence the shorter left hand side whisker for the comparison group).
Fig. 3 appears, then, to provide graphical evidence of the attrition bias
that often affects studies measuring weight loss outcomes (Moroshko
et al., 2011; Paloyo et al., 2014).
6. Conclusions
Commitment devices are recognised as potentially important tools
to promote health behaviour. Yet personal commitment contracts,
which rely solely on self-reputational costs, are under-researched.
Theory predicts that even modest commitment devices can generate
enough psychological tax to bring about behaviour change. A field
experiment with participants in a group-based community weight
management scheme tested this proposition by randomly assigning
personal commitment contracts to participants. A limitation of the
study is attrition on weight loss data, which precluded further analysis
of the causal impact of the contract on health outcomes; and missing
baseline data on initial dietary behaviours which was associated with
some results being sensitive to model specification. Strengths of the
study include the use of objective data to measure participation using
administrative records, avoiding self-reported adherence; and the novel
use of health attitudes data to proxy myopia in health decisions, and
investigate the causal mechanism linking commitment contracts with
behaviour change.
Attendance and completion rates were significantly higher amongst
those who received a simple commitment contract, designed to act as a
signed pledge to oneself to pursue their weight loss goals and complete
the 11-week weight loss intervention. In line with the Thaler and
Shefrin's (1981) planner-doer framework, and the arguments put for-
ward by Benabou and Tirole (2004), these results indicate that a per-
sonal commitment contract was sufficient to promote adherence to an
ongoing health management programme. More limited evidence in-
dicated these positive effects might be stronger for participants ex-
pressing more short-termist health attitudes at the start of the study; but
these sub-group effects are best interpreted as suggestive.
This paper contributes to the literature on commitment devices and
health behaviour change by drawing out novel experimental insights on
a relatively under-researched type of commitment device, and high-
lighting the limitations of theory to understand potential threshold and
substitution effects of commitment devices. Mixed findings on the po-
tential heterogeneous effects of commitment devices calls for further
work to test this fundamental premise of the planner-doer framework –
that commitment devices tackle present bias and myopic decision-
making to deliver health benefits. Future research and theory devel-
opment could usefully address these issues, and identify how diverse
commitment strategies and inducements might be combined to deliver
greater health impact. The results also contribute to the health psy-
chology and public health literatures on adherence and weight man-
agement.
Positive treatment effects on attendance and completion rates jus-
tify a role for commitment devices in improving participation in public
health programmes. The study demonstrates that personal commitment
contracts are low-cost and easily administered to boost engagement
with public health services; and further identifies a target group of
individuals with relatively myopic health attitudes who may especially
benefit from commitment devices that align short term actions with
longer term health goals. These findings are therefore of interest to
policy makers who may seek to add commitment contracts to the be-
haviour change toolkit. Further research could investigate the optimal
design of commitment contracts that are tailored to individual needs,
and qualitative follow up could shed more light on how commitment
contracts are utilised and valued. Future studies could also consider
other biometric data, more intensive follow-up on weight outcomes to
minimise missing data, and behaviour and attitude indicators around
diet and physical activity. This would help develop a comprehensive
picture of how a commitment device can trigger behaviour change, and
the potential for wider and unintended effects such as moral license.
Table A1
Recruitment schedule.
Wave Duration Participants Share Groups Tutors involved
1 January–March 2014 57 29% 1–9 1–5
2 April–August 2014 61 31% 10–19 1–6, 8
3 Sept 2015–March 2016 79 40% 20–27 1, 2, 5, 7, 8
Appendix (Tables A1 to A2)
Table A1 to A2
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
10

References
Andreoni, J., Kuhn, M.A., Sprenger, C., 2015. Measuring time preferences : A comparison
of experimental methods. Journal of Economic Behavior and Organization 116,
451–464. https://doi.org/10.1016/j.jebo.2015.05.018.
Ariely, D., Wertenbroch, K., 2002. Procrastination, deadlines, and performance: Self-
control by precommitment. Psychological Science : A Journal of the American
Psychological Society/APS 13 (2000), 219–224. https://doi.org/10.1111/1467-
9280.00441.
Au, N., Marsden, G., Mortimer, D., Lorgelly, P.K., 2013. The cost-effectiveness of shopping
to a predetermined grocery list to reduce overweight and obesity. Nutrition Diabetes
3 (6), 1–5. https://doi.org/10.1038/nutd.2013.18.
Benabou, R., Tirole, J., 2004. Willpower and personal rules. Journal of Political Economy
112 (4), 848–886.
Blane, D.N., McLoone, P., Morrison, D., Macdonald, S., O'Donnell, C.A., 2017. Patient and
practice characteristics predicting attendance and completion at a specialist weight
management service in the UK: A cross-sectional study. BMJ Open 7, 1–10. https://
doi.org/10.1136/bmjopen-2017-018286.
Brocas, I., Carillo, J.D., Dewatripont, M., 2003. Commitment devices under self-control
problems: an overview. In: Brocas, I., Carillo, J.D. (Eds.), The psychology of economic
decisions vol 2. Oxford University Press, Oxford, pp. 231.
Brown, A., Gouldstone, A., Fox, E., Field, A., Todd, W., Shakher, J., Bellary, S., The, M.M.,
Azam, M., John, R., Jagielski, A., Arora, T., Thomas, G.N., Taheri, S., 2015.
Description and preliminary results from a structured specialist behavioural weight
management group intervention: Specialist Lifestyle Management (SLiM) pro-
gramme. BMJ Open 5, 1–11. https://doi.org/10.1136/bmjopen-2014-007217.
Bryan, G., Karlan, D., Nelson, S., 2010. Commitment devices. Annual Review of
Economics 2, 671–698. https://doi.org/10.1146/annurev.economics.102308.
124324.
Butryn, M.L., Phelan, S., Hill, J.O., Wing, R.R., 2007. Consistent self-monitoring of
weight: A key component of successful weight loss maintenance. Obesity 15 (12),
3091–3096. https://doi.org/10.1002/oby.20584/epdf.
Cavaliere, A., De Marchi, E., Banterle, A., 2014. Healthy – unhealthy weight and time
preference. Is there an association? An analysis through a consumer survey. Appetite
83, 135–143. https://doi.org/10.1016/j.appet.2014.08.011.
Chang, Y.Y.C., Chiou, W.Bin, 2014. Taking weight-loss supplements may elicit liberation
from dietary control. A laboratory experiment. Appetite 72, 8–12. https://doi.org/10.
1016/j.appet.2013.09.021.
Costa-Font, J., Hernandez-Quevedo, C., McDonald, J.T., Variyam, J.N., 2013.
Understanding healthy lifestyles: The role of choice and the environment. Applied
Economic Perspectives and Policyiedomicives and 35 (1), 1–6. https://doi.org/10.
1093/aepp/pps051.
Department of Health, 2011. Healthy lives, healthy people: A call to action on obesity in
England. UK Department of Health.
Downs, J.S., Loewenstein, G., Wisdom, J., 2009. Strategies for promoting healthier food
choices. American Economic Review 99 (2), 159–164. https://doi.org/10.1257/aer.
99.2.159.
Dupas, P., 2011. Health behavior in developing countries. Annual Review of Economics 3
(1), 425–449. https://doi.org/10.1146/annurev-economics-111809-125029.
Elobeid, M.A., Padilla, M.A., McVie, T., Thomas, O., Brock, D.W., Musser, B., Lu, K.,
Coffey, C.S., Desmond, R.A., St-Onge, M.-P., Gadde, K.M., Heymsfield, S.B., Allison,
D.B., 2009. Missing data in randomized clinical trials for weight loss : Scope of the
problem, state of the field, and performance of statistical methods. PLoS ONE 4 (8),
1–11. https://doi.org/10.1371/journal.pone.0006624.
Fan, M., Jin, Y., 2013. Obesity and self-control: Food consumption, physical activity, and
weight-loss intention. Applied Economic Perspectives and Policy 36 (1), 125–145.
https://doi.org/10.1093/aepp/ppt034.
Frederick, S., Loewenstein, G., O'Donoghue, T., 2002. Time discounting and time pre-
ference: A critical review. Journal of Economic Literature 40 (2), 351–401. https://
doi.org/10.1257/002205102320161311.
Gine, X., Karlan, D., Zinman, J., 2010. Put your money where your butt is: A commitment
contract for smoking cessation. American Economic Journal: Applied Economics 2
(October), 213–235.
Glennerster, R., Takavarasha, K., 2013. Running randomized evaluations: A practical
guide. Princeton University Press.
Halpern, S.D., French, B., Small, D.S., Saulsgiver, K., Harhay, M.O., Audrain-McGovern,
J., Loewenstein, G., Brennan, T.A., Asch, D.A., Volpp, K.G., 2015. Randomized trial of
four financial-incentive programs for smoking cessation. New England Journal of
Medicine 372 (22), 2108–2117. https://doi.org/10.1056/NEJMoa1414293.
Health Social Care Information Centre, 2015. HSE 2014 obesity and overweight adult
trends. HSCIC, London.
John, L.K., Loewenstein, G., Troxel, A.B., Norton, L., Fassbender, J.E., Volpp, K.G., 2011.
Financial incentives for extended weight loss: A randomized, controlled trial. Journal
of General Internal Medicine 26 (6), 621–626. https://doi.org/10.1007/s11606-010-
1628-y.
Kiesler, C.A., Sakumura, J., 1966. A test of a model for commitment. Journal of
Table A2
Factors associated with missing weight loss outcomes.
Attrition at week
7
Attrition at week 9
Received a commitment contract −0.420 (0.242) −0.810
⁎⁎
(0.247)
Female −0.146 (0.336) 0.194 (0.300)
Age −0.008 (0.009) −0.015* (0.008)
Overweight −0.643 (0.716) 0.053 (0.610)
Obese −0.452 (0.731) 0.406 (0.622)
Severely obese −1.002 (0.928) −1.686 (1.004)
Myopic health attitudes −0.089 (0.243) 0.103 (0.246)
Exercise −0.185 (0.100) −0.137 (0.084)
Recently experienced a major life
change
−0.329 (0.255) −0.306 (0.247)
Took part in a weight loss programme
before
−0.084 (0.276) 0.465 (0.242)
Takes part in other activities to stay
healthy
0.228 (0.262) 0.096 (0.242)
Attended introductory class −0.793
⁎⁎
(0.243) −0.569* (0.248)
Referred by GP −0.111 (0.278) −0.006 (0.265)
Referred by other health professional −0.231 (0.317) −0.537 (0.291)
Referred by other 1.447 (0.822) 2.460
⁎⁎
(0.860)
Group treatment intensity 0.271 (1.104) 0.225 (1.060)
Group during daytime 0.026 (0.398) 0.083 (0.383)
Group size 0.053 (0.050) −0.028 (0.044)
Tutor 2 −0.398 (0.462) −0.774 (0.433)
Tutor 3 −0.631 (0.434) −1.501
⁎⁎⁎
(0.423)
Tutor 4 −0.745 (0.666) −1.952
⁎⁎
(0.701)
Tutor 5 −0.723 (0.530) −1.554
⁎⁎
(0.477)
Tutor 6 0.701 (0.813) 0.496 (0.935)
Tutor 7 −0.113 (0.742) −1.163 (0.631)
Tutor 8 −0.296 (0.473) −0.946* (0.397)
Recruited to trial in week 2 0.034 (0.470) 0.070 (0.495)
Recruited to trial in week 3 0.242 (0.549) 0.103 (0.523)
Participant in wave 2 −0.167 (0.369) 0.165 (0.395)
Participant in wave 3 −0.204 (0.509) 0.669 (0.455)
Observations 192 192
Pseudo R
2
0.203 0.271
Probit estimates on attrition. Robust standard errors clustered at the level of the individual in parentheses, * p< 0.05,
⁎⁎
p< 0.01,
⁎⁎⁎
p< 0.001.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
11

Personality and Social Psychology 3 (3), 349–353.
Kunreuther, H., Meyer, R., Michel-Kerjan, E., 2013. Overcoming decision biases to reduce
losses from natural catastrophes. In: Shafir, E. (Ed.), The behavioral foundations of
public policy. Princeton University Press, Oxford, pp. 398–413.
Liu, P.J., Wisdom, J., Roberto, C.A., Liu, L.J., Ubel, P.A., 2014. Using behavioral eco-
nomics to design more effective food policies to address obesity. Applied Economic
Perspectives and Policy 36 (1), 6–24. https://doi.org/10.1093/aepp/ppt027.
Loewenstein, G., Asch, D.A., Friedman, J.Y., Melichar, L.A., Volpp, K.G., 2012. Can be-
havioural economics make us healthier? BMJ 344 (May), e3482. https://doi.org/10.
1136/bmj.e3482.
Logue, J., Allardice, G., Gillies, M., Forde, L., Morrison, D.S., 2014. Outcomes of a spe-
cialist weight management programme in the UK national health service: Prospective
study of 1838 patients. BMJ Open 4, 1–8. https://doi.org/10.1136/bmjopen-2013-
003747.
Milkman, K.L., Minson, J.A., Volpp, K.G.M., 2014. Holding the hunger games hostage at
the gym: An evaluation of temptation bundling. Management Science 60 (2),
283–299.
Miller, D.T., Prentice, D.A., 2013. Psychological levers of behaviour change. In: Shafir, E.
(Ed.), The Behavioral Foundations of Public Policy. Princeton University Press, pp.
301–309.
Mochon, D., Schwartz, J., Maroba, J., Patel, D., Ariely, D., 2016. Gain without pain: The
extended effects of a behavioral health intervention. Management Science 63 (1),
58–72. https://doi.org/10.1287/mnsc.2015.2322.
Moroshko, I., Brennan, L., & Brien, P.O. (.2011). ‘Predictors of dropout in weight loss
interventions : A systematic review of the literature’, (2), pp. 912–934. doi: 10.1111/
j.1467-789X.2011.00915.x.
Mullen, E., Monin, B., 2016. Consistency versus licensing effects of past moral behavior.
Annual Review of Psychology 67, 363–385. https://doi.org/10.1146/annurev-psych-
010213-115120.
NHS Digital, 2018. Health survey for england 2017. the Health and Social Care
Information Centre.
Nyer, P.U., Dellande, S., 2010. Public commitment as a motivator for weight loss.
Psychology & Marketing 27 (1), 1–12.
O’ Donoghue, T., Rabin, M., 1999. Doing it now or later. American Economic Review 89
(1), 103–124. https://doi.org/10.1257/aer.89.1.103.
Oliver, A., Ubel, P., 2014. Nudging the obese: A UK-US consideration. Health Economics,
Policy, and Law (April), 1–14. https://doi.org/10.1017/S1744133114000103.
Paloyo, A.R., Reichert, A.R., Reinermann, H., Tauchmann, H., 2014. The causal link be-
tween financial incentives and weight loss: An evidence-based survey of the litera-
ture. Journal of Economic Surveys 28 (3), 401–420. https://doi.org/10.1111/joes.
12010.
Perry, C., Chhatralia, K., Damesick, D., Hobden, S., Volpe, L. (2015).Behavioural insights in
health care: Nudging to reduce inefficiency and waste. The Health Foundation, London.
http://www.health.org.uk/sites/health/files/BehaviouralInsightsInHealthCare.pdf.
Peterson, N.D., Middleton, K.R., Nackers, L.M., Medina, K.E., Milsom, V.A., Perri, M.G.,
2014. Dietary self-monitoring and long-term success with weight management.
Obesity (Silver Spring) 22 (9), 1962–1967. https://doi.org/10.1002/oby.20807.
Relton, C., Strong, M., Li, J., 2011. The “ pounds for pounds ” weight loss financial in-
centive scheme : An evaluation of a pilot in NHS Eastern and Coastal Kent. Journal of
Public Health (Oxford) 33 (4), 536–542. https://doi.org/10.1093/pubmed/fdr030.
Robinson, C., 2012. Healthy foundations segmentation. In: Health survey for England
2011. HSCIC, pp. 1–38.
Rogers, T., Milkman, K.L., Volpp, K.G., 2015. Beyond good intentions: Prompting people
to make plans improves follow-through on important tasks. Behavioral Science and
Policy 1 (2), 33–41.
Rogers, T., Milkman, K.L., Volpp, K.G., 2014. Commitment devices: Using initiatives to
change behavior. JAMA : The Journal of the American Medical Association 311 (20),
2065–2066. https://doi.org/10.1001/jama.2014.3485.
Royer, H., Stehr, M., Sydnor, J., 2015. Incentives, commitments and habit formation in
exercise: Evidence from a field experiment with workers at a fortune-500 company.
American Economic Journal: Applied Economics 7 (3), 51–84. https://doi.org/10.
1257/app.20130327.
Savani, M.M., 2018. The effects of a commitment device on health outcomes:
Reputational commitment and weight loss in an online experiment. International
Journal of Applied Behavioral Economics 7 (4), 1–20. https://doi.org/10.4018/
IJABE.2018100101.
Schulz, K.F., Altman, D.G., Moher, D., 2010. CONSORT 2010 statement: Updated
guidelines for reporting parallel group randomized trials. Open Medicine 4 (1),
E60–E68.
Schwartz, J., Mochon, D., Wyper, L., Maroba, J., Patel, D., Ariely, D., 2014. Healthier by
precommitment. Psychological Science 25 (2), 538–546. https://doi.org/10.1177/
0956797613510950.
Strotz, R.H., 1955. Myopia and inconsistency in dynamic utility maximization. The
Review of Economic Studies 165–180. https://doi.org/10.2307/2295722.
Thaler, R.H., Benartzi, S., 2004. Save more tomorrow™: Using behavioral economics to
increase employee saving. Journal of Political Economy 112 (S1), S164–S187.
https://doi.org/10.1086/380085.
Thaler, R.H., Shefrin, H.M., 1981. An economic theory of self-control. Journal of Political
Economy 89 (2), 392–406.
Thaler, R.H., Sunstein, C.R., 2008. Nudge: Improving decision about health, wealth and
happiness. Penguin Books, London.
Volpp, K.G., John, L.K., Troxel, A.B., Norton, L., Fassbender, J., Loewenstein, G., 2008.
Financial incentive–based approaches for weight loss. JAMA: The Journal of the
American Medical Association 300 (22), 2631–2637. https://doi.org/10.1001/jama.
2008.804.
Williams, B., McVey, D., Davies, L., MacGregor, E. (2011).‘The healthy foundations life-
stage segmentation: Research report no.1: Creating the segmentation using a quan-
titative survey of the general population of England.’, (1), pp. 1–312.
M. Manthri Savani Journal of Behavioral and Experimental Economics 82 (2019) 101457
12