Available via license: CC BY 4.0
Content may be subject to copyright.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
Please cite this article as: Erika L. Kirgios, Organizational Behavior and Human Decision Processes, https://doi.org/10.1016/j.obhdp.2020.09.003
0749-5978/© 2020 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Teaching temptation bundling to boost exercise: A eld experiment
☆
Erika L. Kirgios
a
,
*
, Graelin H. Mandel
b
, Yeji Park
c
, Katherine L. Milkman
a
, Dena M. Gromet
c
,
Joseph S. Kay
c
, Angela L. Duckworth
a
,
d
a
Department of Operations, Information, & Decisions, The Wharton School, University of Pennsylvania, Philadelphia, PA 19104, United States
b
The Wharton School, University of Pennsylvania, Philadelphia, PA 19104, United States
c
Behavior Change for Good Initiative, University of Pennsylvania, Philadelphia, PA 19104, United States
d
Department of Psychology, University of Pennsylvania, Philadelphia, PA 19104, United States
ARTICLE INFO
Keywords:
Behavior change
Temptation bundling
Want-should conicts
Self-control
ABSTRACT
Temptation bundling—pairing a pleasurable indulgence with a behavior that provides delayed
rewards—combats present bias by making behaviors with delayed benets more instantly-gratifying. If people
are sophisticated and capable of following self-set rules to overcome present bias, they could benet from
learning about temptation bundling. Participants in a four-week exercise-boosting program (N =6792) received
either an audiobook with encouragement to temptation bundle, only an audiobook, or neither an audiobook nor
encouragement to temptation bundle. Giving participants audiobooks and encouraging temptation bundling
boosted their likelihood of a weekly workout by 10–14% and average weekly workouts by 10–12% during and up
to seventeen weeks post-intervention. Relative to giving audiobooks alone, encouraging temptation bundling had
a modest positive effect on exercise on the extensive margin. The marginal benet of encouraging temptation
bundling may be small because free audiobooks leak information: Simply providing an audiobook to exercise
program participants suggests they should temptation bundle.
1. Introduction
A dieter is offered chocolate cake at a party. Someone trying to save
more money stares longingly at a fancy new cellphone. A student trying
to improve her grades considers skipping her studies to watch a binge-
worthy TV show instead. These are all examples of want-should con-
icts (Bazerman, Tenbrunsel, & Wade-Benzoni, 1998; Milkman et al.,
2008), in which people are forced to choose between doing what they
want to get instant gratication and what they should to gain long-term
benets. Wants are dened as actions that provide short-term pleasure
but no long-term benets, while shoulds provide long-term benets but
often involve short-term pain. Due to present bias, dened as the ten-
dency to dramatically overweight immediate rewards relative to
delayed rewards, people often choose wants over shoulds in the heat of
the moment, only to later regret their decisions (Milkman et al., 2008).
Unfortunately, evidence shows that over time, repeatedly selecting
wants over shoulds can take a toll on people’s health as well as their
nancial, academic, and professional outcomes (Boals, Vandellen, &
Banks, 2011; Kaur, Kremer, & Mullainathan, 2010; Stavrova & Kokkoris,
2017; Str¨
omb¨
ack, Lind, Skagerlund, V¨
astfj¨
all, & Tingh¨
og, 2017). For
instance, an estimated 40% of premature deaths can be attributed to
repeated decisions that favor wants over shoulds when it comes to eating,
drinking, smoking, exercise, sex, and vehicle safety (Schroeder, 2007).
In light of the large social and personal costs of consistently choosing
wants in the face of want-should conicts, policymakers and behavioral
scientists have tested a wide range of interventions designed to
encourage people to overcome their present bias and make more should
choices (see Duckworth, Milkman, & Laibson, 2018, for a review). Many
such interventions rely on people’s “sophistication” or self-awareness
(O’Donoghue & Rabin, 2011), leaning on their ability to anticipate
future temptations and their willingness to adopt strategies that help
them resist those temptations. Such interventions have encouraged
☆
This article is an invited submission. It is part of a supplemental issue on “Healthy Habits” edited by Katherine L. Milkman, Dilip Soman, and Kevin G. Volpp and
supported by WW. This supplemental issue collects papers by participants in the Roundtable Discussion on Creating Habit Formation for Healthy Behaviors,
organized in late 2019 by the Wharton-Penn Behavior Change for Good Initiative (BCFG) and the Penn Center for Health Incentives and Behavioral Economics
(CHIBE).
* Corresponding author at: 3730 Walnut St, 527.4 Jon M. Huntsman Hall, Philadelphia, PA 19104, United States.
E-mail address: ekirgios@wharton.upenn.edu (E.L. Kirgios).
Contents lists available at ScienceDirect
Organizational Behavior and Human Decision Processes
journal homepage: www.elsevier.com/locate/obhdp
https://doi.org/10.1016/j.obhdp.2020.09.003
Received 28 February 2020; Received in revised form 22 September 2020; Accepted 23 September 2020
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
2
people to pre-commit to should choices such as saving money (Thaler &
Benartzi, 2004), to self-impose penalties if they fail to engage in should
activities such as exercise (Royer, Stehr, & Sydnor, 2015), or to avoid
situations that might make it particularly tempting to choose want ac-
tivities such as screen time over should activities such as studying
(Duckworth, White, Matteucci, Shearer, & Gross, 2016).
These kinds of interventions prioritize helping people engage in
should activities at the expense of wants. But choosing wants versus
shoulds need not always be mutually exclusive.
In the present research, we test the effectiveness of harnessing the
appeal of wants to increase should choices. We teach participants to pair
wants with shoulds to increase their follow-through on should behaviors,
and we examine whether merely providing them with wants in the
company of shoulds is enough to suggest such pairings. First, we examine
whether encouraging people to pair indulgent entertainment with ex-
ercise can durably increase gym attendance using a technique called
“temptation bundling.” In doing so, we shed new light on the degree to
which people can effectively follow self-set, personal rules to overcome
present bias (Ainslie, 2001; Benabou & Tirole, 2004; Grady, Goodenow,
& Borkin, 1988; Thaler, 2000; Wertenbroch, 1998). Second, we explore
whether giving people free entertainment as part of a program to
encourage exercise can implicitly suggest temptation bundling and
therefore boost exercise. Consequently, we illuminate people’s capacity
to make complex inferences from information policymakers leak (Tan-
nenbaum, Valesek, Knowles, & Ditto, 2013; McKenzie, Liersch, & Fin-
kelstein, 2006; McKenzie & Nelson, 2003). In both cases, we test the
extent of individuals’ sophistication in overcoming their present bias,
adding to a growing literature that suggests people are often aware of
present bias and capable of taking steps to reduce its harms (O’Donog-
hue and Rabin, 2011).
As conceived by Milkman, Minson, and Volpp (2014), temptation
bundling pairs wants (e.g., getting a pedicure, watching lowbrow TV,
eating chocolate) with shoulds (e.g., paying taxes, exercising, getting an
eye exam). This pairing makes should activities more enticing and
therefore more likely to be readily executed; it also makes want activities
less wasteful and guilt-inducing. In our experiments, exercise constitutes
the should activity and consuming lowbrow entertainment (in the form
of television, podcasts, and audiobooks) constitutes the want activity.
1
In the single, small (N =226) prior experiment exploring the benets
of temptation bundling, Milkman et al. (2014) found initial evidence for
the short-term effectiveness of this behavior change strategy. However,
because the literature on temptation bundling includes just this small
eld study, a great deal remains unknown about the mechanisms
responsible for its impact, whether it can have durable benets, its
general robustness, and whether people can self-generate the idea to
temptation bundle given the right stimuli, even if they aren’t explicitly
taught how it works.
This paper presents a eld experiment roughly 15 times larger than
the prior study, which tests the benets of merely encouraging temp-
tation bundling rather than externally imposing temptation bundling.
With a seventeen-week follow-up period, we provide the rst evidence
that encouraging temptation bundling and providing a bundle-worthy
want has durable benets, signicantly boosting both the likelihood
and frequency of weekly gym visits for months. Separately, we nd a
small benet of merely encouraging temptation bundling, but it accrues
solely on the extensive margin rather than the intensive margin. Spe-
cically, it leads marginally more people to visit the gym in a given week
during our intervention and it boosts the likelihood of a gym visit
signicantly during our 10 and 17-week post-intervention follow-up
periods. However, encouragement to temptation bundle alone does not
robustly cause people to make more gym visits per week. Our eld
experiment and two follow-up experiments also suggest that people are
sophisticated enough to infer they should temptation bundle from pro-
gramming that “leaks” the idea—specically, an exercise program that
provides the gift of a tempting want in the form of an audiobook.
1.1. Can people self-impose constraints to overcome their present bias?
Research shows that some people are well aware they’re likely to
procrastinate on should activities that can be painful in the present but
benecial in the future in favor of want activities that are pleasurable
now but carry few, if any, long-term benets. Those who recognize their
self-control struggles and are interested in nding ways to prevent self-
control failures have been dubbed “sophisticates” (O’Donoghue &
Rabin, 1999; O’Donoghue and Rabin, 2011). Past research on sophisti-
cates has largely focused on documenting their taste for commitment
devices—technologies that help them avoid future temptations, such as
smaller plates, bank accounts that are inaccessible until a pre-
determined future date, or nes that arise when they fail to achieve
their goals (Ariely & Wertenbroch, 2002; Ashraf, Karlan, & Yin, 2006;
Bryan, Karlan, & Nelson, 2010; Royer, Stehr, & Sydnor, 2015; Laibson,
1997; Wansink & Cheney, 2005). Anticipating that overcoming present
bias will be difcult, sophisticates may also be on the lookout for
behavioral tricks or techniques they can try on their own.
Indeed, a growing literature suggests some sophisticates can suc-
cessfully self-impose behavior change strategies. Teaching people new
skills or practical tricks can empower them to make better choices on
their own (Drexler, Fischer, & Schoar, 2014; Sedlmeier & Gigerenzer,
2001; for a review of “boosts,” see Hertwig & Grüne-Yanoff, 2017).
Evidence suggests people can also craft and successfully follow personal
guidelines that help them exert self-control (Ainslie, 1992; Ainslie,
2001; Benabou & Tirole, 2004; Cheema & Soman, 2006; Grady et al.,
1988; Thaler, 2000; Wertenbroch, 1998). For instance, people can self-
impose constraints on purchases of vice items such as cigarettes (Wer-
tenbroch, 1998), assign activities to categories and restrict spending to
implicit category-level budgets on their own initiative (Thaler, 2000),
and reward themselves for the successful completion of should activities
to encourage their own goal-oriented behaviors (Grady et al., 1988).
These ndings indicate that individuals can be sophisticated and
agentic in their pursuit of behavior change and thus might be able and
eager to self-impose useful new rules that could help them overcome
present bias.
1.2. Temptation bundling
Temptation bundling is one form of personal rule that past research
suggests may appeal to sophisticates hoping to reign in want behaviors
and pursue more should behaviors. Sophisticates may like the idea of
only allowing themselves to enjoy wants, like lowbrow audio-novels or
other forms of entertainment, while simultaneously engaging in a should
behavior, like exercise. One small prior study showed that when access
to a want (a tempting audio novel) was denied to people unless they
engaged in a should behavior (visiting the gym), they exercised more
often (Milkman et al., 2014). In this same study, a survey determined
that the majority of participants would be interested in paying for gym-
only access to wants because they thought this restriction might help
them boost their physical activity.
Although it was not a direct test of temptation bundling (because
wants could be accessed anytime, not only in the presence of shoulds), a
separate eld experiment found that encouraging high school students
to enjoy snacks, music, and colorful markers while completing chal-
lenging math problems led to greater persistence than encouraging this
should behavior without any wants incorporated (Woolley & Fishbach,
2016). Gamication interventions that add features such as points and
leaderboards to the pursuit of should goals in order to make them more
1
Lowbrow entertainment is instantly gratifying but generally lacks long-term
benets (see Milkman, Rogers and Bazerman, 2009). Meanwhile, exercise has
long-term benets but lacks immediate appeal: Exercising more was the most
common New Year’s resolution in 2020 (Ballard, 2020), but people frequently
fail to follow through on this health goal (DellaVigna & Malmendier, 2006).
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
3
immediately enjoyable (and more like wants) have also successfully
increased daily step counts (Patel et al., 2017; Patel et al., 2019; Zuck-
erman & Gal-Oz, 2014). Taken together, these studies suggest that
combining wants with shoulds can promote should behaviors.
These studies all included third parties who bundled wants with
shoulds for participants. We focus on whether it is useful to merely
suggest that people self-impose such combinations. Given that past
research shows that 1) combining wants with shoulds can make shoulds
more appealing and that 2) sophisticates can self-impose constraints to
overcome present bias, there is reason to believe suggesting temptation
bundling might be valuable. However, the only evidence testing this
claim comes from an under-powered treatment in the original tempta-
tion bundling experiment conducted by Milkman et al. in 2014. In this
study, some gym-goers (N =75) received four tempting audiobooks of
their choice (e.g., The Hunger Games, The Da Vinci Code) and were
encouraged to listen to the audiobooks only while exercising at the gym.
Compared to participants in a control condition (N =76) who received
an equally valued gift card to Barnes & Noble and no instructions about
temptation bundling, those encouraged to temptation bundle initially
visited the gym marginally more often. However, these marginal bene-
ts quickly fell below detectable levels. Further, the study’s design
makes it impossible to disentangle the benets of giving people audio-
books from the benets of teaching them about temptation bundling.
In this paper, we test the hypothesis that merely teaching people
about temptation bundling may have durable benets. We also run the
rst well-powered test of the hypothesis that both teaching people about
temptation bundling and providing them with free access to a want that
can be used to temptation bundle has benets. Formally, we hypothesize
the following:
Hypothesis 1. People introduced to the idea of temptation bundling
will engage in more of the should behavior they are seeking to boost (in
this case, exercise).
Hypothesis 2. People introduced to the idea of temptation bundling
and provided with a want (in this case, a free audiobook) to bundle with
the should behavior they are seeking to boost (in this case, exercise) will
engage in more of that should behavior.
1.3. Information leakage
We also investigate whether merely offering people access to bundle-
ready wants when they are aiming to engage in more of a should behavior
can facilitate behavior change by indirectly suggesting the idea of
temptation bundling.
Past research on “information leakage” has shown that people are
often able to correctly infer intentions of policymakers based on context
clues, such that the context “leaks” information that inuences people’s
judgments and subsequent decisions and behaviors. These context clues
may include physical features of the environment such as objects or
images that prime people to think competitively (Kay, Wheeler, Bargh,
& Ross, 2004; Prinsen, de Ridder, & de Vet, 2013); policy structures such
as punitive or rewarding health incentives (Tannenbaum et al., 2013); or
choice or information frames such as which options are listed as the
default (Hilton, 1995; McKenzie, Liersch, & Finkelstein, 2006; McKenzie
& Nelson, 2003; Sher & McKenzie, 2006; Strack, Martin, & Schwarz,
1988). For example, when a default option is present in a choice set,
individuals often infer that policymakers are implicitly recommending
the default option (otherwise, why would it be the “standard” choice?)
and as a result, they are more likely to stick to the default even when
given the opportunity to opt out (McKenzie et al., 2006). Individuals also
(correctly) infer the presence of negative attitudes towards those who
are overweight in an organization based on whether its workplace
wellness programs punish employees for being overweight or, instead,
reward them for being at a weight that the programs deem healthy
(Tannenbaum et al., 2013).
We examine whether information leakage can inspire people to
temptation bundle, such that individuals infer and apply a behavior
change tactic from context clues even with no explicit direction or in-
struction (Ainslie, 1992; Ainslie, 2001; Thaler, 2000; Wertenbroch,
1998). Specically, we propose that simply giving participants a free
audiobook in the context of a program designed to build exercise habits
may lead them to infer and apply a temptation bundling strategy to
exercise more often. Formally, we hypothesize:
Hypothesis 3. Providing people with a want (in this case, a free
audiobook) that can be easily bundled with a should behavior they are
seeking to boost (in this case, exercise) leaks information that the want
ought to be paired with the should to make the should more attractive.
Hypothesis 4. People who receive a want (in this case, a free audio-
book) that can easily be bundled with a should behavior they are seeking
to boost (in this case exercise) will engage in more of that should
behavior.
1.4. Current investigation
We present the results of four experiments that test our hypotheses
about whether people can self-impose or even infer and apply tempta-
tion bundling to boost exercise. Study 1A is a eld experiment in which
we test the effectiveness of giving people a free audiobook and
encouraging a bundling strategy (rather than imposing one) against
simply giving them a free audiobook (H1). We nd that encouraging
temptation bundling only leads to small benets on the extensive rather
than the intensive margin, boosting the likelihood of a weekly gym visit
by a marginal 4% during the intervention and signicantly lifting it by
6% during the ten weeks post-intervention. In Study 1B, we compare
both conditions from Study 1A (teaching participants about temptation
bundling and providing them with a free audiobook as well as simply
giving them a free audiobook) with a control condition in which par-
ticipants are not given an audiobook, allowing us to test H2 and H4. We
nd that teaching participants about temptation bundling and providing
them with a want (i.e., a free audiobook) that can be linked with exercise
increases the likelihood they’ll visit the gym by 10–14% both during and
up to seventeen weeks post-intervention. We further nd that simply
providing participants with a free audiobook in the context of an exer-
cise program has no signicant effect on the likelihood of a weekly gym
visit during the intervention but boosts the likelihood of gym visits by
15% during the four weeks post-intervention. And in two follow-up
laboratory experiments, Studies 2A and 2B, we nd evidence suggest-
ing that participants given a free audiobook by a gym program will infer
they should use a temptation bundling technique to build exercise
habits, consistent with H3. Overall, our ndings contribute important
new insights to scholarly debates regarding (a) people’s degree of so-
phistication and capacity to self-impose rules to counter present bias and
(b) their ability to make useful inferences about leaked information.
Our study focuses on boosting gym attendance for two key reasons.
First, increasing physical activity is an important goal for many in-
dividuals as well as policymakers (Guthold, Stevens, Riley, & Bull,
2018). Fewer than half of American adults exercise sufciently and 30%
of American adults report engaging in no exercise whatsoever (Physical
Activity Guidelines, 2018) . Recent estimates suggest that physical
inactivity accounts for 9% of premature mortality and a 25% increase in
worldwide physical activity levels would be enough to avert 1.3 million
deaths each year (Lee et al., 2012). Thus, our study focuses on an
important policy problem. Further, many past studies of behavior
change have focused on increasing exercise (Acland & Levy, 2015;
Babcock, Bedard, Charness, Hartman, & Royer, 2015; Charness &
Gneezy, 2009; Condliffe, Isgin, & Fitzgerald, 2017; Royer et al., 2015;
DellaVigna & Malmendier, 2006), including the original study of
temptation bundling (Milkman et al., 2014). Testing the effectiveness of
our interventions in a gym setting makes it easier to directly compare
their impact with that of other behavioral interventions.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
4
2. Study 1: A large-scale, preregistered eld experiment
In Study 1, we conducted a large-scale eld experiment to test Hy-
potheses 1, 2 and 4. We rst report the results of Study 1A, a preregis-
tered experiment in which we evaluate whether encouraging
participants to temptation bundle and giving them something appealing
to bundle with exercise (a code to download a free audiobook from
Audible.com) boosts gym visits relative to simply giving participants the
same temptation (a free audiobook) with no further explanation (H1). In
Study 1B, we pool a quasi-experimental control condition with our
Study 1A data. In this control condition, participants did not receive a
free audiobook or encouragement to temptation bundle. Examining this
control condition allows us to test whether encouraging temptation
bundling and providing participants with a free audiobook increases
exercise (H2) and whether simply providing participants with a free
audiobook when they sign up for an exercise-boosting program at their
gym can increase exercise (H4).
2.1. Study 1A: Does encouraging people to temptation bundle and giving
them a free audiobook boost gym visits relative to simply giving them a free
audiobook?
In Study 1A, we conducted a preregistered eld experiment to test
the effectiveness of encouraging people to only enjoy a temptation while
working out at the gym.
Method. We partnered with 24 Hour Fitness, one of the largest gym
chains in the United States, to conduct a eld experiment. All active
adult members of 24 Hour Fitness (over four million people) were
eligible to participate. To access their gym, 24 Hour Fitness members
had to check in by presenting an ID to staff at the front desk, swiping or
scanning a member card, or using a ngerprint reader and unique check-
in code. We used check-in data provided by 24 Hour Fitness to track
participants’ gym visits before, during, and after a four-week interven-
tion period.
Participant recruitment. From March 21, 2018, through January 31,
2019, members of 24 Hour Fitness gyms were invited to join the “StepUp
Program” through a 24 Hour Fitness marketing campaign that included
emails, gym app notications, social media advertisements, phone calls,
postcards, and posters in gyms, as well as on-site, in-person recruitment.
All recruitment materials advertised the StepUp Program as “a habit-
building, science-based workout program” and informed gym mem-
bers that they could sign up for free and earn Amazon cash rewards for
participating. Members were also informed that simply registering for
the StepUp Program would make them eligible for the chance to win a
$50 Amazon gift card. Our study was one of 20 preregistered studies
embedded in the StepUp Program (see Milkman et al., 2020, for a meta-
analysis of all 54 study conditions embedded in these 20 studies).
Participant enrollment. Gym members signed up online for StepUp
via the program’s website. After consenting to participate in the
research, participants provided their gym check-in code and date of
birth to verify their gym membership and allow us to sync their online
enrollment data with their gym visit data. Participants also provided
their name, email address, and phone number upon enrollment. They
received an SMS code to verify they could receive text messages from
StepUp before being randomized to a study condition (see page 7 of the
Online Supplement for the full participant enrollment process).
Experimental design. StepUp registrants were randomly assigned
with equal probability to one of two experimental conditions that were
designed and preregistered to evaluate the impact of teaching gym
members about temptation bundling (http://bit.ly/2NUkEBE; this is
referred to as the "Free Audiobook Experiment" in Fig. 2). As detailed
below, the free audiobook with encouragement to temptation bundle con-
dition introduced and explained the benets of temptation bundling and
provided participants with a code that would allow them to download a
free audiobook from Audible.com. In contrast, the simple free audiobook
condition only provided participants with this Audible.com download
code.
Experimental conditions. The initial registration process was the
same for all participants regardless of experimental condition: After
enrolling in the 28-day StepUp Program, participants were informed
that they had earned 100 points for registering and would receive 300
StepUp points each time they visited the gym, redeemable for Amazon
cash after the program’s conclusion. Participants learned that points
were redeemable at a rate of $5 for 7000 points ($0.22 per gym visit).
They were then prompted to plan the dates and times when they would
work out at the gym each week during the StepUp Program and learned
that they would receive text message reminders to exercise 30 minutes
prior to each scheduled workout for the program’s 28-day duration.
Participants then answered a multiple-choice question about what
they normally do at the gym (“attend tness classes,” “swim,” “use
cardio equipment,” “lift weights,” and/or “other”). Next, they were
informed that as a special promotion, the StepUp Program had partnered
with Audible.com to share one free audiobook with them, and at the end
of registration, they would be provided with a code that would allow
them to download their free audiobook.
Up to this point, participants in each of our two experimental con-
ditions experienced identical registration processes. Their treatments
diverged only after they learned they would receive a free audiobook
code. Participants randomly assigned to the simple free audiobook con-
dition (N =1604) received their code for a free audiobook and were told
their registration process was complete. Meanwhile, participants who
were randomly assigned to the free audiobook with encouragement to
temptation bundle condition (N =1685) received additional content
during registration. After learning that they would receive a free
audiobook from Audible.com, they were told that “A secret to making
working out at the gym into an enjoyable, fun habit is to make a simple
rule: I only let myself enjoy my favorite TV shows, audiobooks, or
podcasts on my smartphone when exercising. Scientists call this ‘temp-
tation bundling.’” Above this text, participants saw a cartoon that
visually depicted the concept of temptation bundling (see Fig. 1). On the
following screen, participants saw a short video about temptation
bundling (see http://bit.ly/2GhdbrM).
Participants were then asked to report what type of entertainment
would make them crave their workouts most if it were paired with
Fig. 1. Participants in the free audiobook with encouragement to temptation
bundle condition saw this cartoon explaining the concept of temptation
bundling during the StepUp Program registration process.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
5
exercise (audiobooks, TV shows, or podcasts) and their preferred genre
of that form of entertainment (thrillers, science ction, fantasy,
romance, literary ction, or dramas). Based on their responses, partici-
pants were offered a customized recommendation for media they could
use to temptation bundle (e.g., someone who reported preferring sci-
audiobooks was told that they might enjoy listening to The Hand-
maid’s Tale, Dune, The Hunger Games, or The Water Knife while exer-
cising). These recommendations were based on the results of a survey of
Amazon Mechanical Turk workers (N =100) in which we asked them to
share their favorite audiobooks, TV shows, and podcasts by genre.
Finally, participants assigned to the free audiobook with encouragement to
temptation bundle condition received their code for a free audiobook, and
were encouraged to sign a pledge promising to only use their favorite
form of media at the gym.
Starting the day after they registered for StepUp, participants in both
experimental conditions received text message reminders to exercise 30
minutes prior to each scheduled workout and weekly emails with their
workout schedule during the four-week program. Some of these text
messages and emails also included reminders to download their free
audiobook from Audible.com. Additionally, in the free audiobook with
encouragement to temptation bundle condition, all emails and text mes-
sages included reminders to temptation bundle and reminders of par-
ticipants’ pledge to temptation bundle if they had signed the pledge. At
the end of the four weeks, all participants received a goodbye text
message conrming their completion of the program. Descriptions of
text messages and emails sent across conditions can be found in the
Online Supplement on pages 15–17 and all study materials are in the
Online Supplement on pages 8–14.
Participants. Following our preregistered analysis plan, we only
analyzed data from participants who indicated (prior to randomization)
that they typically use cardio machines and/or lift weights at the gym.
This is because participants who prefer to swim or take tness classes at
the gym could not easily take advantage of temptation bundling. Fig. 2
presents a diagram of the ow of study participants in Study 1A.
Participants were also dropped from our analyses if they requested to
withdraw from the study or signed up for the program twice. Specif-
ically, nine participants in the simple free audiobook condition and eight
participants in the free audiobook with encouragement to temptation bundle
condition were excluded from our analyses for these reasons.
After excluding the 955 participants who did not indicate that they
typically use cardio machines and/or lift weights at the gym and the 17
participants who signed up twice or requested to withdraw their data,
we were left with a total of 2,334 24 Hour Fitness members in this study
(63% female, average age =38.84, SD
age
=13.16; free audiobook with
encouragement to temptation bundle condition: N =1211; simple free
audiobook condition: N =1123). These participants were residents of 18
states (57% Californian, 16% Texan, 6% Coloradan, 4% Washingtonian,
4% Oregonian, and 14% from other states). Using Census data, we
inferred participants’ race based on their rst and last names (following
Morton, Zettelmeyer, & Silva-Risso, 2003 and Berger & Milkman, 2012).
This technique suggests that our study participants were 50% White,
22% Hispanic, 12% Asian, 2% Black, and 14% Other or Unknown.
Balance checks reported in Table 1 suggest that random assignment was
successful, as we see no difference in observable participant character-
istics across experimental conditions.
Statistical analyses. All of our analyses in Study 1A followed our
preregistered analysis plan: http://bit.ly/2NUkEBE.
Dependent variables. Our primary outcome of interest was gym
attendance, which we measured in two ways. First, we examined the
number of gym visits made by each study participant per week
(following Acland & Levy, 2015; Charness & Gneezy, 2009; Milkman,
Minson, & Volpp, 2014; Royer et al., 2015; Beshears, Lee, Milkman,
Mislavsky, & Wisdom, 2020). Second, we examined an indicator for
whether each participant visited the gym at least once in a given week
([1 if yes, 0 if no]; following Royer et al., 2015; Beshears et al., 2020).
We analyzed these outcome variables during the four-week StepUp
Program, which we refer to as our “intervention period,” and for two
different, preregistered post-intervention periods: four weeks and 10
weeks post-intervention.
2
24 Hour Fitness provided us with up to 52
weeks of pre-intervention gym visit data for each participant, which
allowed us to control for participants’ pre-intervention gym attendance
(following our preregistered analysis plan) and substantially increased
our statistical power.
To evaluate the effect of explaining temptation bundling and
encouraging participants to commit to bundling, we compared the gym
attendance of participants in the free audiobook with encouragement to
temptation bundle condition to the simple free audiobook condition. We
relied on a difference-in-difference regression framework, including
data on participants’ gym visits pre-, during-, and post-intervention and
including participant xed effects in our analyses.
Regression specications. To measure the effect of the free audiobook
with encouragement to temptation bundle condition over and above the
simple free audiobook condition, we ran the following ordinary least
squares regression model, with clustered standard errors by participant:
yit =β0+β1treatmenti*during_interventionit
+β2treatmenti*post_interventionit +CXi+DZt+
ε
it
where i indexes participants and t indexes weeks. The left-hand side yit is
either the number of gym visits made by participant i in week t, or a
binary indicator for whether participant i visited the gym in week t. On
the right-hand side of the equation is an indicator for a participant’s
experimental condition interacted with an indicator for whether week t
is an intervention week (1 if yes, 0 if no) and an indicator for a partic-
ipant’s experimental condition interacted with an indicator for whether
week t is a post-intervention week (1 if yes, 0 if no). We also controlled
for week xed effects (denoted by DZt terms) and participant xed ef-
fects (denoted by CXi terms).
3
Results
StepUp Program registration survey results. We found that 65.4% of
participants in our free audiobook with encouragement to temptation bundle
condition completed all components of the StepUp enrollment process
and received their free audiobook code at the end of the enrollment
survey, a percentage that is statistically indistinguishable (p =.816)
from the 65.9% of participants in our simple free audiobook condition
who completed all components of the StepUp enrollment process and
received their free audiobook code. Furthermore, the majority of par-
ticipants in the free audiobook with encouragement to temptation bundle
condition (78.6%) signed a pledge to temptation bundle during the four-
week StepUp Program. Over a quarter of the participants in the free
audiobook with encouragement to temptation bundle condition (29.9%)
indicated that audiobooks were their preferred type of entertainment to
bundle with exercise. Participants downloaded their free Audible
audiobook at similar (albeit low) rates across the two experimental
conditions: 14% of participants in the free audiobook with encouragement
to temptation bundle condition and 13% of participants in the simple free
audiobook condition downloaded their free audiobooks, z =0.815, p =
.415.
2
We also report supplemental analyses that include 17 weeks of post-
intervention data because we were able to obtain complete data on all study
participants’ gym attendance for up to 17 weeks post-intervention.
3
Data from Study 1A and Study 1B cannot be posted publicly according to
the researchers’ agreement with 24 Hour Fitness. Researchers interested in the
data used in Study 1A and Study 1B should contact the corresponding author,
Erika Kirgios (ekirgios@wharton.upenn.edu), who upon request, will ask 24
Hour Fitness for approval to share de-identied study data. If approval is ob-
tained, the researcher(s) will be granted access to data les and analysis code
posted at https://osf.io/ubzh6/. De-identied data and analyses from Studies
2A and 2B are posted at https://osf.io/8jby5/?view_only=2b3e560987af483
1b5043a50f01678fb.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
6
The impact of encouraging temptation bundling. Participants in
Study 1A visited the gym an average of 1.67 times per week during the
StepUp Program, and, on average, 59.6% of participants visited the gym
at least once a week. Fig. 3 depicts participants’ extra weekly gym visits
during and after the StepUp Program by experimental condition. As this
gure illustrates, the two experimental conditions produced a similar
number of workouts during the StepUp Program, but directionally,
participants in the free audiobook with encouragement to temptation bundle
condition were more likely to visit the gym after the program ended. To
evaluate the signicance of these patterns, we turn to our preregistered
regression models described in the section above entitled Regression
specications.
As reported in Table 2, Model 1, participants in the free audiobook
with encouragement to temptation bundle condition were a marginal,
regression-estimated 2.2 percentage points more likely to visit the gym
at least once in a given week during the intervention period than par-
ticipants in the simple free audiobook condition (SE =0.013, p =.095; a
4.3% increase). However, as Table 2, Model 1 reports, we found no
difference in participants’ average estimated weekly gym visits across
experimental conditions during the intervention period (b = − 0.001, SE
=0.051, p =.989).
We next examined differences between conditions after the conclu-
sion of the StepUp Program. During this four-week follow-up period, as
reported in Table 2, Model 3, participants in the free audiobook with
encouragement to temptation bundle condition were also a marginal,
regression-estimated 2.5 percentage points more likely to visit the gym
in a given week than participants in the simple free audiobook condition
(SE =0.014, p =.080; a 4.9% increase). Furthermore, as reported in
Table 2, Model 4, participants went to the gym a marginal, regression-
estimated 0.09 more times per week (equivalent to a 6.5% lift in
average weekly gym visits) in the free audiobook with encouragement to
temptation bundle condition than in the simple free audiobook condition
during the four weeks post-intervention (SE =0.049, p =.081).
We saw similar results when we extended our analyses to a 10-week
follow-up period. As reported in Table 2, Model 5, participants who were
encouraged to temptation bundle visited the gym at least once a week at
a regression-estimated 3.2 percentage point higher rate than others (SE
=0.01, p =.012; a 6.2% increase). However, as reported in Table 2,
Model 6, the marginal increase in participants’ total number of weekly
gym visits produced by encouragement to temptation bundle in our four-
week follow-up period dropped below marginal signicance levels when
our analyses included gym visits up to 10 weeks post-intervention (b =
0.071, SE =0.045, p =.113).
Overall, these results offer some evidence that participants who were
educated about temptation bundling increased their exercise compared
to those who did not learn about temptation bundling, supporting Hy-
pothesis 1. Although small, the effects of educating participants about
temptation bundling relative to simply giving them a free audiobook
were durable, and surprisingly, seemed stronger post-intervention than
during the program itself. Teaching people about temptation bundling
seemed to be most effective at dissuading them from entirely aban-
doning the gym, as the effects of learning about temptation bundling
Fig. 2. Flow of study participants in Study 1A.
Table 1
Balance table for Study 1A comparing participant characteristics across
treatments.
Study 1A
Simple free
audiobook
Free audiobook with
encouragement to
temptation bundle
p-value
from F-
test
Age 38.7 (13.1) 39.0 (13.2) 0.615
Average number of weekly
gym visits in the four
weeks before joining
StepUp
1.2 (1.5) 1.2 (1.4) 0.671
Weeks of gym
membership prior to
joining StepUp
35.8 (20.4) 36.8 (19.8) 0.262
Female 62.1% 63.0% 0.630
White 50.3% 49.6% 0.688
Black 1.9% 2.0% 0.652
Asian 12.3% 12.3% 0.873
Hispanic 21.7% 22.8% 0.527
Other race 13.7% 13.3% 0.887
Sample size 1,123 1,211
Note. Participants’ age, gender, and weeks of gym membership prior to joining
StepUp were provided by 24 Hour Fitness. Each participant’s race was inferred
using their rst name, last name, and Census data (following Morton, Zettel-
meyer, and Silva-Risso, 2003 and Berger & Milkman, 2012). Each participant’s
average number of weekly gym visits in the four weeks before joining StepUp
was calculated from 24 Hour Fitness data, and for weeks when participants were
not 24 Hour Fitness members, they were presumed to make no gym visits.
Standard deviations for means are reported in parentheses. F-tests were con-
ducted as a joint test of equality across all treatments.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
7
were strongest when we focused on whether participants visited the gym
at all in a given week.
Discussion. The results of Study 1A suggest that the benets of ed-
ucation about temptation bundling are small and accrue primarily in the
long term. Moreover, encouragement to temptation bundle seems to
operate primarily on the extensive margin, increasing the likelihood that
people will go to the gym at least once a week, but exerting less inuence
on their weekly visit total. Although our treatment effect is smaller than
in the Milkman et al. (2014) experiment, this is likely due to the dif-
ferences between the control conditions used in the two experiments. In
Milkman et al. (2014), participants in the control condition received a
gift certicate to Barnes & Noble that was of equal value to the audio-
books on offer in the treatment condition. In our experiment, partici-
pants received a code that would allow them to download a free
audiobook from Audible.com in both conditions. Merely giving partici-
pants a free audiobook could have helped them exercise more often by
leaking information about the intentions of the exercise program’s cre-
ators (per Hypothesis 3): When participants signed up for a program
offered by 24 Hour Fitness to help them exercise more regularly and
received a free audiobook as part of that program, they assumed the
audiobook was intended to help them exercise.
There are two results providing initial support for this contention.
First, Fig. 4 indicates that participants in both conditions increased their
gym visits during the StepUp intervention period relative to baseline.
Second, ndings from a follow-up, preregistered survey of a separate
sample of gym-goers
4
(N =75) point to the possibility that the free
audiobooks leaked information. In this survey, participants saw
screenshots of all study stimuli included in the simple free audiobook
condition of the StepUp Program. Participants were asked to predict
how StepUp Program designers intended for the free audiobook to be
used, rst in an open-ended response—intended to get at their unbiased
opinion—and then in a multiple-choice question in which they indicated
whether the audiobook was intended for “entertainment at the gym,”
“for entertainment outside the gym,” or “whenever they wanted.” Three
independent coders blind to our hypotheses indicated whether each
open-ended response mentioned using the audiobooks for entertainment
at the gym or not. Sixty out of 75, or 80% of participants, mentioned that
the audiobook was intended for use at the gym in their free responses.
5
Furthermore, in a subsequent multiple choice question, 80% of
Fig. 3. Flow of study participants in Study 1B.
Table 2
Regression-estimated effects of providing encouragement to temptation bundle in Study 1A during and post-intervention.
Time period of interest Four-week intervention period Four weeks post-intervention Ten weeks post-intervention
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Dependent variable Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Free audiobook with encouragement to temptation
bundle ×Indicator for time period of interest
0.022†
(0.013)
−0.001
(0.051)
0.025†
(0.014)
0.086†
(0.049)
0.032*
(0.013)
0.071
(0.045)
Fixed effects for participants? Yes Yes Yes Yes Yes Yes
Fixed effects for weeks since start of intervention? Yes Yes Yes Yes Yes Yes
N of participants 2334 2334 2334 2334 2334 2334
N of observations 94,146 94,146 103,482 103,482 117,486 117,486
Adjusted R
2
0.437 0.542 0.428 0.535 0.422 . 528
Note. The table reports the results of preregistered ordinary least squares (OLS) regressions predicting study participants’ likelihood of visiting the gym at least once in a
given week during the four-week StepUp Program intervention period (Model 1), the four-week post-intervention period (Model 3), and the 10-week post-intervention
period (Model 5). The table also reports the results of OLS regressions predicting study participants’ total weekly workouts during the four-week intervention period
(Model 2), the four-week post-intervention period (Model 4), and the 10-week post-intervention period (Model 6). In all models, the key predictor variable is an
indicator for being in the free audiobook with encouragement to temptation bundle experimental condition interacted with an indicator for the time period of interest
(either the four-week intervention period, the four-week post-intervention period, or the 10-week post-intervention period).
Standard errors clustered by participant are reported in parentheses. †p <0.10, * p <.05, ** p <.01, *** p <.001.
4
https://aspredicted.org/g39ff.pdf.
5
Each of the three RAs was given the same Word document containing in-
structions about how to code the free responses. The text of this Word document
is available upon request.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
8
participants indicated that they thought the free audiobook was meant
to be used for entertainment at the gym, even though they were given no
information about temptation bundling.
6
This is a signicantly larger
proportion than would be expected by chance (z =5.60, p <.001) and
suggests that simply providing a free audiobook in this context led
participants to infer that they should temptation bundle.
To test whether merely providing participants with a free audiobook
in a workout context is sufcient to promote behavior change via in-
formation leakage (Hypothesis 4), we make use of our second, quasi-
experimental control condition from the StepUp mega-study. Specif-
ically, the StepUp mega-study included a control condition in which
participants were exposed to incentives, planning, and pre-workout re-
minders, exactly like participants in Study 1A, but were not given a free
audiobook. In Study 1B, we compare each of the two free audiobook
conditions from Study 1A to this control condition to assess the effect of
giving participants a free audiobook on gym visits.
2.2. Study 1B: Does simply giving participants a free audiobook boost gym
visits?
In Study 1B, we explored whether giving free audiobooks to gym
members in the context of an exercise-promoting program increased
their gym attendance to test Hypothesis 4, and we also examined the
benets of providing both a free audiobook and encouragement to
temptation bundle to test Hypothesis 2. We conducted parallel analyses
to those presented in Study 1A (closely following our Study 1A prereg-
istration, though Study 1B was not preregistered), this time comparing
gym attendance across three conditions: control, simple free audiobook,
and free audiobook with encouragement to temptation bundle.
Methods. As in Study 1A, all participants were members of 24 Hour
Fitness who enrolled in the StepUp Program. All recruitment and
enrollment details were identical to Study 1A.
Experimental design. The simple free audiobook and free audiobook
with encouragement to temptation bundle experimental conditions were
described in Study 1A. The control condition in the StepUp mega-study
had all of the same features as our two temptation bundling experi-
mental conditions but did not describe a partnership with Audible.com,
ask participants about what they typically do at the gym, or give par-
ticipants a code that they could use to download a free audiobook from
Audible.com.
Randomization. To ensure as many as possible of the 20 separately
preregistered experiments in the StepUp mega-study would be
adequately powered, StepUp Program enrollees were randomly assigned
to experimental conditions with time-varying probabilities. At any given
moment, the plurality of participants (40–60%) were randomly assigned
with equal probability to one of the conditions in a given, specic
experiment (one of the 20 experiments that were separately preregis-
tered) within the overall mega-study, and the remaining participants
were assigned across the other experimental conditions with equal
probability. After a prespecied number of participants enrolled in
StepUp, the probabilities shifted and a new experiment received the
plurality of participants. The probabilities shifted a total of 27 times;
each time denes a randomization cohort (see Fig. 4 for a owchart
depicting the randomization process in greater detail).
Since the two experimental conditions in which participants were
given free audiobooks were designed as part of a single experiment,
participants always had an equal probability of being assigned to either
of these two conditions in Study 1A. However, the control condition we
analyze in Study 1B was not designed as part of this single experiment,
so at any given time, the probability of being assigned to this condition
differed from the probability of being assigned to either of the free
audiobook conditions. As a result, comparisons between our free audio-
book experimental conditions and the control condition require sample
weighting to ensure balance, making these analyses quasi-experimental.
Below, in a section of our paper entitled Sample weighting, we provide
detailed information about the weights we assigned to participants in
our analyses to account for the different, time-varying probabilities of
being assigned to the control versus free audiobook conditions in our
experiment.
Experimental conditions. As noted above, participants were ran-
domized with unequal probabilities into one of three different condi-
tions: the control condition, the simple free audiobook condition, or the
free audiobook with encouragement to temptation bundle condition. Par-
ticipants assigned to the control condition experienced the same regis-
tration process as participants in the two free audiobook conditions
(described in Study 1A), but they did not receive a code that would allow
them to download a free audiobook or information about temptation
bundling. Every other detail of the participants’ experience was iden-
tical across the control and free audiobook conditions.
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
1234 1 2 3 45678 9 10
dednEpUpetSecniSskeeWpUpetSotniskeeW
Average Weekly Likelihood of Gym Visits
During and Post-intervention Minus the
Average Likelihood of Visits at Baseline
Simple Free Audiobook Free Audiobook with Encouragement to Temptation Bundle
After StepUp
During StepUp
Fig. 4. This graph depicts Study 1A participants’ average likelihood of visiting the gym in a given week during and post-intervention by experimental condition after
subtracting off their baseline gym attendance (averaged over four weeks pre-intervention). Encouraging participants to temptation bundle leads to a durable boost in
average weekly likelihood of gym visits (as shown in the graph), but has negligible effects on average total weekly gym visits. Note: When participants were not
members of 24 Hour Fitness during the entire, four-week pre-intervention baseline period, zero gym visits were assumed for any missing weeks to ensure
balanced panel.
6
Meanwhile, 14.7% and 5.3% of participants indicated that they thought the
free audiobook was intended for use “whenever they wanted” or “outside of the
gym,” respectively.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
9
Descriptions of text messages and emails sent across conditions can
be found in the Online Supplement on pages 15–17, and all study ma-
terials (including screenshots of the enrollment and registration process
in each experimental condition) are presented in the Online Supplement
on pages 8–14.
Participants. As in Study 1A, participants were dropped from the
analyses if they requested to withdraw from the study or signed up for
the program twice. In total, this led us to drop 30 participants from the
control condition, nine participants from the simple free audiobook con-
dition and eight participants from the free audiobook with encouragement
to temptation bundle condition.
After these exclusions, a total of 6,792 24 Hour Fitness members
were assigned to Study 1B’s three different experimental conditions
(64% female, average age =39.26, SD =13.26; simple free audiobook: N
=1604; free audiobook with encouragement to temptation bundle: N =
1685; control: N =3503). These participants were residents of 26 states
(55% Californian, 15% Texan, 6% Coloradan, 4% Washingtonian, 4%
Oregonian, and 15% from other states). As in Study 1A, we used Census
data to infer participants’ race based on their rst and last names
(following Morton et al., 2003 and Berger & Milkman, 2012). This
technique suggests that our study participants were 50% White, 22%
Hispanic, 11% Asian, 3% Black, and 14% Other or Unknown. Balance
checks reported in Table 3 suggest that random assignment was suc-
cessful, as we see no difference in observable participant characteristics
across experimental conditions after including sample weights
(described below) to account for imbalanced random assignment.
Note that in comparing the two free audiobook conditions to the
control condition, we no longer exclude participants from our study
whose self-reported typical activities at the gym were incompatible with
temptation bundling because these screening questions were not asked
of participants in the control condition.
7
This is why Study 1B includes
955 more participants in the two free audiobook conditions (N =3289)
than the total participant count in Study 1A (N =2334). Naturally, the
inclusion of additional participants who could not plausibly temptation
bundle vastly reduces our power to detect the effects of encouraging
temptation bundling and test Hypothesis 1, which our preregistered
Study 1A was designed to measure. We therefore rely on Study 1B solely
for the exploration of Hypotheses 2 and 4.
Statistical analyses. We relied on the same general analysis frame-
work described in Study 1A, but this time we compared the two free
audiobook conditions to the control condition. We accounted for the
quasi-experimental nature of this comparison by including sample
weights to adjust for the different probabilities of assignment to condi-
tions across randomization cohorts in our regression analyses and by
including cohort-by-week xed effects, as detailed below.
Regression specications: free audiobook conditions versus control. To
measure the effect of offering participants either a free audiobook or a
free audiobook with encouragement to temptation bundle, we relied on the
following ordinary least squares regression model, with standard errors
clustered by participant and sample weights as described in the Sample
weighting section below:
yit =β0+∑G
g=1βg
1dg
i×during_interventionit +∑G
g=1βg
2dg
i
×post_interventionit +∑C
c=1∑T
t’=2γc,t’dc,t’
it +CXi+
ε
it,
where i indexes participants, t indexes weeks relative to the start of the
intervention, g indexes experimental condition (our control condition
was set as the omitted reference group, and G =2 to account for the two
free audiobook conditions), and c indexes randomization cohorts. The
left-hand side, yit , represents one of the two dependent variables dis-
cussed above: either the number of gym visits made by participant i in
week t, or a binary variable indicating whether participant i visited the
gym in week t (1 =yes, 0 =no). On the right-hand side of the equation
are two interaction terms: rst, an interaction between an indicator, dg
i,
for whether participant i was assigned to condition g and an indicator for
whether week t was during the intervention period for participant i, and
second, an interaction between dg
i and an indicator for whether week t
was during the post-intervention period of interest for participant i. We
controlled for cohort-by-week interactions (as denoted by dc,t’
it terms)
and participants xed effects (as denoted by CXi terms).
The coefcients βg
1and βg
2 refer to the treatment effects of experi-
mental condition g compared to the control condition during the inter-
vention period and the post-intervention period of interest, respectively.
Sample weighting. To account for compositional differences across
randomization cohorts when comparing the two free audiobook condi-
tions to the control condition, we sample-weighted each observation in
our regressions such that each experimental group was equally weighted
within a cohort, and each cohort was weighted proportionally to its
duration. The weight specication used for each observation was as
follows:
wg,c
it =1
Ti
*1
Ng,c*1
G*Lc
L
where i indexes participants; t indexes weeks relative to the start of the
intervention; g indexes experimental condition; c indexes randomization
cohort; Ti indicates the length, in weeks, of the panel data for participant
i; Ng,c indicates the number of participants assigned to experimental
Table 3
Balance table for Study 1B comparing participant characteristics across
treatments.
Study 1B
Control Simple free
audiobook
Free audiobook
with encourage-
ment to
temptation bundle
p-value
from F-
test
Age 39.2
(13.3)
39.6 (13.4) 39.6 (13.3) 0.294
Average number of
weekly gym visits in
the four weeks
before joining
StepUp
1.3
(1.5)
1.2 (1.5) 1.2 (1.4) 0.084
Weeks of gym
membership prior
to joining StepUp
35.4
(20.3)
35.9 (20.3) 36.9 (19.9) 0.418
Female 57.6% 63.5% 63.6% 0.189
White 43.4% 50.7% 49.8% 0.845
Black 2.8% 2.2% 2.4% 0.888
Asian 9.5% 12.9% 11.8% 0.532
Hispanic 19.5% 20.6% 21.8% 0.769
Other race 24.6% 13.5% 14.2% 0.190
Sample size 3503 1604 1685
Note. Participants’ age, gender, and weeks of gym membership prior to joining
StepUp were provided by 24 Hour Fitness. Each participant’s race was inferred
using their rst name, last name, and Census data (following Morton, Zettel-
meyer, and Silva-Risso, 2003 and Berger & Milkman, 2012). Each participant’s
average number of weekly gym visits in the four weeks before joining StepUp
was calculated from 24 Hour Fitness data, and for weeks when participants were
not 24 Hour Fitness members, they were presumed to make no gym visits.
Standard deviations for means are reported in parentheses. F-tests were con-
ducted as a joint test of equality across all treatments. Ns for simple free audiobook
and free audiobook with encouragement to temptation bundle conditions are larger
than Ns reported in Table 1 (Study 1A) as we no longer exclude participants who
did not select cardio equipment and/or weight lifting as their usual form of
exercise.
7
Note that these exclusions were a preregistered feature of Study 1A, but
would bias our analyses in Study 1B.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
10
condition g during cohort c
8
; G indicates the number of experimental
conditions in the analysis; Lc indicates the length of cohort c, in weeks;
and L is the overall length of the recruitment period, in weeks.
The inclusion of these sample weights allows us to control for both
compositional differences between cohorts and for each cohort’s timing.
By including the inverse of the number of participants assigned to
treatment g in cohort c (1/Ng,c) in our weighting specication, we weight
each treatment equally within a cohort (rather than overweighting
treatments that were assigned more participants). For example, if one-
fourth of our participants in January were assigned to our control con-
dition and three-fourths were assigned to our two free audiobook con-
ditions, we would weight each participant in the control condition three
times as heavily as each participant in our free audiobook conditions in
January. Similarly, by including the relative duration of each cohort
(Lc/L) in our weighting formula, we weight each cohort in proportion to
its duration. This approach ensures any imbalance in random assign-
ment will not bias our regression-estimated effects of the free audiobook
conditions.
Results. Participants in Study 1B visited the gym an average of 1.57
times per week during the intervention, and on average, 57.7% of them
visited the gym in a given week. Fig. 5 depicts participants’ extra gym
attendance during and post-intervention by experimental condition
after accounting for their baseline gym attendance. As this gure illus-
trates, participants in both the simple free audiobook condition and the
free audiobook with encouragement to temptation bundle condition exer-
cised directionally more than those in the control condition during and
after the intervention. To evaluate the signicance of these differences,
we compared the change in frequency of gym visits for participants in
each of the two free audiobook treatments to participants in the control
condition using the regression models described in Regression specica-
tions: free audiobook conditions versus control.
As Table 4, Models 1–2 report, we found that during the intervention
period, participants in the simple free audiobook condition visited the
gym an estimated 0.23 times more per week (SE =0.114, p =.040; a
17.6% increase) than participants in the control condition, supporting
Hypothesis 4, though they were not signicantly more likely to visit the
gym in a given week (b =0.021, SE =0.033, p =.528). We nd that
during the intervention period, participants in the free audiobook with
encouragement to temptation bundle condition visited the gym a marginal,
regression-estimated 0.14 times more per week (SE =0.080, p =.076; a
10.0% increase) than participants in the control condition, and were a
regression-estimated 6.7 percentage points more likely to visit the gym
in a given week (SE =0.024, p =.005; a 13.6% increase), consistent with
Hypothesis 2.
As Table 4, Models 3–4 report, we observe slightly larger benets of
the two free audiobook conditions in the four weeks post-intervention.
During this period, participants in the simple free audiobook condition
made an estimated 0.24 more weekly gym visits, on average (SE =
0.097, p =.015; an 18.4% increase), than participants in the control
condition and were a regression-estimated 7.6 percentage points more
likely to visit the gym in a given week (SE =0.030, p =.013; a 15.4%
increase), supporting Hypothesis 4. Similarly, participants in the free
audiobook with encouragement to temptation bundling condition made an
estimated 0.16 more average weekly gym visits (SE =0.079, p =.041; a
12.2% increase) than participants in the control condition and were a
regression-estimated 6.4 percentage points more likely to visit the gym
in a given week (SE =0.026, p =.012; a 13.0% increase), supporting
Hypothesis 2.
When we run the same analyses but instead look at a 10-week post-
intervention follow-up period, the results are remarkably consistent, as
reported in Table 4, Models 5–6. Participants in the simple free audiobook
condition made a directional but insignicant estimated 0.14 more
weekly gym visits (SE =0.094, p =.127; a 10.7% increase) than par-
ticipants in the control condition, and were directionally but insigni-
cantly more likely to visit the gym in a given week (b =0.035; SE =
0.025, p =.155; a 7.1% increase). Participants in the free audiobook with
encouragement to temptation bundle condition made an estimated 0.13
more weekly gym visits (SE =0.066, p =.046; a 9.9% increase) than
participants in the control condition and were a regression-estimated 5.1
percentage points more likely to visit the gym each week than partici-
pants in the control condition during the 10-week post-intervention
period (SE =0.019, p =.006; a 10.3% increase), supporting Hypothe-
sis 2.
Although we focus on our preregistered analyses in Study 1A to
compare the two free audiobook conditions, we also ran Wald tests to
compare the regression-estimated impact of the simple free audiobook
condition with that of the free audiobook with encouragement to temptation
bundle condition in Study 1B. These comparisons differ substantially
from the comparisons presented in Study 1A because for comparability
with our control condition, we needed to include in our analyses 955
additional participants who reported that they did not typically use
cardio equipment or lift weights at the gym (but rather swam or took
classes) and thus would not be able to temptation bundle while exer-
cising. Unsurprisingly, when diluted by an inux of 41% more study
participants who could not benet from temptation bundling (and
whom our preregistered Study 1A analyses excluded), our simple free
audiobook and free audiobook with encouragement to temptation bundle
conditions no longer differ signicantly in any comparisons (in all Wald
tests, p’s >0.10).
Alternative post-intervention follow-up periods. To ensure the
robustness of our Study 1 regressions presented in Tables 2 and 4 to
alternative follow-up periods after the conclusion of our intervention,
we ran supplementary analyses that are detailed in the Online Supple-
ment. Specically, we extended our post-intervention analyses to
include all 17 weeks of post-intervention data made available to us by 24
Hour Fitness, and we found that our results are remarkably robust in this
lengthier follow-up period, as reported in Online Supplement Tables S1
and S2.
Heterogeneity Analyses. We tested whether our estimated treatment
effects varied by gender, age, race, exercise frequency during the four
weeks pre-intervention, or length of gym membership. We found that
none of these participant characteristics were robustly associated with
any change in our estimated treatment effect.
9
Discussion. In Study 1B, we nd that offering participants access to
free audiobooks, regardless of whether temptation bundling was
explicitly encouraged, signicantly increased their gym visits during the
StepUp Program and for at least 10 weeks after it concluded, supporting
8
During high-volume registration periods, bugs in our system led to mistakes
in random assignment for 6.6% of the participants in the overall StepUp mega-
study. These individuals were accidentally assigned to treatment conditions
that differed from their intended treatment condition according to the pre-
dened randomization matrix. Our weighting accounts for this bug because
we dene N
g,c
as the number of participants who were actually assigned to
condition g during cohort c, instead of the number of participants who were
intended to be assigned to condition g during cohort c according to the
randomization matrix.
9
We did nd that the estimated effects of our simple free audiobook condition
on whether a participant visited the gym in a given week were larger for men
than for women, but only during the four (p =.029) and ten weeks (p =.014)
post-intervention.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
11
both Hypotheses 2 and 4.
10
The results of Study 1B suggest that when people were offered a free
audiobook by their gym as part of a program designed to boost their
exercise, the audiobook’s intended use was clear: It was meant to be
enjoyed during workouts at the gym. It may be that there was little
added benet of explaining temptation bundling to participants because
providing the free audiobook in this context was sufciently instructive,
consistent with Hypothesis 3.
Of course, there are alternative explanations for the results of Study
1B. Receiving the free audiobook could have induced positive feelings,
“warm glow,” or gratitude toward the gym for participants, which made
gym visits more appealing (Andreoni, 1995; Armenta, Fritz, & Lyubo-
mirsky, 2017; Falk, 2007; Isen, Shalker, Clark, & Karp, 1978; List &
Lucking-Reiley, 2002). Listening to or recalling the free audiobook
outside the gym may also have kept the gym top-of-mind for partici-
pants, reminding them to visit the gym more often (Bordalo, Gennaioli,
& Shleifer, 2013; Chetty, Looney, & Kroft, 2009; Dohmen, Falk, Huff-
man, & Sunde, 2006; Finkelstein, 2009).
To assess the plausibility of these alternative explanations for the
-0.15
-0.05
0.05
0.15
0.25
1234123456789 10
dednEpUpetSecniSskeeWpUpetSotniskeeW
Average* Weekly Likelihood of Gym Visits
During and Post-intervention Minus the Average
Likelihood of Visits at Baseline
Control Simple Free Audiobook Free Audiobook with Encouragement to Temptation Bundle
During StepUp After StepUp
Fig. 5. This graph depicts Study 1B participants’ average weekly gym visits during and post-intervention by experimental condition after subtracting off their
baseline gym attendance (averaged over four weeks pre-intervention). Note: When participants were not members of 24 Hour Fitness during the entire, four-week
pre-intervention baseline period, zero gym visits were assumed for any missing weeks to ensure a balanced panel. Average weekly gym visits were weighted to
account for the different probabilities of being assigned to different conditions as described in Sample weighting.
Table 4
Regression-estimated effects of providing a free audiobook and of providing encouragement to temptation bundle in Study 1B during and post-intervention.
Time period of interest Four-week intervention period Four weeks post-intervention Ten weeks post-intervention
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Dependent variable Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Any weekly gym
visit? (Y/N)
Total weekly
gym visits
Simple free audiobook ×Indicator for time period of
interest
0.021
(0.033)
0.235*
(0.114)
0.076*
(0.030)
0.236*
(0.097)
0.035
(0.025)
0.143
(0.094)
Free audiobook with encouragement to temptation
bundle ×Indicator for time period of interest
0.067**
(0.024)
0.142†
(0.080)
0.064*
(0.026)
0.161*
(0.079)
0.051**
(0.019)
0.131*
(0.066)
Fixed effects for participants? Yes Yes Yes Yes Yes Yes
Fixed effects for weeks since start of intervention ×
Cohort interaction?
Yes Yes Yes Yes Yes Yes
N of participants 6792 6792 6792 6792 6792 6792
N of observations 270,892 270,892 298,060 298,060 338,812 338,812
Adjusted R
2
0.451 0.570 0.431 0.551 0.424 0.539
Note. The table reports the results of ordinary least squares (OLS) regressions predicting study participants’ likelihood of visiting the gym at least once in a given week
during the four-week StepUp Program intervention period (Model 1), the four-week post-intervention period (Model 3), and the 10-week post-intervention period
(Model 5). The table also reports the results of OLS regressions predicting study participants’ total weekly workouts during the four-week intervention period (Model
2), the four-week post-intervention period (Model 4), and the 10-week post-intervention period (Model 6). In all models, the key predictor variables are an indicator for
being in the simple free audiobook condition interacted with an indicator for the time period of interest and an indicator for being in the free audiobook with encour-
agement to temptation bundle condition interacted with an indicator for the time period of interest.
Standard errors clustered by participant are reported in parentheses. †p <.10, * p <.05, ** p <.01, *** p <.001.
10
Interestingly, the effects of our temptation bundling and free audiobook
conditions increased in magnitude and signicance during the four-week post-
intervention period (see Table 4). This pattern of results may stem from a
ceiling effect at the outset of the intervention period. Across all conditions,
participants tended to concentrate their visits during the rst few weeks of the
intervention period, and then spread visits more evenly across weeks during the
post-intervention period. This may have made it difcult for differences be-
tween conditions to emerge during the intervention period. A ceiling effect
could also explain why, in Study 1A, the benets of explaining temptation
bundling relative to simply offering participants a free audiobook grew in the
post-intervention period (see Table 2).
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
12
benets of simply providing free audiobooks and to test Hypothesis 3,
we conducted two follow-up scenario experiments.
3. Study 2: Follow-up lab studies
3.1. Study 2A
In this study, we tested Hypothesis 3: that providing people with a
want (in this case, a free audiobook) that can be easily bundled with a
should behavior they are seeking to boost (in this case, exercise) leaks
information that the want ought to be paired with the should to make the
should more attractive. We also explored alternative explanations for
why providing a free audiobook to StepUp participants might have
increased gym visits.
Methods: At an East Coast university that includes gym membership
as part of its tuition and fees, 304 students with university gym mem-
berships completed an online survey in exchange for $5 and a potential
$1 bonus, both paid via Amazon gift card.
Procedure. This experiment was a three-condition scenario study. In
all conditions, participants rst read a brief description of the StepUp
Program. They were then told that after learning more about StepUp,
they would be asked to make predictions about how participants reacted
to the program.
Survey participants were randomly assigned to view screenshots of
the actual StepUp registration content viewed by participants in one of
Study 1’s three experimental conditions: the simple free audiobook con-
dition, the free audiobook with encouragement to temptation bundle con-
dition, or the control condition. Participants received a $1 bonus if they
correctly answered comprehension check questions about the content
they viewed.
After viewing StepUp screenshots and completing comprehension
check questions, participants predicted how many days a week they
thought StepUp participants visited the gym during the StepUp Pro-
gram. Then, participants rated the likelihood that StepUp participants
visited the gym because they (1) “wanted to listen to an audiobook while
working out at the gym,” (2) “were reminded to visit the gym whenever
they listened to an audiobook,” (3) “felt grateful to the gym,” (4) “felt
positively about the gym,” or (5) “wanted to earn money for going to the
gym.” Each rating was collected on a 5-point scale (from “Very Unlikely”
to “Very Likely”).
Finally, participants indicated how frequently they exercised and, for
both free audiobook conditions, if and where they would listen to their
free audiobook if they were in the StepUp Program. Participants also
answered optional questions about their age, gender, and ethnicity.
Complete survey materials and screenshots are available in the Online
Supplement on pages 18–35.
Results and discussion
Ninety-eight percent of survey-takers passed our comprehension
check.
11
To assess the possibility that gratitude or positive feelings to-
ward the gym generated by simply receiving a free audiobook would
increase gym members’ exercise, we compared survey participants’
ratings of these potential motivators across our control and simple free
audiobook conditions. As depicted in Fig. 6, we did not nd any evidence
that gym members thought the receipt of a simple free audiobook in the
StepUp program would create gratitude or positive feelings that would
boost exercise over and above the control condition (all p’s from pairwise
comparisons >0.35).
However, as depicted in Fig. 6, and consistent with Hypothesis 3,
12
we nd participants did anticipate that receiving a simple free audiobook
in the StepUp program would cause gym members to temptation bundle
more (Δ =0.398; t(204) =2.55, p =.011) and remind them to attend the
gym more (Δ =0.523; t(204) =3.19, p =.002), which would both, in
turn, boost their exercise. The free audiobook with encouragement to
temptation bundle condition generally moved in parallel with the simple
free audiobook condition, though participants expected gym members to
temptation bundle more (Δ =0.427; t(197) =2.82, p =.005) and think
about the gym more (Δ =0.459; t(197) =2.79, p =.006) in the free
audiobook with encouragement to temptation bundle condition than in the
simple free audiobook condition. These results are consistent with the
larger long-term boost in weekly gym visits in the free audiobook with
encouragement to temptation bundle condition relative to the simple free
audiobook condition detected in Study 1A and consistent with Hypoth-
esis 1. Pairwise comparisons between all experimental conditions are
reported in Online Supplement Table S3 and depicted in Fig. 6.
3.2. Study 2B
Study 2A demonstrated that gym members inferred StepUp partici-
pants were likely motivated to visit the gym in order to temptation
bundle their audiobook with their workouts, consistent with Hypothesis
3. In Study 2B, we build on this nding to assess whether this inference
was driven by the fact that the entity providing the want item for
bundling with a should was also encouraging engagement in the should
behavior. An information leakage account would suggest that partici-
pants only infer they should temptation bundle if they receive their free
audiobook from an exercise-related context. Thus, testing whether
different inferences are made about the gift of an audiobook when its
provider is encouraging exercise versus some unrelated behavior can
offer further insight into the validity of Hypothesis 3.
Methods. Participants included 602 Prolic users who completed a
preregistered online experiment in exchange for $0.50.
13
Procedure. This experiment was a two-condition scenario study.
Participants were randomly assigned to one of two conditions: the ex-
ercise program condition and the reading program condition. In the exer-
cise program condition, participants were instructed to imagine they
joined StepUp, a free 28-day workout rewards program designed by
scientists to help people build exercise habits. Participants in the reading
program condition were asked to imagine they joined ReadUp, a free 28-
day virtual book club designed by scientists to help people build reading
habits. In both conditions, participants were told they received a free
audiobook when they signed up for their 28-day program.
After reading these instructions and completing comprehension
check questions, participants were asked to indicate the likelihood that
they would listen to their free audiobook while engaging in a series of
different activities on a scale from -3 (“very unlikely”) to 3 (“very
likely”). The activities were presented in random order and included
exercising, taking a break from work, commuting, running errands,
doing chores, other and “whenever, no preferences.” Participants then
indicated when they thought program designers most hoped they would
listen to their free audiobook, selecting one option from the aforemen-
tioned list of choices.
Participants also answered optional questions about their age,
gender, and ethnicity at the end of the survey. Complete survey mate-
rials and screenshots are available in the Online Supplement on pages
36–40.
Results and discussion
Following our preregistration, we analyzed whether participants’
reported likelihood of listening to the free audiobook while exercising
differed across the two conditions. Participants in the exercise program
11
We include the data from all survey takers, including those who failed the
comprehension check, in our main analyses. Our results do not change when we
exclude the six participants who failed the comprehension check.
12
Note that this hypothesis to explain our Study 1B results was generated ex
post rather than ex ante.
13
This study was preregistered on AsPredicted.org: https://aspredicted.org/p
g8g3.pdf.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
13
condition were more likely to indicate that they would listen to their
audiobook while exercising than those in the reading program condition
(Δ =0.629; t(600) =3.50, p <.001). Furthermore, 66.2% of partici-
pants in the exercise program condition indicated that program designers
most hoped they would listen to their free audiobook while exercising,
while only 2.64% made this choice in the reading program condition (z
(600) =16.35, p <.001).
We also tested whether likelihood of listening to the audiobook while
exercising was mediated by inferences about program designers’ in-
tentions (following our preregistration). First, we documented a signif-
icant main effect of assignment to the exercise program condition on
participants’ perception that program designers intended for the
audiobook to be used while exercising (b =0.640, SE =0.028, p <.001).
The relationship between indicating that the audiobook was intended
for use while exercising and participants’ reported likelihood of
listening to the audiobook while exercising was also signicant (b =
0.926, SE =0.254, p <.001). Consistent with our mediation hypothesis,
the effect of assignment to the exercise program condition on study par-
ticipants’ likelihood of using the audiobook while exercising (b =0.630,
SE =0.180, p <.001) was eliminated when controlling for participants’
inferences about whether the audiobook was meant to be used while
exercising (b =0.037, SE =0.241, p =.877). A Sobel test conrmed that
this reduction in effect size was signicant (b =0.593, SE =0.165, p <
.001), and a 5,000-sample bootstrap analysis (MacKinnon, Fairchild, &
Fritz, 2007; Shrout & Bolger, 2002) also produced a 95% bias-corrected
condence interval for the size of the indirect effect that excluded zero
(95% CI [0.287, 0.919]).
Consistent with an information leakage account and Hypothesis 3,
these results demonstrate that participants infer they are meant to use
their free audiobook while exercising when the free audiobook is pro-
vided in the context of an exercise program, but not when it is provided
in a context unrelated to exercise. The supplier of the audiobook leaked
information about its intended purpose, which may help explain why
participants in our eld experiment who received a free audiobook (but
no encouragement to temptation bundle) from a program designed to
boost their exercise increased their gym attendance.
4. General discussion
We provide evidence that people can successfully self-impose
temptation bundling as a behavior change strategy, bundling wants
with shoulds to increase their adherence to should behaviors. In one of
the largest-ever eld experiments studying exercise frequency, we nd
that teaching people about temptation bundling and providing them
with unrestricted access to tempting content can durably increase their
gym visits. Specically, pairing a free audiobook with encouragement to
temptation bundle increased the likelihood of a weekly gym visit by
10–14% and average weekly gym visits by 10–12% during and for up to
seventeen weeks after a four-week intervention period.
We also provide suggestive evidence that people can infer and apply
behavior change strategies on their own when given a gift that leaks
information about its intended use. This may explain why encouraging
temptation bundling and providing a free audiobook only had modest
benets over and above giving our study participants a free audiobook
with no encouragement to bundle. If it was obvious that the gift of a free
audiobook from their gym meant temptation bundling was encouraged,
then explicit encouragement to bundle would be of little value. Indeed,
simply giving gym members free audiobooks boosted gym visits by 11%
during the four-week intervention and by 18% for four weeks post-
intervention. Follow-up lab studies provide support for our informa-
tion leakage account, demonstrating that participants likely inferred
they should temptation bundle and helping to rule out alternative ex-
planations for the benets of giving away free audiobook (e.g., reci-
procity or warm glow). Our ndings suggest both that people can be
highly sophisticated in pursuing behavior change and that extremely
low-touch interventions can successfully promote change. In fact, we
nd evidence that a well-timed gift can be enough to help people
improve their habits.
Overall, our ndings suggest that practitioners who wish to
encourage the adoption of should behaviors may be able to effect change
by providing bundle-ready wants in the right context and encouraging
temptation bundling. Importantly, the effectiveness of this strategy
seems to be remarkably robust across various subpopulations, suggest-
ing that many people can benet from learning how to temptation
bundle and receiving a bundle-ready want. Finally, it’s likely that the
more accessible practitioners can make bundle-ready wants to partici-
pants, the better.
It’s worth noting that our ndings with respect to Hypothesis 2 are
consistent with those of Milkman and colleagues’ 2014 temptation
bundling study. In one experimental condition, Milkman et al. (2014)
Fig. 6. Study 2A participants’ ratings of the likelihood that each of four possible motivators encouraged participant gym visits during the StepUp Program by
experimental condition. The error bars depict 95% condence intervals.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
14
provided participants with iPods equipped with free audiobooks (so no
initiative was required on the part of participants to obtain the books),
monitored an initial workout while participants temptation bundled,
explained temptation bundling in a live laboratory session, and pro-
duced an initial (marginal) 0.27 extra gym visits per week relative to a
control group that was given a gift certicate to Barnes & Noble (not an
audiobook) and no explanation of temptation bundling. Meanwhile, our
intervention was fully digital and provided participants with a code to
download a free audiobook from Audible.com on their own. Our
temptation bundling intervention induced an extra 0.13–0.16 gym visits
per week compared with a control group given no instructions about
temptation bundling and no free audiobook. The fact that our digital
intervention was roughly half as effective as this face-to-face interven-
tion where 100% of participants received audiobooks seems reasonable.
Interestingly, the benets of our digital intervention were far longer-
lasting.
14
The benets of encouraging people to temptation bundle that we
document are consistent with the literature suggesting some people have
the capacity to set and stick to personal rules even without employing
commitment devices (Ainslie, 1992; Ainslie, 2001; Thaler, 2000; Wer-
tenbroch, 1998). We demonstrate that teaching people about a strategy
intended to help them avoid the pitfalls of temptation and fulll their
long-term goals can produce lasting behavior change. These ndings
add support for the theory that people who are sophisticated about their
present bias can self-impose rules or strategies to help them exert self-
control.
Among our most intriguing ndings is the suggestive evidence that
explicit instructions about how temptation bundling works may not be
necessary when the technique is suggested implicitly. Our results indi-
cate that providing a tempting want in the context of a program designed
to increase a should behavior (i.e., a gym handing out free audiobooks)
may be sufcient to inspire temptation bundling. This nding aligns
with and builds on prior work on information leakage (Tannenbaum
et al., 2013; McKenzie et al., 2006; McKenzie & Nelson, 2003; Sher &
McKenzie, 2006), suggesting that goal-seekers can be sophisticated
about gleaning useful recommendations from their environments.
Future work conrming this hypothesized mechanism for Study 1B’s
ndings would be valuable. The design of our eld experiment made it
impossible to verify whether participants who received a free audiobook
actually used it to temptation bundle, although comparable download
rates across experimental conditions are suggestive of similar usage
patterns whether or not bundling was encouraged. We also could not
fully rule out gratitude or warm glow effects as drivers of gym visits in
the free audiobook conditions without a control condition where par-
ticipants received a cash gift of equal monetary value to a free audio-
book. While the results of Studies 2A and 2B suggest our eld
experiment participants likely inferred they should use the free audio-
book at their gym, further conrmation would be useful.
Future work might also explore why teaching participants about
temptation bundling appears particularly benecial in the long run. We
speculate that participants who were taught about temptation bundling
could apply this concept more generally than those who were only
provided with a free audiobook. Teaching participants how to tempta-
tion bundle may have more rmly cemented the idea of pairing exercise
with temptations than merely receiving a free audiobook from an ex-
ercise program (Hertwig & Grüne-Yanoff, 2017). Furthermore, partici-
pants who were taught how to temptation bundle may have enacted a
broader and more exible personal rule, which produced stickier
behavior change (Beshears et al., 2020; Mollick & Rothbard, 2014;
Woolley & Fishbach, 2016). Because participants taught to temptation
bundle were encouraged to bundle whatever media they most preferred
(e.g., TV shows, podcasts) with workouts, they may have bundled ex-
ercise with their favorite want, not only with audiobooks. In contrast,
participants who simply received a free audiobook might have formed a
narrower personal rule which only pertained to pairing the gifted
audiobook with exercise. These participants might not have considered
how to make temptation bundling work best for them personally (e.g.,
trying out an enjoyable podcast instead of another audiobook) or
repeatedly (e.g., they may not have thought to download a new audio-
book after the rst concluded).
15
Although a strength of our research is the focus on an objective
outcome variable measured unobtrusively over many months’ time, this
measure is nonetheless imperfect. We could not, for instance, measure
the impact of our interventions on exercise quality or intensity. It is
possible that temptation bundling could be a distraction from exercise,
decrease the intensity of exercise, or crowd out certain forms of exercise
in favor of those that are more easily bundled with a temptation. On the
other hand, prior work suggests that people persist more on activities
they nd fun (Woolley & Fishbach, 2016), so temptation bundling may
have increased the quality and/or intensity of exercise by making it
more enjoyable. Future research should explore how temptation
bundling inuences the quality and intensity of engagement in the target
should activity.
CRediT authorship contribution statement
Erika L. Kirgios: Conceptualization, Methodology, Formal analysis,
Investigation, Writing - original draft, Writing - review & editing.
Graelin H. Mandel: Methodology, Investigation, Formal analysis,
Writing - original draft, Writing - review & editing, Visualization. Yeji
Park: Methodology, Formal analysis, Investigation, Writing - review &
editing, Visualization. Katherine L. Milkman: Conceptualization,
Methodology, Investigation, Writing - review & editing, Supervision,
Resources. Dena M. Gromet: Conceptualization, Methodology, Inves-
tigation, Writing - review & editing. Joseph S. Kay: Conceptualization,
Methodology, Investigation, Writing - review & editing. Angela L.
Duckworth: Conceptualization, Methodology, Investigation, Writing -
review & editing, Resources.
Acknowledgements
We thank 24 Hour Fitness and Audible for partnering with the
Behavior Change for Good Initiative at the University of Pennsylvania to
make this research possible. We thank the members of the Duckworth-
Milkman Lab Meeting for their insightful feedback on this work. We
are grateful to Kasandra Brabaw and Michelle Shih for their editorial
assistance on the manuscript, Meghan Chung, Karen Herrera, and Jor-
dyn Schor for their research assistance, and to the Wharton Behavioral
Lab for their assistance in gathering data for this research.
Funding
Support for this research was provided in part by the Robert Wood
Johnson Foundation, the AKO Foundation, John Alexander, Mark J.
Leder, and Warren G. Lichtenstein. This material is based upon work
supported by the National Science Foundation Graduate Research
14
Our effects were likely long-lasting in part because the original temptation
bundling study was interrupted by a gym closure for Thanksgiving break. But
24 Hour Fitness gyms rarely close. Furthermore, our participants were recruited
over a roughly one-year period with staggered start dates. Thus, any disruptions
to participants’ schedules or to their consumption of tempting audiobooks
induced by holidays would have been smoothed across participants.
15
This could also help explain why encouraging temptation bundling and
providing a free audiobook produced far more durable benets in Study 2B
than in the 2014 study by Milkman et al. Participants in the Milkman et al.
(2014) study of temptation bundling were specically instructed to listen to
their free audiobook while visiting the gym, so they might not have formed a
exible and adaptable temptation bundling habit.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
15
Fellowship under Grant No. DGE-1845298. The views expressed here do
not necessarily reect the views of these entities. Audible provided free
audiobooks for the eld experiment.
Declaration of Competing Interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
References
2018 Physical Activity Guidelines Advisory Committee Scientic Report A-4. (2018).
Retrieved from https://health.gov/paguidelines/second-edition/report/pdf/PAG_
Advisory_Committee_Report.pdf.
Acland, D., & Levy, M. R. (2015). Naivet´
e, projection bias, and habit formation in gym
attendance. Management Science, 61(1), 146–160. https://doi.org/10.1287/
mnsc.2014.2091.
Ainslie, G. (1992). Studies in rationality and social change. Picoeconomics: The strategic
interaction of successive motivational states within the person. Cambridge, England:
Cambridge University Press.
Ainslie, G. (2001). Breakdown of will. Cambridge, England: Cambridge University Press.
Andreoni, J. (1995). Cooperation in public-goods experiments: Kindness or confusion?
American Economic Review, 85(4), 891–904.
Ariely, D., & Wertenbroch, K. (2002). Procrastination, deadlines, and performance: Self-
control by precommitment. Psychological science, 13(3), 219–224.
Armenta, C. N., Fritz, M. M., & Lyubomirsky, S. (2017). Functions of positive emotions:
Gratitude as a motivator of self-improvement and positive change. Emotion Review, 9
(3), 183–190. https://doi.org/10.1177/1754073916669596.
Ashraf, N., Karlan, D., & Yin, W. (2006). Tying Odysseus to the mast: Evidence from a
commitment savings product in the Philippines. The Quarterly Journal of Economics,
121(2), 635–672.
Babcock, P., Bedard, K., Charness, G., Hartman, J., & Royer, H. (2015). Letting down the
team? Social effects of team incentives. Journal of the European Economic Association,
13(5), 841–870. https://doi.org/10.1111/jeea.12131.
Ballard, Jamie. Exercising more and saving money are the most popular 2020 new year’s
resolutions. YouGov, YouGov, 2 Jan. 2020, today.yougov.com/topics/lifestyle/artic
les-reports/2020/01/02/new-years-resolutions-2020-health-nance.
Bazerman, M. H., Tenbrunsel, A. E., & Wade-Benzoni, K. (1998). Negotiating with
yourself and losing: Making decisions with competing internal preferences. Academy
of Management Review, 23(2), 225–241.
B´
enabou, R., & Tirole, J. (2004). Willpower and personal rules. The American Economic
Review, 112(4), 848–886.
Berger, J., & Milkman, K. L. (2012). What makes online content viral? Journal of
Marketing Research, 49(2), 192–205.
Beshears, J., Lee, H. N., Milkman, K. L., Mislavsky, R. & Wisdom, J. (2020). Creating
exercise habits using incentives: The tradeoff between exibility and routinization.
Manuscript in preparation.
Boals, A., Vandellen, M. R., & Banks, J. B. (2011). The relationship between self-control
and health: The mediating effect of avoidant coping. Psychology & Health, 26(8),
1049–1062. https://doi.org/10.1080/08870446.2010.529139.
Bordalo, P., Gennaioli, N., & Shleifer, A. (2013). Salience and consumer choice. Journal of
Political Economy, 121(5), 803–843. https://doi.org/10.1086/673885.
Bryan, G., Karlan, D., & Nelson, S. (2010). Commitment devices. Annual Review of
Economics, 2(1), 671–698.
Charness, G., & Gneezy, U. (2009). Incentives to exercise. Econometrica, 77(3), 909–931.
Cheema, A., & Soman, D. (2006). Malleable mental accounting: The effect of
exibility on the justication of attractive spending and consumption decisions.
Journal of Consumer Psychology, 16(1), 33–44. https://doi.org/10.1207/
s15327663jcp1601_6.
Chetty, R., Looney, A., & Kroft, K. (2009). Salience and taxation: Theory and evidence.
The American Economic Review, 99(4), 1145–1177.
Condliffe, S., Is¸gın, E., & Fitzgerald, B. (2017). Get thee to the gym! A eld experiment on
improving exercise habits. Journal of Behavioral and Experimental Economics, 70,
23–32. https://doi.org/10.1016/j.socec.2017.07.007.
DellaVigna, S., & Malmendier, U. (2006). Paying not to go to the gym. American Economic
Review, 96(3), 694–719.
Dohmen, T., Falk, A., Huffman, D., & Sunde, U. (2006). Seemingly irrelevant events
affect economic perceptions and expectations: The FIFA World Cup 2006 as a natural
experiment. https://ssrn.com/abstract=928830.
Drexler, A., Fischer, G., & Schoar, A. (2014). Keeping it simple: Financial literacy and
rules of thumb. American Economic Journal: Applied Economics, 6(2), 1–31. https://
doi.org/10.1257/app.6.2.1.
Duckworth, A. L., White, R. E., Matteucci, A. J., Shearer, A., & Gross, J. J. (2016).
A stitch in time: Strategic self-control in high school and college students. Journal
of Educational Psychology, 108(3), 329–341. https://doi.org/10.1037/
edu0000062.
Duckworth, A. L., Milkman, K. L., & Laibson, D. (2018). Beyond willpower: Strategies for
reducing failures of self-control. Psychological Science in the Public Interest, 19(3),
102–129. https://doi.org/10.1177/1529100618821893.
Falk, A. (2007). Gift exchange in the eld. Econometrica, 75(5), 1501–1511. https://doi.
org/10.1111/j.1468-0262.2007.00800.x.
Finkelstein, A. (2009). E-ZTAX: Tax salience and tax rates. Quarterly Journal of Economics,
124(3), 969–1010. https://doi.org/10.1162/qjec.2009.124.3.969.
Grady, K. E., Goodenow, C., & Borkin, J. R. (1988). The effect of reward on compliance
with breast self-examination. Journal of Behavioral Medicine, 11(1), 43–57. https://
doi.org/10.1007/bf00846168.
Guthold, R., Stevens, G. A., Riley, L. M., & Bull, F. C. (2018). Worldwide trends in
insufcient physical activity from 2001 to 2016: A pooled analysis of 358
population-based surveys with 1⋅9 million participants. The Lancet Global Health, 6
(10). https://doi.org/10.1016/s2214-109x(18)30357-7.
Hertwig, R., & Grüne-Yanoff, T. (2017). Nudging and boosting: Steering or empowering
good decisions. Perspectives on Psychological Science, 12(6), 973–986. https://doi.
org/10.1177/1745691617702496.
Hilton, D. J. (1995). The social context of reasoning: Conversational inference and
rational judgment. Psychological Bulletin, 118(2), 248–271. https://doi.org/10.1037/
0033-2909.118.2.248.
Isen, A. M., Shalker, T. E., Clark, M., & Karp, L. (1978). Affect, accessibility of material in
memory, and behavior: A cognitive loop? Journal of Personality and Social Psychology,
36(1), 1–12. https://doi.org/10.1037/0022-3514.36.1.1.
Kaur, S., Kremer, M., & Mullainathan, S. (2010). Self-control and the development of
work arrangements. American Economic Review, 100(2), 624–628. https://doi.org/
10.1257/aer.100.2.624.
Kay, A. C., Wheeler, S. C., Bargh, J. A., & Ross, L. (2004). Material priming: The inuence
of mundane physical objects on situational construal and competitive behavioral
choice. Organizational behavior and human decision processes, 95(1), 83–96.
Laibson, D. (1997). Golden eggs and hyperbolic discounting. The Quarterly Journal of
Economics, 112(2), 443–478.
Lee, I.-M., Shiroma, E. J., Lobelo, F., Puska, P., Blair, S. N., & Katzmarzyk, P. T. (2012).
Effect of physical inactivity on major non-communicable diseases worldwide: An
analysis of burden of disease and life expectancy. The Lancet, 380(9838), 219–229.
https://doi.org/10.1016/s0140-6736(12)61031-9.
List, J. A., & Lucking-Reiley, D. (2002). The effects of seed money and refunds on
charitable giving: Experimental evidence from a university capital campaign. Journal
of Political Economy, 110(1), 215–233. https://doi.org/10.1086/324392.
McKenzie, C. R., Liersch, M. J., & Finkelstein, S. R. (2006). Recommendations implicit in
policy defaults. Psychological Science, 17(5), 414–420. https://doi.org/10.1111/
j.1467-9280.2006.01721.x.
McKenzie, C. R., & Nelson, J. D. (2003). What a speaker’s choice of frame reveals:
Reference points, frame selection, and framing effects. Psychonomic Bulletin &
Review, 10(3), 596–602.
Milkman, K. L., Gromet, D., Ho, H., Kay, J. S., Lee, T., Pandilosky … Duckworth, A. L.
(2020). A mega-study approach to evaluating interventions. Working paper.
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. https://doi.org/10.1287/mnsc.2013.1784.
Milkman, K. L., Rogers, T., & Bazerman, M. H. (2008). Harnessing our inner angels and
demons: What we have learned about want/should conicts and how that
knowledge can help us reduce short-sighted decision making. Perspectives on
Psychological Science, 3(4), 324–338.
Milkman, K. L., Rogers, T., & Bazerman, M. H. (2009). Highbrow lms gather dust: Time-
inconsistent preferences and online DVD rentals. Management Science, 55(6), 1047–1059.
Mollick, E. R., & Rothbard, N. (2014). Mandatory fun: Gamication and the impact of
games at work. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2277103.
Morton, F. S., Zettelmeyer, F., & Silva-Risso, J. (2003). Consumer information and
discrimination: Does the internet affect the pricing of new cars to women and
minorities? Quantitative Marketing and Economics, 1(1), 65–92.
O’Donoghue, T., & Rabin, M. (1999). Doing it now or later. American Economic Review, 89
(1), 103–124.
O’Donoghue, T., & Rabin, M. (2011). Doing it now or later. In: C. F. Camerer, G.
Loewenstein, & M. Rabin (Eds.), Advances in behavioral economics (pp. 223–251).
https://doi.org/10.2307/j.ctvcm4j8j.12.
Patel, M. S., Benjamin, E. J., Volpp, K. G., Fox, C. S., Small, D. S., Massaro, J. M., …
Murabito, J. M. (2017). Effect of a game-based intervention designed to enhance
social incentives to increase physical activity among families. JAMA Internal
Medicine, 177(11), 1586–1593. https://doi.org/10.1001/
jamainternmed.2017.3458.
Patel, M. S., Small, D. S., Harrison, J. D., Fortunato, M. P., Oon, A. L., Rareshide, C. A. L.,
… Hilbert, V. (2019). Effectiveness of behaviorally designed gamication
interventions with social incentives for increasing physical activity among
overweight and obese adults across the United States. JAMA Internal Medicine, 179
(12), 1624–1632. https://doi.org/10.1001/jamainternmed.2019.3505.
Prinsen, S., de Ridder, D. T., & de Vet, E. (2013). Eating by example. Effects of
environmental cues on dietary decisions. Appetite, 70, 1–5.
Royer, H., Stehr, M., & Sydnor, J. (2015). Incentives, commitments, and habit formation
in exercise: Evidence from a eld experiment with workers at a Fortune-500
company. American Economic Journal: Applied Economics, 7(3), 51–84.
Schroeder, S. A. (2007). We can do better—Improving the health of the American people.
New England Journal of Medicine, 357(12), 1221–1228. https://doi.org/10.1056/
nejmsa073350.
Sedlmeier, P., & Gigerenzer, G. (2001). Teaching Bayesian reasoning in less than two
hours. Journal of Experimental Psychology: General, 130(3), 380–400. https://doi.org/
10.1037/0096-3445.130.3.380.
Sher, S., & McKenzie, C. R. (2006). Information leakage from logically equivalent frames.
Cognition, 101(3), 467–494.
Stavrova, O., & Kokkoris, M. D. (2017). Struggling to be liked: The prospective effect of trait
self-control on social desirability and the moderating role of agreeableness. International
Journal of Psychology, 54(2), 232–236. https://doi.org/10.1002/ijop.12444.
E.L. Kirgios et al.
Organizational Behavior and Human Decision Processes xxx (xxxx) xxx
16
Strack, F., Martin, L. L., & Schwarz, N. (1988). Priming and communication: Social
determinants of information use in judgments of life satisfaction. European Journal of
Social Psychology, 18(5), 429–442.
Str¨
omb¨
ack, C., Lind, T., Skagerlund, K., V¨
astfj¨
all, D., & Tingh¨
og, G. (2017). Does self-
control predict nancial behavior and nancial well-being? Journal of Behavioral and
Experimental Finance, 14, 30–38. https://doi.org/10.1016/j.jbef.2017.04.002.
Tannenbaum, D., Valasek, C. J., Knowles, E. D., & Ditto, P. H. (2013). Incentivizing
wellness in the workplace. Psychological Science, 24(8), 1512–1522. https://doi.org/
10.1177/0956797612474471.
Thaler, R. H. (2000). Mental accounting matters. Journal of Behavioral Decision Making,
12(3), 183–206.
Thaler, R. H., & Benartzi, S. (2004). Save More Tomorrow™: Using behavioral economics
to increase employee saving. Journal of Political Economy, 112(S1). https://doi.org/
10.1086/380085.
Wansink, B., & Cheney, M. M. (2005). Super bowls: serving bowl size and food
consumption. Jama, 293(14), 1727–1728.
Wertenbroch, K. (1998). Consumption self-control by rationing purchase quantities of
virtue and vice. Marketing Science, 17(4), 317–337. https://doi.org/10.1287/
mksc.17.4.317.
Woolley, K., & Fishbach, A. (2016). For the fun of it: Harnessing immediate rewards to
increase persistence in long-term goals. Journal of Consumer Research, 42(6),
952–966. https://doi.org/10.1093/jcr/ucv098.
Zuckerman, O., & Gal-Oz, A. (2014). Deconstructing gamication: Evaluating the
effectiveness of continuous measurement, virtual rewards, and social comparison for
promoting physical activity. Personal and Ubiquitous Computing, 18(7), 1705–1719.
https://doi.org/10.1007/s00779-014-0783-2.
E.L. Kirgios et al.
