ArticlePDF Available

Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems: Evidence-Based Interventions and Theory

Authors:

Abstract and Figures

We present a series of mHealth applications and studies pursued as part of the Fittle+ project. This program of research has the dual aims of (1) bringing scalable evidence-based behavior-change interventions to mHealth and evaluating them and (2) developing theoretically based predictive models to better understand the dynamics of the impact of these interventions on achieving behavior-change goals. Our approach in the Fittle+ systems rests on the idea that to master the complex fabric of a new healthy lifestyle, one must weave together a new set of healthy habits that over-ride the old unhealthy habits. To achieve these aims, we have developed a series of mHealth platforms that provide scaffolding interventions: Behavior-change techniques and associated mHealth interactions (e.g., SMS reminders; chatbot dialogs; user interface functionality; etc.) that provide additional support to the acquisition and maintenance of healthy habits. We present experimental evidence collected so far for statistically significant improvements in behavior change in eating, exercise, and physical activity for the following scaffolding interventions: guided mastery, teaming, self-affirmation, and implementation intentions. We also present predictive computational ACT-R models of daily individual behavior goal success for data collected in guided mastery and implementation intention studies that address goal-striving and habit formation mechanisms.
Content may be subject to copyright.
Scaffolding the Mastery of Healthy Behaviors
with Fittle+ Systems: Evidence-Based
Interventions and Theory
Peter Pirolli,
1
G. Michael Youngblood,
2
Honglu Du,
3
Artie Konrad,
4
5Les Nelson,
2
and Aaron Springer
5
1
Institute for Human and Machine Cognition
2
Palo Alto Research Center
3
Ant Financial
4
Facebook
5
10University of California, Santa Cruz
We present a series of mHealth applications and studies pursued as part
of the Fittle+ project. This program of research has the dual aims of (1)
bringing scalable evidence-based behavior-change interventions to
mHealth and evaluating them and (2) developing theoretically based
15predictive models to better understand the dynamics of the impact of
these interventions on achieving behavior-change goals. Our approach in
the Fittle+ systems rests on the idea that to master the complex fabric of
a new healthy lifestyle, one must weave together a new set of healthy
habits that over-ride the old unhealthy habits. To achieve these aims, we
20have developed a series of mHealth platforms that provide scaffolding
interventions: Behavior-change techniques and associated mHealth interac-
tions (e.g., SMS reminders; chatbot dialogs; user interface functionality;
etc.) that provide additional support to the acquisition and maintenance
of healthy habits. We present experimental evidence collected so far for
25statistically significant improvements in behavior change in eating,
Peter Pirolli (ppirolli@ihmc.us), is a cognitive psychologist; he is a Senior Research Scientist at the
Institute for Human and Machine Cognition. G. Michael Youngblood (michael.youngblood@parc.
com) is a computer scientist; he is a Principal Scientist at the Palo Alto Research Center. Honglu Du
(honglu.dhl@antfin.com) is a computer scientist; he is Staff User Research at the Ant Financial
Services Group. Artie Konrad (akonrad@fb.com) is a mixed methods researcher; he is a UX
Researcher at Facebook. Les Nelson (nelson@parc.com) is a mixed methods researcher; he is a Senior
Researcher at the Palo Alto Research Center. Aaron Springer (alspring@ucsc.edu) is a psychology
doctoral student at the University of California, Santa cruz.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.
com/hhci.
HUMANCOMPUTER INTERACTION, 2018, Volume 00, pp. 134
Copyright © 2018 Taylor & Francis Group, LLC
ISSN: 0737-0024 print / 1532-7051 online
DOI: https://doi.org/10.1080/07370024.2018.1512414
1
exercise, and physical activity for the following scaffolding interventions:
guided mastery, teaming, self-affirmation, and implementation inten-
tions. We also present predictive computational ACT-R models of
daily individual behavior goal success for data collected in guided mas-
30tery and implementation intention studies that address goal-striving and
habit formation mechanisms.
1. INTRODUCTION
With the recent thrust to develop precision medicine there are opportunities to
apply social and behavioral science to the challenges of helping people develop
35healthier lifestyles that include better diet, increased physical activity, better sleep,
and greater resilience to stress, etc. (Riley, Nilsen, Manolio, Masys, & Lauer, 2015).
As highlighted in Riley et al. (2015), 70% of healthcare costs are due to changeable
behavior (e.g., diet, fitness, smoking), and behavioral and environmental factors
account for more deaths than genetics. Mobile health (mHealth) systems including,
40more generally, pervasive technologies such as the Internet of Things, offer novel
ways for supporting people in changing their behavior in the actual ecology of their
everyday environments (Heron & Smyth, 2010). A recent comprehensive review of
mHealth (Silva, Rodrigues, De La Torre Díez, López-Coronado, & Saleem, 2015)
resolved that mHealth services and applications are already playing a very important
45and determinant role in restructuring the old healthcare services and systems that
are still based on the physical relationship between patient and physician. mHealth
provides a path for translating evidence-based interventions (EBIs) onto delivery
systems that are replicable, scalable, and sustainable, with great economies of scale
for healthcare delivery (Rotheram-Borus, Swendeman, & Chorpita, 2012). For the
50field of human-computer interaction, and the social and behavioral sciences more
generally, mHealth also provides new opportunities to develop theory about inter-
ventions in the ecology of everyday life, with a focus on meaningful behavior
(Baumeister, Vohs, & Funder, 2007; Harari et al., 2016).
In this article, we present a series of mHealth applications and studies pursued
55as part of the Fittle+ project. This program of research has the dual aims of (1)
bringing scalable evidence-based behavior-change interventions to mHealth and
evaluating them (Rotheram-Borus et al., 2012) and (2) developing theoretically
based predictive models to better understand the dynamics of the impact of these
interventions on achieving behavior-change goals (Spruijt-Metz et al., 2015). The
60empirical studies and predictive models have been presented previously in disparate
sources. Here we present, for the first time, the broader unifying concept of
scaffolding interventions: Behavior-change techniques and associated mHealth interac-
tions (e.g., SMS reminders; chatbot dialogs; user interface functionality; etc.) that
provide additional support to the acquisition and maintenance of healthy habits. The
65general programmatic approach we have pursued involves selecting scaffolding
2P. Pirolli et al.
interventions from the evidence base on behavior change techniques, refining these
for mobile health delivery, and understanding their dynamic effects on behavior
change using computational cognitive models developed in ACT-R (Adaptive Con-
trol of Thought-Rational; Anderson, 2007).
701.1. Overview of Fittle Functionality
We use the term Fittle+ for a project that has explored several systems that
evolved as variations of the Fittle mobile phone application (Du, Youngblood, &
Pirolli, 2014). As background for our studies, it is useful to understand the core
functionality of Fittle. Figure 1 shows screen images from that initial Fittle system
75presented in Du et al. (2014). Users selected for themselves, or were assigned, to
multi-week challenge program involving sets of related daily behavior-change goals.
A user could start, join, or be assigned to a team (Figure 1a). The primary Fittle
screen is the Fittle dashboard (Figure 1c). The Fittle dashboard consists of three
parts. The top portion shows the daily behavior-change goal icons with their
80completion status shown below as a set of circles similar to a horizontal traffic
light. Tap-selection of the goal icons (as shown in Figure 1c under the blue-backed
area) accesses the title, basic reminder details, substitution tasks, detailed informa-
tion about the goal in an electronic card (which may include video demonstration,
images, detailed instructions, substitution suggestions, background information,
85external reference links, links to other related Fittle cards, and more), and the ability
to self-report completion or submit a multi-media post to the team feed. The middle
section of the dashboard in Figure 1c provides visual analytics showing the users
and the teams goal accomplishment this week. The weekly goal set view, as
illustrated in Figure 1d, shows the user all of the behavior-change goals that Fittle
90will schedule the week. The user can interact and share with his or her team
(Figure 1e). Users can share information and multi-media (e.g., photos) with the
FIGURE 1. Fittle® Mobile Application(A) Teams Available, (B) the Details of a Team, (C)
Activity Information, (D) Overall Goals for This Week, and (E) the Team-Based Social
Activity Feed.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 3
team. Users can give high fives to other users in the same team and comment on
each others posts. All the posts, comments, and high fives are public information in
the team. Users may also communicate directly with each other through a peer-to-
95peer messaging system.
2. THEORETICAL BACKGROUND
2.1. From Good Intentions to Healthy Habits
Our approach in the Fittle+ systems rests on the idea that to master the
complex fabric of a new healthy lifestyle, one must weave together a new set of
healthy habits that over-ride the old unhealthy habits. Fogg (Fogg, 2009; Fogg &
100Hreha, 2010) has provided useful summaries and practice-oriented methods for
decomposing larger lifestyle changes into tiny habits,(behaviors) to be changed,
and for matching target behaviors with solutions for achieving those behaviors. The
path to developing healthy habits is a difficult one, and that path typically begins
with the adoption of intentions to change, the setting of goals, and repeated striving
105to achieve those goals (Graybiel, 2008; James, 1890; Wood & Neal, 2007).
2.2. Individual Behavior Change Theory and Interventions
A summary of the current state-of-the-art in behavior change theory and
techniques is well beyond the scope of this article. But a recent, multi-year effort to
produce such a summary (Michie et al., 2013; Michie, West, Campbell, Brown, &
110Gainforth, 2014) identified 83 theories, 26 mechanisms of action, 93 behavior change
techniques, and 1725 theoretical constructs. A recent meta-analysis (Samdal, Eide,
Barth, Williams, & Meland, 2017) of the literature on behavior change techniques
identified in Michie et al. (2014) summarizes the evidence on which techniques
produce reliable effects along with effect size estimates. As discussed below, our
115Fittle systems have incorporated five of the nine behavior change techniques asso-
ciated with long-term behavior change in diet and physical activity found in Samdal
et al. (2017; Table 3), as well as implementation intentions and self-affirmation
techniques that also have robust effects (Cohen & Sherman, 2014; Epton, Harris,
Kane, Van Koningsbruggen, & Sheeran, 2015; Gollwitzer & Sheeran, 2006).
120Individual-level health behavior theories (Brewer & Rimer, 2008) include the
Transtheoretical Model (Bridle et al., 2005), the Health Belief Model (Harrison,
Mullen, & Green, 1992), Goal Setting Theory (Locke & Latham, 2002), and the
Theory of Planned Behavior (TPB) (Ajzen, 1991,1998). Applied techniques, such as
altering the environment (Fogg, 2009; Fogg & Hreha, 2010; Sallis & Glanz, 2009;
125Wansink & Sobal, 2007) to no longer trigger old unhealthy habits or trigger healthier
habits, or psychological therapies such as Motivational Interviewing (Miller & Roll-
nick, 2013) and Cognitive Behavioral Therapy (Cooper, Fairburn, & Hawker, 2003)
build upon these theoretical foundations. None of these theories is specified as the
4P. Pirolli et al.
kind of fine-grained predictive and dynamic model of behavior and intervention
130effects that is necessary for engineering mHealth applications (Riley et al., 2011).
As a theoretical blueprint for supporting the goal-striving phase of habit
formation in Fittle+ applications, we were guided by the TPB (Ajzen, 1991). TPB
has been studied extensively (Brewer & Rimer, 2008) and meta-analyses support the
efficacy of the approach in predicting behavior at a coarse-grained level (Armitage &
135Conner, 2001). Meta-analyses also suggest that TPB is credited more often in
successful Internet-based health interventions than other theoretical models
(Webb, Joseph, Yardley, & Michie, 2010). The TPB (Figure 2) proposes that the
predictors of a person engaging in a target behavior include the persons intention to
do the desired behavior and their perceived control over the behaviorwhether the
140person perceives themselves as being in control of doing the target behavior. In
turn, the predictors of intention are attitudes, subjective norms, and (again) per-
ceived behavioral control. Attitudes are whether a person is in favor of doing the
behavior. Subjective norms are how much the person perceives social pressure to do
the behavior. Attitudes, subjective norms, and perceived behavioral control are all
145forms of expectancy-value judgments deriving from beliefs about outcomes, sig-
nificant referents, and specific facilitating/inhibiting factors, respectively.
2.3. Scaffolding Interventions
We propose the concept of scaffolding interventions: Behavior-change techniques
(e.g., Michie et al., 2013) and associated mHealth interactions (e.g., SMS reminders;
chatbot dialogs; user interface functionality; etc.) that provide additional support to
150the acquisition and maintenance of healthy habits. The Fittle+ systems have core
challenge programs containing daily goals that form a backbone of the Fittle
+ mHealth process, and the scaffolding interventions provide additional supports
FIGURE 2. The Theory of Planned Behavior.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 5
to that core. The term scaffolding is appropriated from the learning sciences, where
it is used to describe support techniques used during the learning process that are
155tailored to the needs of individual students with the intention of helping the student
achieve their learning goals in the context of performance (Reiser & Tabak, 2014).
In the learning sciences, scaffolding has been defined as software toolsthat
enable students to deal with more complex content and skill demands than they
could otherwise handle(Reiser, 2004, p. 273) with the intent that the support not
160only assists learners in accomplishing tasks but also enables them to learn from the
experience(Reiser, 2004, p. 275). Analogously, the Fittle+ systems provide scaf-
folding to support the building and strengthening of healthy habits by providing a
platform for intervention content, methods, and tools that are delivered in the
context of the complex ecology of everyday life. The notion of scaffolding is also
165related to the concepts of shaping, fading, and chaining in classic behavior mod-
ification (Martin & Pear, 2016).
Although our studies have typically focused on experimental studies of just a
few scaffolding interventions at a timein order to better understand themwe
believe the ultimate platform will have a smorgasbord of scaffolding interventions
170from which to select and tailor to each individual for maximum support and impact.
The meta-analysis in Samdal et al. (2017) of behavior change techniques indicates
that including more behavior-change techniques is more likely to produce larger
changes in physical activity and diet in both the short term (<12 weeks) and long
term (12 weeks). Figure 3 summarizes the scaffolding interventions explored in
175Fittle+ . Guided mastery, teaming, and self-affirmation interventions are all targeted
at strengthening behavior-change intentions. Implementation intentions (in addition
FIGURE 3. Interventions Studied in the Fittle+ Project.
6P. Pirolli et al.
to guided mastery) are targeted at translating intentions into actual behavior. The
interventions chosen so far for exploration in the Fittle+ project were selected
based on bodies of evidence supporting their efficacy, and to provide a set of
180interventions that targeted increasing intention-to-change as well as that translation
of intentions into behavior.
In our studies, we explored (Figure 3) guided mastery, teaming, self-affirma-
tion, and implementation intentions. For reference, the scaffolding interventions we
studied are mapped on to the Michie et al. (2013) Behavior Change Taxonomy in
Figure 4. The meta-analysis in Samdal et al. (Table 3, 2017) suggests that BCTs
1851.1 Goal-setting behavior, BCT 8.7 Graded tasks, BCT 1.5 Review behavior goals,
BCT 2.3 Self-monitoring, and BCT 3.1 Social support, have significant effects on
long-term outcomes. Meta-analysis (Epton et al., 2015) also suggests that self-
affirmations (BCT 13.4 Valued self-identity) have small but reliable effects on
intentions and behavior . The meta-analysis of Gollwitzer and Sheeran (2006)
190suggests medium to large effects of implementation intentions (BCT 1.4 Action
planning).
2.4. A Brief History of Fittle+ Systems
Like many software-based research projects spanning many years, Fittle has
grown through many interations and side projects. Fittle began in 2012 as nudg
and for many years we referred to the core platform as the Nudg Platformin
195many publications and patents. The initial struggle was between implementation of
the mobile app portion in the emerging HTML5 hybrid methodology or going with
a native experience either on iOS or Android, but eventually both. In those early
FIGURE 4. Scaffolding Interventions and Their Mapping to the Behavior Change
Taxonomy.
Scaffolding Intervention Behavior Change Taxonomy Fittle+ System
Guided mastery BCT 1.1 Goal-setting behavior
BCT 8.7 Graded tasks
BCT 1.5 Review behavior goals
BCT 2.3 Self-monitoring
BCT 2.2 Feedback on behavior
DStress
Fittle
Team social support BCT 3.1 Social support Fittle
Self-affirmation BCT 13.4 Valued self-identity PARC Coach
Implementation intention BCT 1.4 Action planning PARC Coach
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 7
days, the native experience won out and the largely more popular iOS on the iPhone
resonated more with our target population.
200The first app instance of Fittle was indeed a version of nudg written in native
Objective-C for iOS on iPhone and is shown in Figure 5. The backend server
system was developed with Django (initially version 1.4) as a RESTful service. This
version focused on one exercise and one nutrition goal weekly. It included the social
aspect, which has always been core to Fittle. The social aspect in testing proved to
205be what drove repeated use of the app, but the limited behavior tasks were not
compelling, nor did they push the user to improve over time. It was precisely this
feedback that led to an offshoot, called DStress, to explore programs of behavior
change-inducing activities that progress and adapt.
DStress, was developed as an HTML5 application delivered on mobile devices.
That work and our experiences with nudg led to adopting health and wellness
210programs of many varieties that often incorporated both exercise and nutrition.
FIGURE 5. The First Versions of Fittle Were Called Nudg and Featured One Exercise and
One Nutrition Goal Each Week.
8P. Pirolli et al.
These were offered in a new native iOS app adopting the name Fittle in 2013 with
the interfaces shown in Figure 1.
That core Fittle version was adapted, expanded, and used for most of the
experiments discussed in this article, including an exploration of the use of the
215Fittlebot virtual coach (Lukin, Youngblood, Du, & Walker, 2014). An Android
version was developed in 2014. The platform and commercial use of the name
Fittle® were transferred to internal wellness program divisions of Xerox in 2015,
which eventually separated as Conduent, Inc. in 2017. Those versions were used to
deliver health and wellness programs over mobile, and eventually the ideas were
220incorporated in other products at various levels. PARC and our research colleagues
continued to use that platform until late 2016, sometimes under specific program
names such as Nutriwalking (a wellness program that incorporated better nutrition
with a walking-based exercise program), when the accumulation of technical debt
finally made maintaining and adapting it too difficult for our small research team. As
225described in this article, research accomplished with this platform covered the social
aspects of teams, self-affirmation, dynamic and adaptive user-centric activities, and
coaching with over a thousand different users.
In mid 2016, we decided a new research version of Fittle was to be created that
could support more adaptive activity programs and personalized, just-in-time adap-
230tive interventions and coaching. The target app design is shown in Figure 6. Being
multi-platform with a small engineering/research team led to adopting the latest
HTML5 hybrid systems with a React-based (https://reactjs.org/) front-end and
Meteor (https://www.meteor.com/) back-end. Essentially, JavaScript everywhere.
This provides a consistent experience everywhere, is easier to maintain, and easier to
235adapt and integrate with other services. Early versions of this software were used in
the self-affirmation studies discussed below with the core coaching interaction
engine. Because of this initial focus, this new platform was often called the PARC
Coach. The first full versions of this new platform started use in 2017 with two
primary variations: one targeted for an elderly population and one for research
FIGURE 6. The 2016 Re-Design for the New Fittle Built with HTML5 Technologies.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 9
240partners in Hawaii. Experiments planned for the next few years are using these
versions. In April 2018, PARC released the new Fittle platform code publicly for
free research use on Github (https://github.com/PARC/fittle).
2.5. ACT-R Models of Scaffolding Intervention Effects on Healthy
Habits
Although the TPB has provided a framework for the Fittle+ selection of
245mHealth interventions (i.e., self-affirmation, implementation intentions, guided
mastery, teaming), we rely on recent neurocognitive theory and models to provide
a deeper, more precise, and more dynamical psychological understanding of the
effects of scaffolding interventions on behavior change and habit formation. This is
because TPB is not specified as a theory capable of addressing the finer-grained,
250dynamical changes in behavior responsive to repeated interactions with rich, mixed,
content, and interactions over many days and weeks. Finer-grained, dynamical
models are needed if we want to understand the richer traces of moment-by-
moment or daily behavior changes that can now be traced using pervasive technol-
ogy (Spruijt-Metz et al., 2015). We have used the ACT-R theory (Anderson, 2007)to
255begin to refine macro-theories such as the TPB (Ajzen, 1991), Social Cognitive
Theory (Bandura, 1998), and implementation intentions (Gollwitzer, 1999).
Habits are only gradually learned through the association of specific behaviors to
triggering cues in the environment. Triggering cues can include features of physical
settings (including cues produced by digital means, such as reminders) and previous
260actions. A long history of psychological research suggests that there are dual systems
involved in habit acquisition and strengthening (Graybiel, 2008; James, 1890; Wood &
Neal, 2007). First, a deliberative or controlled goal-striving process motivates and
guides seminal attempts at behavior in the relevant contexts. Second, habit learning
and strengthening processes form new habits through repeated practice, and habitual
265behaviors are executed without effortful, controlled, goal striving. Habit formation
typically depends on a long period of goal-mediated, consciously controlled, explora-
tion, repetition, and practice of behavior (System 2; Kahneman, 2011), but well-
practiced habits appear to be performed automatically without mediating goals,
motivation, or deliberative thought (System 1; Kahneman, 2011).
270The ACT-R theory (Anderson, 2007) provides a deeper understanding of the
dual-system framework for habit formation. We have used ACT-R to develop
refined predictive computational models of our interventions for self-efficacy and
implementation intentions. ACT-R (Anderson, 2007; Anderson et al., 2004)isa
unified theory of how the structure and dynamics of the brain give rise to the
275functioning of the mind. The ACT-R simulation environment is a computational
architecture that supports the development of models.
ACT-R (Figure 7) is composed of modules processing different kinds of content.
Recent summaries of the theory can be found in Anderson (2007) and the ACT-R
web site (http://act-r.psy.cmu.edu). Each ACT-R module is devoted to processing a
10 P. Pirolli et al.
280particular kind of information. Each module is associated with a buffer, each module
accesses and deposits information via those buffers, and the processing of informa-
tion across modules and the execution of behavior is coordinated through a
centralized production module. The production module can only respond to the
contents of the buffers.
Important to our models are the modules that process goals and declarative
285memory. The ACT-R goal buffer keeps track of the active goals (when behaviors are
being executed). An ACTR declarative memory module and its associated retrieval buffer model
the retrieval of knowledge and past experiences from long-term declarative memory.
The declarative module is where goal-striving intentions are stored (before they become
active goals in the goal buffer), where memories of past goal-striving experiences are
290stored, and where implementation intentions are stored. As summarized below, learning
and recall mechanisms associated with declarative memory are crucial to predicting the
impact of past goal-striving experience on self-efficacy, which in turn affects intentional
effort, and ultimately behavior change. Declarative memory mechanisms are also
implicated in the effects of reminders on implementation intentions and their effects
295on goal-striving success. Memory activation mechanisms determine the retrieval of past
experiences in current contexts, and base-level learning mechanisms determine how
practice and forgetting affect the levels of memory activation.
FIGURE 7. The ACT-R Theory of Cognitive Architecture. Modules (rectangles) process
different kinds of content. Modules are associated with buffers (shaded circles) in which
they access and deposit information. The central production module coordinates activities
across modules via those buffers.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 11
ACT-R learning mechanisms associated with the production module are cen-
tral to the acquisition and strengthening of new habits. Important to the ACT-R
300model of habit formation is the mechanism of production compilation (Anderson, 2007;
Dayan, 2009; Taatgen, 2004), by which new habits (technically: production rules) are
acquired. The mechanism works to create new habits that eliminate internal cogni-
tive processing, such as the need to retrieve information from the declarative
module or set and maintain sequences of goals. ACT-R utility learning is a variety
305of reinforcement learning similar to temporal-difference learning (Sutton & Barto,
1998) and Rescorla-Wagner learning (Rescorla & Wagner, 1972). The utility of a
new habit is gradually adjusted until it matches the average reward. Through this
gradual effect of reinforcement learning, a new habitual behavior can come to
supersede old habits and more effortful goal-striving behaviors.
3103. FITTLE+ SYSTEMS AND STUDIES
3.1. Guided Mastery
As noted, the path to healthy habits typically involves a phase in which
intentions and goals motivate and guide repeated enactments of the desired beha-
vior until the behavior becomes habitual in desired contexts (Wood & Neal, 2007).
315Research shows that specific, challenging goals consistently lead to higher perfor-
mance than exhortations to do ones best (Locke & Latham, 2002).
1
One key
principle in the design of Fittle+ systems is to have our users continuously engaged
in achieving well-specified behavior-change goals and to provide interactions and interven-
tions that promote success in achieving those goals. Another key principle is to
320organize related behavior-change goals into challenge programs and to support guided
mastery whereby users achieve progressively more difficult or complex goals over
time (usually weeks to months).
As suggested in Figure 1, Fittle users are asked to set behavioral goals, self-
monitor achievement of those goals, and are provided feedback on their goal
325progress. These behavior-change techniques have been found to be among the
top most effective for long-term improvement in diet and physical activity (Samdal
et al., 2017). Using tasks that increase in difficulty from simple to hard (graded tasks)
is also among the top behavior change techniques for diet and physical activity
(Samdal et al., 2017).
330Fittle+ challenge programs have been developed by members of our team who
are subject matter experts in program design to foster healthy nutrition and physical
activity/exercise habits. Challenge programs are aimed at small groups of people
(teams) with similar behavior-change intentions (i.e., goal alignment) and similar
entry-level capabilities. Many of the challenge programs are designed to support
1
The notion of specific goals is consistent with the definition of SMART goals (specific, measurable, achievable,
realistic, relevant and timed).
12 P. Pirolli et al.
335guided mastery by having a static or dynamically computed progression of behavior-
change goals that increase in difficulty, or habits that build upon each other. Fittle
+ applications support the user in mastering each goal before progressing to the
next. This approach is motivated by the success of mastery learning in education
(Bloom, 1968) and cognitive tutoring systems (Anderson, Boyle, Corbett, & Lewis,
3401990; Anderson, Boyle, & Reiser, 1985). The approach is also motivated by the
success of guided enactive mastery (Bandura, 1998) in improving self-efficacy.
For instance, the original Fittle application (Honglu Du et al., 2014) included a
NutriWalking program targeting sedentary people who wanted to increase their
physical activity and eat better. The program includes nutrition activities: eat slowly,
345add a serving of vegetables (or a different vegetable if a vegetarian), add a small
healthy meal, and keep a food diary. The original NutriWalking program also
focused on getting participants to walk more by starting with 15 minutes 3 times
a week on flat surfaces and ramping up to 45 minutes 5 times a week on inclined
surfaces with some exercises (e.g., jumping jacks) or short jogging sessions added to
350the walk. Within each activity class, daily goals are assigned, and those goals progress
in difficulty. A specialized version of the Fittle appcalled NutriWalkingperso-
nalized the users daily goals within the program based on model-based estimates of
the users current aerobic capabilities (Mohan, Venkatakrishnan, Silva, & Pirolli,
2017). The DStress app aimed to improve peoples resilience to stress through
355physical exercise and meditation, and it dynamically modified individual daily
exercise goals based on the difficulty of exercise goals completed on earlier days.
3.2. Guided Mastery as an Intervention to Increase Self-efficacy
Guided mastery is also generally considered a therapeutic method of assisting
people in raising their self-efficacy (i.e., perception that a task can be accomplished)
360so they are motivated to attempt, and subsequently accomplish, progressively more
difficult tasks that are involved in the implementation of behavioral therapies. For
instance, exposure to progressively greater anxiety-provoking situations is the treat-
ment of choice for individuals who evidence problems associated with anxiety
disorders (e.g., panic disorder with agoraphobia). In the utilization of guided
365mastery, a therapist might encourage and assist the individual in accomplishing a
situation that is associated with a low degree of anxiety (e.g., walking in the
shopping center with a friend) before moving to situations originally associated
with a high degree of anxiety (e.g., walking in the shopping center by oneself). As
discussed in greater detail below in the context of the DStress app, we have studied
370and modeled the effects of guided mastery on self-efficacy.
3.3. Teaming
Support for interaction among a team of people engaged in the same behavior-
change challenge program was one of the core interventions used in Fittle. Social
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 13
support is also among the top six behavior-change techniques in terms of effect size
375(Samdal et al., 2017). Forms of social support can be found in many commercial
smartphone-based health behavior change applications, such as MyFitnessPal,
WeightWatchers, etc. However, social support in those applications is built around
the users personal social network, such as Facebook or Twitter friends. In Fittle
+ we have adopted the principle that teams will need to be formed around the
380challenge programs (community of interest) rather than formed around existing
social networks. One reason for this strategy is scalabilityconvincing members of
a social cluster to engage in a program is likely to be less efficient than casting a wide
net for people interested in a behavior-change challenge program and then forming
teams.
385Social support has been shown to have health benefits in its own right
(Callaghan & Morrissey, 1993) and increases participation in exercise programs
(Richardson et al., 2010). Social- and group/team-based behavior-change techni-
ques have been shown to be effective in supporting behavior change in long-
term weight loss (Wing & Jeffery, 2001). Social support may potentially help
390sustainengagementwithhealthbehaviorchange interventions, and consequently
increase efficacy. An observational study of more than 80,000 users in the
contextofaweb-basedhealthpromotionintervention revealed that increased
social ties within this challenge community directly predicted online engagement
and activity completion (Poirer & Cobb, 2012). Similarly, a recent study com-
395pared a structured physical activity intervention comprising education, activity
monitoring, and online social networking via a Facebook group versus an
education-only control showed that online social networking lowered attrition
rates in the program (Cavallo et al., 2014). Likewise, in another study, it was
found that weight loss in a 6-month, remotely delivered weight loss intervention
400was strongly associated with engagement within an online Twitter-based social
network wherein participants provided each other with informational support
(Turner-McGrievy & Tate, 2013).
3.4. Self-affirmation
Self-affirmation interventions typically ask people to think or write about their
405core values. This technique has been shown to improve health, education, and
relationship outcomes with benefits lasting months to years (Cohen & Sherman,
2014). According to many theorists (Cohen & Sherman, 2014), people are motivated
to defend their global sense of self-worth. Health communications, such as the
psycho-educational materials presented in Fittle+, can be perceived as threatening to
410the self-worth of people who are unhealthy, which elicits a defensive resistance to
processing those communications. It is hypothesized that self-affirming in one
domain (e.g., by recalling ones history of kindness to others) boosts global self-
worth and reduces the need to be threatened in another domain, such as health
behavior change.
14 P. Pirolli et al.
4153.5. Implementation Intentions
Implementation intentions are mental representations of simple plans to
translate goal intentions into behavior under specific conditions (Gollwitzer, 1993,
1999). Interventions designed to foster the setting of implementation intentions
typically ask people to specify when, where, how, and (sometimes) with whom to act
420on a goal intention by using if-then statements of the form: If I encounter situation
Sthen I will initiate action A.It is argued (Wieber, Thürmer, & Gollwitzer, 2015)
that one reason to focus intervention efforts on implementation intentions rather
than goal intentions is that medium-to-large changes in commitment to goal inten-
tions (d= 0.66) only lead to small-to-medium changes in behavior (d= 0.33) (T. L.
425Webb & Sheeran, 2006), but implementation intentions have medium-to-large
effects on goal attainment (d= 0.65) (Gollwitzer & Sheeran, 2006).
Wieber et al. (2015) review the experimental literature and studies of physio-
logical correlates to bolster the hypothesis that two processes are involved in the
effectiveness of implementation intentions: (1) the mental representation of situa-
430tions in which the intended behavior is to take place becomes more accessible and
activates the goal intention and (2) a strong associative link between a mental
representation of the situation and intended behavioral action effects a heightened
readiness to perform the action and the action takes less effort.
4. RESULTS
4354.1. DStress: An Adaptive Algorithm to Build Self-Efficacy through
Guided Mastery
One of the major promises of mHealh is the potential for individualization
using adaptive algorithms. DStress (Konrad et al., 2015) is a web- and mobile-based
system that provides automated coaching on exercise and meditation goals aimed at
440reducing perceived stress. The coaching algorithms modulate the difficulty of daily
exercise and meditation goals based on individualsperformance on the immediately
preceding goals. Goal-Setting Theory (Locke & Latham, 2002) predicts that goals
need to be challenging enough to be motivating. Self-efficacy (Bandura, 1998)
predicts that goals that are perceived as too difficult are unlikely to be attempted.
445Greater levels of self-efficacy lead to greater likelihoods of achieving a goal. The
Attributional Theory of Performance (Kukla, 1972) proposes that the level of
intended effort motivating a performance will increase with the difference between
self-efficacy and the perceived difficulty of achieving a goal. Suggested goals for
users should be difficult enough to be motivating, but easy enough to be success-
450fully achieved.
The Konrad et al. (2015) study took place over 28 days. Figure 8 presents
screen shots from the DStress system. Participants were sent an email every
morningwitharemindertologintoDStress.OntheDStresshomescreen
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 15
(Figure 8a), users were presented with their current goals, as well as previous
455activities and their completion status. Pictures and detailed instructions of how
to safely and properly perform each activity were available (Figure 8b). Partici-
pants were asked to report whether or not they performed their goal for the day
and an email reminder was sent in the evening if they failed to report
(Figure 8c)
The DStress programs (Konrad et al., 2015) interleaved Exercise Days with
460Meditation Days and one Rest Day per week. Exercises assigned to individuals came
from a pool of exercises that varied in difficulty (always two sets of upper body,
lower body, and circuit training exercises). Fourty-six exercises in total were devel-
oped by working with three certified trainers, and each one was independently rated
by the trainers for difficulty. If a person successfully completed all exercises assigned
465for a day, an algorithm advanced them to the more difficult level. If they did not
succeed at the exercise activities, then they were regressed to an easier level.
Additionally, if a person was unable to get to the gym that day to complete the
gym-based exercises, the system would propose exercises that could instead be
performed at home. Thus, the location accessibility of the exercises adapted based
470on the users success with getting to the gym. Meditation activities were also adjusted
dynamically by increasing or decreasing the number of minutes assigned to meditate
(using standard mindfulness meditation instructions).
FIGURE 8. The DStress Application for Reducing Stress: (A) the Home Screen Showing
Daily Goals and Part Adherence, (B) Instruction Screen, and (C) Reporting Screen.
(c)(b)(a)
16 P. Pirolli et al.
The Konrad et al. (2015) experiment compared three groups of adult partici-
pants: (1) a DStress-adaptive group (N= 19) who used the adaptive coaching system
475in which goal difficulties adjusted to the user based on past performance, (2) an
Easy-fixed group (N= 24) in which the difficulty of daily goals increased at the same
slow rate for all and (3) a Difficult-fixed group (N= 22) in which the goal difficulties
increased at a greater rate. The participants in the DStress-adaptive group self-
reported significantly lower stress levels compared to the Easy-fixed and Difficult-
480fixed groups.
Here, the focus is on the success rates in performing assigned daily goals. By
the end of the 28 days study, the DStress-adaptive group was reporting significantly
higher level of activity completions than the other two groups (Figure 9a) even
though they were performing more difficult exercises than the Easy-fixed group.
485This increased ability to tackle more difficult goals is consistent with a build-up in
self-efficacy.
4.2. Effects of Teaming
Du et al. (2014) performed an eight week field study of N= 19 adult
participants using an early version of Fittle that provided guided mastery support
for diet, physical activity, and stress-reduction. A hierarchical regression analysis
490indicated that 37% of the variance in success on daily goals was attributable to
group membership. Content analysis of the online messaging among team members
suggested that high performance groups were more social (commenting on, teasing,
FIGURE 9. Summary Data from Konrad Et Al (KONRAD ET AL., 2015) and the ACT-R
Model of Predicted Success Based on Tracing the Experiences on Individual Participants: (A)
the Mean Rate of Participants Successfully Completing Assigned Exercises, (B) the Mean
ACT-R Predicted Success Rates.
(
a
)(
b
)
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 17
chatting with one another), more supportive (with informational, emotional, or
motivational support), and shared more.
495Following that study, Du, Venkatakrishnan, Youngblood, Ram, and Pirolli
(2016) performed an experiment testing the hypothesis that social support, in the
form of small groups or teams, would improve the achievement of daily behavior
change goals and reduce the attrition typically observed in digital health programs
(Eysenbach, 2005) relative to people working on behavior change in a solitary
500manner. Figure 1 shows several screenshots of the version of Fittle also used in
Du et al. (2016). Users could join teams (Figure 1a) and see profiles of their
teammates (Figure 1b). Figure 1e shows the team feed where users can share
multimedia posts with the team at any time. Users may also communicate directly
with each other through a peer-to-peer messaging system. All teams also include a
505simple artificial intelligence (AI) agent as a member, named FittleBot. FittleBot
provides daily tips to the team relevant to their activities, previews the activities
for the week, and comments on the daily activities of the team members as a group.
Over the course of an eight-week study with N= 124 participants, Du et al.
(2016) found a significant difference in the proportion of daily behavioral goals
510achieved between people working solo or in teams comprised of 39 people in each
team (Figure 10). A survival analysis also showed that people working in teams were
66% more likely continue engagement with Fittle when working in teams as
opposed to solo (Figure 11).
Real world behavior-change programs, including commercial ones such as
Weight Watchers, often employ small groups as part of their technique. These
515preliminary studies of Fittle system suggest that online teams can have a substantial
impact on individual achievement. What remains to be understood are the factors of
group composition and group interaction that result in good versus poor groups.
FIGURE 10. Effects of Participating in Small Teams versus Participating Solo in the Nutri-
Walking mHealth Program.
18 P. Pirolli et al.
4.3. Effects of Self-affirmation
A large body of research (Cohen & Sherman, 2014) has shown that self-
520affirmation techniques can produce substantial improvements in behavior change in
health, education, and relationship outcomes that can last for months to years. Self-
affirmation exercises typically involve writing about, or rating, things that a person
values (e.g., family, career) (McQueen & Klein, 2006). The exercises appear to boost
individuals self-worth and make them more resilient in the face of threatssuch as
525health messaging or fear of failureand this can produce ongoing positive feedback.
Brain imaging studies (Cascio et al., 2016) support the ideas that self-affirmation (a) is
rewarding (and pleasurable) as it produces increased activity in the valuation systems
(ventral striatum and ventral medial prefrontal cortex) when participants reflect on
future-oriented core values (compared with everyday activities) and (b) focuses
530attention on thinking about the self when reflecting on future-oriented core values,
as indicated by increased activity in the medial prefrontal cortex and posterior
cingulate cortex. Self-affirmation increases the level of activity in the ventral medial
prefrontal cortex (VMPFC), and the level of VMPFC activity during exposure to
health messages predicts subsequent behavior change (Falk et al., 2015).
535One problem with translating self-affirmation interventions to mobile phone
interaction is the length of the exercises in terms of the amount of writing typically
requested (e.g., a short essay) or the number of items to be rated. The exercises are
appropriate for laboratory, classroom, or face-to-face clinical settings, where paper-
and-pen or desktop computer interaction are standard. An advantage of the mobile
540platform is that it affords more intensive and pervasive interaction. Springer et al.
(2018) explored an approach in which initial self-affirmation exercises could be
coupled with additional dosing of self-affirmation (boosters) throughout the study.
FIGURE 11. Survival Estimates for Continued Engagement with the NutriWalking Program.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 19
Springer et al. (2018) examined the role of self-affirmation dosage in behavior
change and health goal adherence. As far as we are aware, this was the first usage of
545self-affirmation in a mobile health intervention; this presented challenges in the
form of changing contexts in which users completed these self-affirmation exercises
and challenges in designing self-affirmation exercises for the slower input affor-
dances on mobile phones.
This mobile health intervention using self-affirmation exercises, PARC Coach,
550was deployed for N= 127 users who were attempting to improve their physical
health. The PARC Coach daily goal was to consume the recommended amount (5)
of fruit and vegetable servings, as suggested by the National Health Service (NHS,
2015). Before starting the study, all participants reported that they were not
currently consuming the recommended amount of fruit and vegetable servings.
555Participants were divided into four conditions, in a 2 × 2 experimental design
where the independent variables were the completion of an initial self-affirmation
exercise and the weekly completion of mobile self-affirmation boosters. The initial
self-affirmation conditions completed a self-affirmation essay manipulation where
the users wrote about a cherished value they hold and then were shown threatening
560health information motivating them to change behavior (McQueen & Klein, 2006).
Since the effects of this form of self-affirmation are well established, these condi-
tions allowed us to contrast typical self-affirmation doses with the conditions using
self-affirmation boosters. Springer et al. (2018) designed short self-affirmation
boosters that consisted of brief self-affirmation exercises and health information
565that could be quickly and easily completed on a mobile phone. Few previous studies
(Cohen et al., 2009; Nelson et al., 2014) have examined the role of multiple self-
affirmations in a short time period and none in the context of health behavior
change.
Analysis of 3556 observations from the N= 127 participants indicated that
570higher doses of self-affirmation resulted in statistically greater rates of achieving
behavior goals.
Participants who received the maximum dosage of self-affirmation (both initial
and booster self affirmation exercises) successfully met over 12% more of their daily
fruit and vegetable consumption goals when compared with control (Figure 12).
575The findings here were twofold: increased self-affirmation dosage results in
increased behavior change and self-affirmations can be adapted into a form that is
deliverable through mobile devices. This work opened the door to broader usage of
self-affirmation in mobile health interventions.
4.4. Effects of Implementation Intentions and Reminders
Research (Wieber et al., 2015) implicates human associative memory mechan-
580isms in the way that implementation intentions produce effects. Based on the ACT-
R theory of human memory and cognition, we hypothesized that the strength of
implementation intention effects could be manipulated in predictable ways using
20 P. Pirolli et al.
reminders delivered by a mobile health (mHealth) application. Reminders delivered
by mobile devices (e.g., SMS text messages) are common (e.g., Prestwich, Perugini,
585& Hurling, 2010), but there is no theoretical understanding of their effects.
The ACT-R theory of declarative memory includes a base-level learning mechan-
ism that accounts for practice (e.g., reminding) effects and forgetting effects.
Declarative memories have a base-level activation value that predicts their prob-
ability and speed of retrieval. The dynamics of base-level memory activation are
590predicted to be related to frequency and timing of reminders, as well as the
frequency and timing of actual use of the implementation intentions in performing
behavior. The ACT-R base-level learning mechanisms predict that each time an
implementation intention is formulated, reminded, or put into practice it receives an
increment of activation (a practice effect). However, each increment of activation
595decays as a power function of time (the forgetting effect). The rate of decay of each
increment of activation depends on the strength of activation at the time of the
reminding or practice: At longer intervals between remindings or practice, the
activation levels are lower and subsequent forgetting occurs less quickly (the spacing
effect). When reminding or practice is spaced closely, the forgetting occurs more
600quickly.
Pirolli et al. (2017) presents an experiment that manipulated the effects
implementation intentions on daily behavioral goal success by controlling and
manipulating the scheduled delivery of reminders about those implementation
intentions. All participants were asked to choose a healthy behavior goal associated
605with Eat Slowly, Walking, or Eating More Vegetables, and were asked to set
implementation intentions. N= 64 adult participants were in the study for
28 days. Participants were stratified by pre-experimental levels of self-efficacy and
assigned to one of two reminder conditions: Reminders-presented versus Remin-
ders-absent. Self-Efficacy and Reminder conditions were crossed. Nested within the
610Reminders-presented condition was a crossing of Frequency of reminders sent (High,
Low) by Distribution of reminders sent (Distributed, Massed). Participants in the
FIGURE 12. Percentage of Fruit and Vegetable Consumption Goals Met by Condition.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 21
Low Frequency condition got 7 reminders over 28 days; those in the High Fre-
quency condition were sent 14. Participants in the Distributed conditions were sent
reminders at uniform intervals. Participants in the Massed Distribution conditions
615were sent reminders in clusters.
Figure 13 shows the predicted effects of the Pirolli et al. reminding schedules
on the base-level activation of implementation intentions. Each peak in Figure 13
corresponds to a day on which a reminder was presented. Reminders are predicted
to boost up the base-level activation of participantsimplementation intentions but
620the activation decays without further practice, and distributed presentations of
reminders are forgotten less quickly Manipulation of the declarative memory activa-
tion of implementation intentions was expected to affect the goal-striving phase of
habit formation.
Each plot in Figure 13 also presents the predicted mean activation level of
the implementation intention over the full 28 days for each condition (upper left
625corner of each plot). Note that at Low Frequency of reminders, the mean
activation level in the Massed condition is greater than that of the Distributed
FIGURE 13. Simulated Base-Level Learning of Implementation Intentions as a Function of
Different Reminder Schedules.
Distributed
Massed
0.00.51.01.5
Low Frequency
Day
Activation
X = 0.66
High Frequency
Day
Activation
X = 1.04
Day
Activation
X = 0.71
0 5 10 15 20 25 0 5 10 15 20 25
0.00.51.01.5
0 5 10 15 20 25
0.00.51.01.5
0 5 10 15 20 25
0.00.51.01.5
Day
Activation
X = 1.00
22 P. Pirolli et al.
condition, but at High Frequency the mean activation of the Distributed condi-
tion is greater than the Massed condition. Thus, there is a predicted interaction
of reminder Distribution (Massed, Distributed) by Frequency (Low, High), and
630specifically the average activation levels for the implementation interventions are
predicted to be High Frequency-Distributed >High Frequency-Massed >Low
Frequency-Massed >Low Frequency-Distributed. Behavior-change data were
used to test for this predicted interaction and the specific ordering of success
rates predicted by the model.
635There was a significant overall effect of reminders on achieving a daily
behavioral goal. As predicted by ACT-R (Figure 13), there was a statistical interac-
tion of reminder Frequency by Distribution on daily goal success. The total number of
times a reminder was acknowledged as received by a participant had a marginal
effect on daily goal success and the time since acknowledging receipt of a reminder
640was highly significant.
5. ACT-R MODELS
5.1. ACT-R Model of Self-Efficacy Using DStress
The ACT-R model for this study (Pirolli, 2016a) assumes that self-efficacy and
intended effort are fundamentally the result of declarative memory processes. Past
645experiences of efficacy at behaviors similar to a target goal are retrieved to produce
assessments of self-efficacy and intended effort for the new goal. Consequently, the
dynamics of self-efficacy and performance exhibit the dynamics of the underlying
memory mechanisms, and exhibit well-known memory phenomena.
In outline, the model involves the following steps:
650
Abehavioral goal is considered for doing activities that are believed to have some
level of difficulty to being performed.
Blended memory retrieval to form an assessment of self-efficacy. Successful experiences are
recalled involving activities similar to the goal activities. A composite assessment is
produced characterizing the difficulty levels of exercises achieved in past experi-
655ences, and this is mapped onto an assessment of self-efficacy.
Blended memory retrieval to form an intended effort level. Experiences are recalled with
similar levels of self-efficacy and perceived goal difficulty. This process produces
an assessment of intended effort levels required to achieve success in those past
experiences.
660
Predicting success. Based on the goal difficulty, perceived self-efficacy, and intended
effort, the model makes a prediction about the likelihood of success.
Choose to do it (or not). If the expected probability of success is above a threshold it
is attempted.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 23
Store new experiences. If the activity is attempted, the experience is stored in memory
665and influences future attempts.
Figure 9b shows summary data from the ACT-R model. The model produces a
predicted success or failure for each and every exercise on every exercise day for
every participant in the Konrad et al. (2015) study. Each point in Figure 9b pools
the predicted success data by day and by group (DStress-adaptive, Easy-fixed,
670Difficult-fixed). In fitting the ACT-R model to the data, we used the default
parameter settings suggested for ACT-R simulations. That is, no free parameters
were estimated while fitting the model, and the parameters did not differ from
individual to individual. For the data displayed in Figure 9,RMSE = = 0.083.
5.2. ACT-R Model of Implementation Intentions
675As discussed above, ACT-R theory suggests that behavior change involves dual
systems: the goal-striving system dependent on declarative memory plus a habit-
forming system that acquires more automatic procedures for performance. Pirolli
et al. (2017) fit an ACT-R model dual-system model to the implementation intention
experiment. The model includes (a) the goal-striving system dependent on imple-
680mentation intentions in declarative memory whose strength is affected by reminders
and (b) the habit/reinforcement learning system that acquires new behavioral
routines through repeated performance.
Figure 14 plots the goal success predictions of this dynamical ACT-R model
against the observed data in Pirolli et al. (2017) as functions of past goal adherence
685and reminders acknowledged. Recency in Figure 14 means the number of days since
the last event (performing a goal or acknowledging a reminder) and frequency refers
to the cumulative total number of events. The points are the observed probabilities
of success at achieving a goal behavior and the lines are the model predicted
probabilities. The model fits are to each individuals daily data for the entire course
690of the experiment. Each point is the mean of the observed individual daily success
for participants at a given level of recency or frequency, and similarly the lines are
the means of the models predictions for each individual on each day, pooled by
level of frequency and recency. Overall, the cumulative total frequency of reminders
is predictive of an increase in goal success, the time since last reminder (recency) is
695predictive of a decrease in goal success, and the total frequency of past successful
goal performance (adherence) is predictive of an increase in success, and time since
last successful performance (adherence recency) predict a decrease in goal success.
For these implementation intention data, the ACT-R model is more complex
than the one used to model the DStress data. Five ACT-R parameters representing the
implicit reward for achieving a behavior goal, the utility value of a new habit, the utility
700learning rate, and scaling and slope parameters for base-level learning in declarative
memory were estimated simultaneously along with six scaling and weight parameters
that mapped ACT-R memory activation and production utility values onto the
24 P. Pirolli et al.
empirical goal-success rates. These parameters were not varied for each individual
simulated. The Brier score on the fit of the model to the data was 0.1724. Full details
705on the model an parameter estimation can be found in Pirolli et al. (2017). Overall the
dual-system ACT-R model theory provides a good fit to the individual-level data.
Overall, across the DStress and implementation intention models, ACT-R
provides an integrated computational account that refines self-efficacy, Goal
Setting Theory, the Attributional Theory of Performance, and habit formation
710predictions.
FIGURE 14. Fit of the ACT-R Dual-System Model to Daily Success in Performing Behavior
Goals.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 25
6. GENERAL DISCUSSION
There are two main contributions of this article. First, we propose the concept
of scaffolding interventions as digitally delivered interactions that instantiate beha-
vior change techniques that support people in developing healthy habits. Second, we
715present predictive models developed within a unified computational cognitive
architecture that combines multiple psychological mechanisms involved in two
independent studies of different scaffolding interventions. We also propose that
this work serves as a launching point for richer exploration of how to develop more
sophisticated, dynamic, and intelligent coaching support for healthy lifestyle change.
7206.1. Scaffolding Interventions
The Fittle+ systems and studies have been aimed at exploring and under-
standing mHealth behavior change systems as a process of building a set of healthy
habits. The approach involves a combination of behavior-change techniques
(Michie et al., 2014) found to be effective (Cohen & Sherman, 2014; Gollwitzer &
725Sheeran, 2006; Samdal et al., 2017) implemented as specific scaffolding interventions
on an mHealth platform. The scaffolding interventions are aimed at increasing
peoples adherence to specific behavior-change goals, with the ultimate goal of
helping them develop healthy habits. Many mHealth systems (Consolvo, Everitt,
Smith, & Landay, 2006; Consolvo et al., 2008; Consolvo, McDonald, & Landay,
7302009; Klasnja, Consolvo, & Pratt, 2011) focus on one or just a few variants of
behavior (e.g., increased physical activity) and on one or a few intervention techni-
ques. Our approach can be viewed as drawing upon traditions in education, classical
behaviorism, and cognitive skill acquisition that view desirable complex behaviors as
systems that can be decomposed into elements that can be built up and organized
735through various interventions and scaffolding that may be subsequently faded
(although not necessarily).
Each of the particular scaffolding interventions presented in this article could
benefit from further research. Particularly intriguing is the teaming intervention. As
noted above, Du et al. (2014) found that 37% of the variance in individualsgoal
740achievement success was due to variation in the teams they were in. But we do not
understand what social or communicative interactions are causal, nor the underlying
mechanisms involved. A deeper understanding of these mechanisms could lead to
interesting interventions focused on team formation and team dynamics that pro-
duce large improvements in healthy behavior. Such research would also be an
745opportunity for pushing computational theories of individual cognition, such as
ACT-R, further into the realm of social interaction and social psychology.
Digital health platforms have the potential to support new empirical methods
that could revolutionize the study and optimization of scaffolding interventions,
accelerating the attempts to harmonize the vast literature on behavior-change
750techniques and yield scalable digital health platforms (Michie et al., 2013; Samdal
26 P. Pirolli et al.
et al., 2017). One such method is the Multiphase Optimization Strategy (MOST;
Collins, Murphy, & Strecher, 2007) that could be used to efficiently select and refine
scaffolding interventions based on their effectiveness. Another is the Sequential
Multiple Assignment Randomized Trial method (SMART; Collins et al., 2007)
755which is suited for understanding sequentially delivered interventions, and for
tailoring time-varying adaptive interventions. Micro-randomized trials (Klasnja
et al., 2015) combine aspects of within-participant and between-participant experi-
mental designs in ways that greatly increase statistical power. Each of these methods
extends traditional experimental designs and statistical techniques in ways that are
760more congruent with how digital platforms actually operate and collect rich data.
They offer new ways of studying multiple scaffolding interventions or variants in
more efficient ways than standard randomized controlled trial designs. Machine
learning techniques, such as reinforcement learning (Yom-Tov, Kozdoba, Mannor,
Tennenholtz, & Hochberg, 2017), can also be applied to better understand and
765optimize the effects of specific contextually tailored interventions.
Scaffoldingin the learning sciences may typically imply the intention that
the scaffolding supports for learning will eventually be withdrawn (Reiser, 2004;
Reiser & Tabak, 2014). However, many interventions that may have been
originally intended as educational scaffolding have become part of our everyday
770environment, such as calculators and structured code editors. It is conceivable
(for instance, see Stawarz, Cox, & Blandford, 2015) that new behaviors may be
supported with scaffolding (e.g., reminders), but then be triggered only if that
scaffolding provides the necessary cues in the environment (e.g., will not occur
unless a reminder is sent). In other words, the new habit is contingent on the
775scaffolding cues, and the intervention cannot be removed without affecting the
habits performance. This is not necessarily a problem, as much of our modern
life depends on continuous support from digital infrastructure, but it is impor-
tant to realize that maintaining a habit as scaffolding interventions are removed
may require attention over and above the initial building up of the habit in the
780presence of scaffolding interventions.
Much remains to be understood about how and when to combine scaffold-
ing interventions. It is possible that scaffolding interventions may combine to
produce benefits beyond the sum of their independent effects. There may be
person-dependent interactions. There may even be negative interactions among
785scaffolding interventions. Just as there are sometimes drug-drug interactions that
must be avoided in medical treatment, there may be particular combinations of
interventions that cancel out each others effects, or perhaps have negative
consequences. One can also imagine that piling up scaffolding interventions in
an engagement with users might have an overall diminishing returns (or negative
790returns)e.g., reminders associated with different interventions might increase
in number to the point that people become annoyed and inattentive, or com-
pletely disengaged.
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 27
6.2. Predictive Theory and Modeling
As part of this effort, we have begun to develop a predictive modeling
795approach that builds upon well-established computational cognitive theory. In
essence, we have taken the ACT-R theory of cognitive skill acquisition (Anderson,
1982; Anderson, 1987; Anderson, Conrad, & Corbett, 1989) and extended it to habit
formation in the ecology of everyday life. As argued elsewhere, such fine-grained
predictive models will need to be developed to better predict and control fine-
800grained, dynamic digital, and mHealth interventions (Riley et al., 2011). Progress in
predictive modeling (Martín et al., 2014; Pirolli, 2016a) should lead to the develop-
ment of user modeling approaches that support the personalization of interventions
(Spruijt-Metz et al., 2015). This also opens up a path for scientific psychology to
extend laboratory-developed theories and models out into the ecology of everyday
805life, where meaningful behavior happens (Baumeister et al., 2007).
To repeat an argument made elsewhere (Pirolli, 2016a,2016b; Pirolli et al., 2017),
the motivation for developing predictive theories of scaffolding interventions by
extending theories of the human cognitive architecture rests on four theses (Anderson,
2002;Newell,1990): (a) the Integration Thesis that cognitive architectures provide a
810unified account of how the mind functions and can provide a basis for an integration
across specialized theories and techniques of behavior change, (b) the Decomposition
Thesis that longer-term behavior change can be decomposed to learning and interven-
tion events occurring at a much finer granularity of time, (c) the Modeling Thesis that
models in cognitive architectures can provide a basis for bridging those events at the
815small scale to the dynamics of behavior change occurring at the large scale, and (d) the
Relevance Thesis, that longer term changes and outcomes can be improved by modeling
and predicting specific scaffolding interventions in contexts that are occurring at the
smaller time scales in the everyday lives of people wishing to change.
HCI has had a long history of appropriating psychological science to improve
820personal computing (Card, Moran, & Newell, 1983) and, in return, being a crucible
for the advancement of psychology. With mHealth, and digital health more gen-
erally, there is an opportunity to develop a new science and engineering of perso-
nalized computing that supports long-term improvements in lifestyle and health.
ACKNOWLEDGMENTS
825We would like to thank those who collaborated on Fittle+ research: Shane Ahern,
Victoria Bellotti, Nicole Crenshaw, Jacqueline LeBlanc, Pai Liu, Shiwali Mohan,
Ashwin Ram, Frank Rolek, Jonathan Rubin, Michael Silva, Simon Tucker, Anusha
Venkatakrishnan, Jesse Vig, Steve Whittaker, and Rong Yang. Fittle®is a registered
trademark of Palo Alto Research Center, Inc.
28 P. Pirolli et al.
830FUNDING
This material is based upon work supported by the National Science Foundation,
Directorate for
Computer and Information Science and Engineering, Smart and Connected Health
program, under Grant No. 1346066 to Peter Pirolli and Michael Youngblood.
835
REFERENCES
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision
Processes,50, 179211. doi:10.1016/0749-5978(91)90020-T
Ajzen, I. (1998). Models of human social behavior and their application to health psychol-
840ogy. Psychology and Health,13, 735740. doi:10.1080/08870449808407426
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review,89, 369406.
doi:10.1037/0033-295X.89.4.369
Anderson, J. R. (1987). Skill acquisition: compilation of weak-method problem situations.
Psychological Review,94(2), 192-210. doi:10.1037/0033-295X.94.2.192
845Anderson, J. R. (2002). Spanning seven orders of magnitude: A challenge for cognitive
modeling. Cognitive Science,26(1), 85112. doi:10.1207/s15516709cog2601_3
Anderson, J. R. (2007). How can the human mind occur in the physical universe? Oxford, UK:
Oxford University Press.
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004). An
850integrated theory of mind. Psychological Review,11(4), 10361060. doi:10.1037/0033-
295X.111.4.1036
Anderson, J. R., Boyle, C. F., Corbett, A. T., & Lewis, M. W. (1990). Cognitive modelling and
intelligent tutoring. Artificial Intelligence,42,749. doi:10.1016/0004-3702(90)90093-F
Anderson, J. R., Boyle, C. F., & Reiser, B. J. (1985). Intelligent tutoring systems. Science,228,
855456462. doi:10.1126/science.228.4698.456
Anderson, J. R., Conrad, F. G., & Corbett, A. T. (1989). Skill acquisition and the LISP tutor.
Cognitive Science,13,467505. doi:10.1207/s15516709cog1304_1
Armitage, C. J., & Conner, M. (2001). Efficacy of the theory of planned behavior: A meta-
analytic review. British Jorunal of Psychology,40, 471499. doi:10.1348/014466601164939
860Bandura, A. (1998). Self-efficacy: the exercise of control. New York, NY: W.H. Freeman.
Baumeister, R. F., Vohs, K. D., & Funder, D. C. (2007). Psychology as the science of self-
reports and finger movements: Whatever happened to actual behavior? Perspectives on
Psychological Science : A Journal of the Association for Psychological Science,2(4), 396403.
doi:10.1111/j.1745-6916.2007.00051.x
865Bloom, B. (1968). Learning for mastery. Evaluation Comment (UCLA-CSIEP),1(2), 112.
Brewer, N. T., & Rimer, B. K. (2008). Perspectives on health behavior theories that focus on
individuals. In K. Glanz, B. K. Rimer, & K. Viswanath (Eds.), Health behavior and health
education: Theory, research, and practice, (4th ed.). (pp. 149166). San Francisco, CA: Jossey-
Bass.
870Bridle, C., Riemsma, R. P., Pattenden, J., Sowden, A. J., Mather, L., Watt, I. S., & Walker, A.
(2005). Systematic review of the effectiveness of health behavior interventions based on
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 29
the transtheoretical model. Psychology and Health,20, 283301. doi:10.1080/
08870440512331333997
Callaghan, P., & Morrissey, J. (1993). Social support and health: A review. Journal of Advanced
875Nursing,18(2), 203210. doi:10.1046/j.1365-2648.1993.18020203.x
Card, S. K., Moran, T. P., & Newell, A. (1983). The psychology of human-computer interaction.
Hillsdale, New Jersey: Lawrence Erlbaum Associates.
Cascio, C. N., ODonnell, M. B., Tinney, F. J., Lieberman, M. D., Taylor, S. E., Strecher, V.
J., & Falk, E. B. (2016). Self-affirmation activates brain systems associated with self-
880related processing and reward and is reinforced by future orientation. Social Cognitive and
Affective Neuroscience,11(4), 621629. doi:10.1093/scan/nsv136
Cavallo, D. N., Tate, D. F., Ward, D. S., DeVellis, R. F., Thayer, L. M., & Ammerman, A. S.
(2014). Social support for physical activity-role of Facebook with and without structured
intervention. Translational Behavioral Medicine,4(4), 346354. doi:10.1007/s13142-014-
8850269-9
Cohen, G. L, Garcia, J, Purdie-Vaughns, V, Apfel, N, & Brzustoski, P. (2009). Recursive
processes in self-affirmation: intervening to close the minority achievement gap. Science,
324(5925), 400-403. doi: 10.1126/science.1170769
Cohen, G. L., & Sherman, D. K. (2014). The psychology of change: Self-affirmation and
890social psychological interventions. Annual Review of Psychology,65, 333371. doi:10.1146/
annurev-psych-010213-115137
Collins, L. M., Murphy, S. A., & Strecher, V. (2007). The multiphase optimization strategy
(MOST) and the sequential multiple assignment randomized trial (SMART): New
methods for more potent eHealth interventions. American Journal of Preventive Medicine,
89532(5 Suppl), S112118. doi:10.1016/j.amepre.2007.01.022
Consolvo, S., Everitt, K., Smith, I., & Landay, J. A. (2006). Design requirements for
technologies that encourage physical activity. Paper presented at the Proceedings of the
SIGCHI Conference on Human Factors in Computing Systems, Montréal, Québec, Canada.
Consolvo, S., Klasnja, P., McDonald, D. W., Avrahami, D., Froehlich, J., LeGrand, L.,
900Landay, J. A. (2008). Flowers or a robot army?: Encouraging awareness & activity with
personal, mobile displays. Paper presented at the Proceedings of the 10th international conference
on Ubiquitous computing, Seoul, Korea.
Consolvo, S., McDonald, D. W., & Landay, J. A. (2009). Theory-driven design strategies for
technologies that support behavior change in everyday life. Paper presented at the Proceed-
905ings of the SIGCHI Conference on Human Factors in Computing Systems, Boston, MA, USA.
Cooper, Z, Fairburn, C. G, & Hawker, D. M. (2003). Cognitive-behavioral treatment of obesity.
New York, NY: Guilford Press.
Dayan, P. (2009). Goal-directed control and its antipodes. Neural Networks,22, 213219.
doi:10.1016/j.neunet.2009.03.004
910Du, H., Venkatakrishnan, A., Youngblood, G. M., Ram, A., & Pirolli, P. (2016). A group-
based mobile application to increase adherence in exercise and nutrition programs: A
factorial design feasibility study. JMIR Mhealth Uhealth,4(1), e4. doi:10.2196/
mhealth.4900
Du, H., Youngblood, G. M., & Pirolli, P. (2014). Efficacy of a smartphone system to
915support groups in behavior change programs. Paper presented at the Proceedings of the
Wireless Health 2014 on National Institutes of Health, Bethesda, MD, USA. http://dl.acm.
org/citation.cfm?doid=2668883.2668887
30 P. Pirolli et al.
Epton, T., Harris, P. R., Kane, R., Van Koningsbruggen, G. M., & Sheeran, P. (2015). The
impact of self-affirmation on health-behavior change: A meta-analysis. Health Psychology,
92034(3), 187196. doi:10.1037/hea0000116
Eysenbach, G. (2005). The law of attrition. Journal of Medical Internet Research,7(1), e11.
doi:10.2196/jmir.7.1.e11
Falk, E. B., ODonnell, M. B., Cascio, C. N., Tinney, F., Kang, Y., Lieberman, M. D.,
Strecher, V. J. (2015). Self-affirmation alters the brains response to health messages and
925subsequent behavior change. Proceedings of the National Academy of Sciences,112(7), 1977
1982. doi:10.1073/pnas.1500247112
Fogg, B. J. (2009). The behavior grid: 35 ways behavior can change. Paper presented at the
Persuasive 2009, Claremont, CA.
Fogg, B. J, & Hreha, J. (2010). Behavior wizard: a method for matching target behaviors with solutions.
930(Eds.), (6137, pp. 117131). Germany: Springer Berlin Heidelberg.
Gollwitzer, P. M. (1993). Goal achievement: The role of intentions. European Review of Social
Psychology,4(1), 141185. doi:10.1080/14792779343000059
Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. American
Psychologist,54(7), 493503. doi:10.1037/0003-066X.54.7.493
935Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A
meta-analysis of effects and processes. Advances in Experimental Social Psychology,42,668675.
Graybiel, A. M. (2008). Habits, rituals, and the evaluative brain. Annual Review of Neuroscience,
31, 359387. doi:10.1146/annurev.neuro.29.051605.112851
Harari, G. M., Lane, N. D., Wang, R., Crosier, B. S., Campbell, A. T., & Gosling, S. D.
940(2016). Using smartphones to collect behavioral data in psychological science: Oppor-
tunities, practical considerations, and challenges. Perspectives on Psychological Science : a
Journal of the Association for Psychological Science,11(6), 838854. doi:10.1177/
1745691616650285
Harrison, J. A., Mullen, P. D., & Green, L. W. (1992). A meta-analysis of studies of the
945health belief model. Health Education Research,7, 107116.
Heron, K. E., & Smyth, J. M. (2010). Ecological momentary interventions: Incorporating
mobile technology into psychosocial and health behaviour treatments. British Journal of
Health Psychology,15(Pt 1)), 139. doi:10.1348/135910709X466063
James, W. (1890). The principles of psychology: In two volumes. New York, NY: Henry
950Holt and Company.
Kahneman, D. (2011). Thinking, fast and slow. New York. Farar, Straus and Giroux.
Klasnja, P., Consolvo, S., & Pratt, W. (2011). How to evaluate technologies for health
behavior change in HCI research. Paper presented at the Proceedings of the SIGCHI Conference
on Human Factors in Computing Systems, Vancouver, BC, Canada.
955Klasnja, P., Hekler, E. B., Shiffman, S., Boruvka, A., Almirall, D., Tewari, A., & Murphy, S.
A. (2015). Micro-randomized trials: An experimental design for developing just-in-time
adaptive interventions. Health Psychology: Official Journal of the Division of Health Psychology,
American Psychological Association,34(S), 12201228. doi:10.1037/hea0000305
Konrad, A., Bellotti, V., Crenshaw, N., Tucker, S., Nelson, L., Du, H., Whittaker, S.
960(2015). Finding the adaptive sweet spot: balancing compliance and achievement in
automated stress reduction. Paper presented at the SIGCHI Conference on Human Factors in
Computing Systems (CHI 2015), Seoul, Korea.
Kukla, A. (1972). Foundations of an attributional theory of performance. Psychological Review,
79(6), 454470. doi:10.1037/h0033494
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 31
965Locke, E. A., & Latham, G. P. (2002). Building a practically useful theory of goal setting and
task motivation: A 35-year odyssey. American Psychologist,57(9), 705717. doi:10.1037/
0003-066X.57.9.705
Lukin, S. M., Youngblood, G. M., Du, H., & Walker, M. A. (2014). Building community and
commitment with a virtual coach in mobile wellness programs. Paper presented at the 14th
970International Conference on Intelligent Virtual Agents (IVA), Boston, MA.
Martín, C. A., Rivera, D. E., Riley, W. T., Hekler, E. B., Buman, M. P., Adams, M. A., &
King, A. C. (2014). A dynamical systems model of social cognitive theory. Paper presented
at the American Control Conference, Portland, OR.
Martin, G., & Pear, J. (2016). Behavior modification: What it is and how to do it (Tenth ed.).
975London and New York, NY: Routledge.
McQueen, A., & Klein, W. M. P. (2006). Experimental manipulation of self-affitmation: A
systematic review. Self and Identity,5, 289354. doi:10.1080/15298860600805325
Michie, S., Richardson, M., Johnston, M., Abraham, C., Francis, J., Hardeman, W., Wood,
C. E. (2013). The behavior change technique taxonomy (v1) of 93 hierarchically
980clustered techniques: Building an international consensus for the reporting of behavior
change interventions. Annals of Behavioral Medicine,46(1), 8195. doi:10.1007/s12160-
013-9486-6
Michie, S., West, R., Campbell, R., Brown, J., & Gainforth, H. (2014). Abc of behavior
change theory. Sutton, Great Britain: Silverback Publishing.
985Miller, W. R., & Rollnick, S. (2013). Motivational interviewing: Helping people change (3rd ed.). New
York, NY, US: Guilford Press.
Mohan, S., Venkatakrishnan, A., Silva, M., & Pirolli, P. (2017). On designing a social coach to
promote regular aerobic exercise.Proceedings of the Twenty-Ninth AAAI Conference on Innovative
Applications (IAAI-17) (pp. 47214727). San Francisco, CA: AAAI.
990Nelson, S. K, Fuller, Joshua A. K, Choi, I, & Lyubomirsky, S. (2014). Beyond self-protec-
tion: self-affirmation benefits hedonic and eudaimonic well-being. Personality and Social
Psychology Bulletin,40(8), 998-1011. doi:10.1177/0146167214533389
Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press.
Pirolli, P. (2016a). A computational cognitive model of self-efficacy and daily adherence in
995mHealth. Translational Behavioral Medicine,6(4), 113. doi:10.1007/s13142-016-0391-y
Pirolli, P. (2016b). From good intentions to healthy habits: Towards integrated computa-
tional models of goals striving and habit formation. The 38th Annual International
Conference of the IEEE Engineering in Medicine and Biology Society (EMBC16). Orlando,
FL: IEEE.
1000Pirolli, P, Mohan, S, Venkatakrishnan, A, Nelson, L, Silva, M, & Springer, A. (2017).
Implementation intention and reminder effects on behavior change in a mobile health
system: a predictive cognitive model. Journal Of Medical Internet Research,19(11), e397. doi:
10.2196/jmir.8217
Pirolli, P., Mohan, S., Venkatakrishnan, A., Nelson, L., Silva, M., & Springer, A. (2017).
1005Implementation intention and reminder effects on behavior change in a mobile health
system: A predictive cognitive model. Journal of Medical Internet Research,19(11), e397.
doi:10.2196/jmir.8217
Poirer, J., & Cobb, N. K. (2012). Social influence as a driver of engagement in a web-based
health intervention. Journal of Medical Internet Research,14(1), e36. doi:10.2196/jmir.1957
32 P. Pirolli et al.
1010Prestwich, A., Perugini, M., & Hurling, R. (2010). Can implemetation intentions and text
messages promote brisk walking? A randomized trial. Health Psychology,29(1), 4059.
doi:10.1037/a0016993
Reiser, B. J. (2004). Scaffolding complex learning: The mechanisms of structuring and
problematizing student work. Journal of the Learning Sciences,13(3), 273304.
1015doi:10.1207/s15327809jls1303_2
Reiser, B. J., & Tabak, I. (2014). Scafolding. In R. K. Sawyer (Ed.), The Cambridge hand-
book of the learning science (Second ed.) (pp. 4462). New York, NY: Cambridge
University Press.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in
1020the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F.
Prokasy (Eds.), Classical conditioning: II. Current research and theory (pp. 6499). New
York, NY: Appleton-Century-Crofts. doi:10.1016/S0079-7421(08)60383-7
Richardson, C. R., Buis, L. R., Janney, A. W., Goodrich, D. E., Sen, A., Hess, M. L.,
Piette, J. D. (2010). An online community improves adherence in an Internet-mediated
1025walking program. Part 1: Results of a randomized controlled trial. Journal of Medical
Internet Research,12(4), e71. doi:10.2196/jmir.1338
Riley, W. T., Nilsen, W. J., Manolio, T. A., Masys, D. R., & Lauer, M. (2015). News from the
NIH: Potential contributions of the behavioral and social sciences to the precision
medicine initiative. Translational Behavioral Medicine,5(3), 243246. doi:10.1007/s13142-
1030015-0320-5
Riley, W. T., Rivera, D. E., Atienza, A. A., Nilsen, W., Allison, S. M., & Mermelstein, R.
(2011). Health behavior models in the age of mobile interventions: Are our theories up
to the task? Translational Behavioral Medicine,1(1), 5371. doi:10.1007/s13142-011-0021-7
Rotheram-Borus, M. J., Swendeman, D., & Chorpita, B. F. (2012). Disruptive innovations
1035for designing and diffusing evidence-based interventions. American Psychologist,67(6),
463476. doi:10.1037/a0028180
Sallis, J. F., & Glanz, K. (2009). Physical activity and food environments: Solutions to the
obesity epidemic. The Milbank Quarterly,87(1), 123154. doi:10.1111/j.1468-
0009.2009.00550.x
1040Samdal, G. B., Eide, G. E., Barth, T., Williams, G., & Meland, E. (2017). Effective behaviour
change techniques for physical activity and healthy eating in overweight and obese
adults; systematic review and meta-regression analyses. The International Journal of Beha-
vioral Nutrition and Physical Activity,14(1), 42. doi:10.1186/s12966-017-0494-y
Service, National Health. (2015). Eat well: why 5 A Day? NHS Choices. London, UK: gov.
1045uk; Dec 12. URL: https://www.nhs.uk/live-well/eat-well/why-5-a-day/ [accessed
2018-06-18]
Silva, B. M. C., Rodrigues, J. J. P. C., De La Torre Díez, I., López-Coronado, M., & Saleem,
K. (2015). Mobile-health: A review of current state in 2015. Journal of Biomedical Infor-
matics,56, 265272. doi:10.1016/j.jbi.2015.06.003
1050Springer, A, Venkatakrishnan, A, Mohan, S, Nelson, L, Silva, M, & Pirolli, P. (2018).
Leveraging self-affirmation to improve behavior change: a mobile health app experi-
ment. Jmir Mhealth Uhealth,6(7), e157. doi: 10.2196/mhealth.9151
Spruijt-Metz, D., Hekler, E., Saranummi, N., Intille, S., Korhonen, I., Nilsen, W., Pavel,
M. (2015). Building new computational models to support health behavior change and
1055maintenance: New opportunities in behavioral research. Translational Behavioral Medicine,
5(3), 335346. doi:10.1007/s13142-015-0324-1
Scaffolding the Mastery of Healthy Behaviors with Fittle+ Systems 33
Stawarz, K., Cox, A. L., & Blandford, A. (2015). Beyond self-tracking and reminders:
Designing smartphone apps that support habit formation. Paper presented at the Proceedings
of the 33rd Annual ACM Conference on Human Factors in Computing Systems, Seoul, Republic
1060of Korea.
Sutton, R. S., & Barto, A. G. (1998). Reinforcement learning. Cambridge, MA: MIT Press.
Taatgen, N. A. (2004). Production compilation: a versatile cognitive mechanism. In Aisb
quarterly, 115((winter)) (pp. 7-7).
Turner-McGrievy, G., & Tate, D. F. (2013). Weight loss social support in 140 characters or
1065less: Use of an online social network in a remotely delivered weight loss intervention.
Translational Behavioral Medicine,3(3), 287294. doi:10.1007/s13142-012-0183-y
Wansink, B., & Sobal, J. (2007). Mindless eating: The 200 daily food decisions we overlook.
Environment and Behavior,39(1), 106123. doi:10.1177/0013916506295573
Webb, T. L., Joseph, J., Yardley, L., & Michie, S. (2010). Using the internet to promote
1070health behavior change: A systematic review and meta-analysis of the impact of
theoretical basis, use of behavior change techniques, and mode of delivery on efficacy.
Journal of Medical Internet Research(1438-8871 (Electronic)). doi:10.2196/jmir.1376
Webb, T. L., & Sheeran, P. (2006). Does changing behavioral intentions engender behavior
change? A meta-analysis of the experimental evidence. Psychological Bulletin,132(2), 249
1075268. doi:10.1037/0033-2909.132.2.249
Wieber, F., Thürmer, J. L., & Gollwitzer, P. M. (2015). Promoting the translation of
intentions into action by implementation intentions: Behavioral effects and physiologi-
cal correlates. Frontiers in Human Neuroscience,9. doi:10.3389/fnhum.2015.00395
Wing, R. R., & Jeffery, R. W. (2001). Benefits of recruiting friends and increasing social
1080support for weight loss and maintenance. Journal of Consulting and Clinical Psychology,67(1),
132138. doi:10.1037/0022-006X.67.1.132
Wood, W., & Neal, D. T. (2007). A new look at habits and the habit-goal interface.
Psychological Review,114(4), 843863. doi:10.1037/0033-295x.114.4.843
Yom-Tov, E., Kozdoba, M., Mannor, S., Tennenholtz, M., & Hochberg, I. (2017). Encoura-
1085ging physical activity in patients with diabetes: Intervention using a reinforcement
learning system. Journal of Medical Internet Research,19(10), e338. doi:10.2196/jmir.7994
34 P. Pirolli et al.
... Commercial systems such as noom [3] aim to provide psychological support via mobile health (mHealth) systems. Research platforms, such as the Fittle+ system [4], have demonstrated the efficacy of translating known behavior-change techniques [5] into personal mHealth applications. Computational cognitive models have successfully simulated the day-by-day individual-level effects of such interventions in Fittle+ experiments [6,7]. ...
... In this paper, we develop a novel recommendation system that recommends exercise goals with high probability of their successful completion by users. To do this, we build upon (1) a 28-day experimental study of a mobile health system called DStress [12] that evaluated a hand-engineered adaptive exercise recommender, and (2) an explanatory computational cognitive model, based on theories described above, fit post-hoc to the DStress data to predict every individual's observed success likelihood for every recommended exercise every day over 28 days [4]. The novel recommender system we described here, trained with user data: (1) recommends exercises in a way that adapts to individual users without prior hand-engineering and (2) more accurately predicts exercise success rates than the previous computational cognitive model. ...
Preprint
Full-text available
BACKGROUND Unhealthy behaviors, e.g., physical inactivity and unhealthful food choice, are the primary healthcare cost drivers in developed countries. Pervasive computational, sensing, and communication technology provided by smartphones and smartwatches have made it possible to support individuals in their everyday lives to develop healthier lifestyles. OBJECTIVE This paper proposes an exercise recommendation system to recommend daily exercises to elderly population. METHODS The system, consisting of two inter-connected recurrent neural networks (RNNs), uses the history of workouts to recommend the next workout activity for each individual. The system then predicts the probability of successful completion of the predicted activity by the individual. RESULTS The prediction accuracy of this interconnected-RNN model is assessed on previously published data from a four-week mobile health experiment and is shown to improve upon previous predictions from a computational cognitive model. The proposed system is able to predict the next exercise for each individual with 80% accuracy. CONCLUSIONS The dual-RNN system for recommending workout exercises along with predicting individual success rates achieves high accuracy for individuals from whom we do not have any training data. The proposed system was validated this achievement by training the proposed model on a set of users and testing on a new set of test users. Future studies will involve combinations of explanatory computational models such as ACT-R and machine learning approaches such as the dual-RNN system to address the shortcomings of existing recommendations systems in need of large sample size.
... Supervision: van der Weegen et al. (2013) showed that users perceive a health coach who supervises self-monitoring as encouraging and motivating. Pirolli et al. (2021) suggest the incorporation of a virtual chatbot that collaborates with the user/ team in terms of tips, previews and comments on daily activities. ...
Conference Paper
Full-text available
Setting and pursuing goals plays a major role in health behavior change. As an elementary component of interventions and mHealth apps, goals can support people in pursuing and attaining a desired outcome or behavior. Despite its widespread use and indicated effects on behavioral performance, there is no guidance on how to design and implement goal setting, planning and evaluation for mHealth nutrition promotion. We investigate goal setting and related components in the context of mHealth and health behavior change in a structured literature review. By utilizing morphological analysis, we develop a design framework for goal setting and pursuit. We validate and refine the framework by application on popular commercial nutrition apps. As a result, we identify the coverage of goal-setting and -pursuit characteristics as well as current gaps in the implementation of commercial mHealth nutrition apps, which offer the potential for future effective development of goal setting in mHealth nutrition promotion.
... Commercial systems such as noom [3] aim to provide psychological support via mobile health (mHealth) systems. Research platforms, such as the Fittle+ system [4], have demonstrated the efficacy of translating known behaviorchange techniques [5] into personal mHealth applications. However, most work in the mHealth domain are limited to Research reported in this paper was supported by the National Institute on Aging of the National Institutes of Health under award number R01AG053163. ...
Article
Full-text available
Recommendation systems play an important role in today’s digital world. They have found applications in various areas such as music platforms, e.g., Spotify, and movie streaming services, e.g., Netflix. Less research effort has been devoted to physical exercise recommendation systems. Sedentary lifestyles have become the major driver of several diseases as well as healthcare costs. In this paper, we develop a recommendation system to recommend daily exercise activities to users based on their history, profiles and similar users. The developed recommendation system uses a deep recurrent neural network with user-profile attention and temporal attention mechanisms. Moreover, exercise recommendation systems are significantly different from streaming recommendation systems in that we are not able to collect click feedback from the participants in exercise recommendation systems. Thus, we propose a real-time, expert-in-the-loop active learning procedure. The active learner calculates the uncertainty of the recommendation system at each time step for each user and asks an expert for recommendation when the certainty is low. In this paper, we derive the probability distribution function of marginal distance, and use it to determine when to ask experts for feedback. Our experimental results on a mHealth and MovieLens datasets show improved accuracy after incorporating the real-time active learner with the recommendation system.
... Commercial systems such as noom [3] aim to provide psychological support via mobile health (mHealth) systems. Research platforms, such as the Fittle+ system [4], have demonstrated the efficacy of translating known behaviorchange techniques [5] into personal mHealth applications. However, most work in mHealth domain are limited to the development of smartphone applications and connecting subjects and expert knowledge. ...
Preprint
Full-text available
Recommendation systems play an important role in today's digital world. They have found applications in various applications such as music platforms, e.g., Spotify, and movie streaming services, e.g., Netflix. Less research effort has been devoted to physical exercise recommendation systems. Sedentary lifestyles have become the major driver of several diseases as well as healthcare costs. In this paper, we develop a recommendation system for daily exercise activities to users based on their history, profile and similar users. The developed recommendation system uses a deep recurrent neural network with user-profile attention and temporal attention mechanisms. Moreover, exercise recommendation systems are significantly different from streaming recommendation systems in that we are not able to collect click feedback from the participants in exercise recommendation systems. Thus, we propose a real-time, expert-in-the-loop active learning procedure. The active learners calculate the uncertainty of the recommender at each time step for each user and ask an expert for a recommendation when the certainty is low. In this paper, we derive the probability distribution function of marginal distance, and use it to determine when to ask experts for feedback. Our experimental results on a mHealth dataset show improved accuracy after incorporating the real-time active learner with the recommendation system.
Article
Full-text available
Background Traditional psychological theories are inadequate to fully leverage the potential of smartphones and improve the effectiveness of physical activity (PA) and sedentary behavior (SB) change interventions. Future interventions need to consider dynamic models taken from other disciplines, such as engineering (eg, control systems). The extent to which such dynamic models have been incorporated in the development of interventions for PA and SB remains unclear. Objective This review aims to quantify the number of studies that have used dynamic models to develop smartphone-based interventions to promote PA and reduce SB, describe their features, and evaluate their effectiveness where possible. Methods Databases including PubMed, PsycINFO, IEEE Xplore, Cochrane, and Scopus were searched from inception to May 15, 2019, using terms related to mobile health, dynamic models, SB, and PA. The included studies involved the following: PA or SB interventions involving human adults; either developed or evaluated integrated psychological theory with dynamic theories; used smartphones for the intervention delivery; the interventions were adaptive or just-in-time adaptive; included randomized controlled trials (RCTs), pilot RCTs, quasi-experimental, and pre-post study designs; and were published from 2000 onward. Outcomes included general characteristics, dynamic models, theory or construct integration, and measured SB and PA behaviors. Data were synthesized narratively. There was limited scope for meta-analysis because of the variability in the study results. Results A total of 1087 publications were screened, with 11 publications describing 8 studies included in the review. All studies targeted PA; 4 also included SB. Social cognitive theory was the major psychological theory upon which the studies were based. Behavioral intervention technology, control systems, computational agent model, exploit-explore strategy, behavioral analytic algorithm, and dynamic decision network were the dynamic models used in the included studies. The effectiveness of quasi-experimental studies involved reduced SB (1 study; P=.08), increased light PA (1 study; P=.002), walking steps (2 studies; P=.06 and P<.001), walking time (1 study; P=.02), moderate-to-vigorous PA (2 studies; P=.08 and P=.81), and nonwalking exercise time (1 study; P=.31). RCT studies showed increased walking steps (1 study; P=.003) and walking time (1 study; P=.06). To measure activity, 5 studies used built-in smartphone sensors (ie, accelerometers), 3 of which used the phone’s GPS, and 3 studies used wearable activity trackers. Conclusions To our knowledge, this is the first systematic review to report on smartphone-based studies to reduce SB and promote PA with a focus on integrated dynamic models. These findings highlight the scarcity of dynamic model–based smartphone studies to reduce SB or promote PA. The limited number of studies that incorporate these models shows promising findings. Future research is required to assess the effectiveness of dynamic models in promoting PA and reducing SB. Trial Registration International Prospective Register of Systematic Reviews (PROSPERO) CRD42020139350; https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=139350.
Preprint
BACKGROUND Background: Traditional psychological theories are inadequate to fully leverage the potential of smartphones and improve effectiveness of physical activity (PA) and sedentary behaviour (SB) change interventions. Future interventions need to consider dynamic models taken from other disciplines such as engineering (e.g., control systems). It is unclear the extent to which such dynamic models have been incorporated in the development of interventions for PA and SB. OBJECTIVE Objectives: This review aims to quantify the number of studies that have used dynamic models to develop smartphone-based interventions to promote PA and reduce SB, describe their features, and where possible evaluate their effectiveness. METHODS Methods: Databases including PubMed, PsychINFO, IEEE Xplore, Cochrane and SCOPUS were searched from inception to 15 May, 2019 using terms related to mobile health, dynamic models, SB and PA. Included studies involved: PA or SB interventions involving human adults; either developed or evaluated, integrated psychological theory with dynamic theories; used smartphones for the intervention delivery; the interventions were either adaptive or JITAIs; included randomized controlled trials (RCTs), pilot RCTs, quasi-experimental, and pre-post study designs; and published from 2000 onwards. Outcomes included general characteristics, dynamic model, theory/construct integrated, and measured SB and PA behaviours. Data were synthesized narratively. There was limited scope for a meta-analysis because of the variability in the study results. RESULTS Results: A total of 1087 publications were screened, with 11 publications describing 8 studies included in the review. All studies targeted PA and four included SB as well. Social Cognitive Theory (SCT) was the major psychological theory upon which the studies were based. Behavioral Intervention Technology, Control Systems, Computational Agent Model, Explit-Explore strategy, Behavioural Analytic Algorithm, and Dynamic Decision Network were dynamic models used by included studies. Effectiveness results for quasi-experimental studies involved reduced SB (one study; p=.08), increased light PA (one study; p=.002), walking steps (two studies; p=.057 and p<.001), walking time (one study; p=.02), moderate-to-vigorous PA (two studies; p=.08 and p=.81), and non-walking exercise time (one study; p=.31). RCT studies showed increased walking steps (one study; p=.003) and walking time (one study; p=.055). To measure activity, five studies used built-in smartphone sensors (i.e. accelerometers), three of which used phone’s GPS as well, and three studies used wearable activity trackers. CONCLUSIONS Conclusion: To our knowledge, this is the first systematic review to report on smartphone-based studies to reduce SB and promote PA with a focus on integrated dynamic models. Current findings highlight the scarcity of dynamic model-based smartphone studies to reduce SB or to promote PA. However, the limited number of studies incorporating these models show promising findings. Future research is required to assess the effectiveness of dynamic models to promote PA and reduce SB. CLINICALTRIAL Trial Registration: International Prospective Register of Systematic Reviews (PROSPERO) CRD42020139350; https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=139350
Chapter
Advances in computational cognitive psychology have played an important role in understanding and engineering human–information interaction systems. These computational models include several addressing the cognition involved in the human sensemaking process, user models that capture the knowledge that humans acquire from interaction, and how people judge the credibility of online Twitter users who influence decision-making. The models presented in this chapter build on earlier information foraging models in which it is important to model individual-level knowledge and experience because these clearly influence human–information interaction processes. This chapter concludes with a discussion of challenges to computational cognitive models as digital information interaction becomes increasingly pervasive and complex.
Article
Background Multiple types of mobile health (mHealth) technologies are available, such as smartphone health apps, fitness trackers, and digital medical devices. However, despite their availability, some individuals do not own, do not realize they own, or own but do not use these technologies. Others may use mHealth devices, but their use varies in tracking health, behaviors, and goals. Examining patterns of mHealth use at the population level can advance our understanding of technology use for health and behavioral tracking. Moreover, investigating sociodemographic and health-related correlates of these patterns can provide direction to researchers about how to target mHealth interventions for diverse audiences. Objective The aim of this study was to identify patterns of mHealth use for health and behavioral tracking in the US adult population and to characterize the population according to those patterns. Methods We combined data from the 2017 and 2018 National Cancer Institute Health Information National Trends Survey (N=6789) to characterize respondents according to 5 mutually exclusive reported patterns of mHealth use for health and behavioral tracking: (1) mHealth nonowners and nonusers report not owning or using devices to track health, behaviors, or goals; (2) supertrackers track health or behaviors and goals using a smartphone or tablet plus other devices (eg, Fitbit); (3) app trackers use only a smartphone or tablet; (4) device trackers use only nonsmartphone or nontablet devices and do not track goals; and (5) nontrackers report having smartphone or tablet health apps but do not track health, behaviors, or goals. Results Being in the mHealth nonowners and nonusers category (vs all mHealth owners and users) is associated with males, older age, lower income, and not being a health information seeker. Among mHealth owners and users, characteristics of device trackers and supertrackers were most distinctive. Compared with supertrackers, device trackers have higher odds of being male (odds ratio [OR] 2.22, 95% CI 1.55-3.19), older age (vs 18-34 years; 50-64 years: OR 2.83, 95% CI 1.52-5.30; 65+ years: OR 6.28, 95% CI 3.35-11.79), have an annual household income of US $20,000 to US $49,999 (vs US $75,000+: OR 2.31, 95% CI 1.36-3.91), and have a chronic condition (OR 1.69, 95% CI 1.14-2.49). Device trackers also have higher odds of not being health information seekers than supertrackers (OR 2.98, 95% CI 1.66-5.33). Conclusions Findings revealed distinctive sociodemographic and health-related characteristics of the population by pattern of mHealth use, with notable contrasts between those who do and do not use devices to track goals. Several characteristics of individuals who track health or behaviors but not goals (device trackers) are similar to those of mHealth nonowners and nonusers. Our results suggest patterns of mHealth use may inform how to target mHealth interventions to enhance reach and facilitate healthy behaviors.
Article
Full-text available
Background: MHealth interventions can help to improve the physical well-being of participants. Unfortunately, mHealth interventions often have low adherence and high attrition. One possible way to increase adherence is instructing participants to complete self-affirmation exercises. Self-affirmation exercises have been effective in increasing many types of positive behaviors. However, self-affirmation exercises often involve extensive essay writing, a task that is not easy to complete on mobile platforms. Objective: This study aimed to adapt a self-affirmation exercise to a form better suited for delivery through a mobile app targeting healthy eating behaviors, and to test the effect of differing self-affirmation doses on adherence to behavior change goals over time. Methods: We examined how varied self-affirmation doses affected behavior change in an mHealth app targeting healthy eating that participants used for 28 days. We divided participants into the 4 total conditions using a 2×2 factorial design. The first independent variable was whether the participant received an initial self-affirmation exercise. The second independent variable was whether the participant received ongoing booster self-affirmations throughout the 28-day study. To examine possible mechanisms through which self-affirmation may cause positive behavior change, we analyzed three aspects of self-affirmation effects in our research. First, we analyzed how adherence was affected by self-affirmation exercises. Second, we analyzed whether self-affirmation exercises reduced attrition rates from the app. Third, we examined a model for self-affirmation behavior change. Results: Analysis of 3556 observations from 127 participants indicated that higher doses of self-affirmation resulted in improved adherence to mHealth intervention goals (coefficient 1.42, SE 0.71, P=.04). This increased adherence did not seem to translate to a decrease in participant attrition (P value range .61-.96), although our definition of attrition was conservative. Finally, we examined the mechanisms by which self-affirmation may have affected intentions of behavior change; we built a model of intention (R²=.39, P<.001), but self-affirmation did not directly affect final intentions (P value range .09-.93). Conclusions: Self-affirmations can successfully increase adherence to recommended diet and health goals in the context of an mHealth app. However, this increase in adherence does not seem to reduce overall attrition. The self-affirmation exercises we developed were simple to implement and had a low cost for both users and developers. While this study focused on an mHealth app for healthy eating, we recommend that other mHealth apps integrate similar self-affirmation exercises to examine effectiveness in other behaviors and contexts. © Aaron Springer, Anusha Venkatakrishnan, Shiwali Mohan, Lester Nelson, Michael Silva, Peter Pirolli.
Article
Full-text available
Background Implementation intentions are mental representations of simple plans to translate goal intentions into behavior under specific conditions. Studies show implementation intentions can produce moderate to large improvements in behavioral goal achievement. Human associative memory mechanisms have been implicated in the processes by which implementation intentions produce effects. On the basis of the adaptive control of thought-rational (ACT-R) theory of cognition, we hypothesized that the strength of implementation intention effect could be manipulated in predictable ways using reminders delivered by a mobile health (mHealth) app. Objective The aim of this experiment was to manipulate the effects of implementation intentions on daily behavioral goal success in ways predicted by the ACT-R theory concerning mHealth reminder scheduling. Methods An incomplete factorial design was used in this mHealth study. All participants were asked to choose a healthy behavior goal associated with eat slowly, walking, or eating more vegetables and were asked to set implementation intentions. N=64 adult participants were in the study for 28 days. Participants were stratified by self-efficacy and assigned to one of two reminder conditions: reminders-presented versus reminders-absent. Self-efficacy and reminder conditions were crossed. Nested within the reminders-presented condition was a crossing of frequency of reminders sent (high, low) by distribution of reminders sent (distributed, massed). Participants in the low frequency condition got 7 reminders over 28 days; those in the high frequency condition were sent 14. Participants in the distributed conditions were sent reminders at uniform intervals. Participants in the massed distribution conditions were sent reminders in clusters. Results There was a significant overall effect of reminders on achieving a daily behavioral goal (coefficient=2.018, standard error [SE]=0.572, odds ratio [OR]=7.52, 95% CI 0.9037-3.2594, P<.001). As predicted by ACT-R, using default theoretical parameters, there was an interaction of reminder frequency by distribution on daily goal success (coefficient=0.7994, SE=0.2215, OR=2.2242, 95% CI 0.3656-1.2341, P<.001). The total number of times a reminder was acknowledged as received by a participant had a marginal effect on daily goal success (coefficient=0.0694, SE=0.0410, OR=1.0717, 95% CI −0.01116 to 0.1505, P=.09), and the time since acknowledging receipt of a reminder was highly significant (coefficient=−0.0490, SE=0.0104, OR=0.9522, 95% CI −0.0700 to −0.2852], P<.001). A dual system ACT-R mathematical model was fit to individuals’ daily goal successes and reminder acknowledgments: a goal-striving system dependent on declarative memory plus a habit-forming system that acquires automatic procedures for performance of behavioral goals. Conclusions Computational cognitive theory such as ACT-R can be used to make precise quantitative predictions concerning daily health behavior goal success in response to implementation intentions and the dosing schedules of reminders.
Article
Full-text available
Background Regular physical activity is known to be beneficial for people with type 2 diabetes. Nevertheless, most of the people who have diabetes lead a sedentary lifestyle. Smartphones create new possibilities for helping people to adhere to their physical activity goals through continuous monitoring and communication, coupled with personalized feedback. Objective The aim of this study was to help type 2 diabetes patients increase the level of their physical activity. Methods We provided 27 sedentary type 2 diabetes patients with a smartphone-based pedometer and a personal plan for physical activity. Patients were sent short message service messages to encourage physical activity between once a day and once per week. Messages were personalized through a Reinforcement Learning algorithm so as to improve each participant’s compliance with the activity regimen. The algorithm was compared with a static policy for sending messages and weekly reminders. Results Our results show that participants who received messages generated by the learning algorithm increased the amount of activity and pace of walking, whereas the control group patients did not. Patients assigned to the learning algorithm group experienced a superior reduction in blood glucose levels (glycated hemoglobin [HbA1c]) compared with control policies, and longer participation caused greater reductions in blood glucose levels. The learning algorithm improved gradually in predicting which messages would lead participants to exercise. Conclusions Mobile phone apps coupled with a learning algorithm can improve adherence to exercise in diabetic patients. This algorithm can be used in large populations of diabetic patients to improve health and glycemic control. Our results can be expanded to other areas where computer-led health coaching of humans may have a positive impact. Summary of a part of this manuscript has been previously published as a letter in Diabetes Care, 2016.
Article
Full-text available
PurposeThis systematic review aims to explain the heterogeneity in results of interventions to promote physical activity and healthy eating for overweight and obese adults, by exploring the differential effects of behaviour change techniques (BCTs) and other intervention characteristics. Methods The inclusion criteria specified RCTs with ? 12 weeks? duration, from January 2007 to October 2014, for adults (mean age ? 40 years, mean BMI ? 30). Primary outcomes were measures of healthy diet or physical activity. Two reviewers rated study quality, coded the BCTs, and collected outcome results at short (?6 months) and long term (?12 months). Meta-analyses and meta-regressions were used to estimate effect sizes (ES), heterogeneity indices (I2) and regression coefficients. ResultsWe included 48 studies containing a total of 82 outcome reports. The 32 long term reports had an overall ES = 0.24 with 95% confidence interval (CI): 0.15 to 0.33 and I2 = 59.4%. The 50 short term reports had an ES = 0.37 with 95% CI: 0.26 to 0.48, and I2 = 71.3%. The number of BCTs unique to the intervention group, and the BCTs goal setting and self-monitoring of behaviour predicted the effect at short and long term. The total number of BCTs in both intervention arms and using the BCTs goal setting of outcome, feedback on outcome of behaviour, implementing graded tasks, and adding objects to the environment, e.g. using a step counter, significantly predicted the effect at long term. Setting a goal for change; and the presence of reporting bias independently explained 58.8% of inter-study variation at short term. Autonomy supportive and person-centred methods as in Motivational Interviewing, the BCTs goal setting of behaviour, and receiving feedback on the outcome of behaviour, explained all of the between study variations in effects at long term. Conclusion There are similarities, but also differences in effective BCTs promoting change in healthy eating and physical activity and BCTs supporting maintenance of change. The results support the use of goal setting and self-monitoring of behaviour when counselling overweight and obese adults. Several other BCTs as well as the use of a person-centred and autonomy supportive counselling approach seem important in order to maintain behaviour over time. Trial RegistrationPROSPERO CRD42015020624
Conference Paper
Full-text available
Our research aims at developing interactive, social agents that can coach people to learn new tasks, skills, and habits. In this paper, we focus on coaching sedentary, overweight individuals to exercise regularly. We employ adaptive goal setting in which the coach generates, tracks, and revises personalized exercise goals for a trainee. The goals become incrementally more difficult as the trainee progresses through the training program. Our approach is model-based-the coach maintains a parameterized model of the trainee's aerobic capability that drives its expectation of the trainee's performance. The model is continually revised based on interactions with the trainee. The coach is embodied in a smartphone application which serves as a medium for coach-trainee interaction. We show that our approach can adapt the trainee program not only to several trainees with different capabilities but also to how a trainee's capability improves as they begin to exercise more. Experts rate the goals selected by the coach better than other plausible goals, demonstrating that our approach is effective.
Conference Paper
Full-text available
Computational models were developed in the ACT-R neurocognitive architecture to address some aspects of the dynamics of behavior change. The simulations aim to address the day-today goal achievement data available from mobile health systems. The models refine current psychological theories of self-efficacy, intended effort, and habit formation, and provide an account for the mechanisms by which goal personalization, implementation intentions, and remindings work.
Article
Smartphones now offer the promise of collecting behavioral data unobtrusively, in situ, as it unfolds in the course of daily life. Data can be collected from the onboard sensors and other phone logs embedded in today’s off-the-shelf smartphone devices. These data permit fine-grained, continuous collection of people’s social interactions (e.g., speaking rates in conversation, size of social groups, calls, and text messages), daily activities (e.g., physical activity and sleep), and mobility patterns (e.g., frequency and duration of time spent at various locations). In this article, we have drawn on the lessons from the first wave of smartphone-sensing research to highlight areas of opportunity for psychological research, present practical considerations for designing smartphone studies, and discuss the ongoing methodological and ethical challenges associated with research in this domain. It is our hope that these practical guidelines will facilitate the use of smartphones as a behavioral observation tool in psychological science.
Chapter
A classroom of 7th grade students is developing a scientific model of the factors influencing water quality in their local stream. They run a dynamic simulation of the model to test it, yet they are able to do so without having to produce sophisticated mathematical representations for these relationships. In another classroom, 11th graders are reading primary historical texts and engaging in argumentation to develop a coherent explanation of events, despite lacking the extensive disciplinary knowledge and experience historians use to analyze primary documents. How can these learners participate in activities that share elements of expert practices, but that call on knowledge and skills that they do not yet possess? These feats are possible in the same way that young children can ride two-wheelers using training wheels before they have mastered balancing, or that construction workers can use scaffolding to work on the penthouse before the lower floors have been fully constructed. The Historical Roots of Scaffolding Drawing on the metaphor of scaffolding in building construction, Wood, Bruner, and Ross (1976) proposed the concept of scaffolding to describe how children, with the help of someone more knowledgeable to share and support their problem solving, can perform more complex tasks than they would otherwise be capable of performing on their own (Palincsar, 1998; Rogoff, 1990). Scaffolding may be provided by a variety of different mechanisms. In the example cited earlier, in the history classroom, scaffolding is provided by interaction with guidance from the teacher and curriculum materials (Reisman, 2012), while in the water quality model, part of the assistance is provided by a supportive software environment (Fretz et al., 2002).