Behavior Change Technologies can address key societal problems – from global warming, to the rising cost of healthcare worldwide, and emerging concerns of the technological age, such as online privacy and the propagation of misinformation online. But are the technologies we develop grounded on theories of behavior change? And, if not, why? In this talk we will argue for the need for theoretically and empirically grounded design, and will present our recent work on making behavioral theory accessible to design teams, along with empirical studies of the adoption, engagement with, and impact of behavior change technologies in the context of health.
** Presentation given at the "Considering Health Behavior Change" Symposium, on Feb 11, 2020, Eindhoven, The Netherlands.
4. tiny share of U.S. workers and use 5 million tractors in place of the horses
and mules of earlier days.
As a result of this transformation, U.S. agriculture has become increasingly
efficient and has contributed to the overall growth of the U.S. economy.
Output from U.S. farms has grown dramatically, allowing consumers to
spend an increasingly smaller portion of their income on food and freeing a
large share of the population to enter nonfarm occupations that have
supported economic growth and development. As a part of the transforma-
tion spurred by technological innovation and changing market conditions,
production agriculture has become a smaller player in the national and rural
economies. While the more broadly defined food and agriculture sector
continues to play a strong role in the national economy, farming has
progressively contributed a smaller share of gross domestic product (GDP)
and employed a smaller share of the labor force over the course of the
century (see box, “Farming’s changing role in the Nation’s economy”). Over
the same period, the share of the U.S. population living on farms also
declined (fig. 1), as did agriculture’s central role in the rural economy; while
farming-dependent counties once comprised most of the rural economy,
only 20 percent of nonmetro counties were considered farming-dependent in
2000 (fig. 2).
The altered role of farming in the overall economy reflects changes at the
farm and farm household level. Since 1900, the number of farms has fallen
by 63 percent, while the average farm size has risen 67 percent (fig. 3).
Farm operations have become increasingly specialized as well (fig. 4)—
from an average of about five commodities per farm in 1900 to about one
per farm in 2000—reflecting the production and marketing efficiencies
gained by concentration on fewer commodities, as well as the effects of
farm price and income policies that have reduced the risk of depending on
returns from only one or a few crops. All of this has taken place with almost
no variation in the amount of land being farmed.
Farm households have adapted as dramatic increases in productivity have
reduced the need for household labor on the farm, and as alternative
employment opportunities have developed in nearby rural and metro
economies. Although measures of off-farm work and income have varied
over the century, making comparisons over time difficult, about a third of
1900
41 percent of workforce
employed in agriculture
1930
21.5 percent of workforce
employed in agriculture
Agricultural GDP as a share
of total GDP, 7.7 percent
1945
16 percent of the total labor
force employed in agriculture
Agricultural GDP as a share
of total GDP, 6.8 percent
1970
4 percent of employed labor
force worked in agriculture
Agricultural GDP as a share
of total GDP, 2.3 percent
2000/02
1.9 percent of employed
labor force worked in
agriculture (2000)
Agricultural GDP as a share
of total GDP (2002),
0.7 percent
Source: Compiled by Economic
Research Service, USDA. Share
of workforce employed in agricul-
ture, for 1900-1970, Historical
Statistics of the United States; for
2000, calculated using data from
Census of Population; agricultural
GDP as part of total GDP, calcu-
lated using data from the Bureau
of Economic Analysis.
Farming’s changing role
in the Nation’s economy
gure 2
onmetro farming-dependent counties, 1950 and 2000
2000
1950
2000
Nonmetro farming-dependent
Other nonmetro
Metro
Source: Economic Research Service, USDA. Farming-dependent counties are defined by ERS.
Dimitri, C., Effland, A., Conklin, N. (2005) The 20th Century Transformation of U.S. Agriculture and Farm Policy,
Economic Information Bulletin Number 3.
US workforce employed in agriculture
dropped from 41% to 1.9% in the 20th
century
2000
Nonmetro farming-dependent
Other nonmetro
Metro
Source: Economic Research Service, USDA. Farming-dependent counties are defined by ERS.
For 1950, at least 20 percent of income in the county was derived from agriculture. For 2000,
either 15 percent or more of average annual labor and proprietors' earnings were derived from
farming during 1998-2000 or 15 percent or more of employed residents worked in farm occupa-
tions. Metro/nonmetro status is based on the Office of Management and Budget (OMB) June
2003 classification.
5. 1200
steps
Physically active jobs make less than 20% of the
occupations (1950s: 50%)
American Heart Association (2015) The Price of Inactivity, Retrieved online from Heart.org, 4 February 2020.
7. 1200
steps
a dose-response association between sitting
time and mortality from all causes and CVD,
independent of leisure time physical
activity.
those who spent most of their time sitting were 50% more likely to
die during the follow-up than those that sit the least, even after
controlling for age, smoking, and physical activity levels.
Katzmarzyk, P. T., Church, T. S., Craig, C. L., & Bouchard, C. (2009). Sitting time and mortality from all causes,
cardiovascular disease, and cancer. Medicine & Science in Sports & Exercise, 41(5), 998-1005.
8. 1200
stepsFrom infectious diseases (as the primary cause
of illness, mortality, and healthcare
expenditures), to chronic, non-
communicable conditions (‘diseases of
lifestyle’)
9. 1200
steps
From cure to prevention
Gordon Brown: ”NHS [National Health Service] of
the future [being] one of patient power, with
patients engaged and taking control over their
own health and healthcare".
13. But, are those technologies grounded on
theories of behavior change?
14. •Cowan et al. (2013): mean behavioural score of 10/100
•Azar et al. (2013): mean behavioural score of 8.1/100
•Riley et al. (2011): the use of theory varied substantially per
domain:
•1/20 of disease management interventions were theory
based,
•0/10 in treatment adherence
•7/12 in weight loss
•5/7 in smoking cessation
Apps lack theoretical content
15. 83 Theories of Behavior & Behavior Change
Davis et al., 2015
"designers and researchers are having a hard time
deciding with confidence which of the theories and
techniques to use in their design and research”
Michie & Prestwich (2010)
Abundance of theories
16. Can we use design cards to make theory more
accessible during design meetings?
17. Design cards as a design tool for providing for knowledge
transfer - the translation of research findings from one discipline
into another (Rogers, 2004)
• they make the design process visible and less abstract,
• they communicate knowledge between the group members
• they increase creativity and idea generation
(Wolfer & Merit 2013)
19. Behavior Change Design Cards
Transtheoretical model / Stages of behavior change
Figure reproduced from Kersten-van Dijk et al., (2017)
Kersten-van Dijk, E. T., Westerink, J. H., Beute, F., & IJsselsteijn, W. A. (2017). Personal informatics, self-insight, and behavior change: A critical
review of current literature. Human–Computer Interaction, 32(5-6), 268-296.ISO 690
20. Behavior Change Design Cards
5 stage of change cards 33 technique cards
Pre-contemplation
The individual has no intention to change
the behavior, yet.
NO, NOT ME.
Designing for
Pre-contemplation
How would you increase the user’s awareness of
the need for change?
How would you lead the user to understand the
right decision (i.e. perceived procs and cons of
behavior)?
person’s belief that they are capable of adopting
a new pattern of behavior)?
Possible Directions:
Show attention to problematic behaviors;
informing about long term consequences;
think of alternative behaviors;
Goal Setting
Setting a goal to increase physical activity by
walking the dog at the park for 3 miles a day, for
3-4 times a week.
the behavior to be achieved or the
outcome of the wanted behavior.
ActionMaintenanceContemplationPreparationPreContemplation
How will you guide the user in setting an
appropriate goal?
What types of goals are you designing for: behavior
(eg. steps) or outcome (e.g. weight loss)?
How will the feedback provided challenge the
Goal setting
HINT:
Goals that are self-set, important to
effective.
ActionMaintenanceContemplationPreparationPreContemplation
21. The Nudge Deck
Nudging
any aspect of the choice architecture that alters people's behavior in a
predictable way without forbidding any option or signicantly changing their
economic incentives
Thaler and Sunstein (2008)
Caraban, A., Karapanos, E., Gonçalves, D., & Campos, P. (2019). 23 Ways to Nudge: A Review of Technology-Mediated Nudging in Human-
Computer Interaction. In Proceedings of CHI’19 (p. 503). ACM.
22. Simply moving bottles of water (instead of soda bottles) so that they were at eye-level in
the kitchens at Google increased water uptake by a whopping 47% (Kuang 2012)
The Nudge Deck
23. we still lack an understanding of how to design
effective technology-mediated nudges
• the why of nudging (i.e., which cognitive biases can nudges
combat),
• the how of nudging (i.e., what exact mechanisms can
nudges employ to incur behavior change)
The 23 ways to nudge framework
Caraban, A., Karapanos, E., Gonçalves, D., & Campos, P. (2019). 23 Ways to Nudge: A Review of Technology-Mediated Nudging in Human-
Computer Interaction. In Proceedings of CHI’19 (p. 503). ACM.
24. CHI, 32 papers Persuasive, 10 Ubicomp, 5 Others, 24
Venues
71
papers selected
Caraban, A., Karapanos, E., Gonçalves, D., & Campos, P. (2019). 23 Ways to Nudge: A Review of Technology-Mediated Nudging in Human-
Computer Interaction. In Proceedings of CHI’19 (p. 503). ACM.
25. Caraban, A., Karapanos, E., Gonçalves, D., & Campos, P. (2019). 23 Ways to Nudge: A Review of Technology-Mediated Nudging in Human-
Computer Interaction. In Proceedings of CHI’19 (p. 503). ACM.
23 Ways to Nudge:
The Framework
What 6 categories
Why 15 cognitive biases
How 23 nudging mechanisms
26. Reminding the consequences
Availability heuristic: our tendency to judge the probability of
occurrence of an event based on the ease at which it can be
recalled
27. The Nudge Deck
Caraban, A., Karapanos, E., Gonçalves, D., & Campos, P. (2019). 23 Ways to Nudge: A Review of Technology-Mediated Nudging in Human-
Computer Interaction. In Proceedings of CHI’19 (p. 503). ACM.
28. Defining the problem and laying out directions for design
“I see these cards as personas, one does not have the motivation… think of someone like
Jennifer”
Provided a common vocabulary
Jump-off ideation: helped shifted focus when discussion was becoming unproductive
“the prize should be something that she likes, based on her preferences … [extended pause] …
let’s see this one”
Participants did not fixate on the examples provided
“I only saw the examples in the beginning. I used to focus on the category and on the questions of
the mechanism selected”
Facilitate collaborative work and support lateral thinking
Design considerations & hints used as heuristics for evaluation
Did cards support the design process?
29. Did cards support the design process?
Figure 4. Participants’ self-reported Self-Efficacy increased after the
design session across both conditions (control vs experimental) and cases
(physical activity vs misinformation).
(control vs Nudge Deck) and case (physical activity vs mis-
information) as independent variables revealed no sıgnıfıcant
main effects for condıtıon (F(1,55) = 0.30, p > .05 , h2
p = .3005,
control: M = 7.54, SD = 1.51, Nudge Deck: M = 7.29, SD
= 1.17) and for the case (F(1,55) = 2.48, p > .05 , h2
p = .04,
physical activity: M = 7.14, SD = 1.37, misinformation: M =
7.66, SD = 1.26), and no interaction effect.
th
in
ca
p
th
(m
H
A
5
5
in
4
th
fe
to
ca
in
to
u
ca
“T
th
H
o
Figure 4. Participants’ self-reported Self-Efficacy increased after the
design session across both conditions (control vs experimental) and cases
(physical activity vs misinformation).
(control vs Nudge Deck) and case (physical activity vs mis-
information) as independent variables revealed no sıgnıfıcant
main effects for condıtıon (F(1,55) = 0.30, p > .05 , h2
p = .3005,
control: M = 7.54, SD = 1.51, Nudge Deck: M = 7.29, SD
= 1.17) and for the case (F(1,55) = 2.48, p > .05 , h2
p = .04,
physical activity: M = 7.14, SD = 1.37, misinformation: M =
7.66, SD = 1.26), and no interaction effect.
Figure 5. Participants reported significantly higher reward/effort trade-
off and perceived expressiveness in the misinformation case. An inter-
action effect between case and conditon was observed in related to per-
ceived expressiveness.
A closer look at the ındıvıdual ıtems of the CSI questıonnaıre
revealed a significant effect of the case (misinformation vs
physical activity) on expressiveness ("I was able to be very
expressive and creative during the activity"; F(1,50) = 5.4, p
< .05 , h2
p = .098) and a significant interaction effect between
case and condition (F(1,50) = 5.5, p < .05 , h2
p = .099)). Sim-
ilarly, we found a significant effect of case on effort/reward
trade off ("What I was able to produce was worth the effort
I had to exert to produce it"; F(1,55) = 4.4, p < .05 , h2
p =
.074). Participants felt more expressive and thought that what
the cards (M = 4.9, STD =1.2). Some participants reported
feeling overwhelmed at the start of the session, struggling
to understand the relation between the different levels of the
cards (i.e., triggers, categories and mechanisms). This feel-
ing of confusion, however, dissipated as participants started
to explore the cards P[7]: “At the beginning it was hard to
understand all the cards and to connect the different types of
cards but after a while you understand and it helps”, P[3]
“The part that was hard was to understand the connection of
the different levels”.
How did the Nudge Deck influence the quality of design
output?
Theoretical Grounding
Given that participants in the experimental condition were
provided with the Nudge Deck, it was natural to expect that
their ideas would have a stronger grounding to theory, as
compared to the control condition, where participants had no
interaction with theoretical content, apart from the definition
of nudging and the provision of one example. As expected,
ideas produced in the experimental condition were rated as
more theoretically grounded than ones produced in the control
condition. A two-way ANOVA with theoretical grounding as
the dependent variable and condition (control vs Nudge Deck)
and case (physical activity vs misinformation) as indepen-
dent variables revealed a significant main effect for condition
(F(1,19) = 6.3, p < .05 , h2
p = .25, control: M = 4.42, SD =
2.38, experimental: M = 6.91, SD = 2.96) as well as for the
case (F(1,19) = 5.6, p < .05 , h2
p = .23, physical activity: M =
4.50, SD = 3.32, misinformation: M = 6.91, SD = 2.96), and
no interaction effect (see figure 6).
For instance, looking at the control condition, participants in
the physical activity case often drew inspiration from their
own personal experience using activity trackers and employed
features such as prompting (N=6), social comparison and
social support (N=6), self-monitoring (N=3), feedback on
performance (N=3), goal-setting (N=2) and rewards (N=1).
While these features are based on theoretically and empirically
grounded behavior change techniques (see [36]), participants
only superficially drew on these techniques without much elab-
oration on their functioning. For instance, when designing a
goal-setting feature, participants did not elaborate on how they
will engage users to self-set a concrete, and challenging yet
attainable goal, which is considered to be a key predictor to the
success of goal-setting [33, 22]. Similarly, when designing for
Figure 6. The Nudge Deck led to more theoretically grounded, fit to con-
text of use, and creative ideas. Ideas from the misinformation case were
more theoretically grounded and creative than ones from the physical
activity case.
social comparison, they didn’t reflect on ways to support ap-
propriate comparisons (e.g., comparisons to users with similar
activity level), that have been found to lead to higher perfor-
mance [23]. In contrast, ideas resulting from the experimental
Fitness to the context of use
Drawing on, and combing different mechanisms during
ideation made participants better able to design solutions that
fit the context of use. A two-way ANOVA with fitness to the
context of use as the dependent variable and condition (control
vs Nudge Deck) and case (physical activity vs misinformation)
as independent variables revealed a significant main effect for
condition (F(1,19) = 8.0, p < .05 , h2
p = .30, control: M = 4.42,
SD = 2.38, experimental: M = 6.91, SD = 2.96) but not for the
case (F(1,19) = 0.4, p > .05 , h2
p = .21, physical activity: M =
4.50, SD = 3.32, misinformation: M = 6.91, SD = 2.96), and
no interaction effect.
We observed that the cards supported participants in search-
ing for strategies and mechanisms that better suit the context
under concern. For instance, when designing interventions
to promote physical activity, participants would narrow their
inquiry based on the most relevant trigger category. Trigger
cards, thus, served as a way to understand the problem they
wanted to solve (e.g., do users fail to perform sufficient lev-
els of physical activity because they lack the motivation, the
ability, or do they simply need a reminder?). Based on the
answer to the above questions, participants would explore the
mechanisms that were most relevant to the identified trigger.
Creativity
A two-way ANOVA with creativity as the dependent variable
and condition (control vs Nudge Deck) and case (physical
Quality of ideas Experienced Creativity Self-efficacy
32. 1200
steps
User engagement
Gouveia, R., Karapanos, E., & Hassenzahl, M. (2015). How do we engage with activity trackers? A longitudinal study of Habito. In
Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing (pp. 1305-1316).
Gouveia, R., Pereira, F., Karapanos, E., Munson, S. A., & Hassenzahl, M. (2016). Exploring the design space of glanceable feedback for
physical activity trackers. In Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing (pp.
144-155).
33. 1200
steps
1. Did people adopt the technology?
2. Was their use of the technology in line with what expected from theory?
3. Can we measure the proximal impact of each engagement on people’s
behaviors?
40. 256 users downloaded Habito over
the course of 10 months
none of these users were recruited or rewarded towards usage
41. 62% (159) stopped using Habito
within their first week of use
97 adopters, which used the app for more than a week
42. 1a. Did all people equally adopt the
technology?
43. stages of behavior change
questionnaire
understanding how different stages of ‘readiness’ impacted adoption
precontemplation currently have no intention of being active
contemplation not active but intend to be soon
preparation trying, but not regularly active
action regularly active, but for less than 6 months
maintenance regularly active for 6 months or more
44. precontemplation
5 of 36, 14%
contemplation
preparation
action
maintenance
14 of 26, 54%
19 of 33, 58%
7 of 24, 29%
4 of 19, 21%
Readiness for use:
motivation and adoption
45. 2. Was their use of the technology in line
with what expected from theory?
46. Figure reproduced from Li et al. (2010)
Li, I., Dey, A., & Forlizzi, J. (2010, April). A stage-based model of personal informatics systems. In Proceedings of the SIGCHI conference on
human factors in computing systems (pp. 557-566).
48. Glances
sessions in which users open and
close Habito with no additional
actions or inputs
57%, 5 sec
Review Engage
22%,12 sec 21%,45 sec
sessions with at least one
additional actions and last
up to 22 seconds
sessions with at least one
additional actions and last
more than 22 seconds
Usage sessions
50. Figure reproduced from Li et al. (2010)
Li, I., Dey, A., & Forlizzi, J. (2010, April). A stage-based model of personal informatics systems. In Proceedings of the SIGCHI conference on
human factors in computing systems (pp. 557-566).
51. Exploring the Design Space of Glanceable
Feedback for Physical Activity Trackers
Ruben Gouveia, Fábio Pereira, Evangelos Karapanos, Sean Munson & Marc Hassenzahl
1200
steps
52. #1 what are some of the attributes that GFI should have for activity
trackers?
53. #1 Abstract
#2 Integrate with existing activities
#3 Support comparison to targets and norms
#4 Actionable
#5 Lead to checking habits
#6 Act as proxy to further engagement