personal informatics and context: using context to reveal factors that affect behavior

Ian Li Personal Informatics+Context Thesis Proposal

Personal Informatics+ContextUsing Context to Reveal Factors that Affect Behavior

Ian Li Anind Dey, CMU, Co-chair Jodi Forlizzi, CMU, Co-chair Niki Kittur, CMU John Stasko, Georgia Tech

Ian Li Personal Informatics+Context Thesis Proposal 2

Alice •  20 years old •  Family history of heart

disease •  Wants to be more active,

but donʼt know how because sheʼs busy


Mon 1573 Tue 4392 Wed 4537 Thu 5842 Fri 10258 Sat 7528 Sun 1368 Mon 1497 Tue 1837


M T W Th F Sa Su M T

Transcribe to Excel


Active

Inactive Inactive



Active

Inactive Inactive


Factors •  Lack of time •  Lack of motivation •  Activities •  Location •  People


Active

Inactive Inactive

Location Office

Activity Shopping

People Family



Problem Pedometer only recorded one type of information.

9

Location Activity People


Physical Activity

Finance

Health

Mood

Electricity

Diabetes

http://personalinformatics.org/tools


Thesis A personal informatics system that allows users to associate context with behavioral information can better reveal factors that affect behavior, compared to systems that only show behavioral information.

11


Thesis Questions

12

Is a PI system with context better at revealing factors that affect behavior? •  YES, I will show this in 3 completed

studies and my proposed work.



Location: Park Activity: Hiking People: Friend Step counts: 7531



UbiComp Sensors

Data mining


Thesis Questions

15

How do we build a PI system with context? •  Alice had to do a lot to get data and reflect

on them. •  Issues collecting data? Reflecting on

data? •  It is not as easy as just automating the

system. •  Whether the system is manual or

automated has an effect on the userʼs awareness.


Introduction

Stage-Based Model of PI

Prototypes

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


Introduction


Prototypes

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


Thesis Questions

18

Is a PI system with context better at revealing factors that affect behavior?

How do we build a PI system with context? •  Create a framework as guide in designing

personal informatics systems. •  Building a PI system involves many parts

each with their own HCI issues.


Survey and Interviews 68 people who use personal informatics

Advertised the survey in blogs about personal informatics.

What tools they use and their problems

19


Sample Questions •  How difficult is it to collect this personal

information? •  How do you explore this collected personal

information? •  What patterns have you found?

Transcript of the survey is at: http://personalinformatics.org/lab/survey

20


Analysis Identified barriers that people experienced.

Affinity diagrams to identify themes.

Derived a model composed of: •  5 stages

21


COLLECTION REFLECTIONPREPARATION INTEGRATION ACTION

5 Stages

22



5 Stages

23



Alice •  Wanted to become

active •  Decided to track her

physical activity •  Chose to track step

counts using a pedometer



26


Transcribe to Excel



Active

Inactive Inactive



The stage when people choose what they are going to do with their new-found understanding of themselves. •  Alerts •  Incentives •  Suggestions



Other research have explored these different stages in isolation •  Collection •  MyLifeBits (Gemmell et al. 2006)

•  SenseCam (Hodges et al. 2006)

•  Reflection •  Casual InfoVis (Pousman et al. 2007)

29



1. Barriers cascade. 2. Stages are iterative.

30



1. Barriers Cascade. Problems in the earlier stages can affect the later stages.

31


1. Barriers Cascade.

32

Active

Inactive Inactive

Location Office

Activity Shopping

People Family



1. Barriers Cascade. Problems in the earlier stages can affect the later stages.

→ Consider all the stages when designing PI systems.

33


2. Stages are Iterative. Users may need to incorporate new types of data, tools, and processes as they progressed through the stages.

34


2. Stages are Iterative.

35


Location: Park Activity: Hiking People: Friend Step counts: 7531


2. Stages are Iterative. Users may need to incorporate new types of data, tools, and processes as they progress through the stages.

→ Flexibility is important, but consider user needs early to minimize missed data.

36


1. Barriers cascade. 2. Stages are iterative. 3. User- or system-driven 4. Uni- or multi-faceted

37



3. User- vs. System-driven The stages can be user-driven, system-driven, or a combination of both.

38


3. User- vs. System-driven Collection Combination

Integration User-driven

Reflection System-driven

39


Transcribe to Excel

Excel graphs


3. User- vs. System-driven Collection System-driven

Integration System-driven

Reflection System-driven

40


3. User- vs. System-driven The stages can be user-driven, system-driven, or a combination of both.

→ Explore the tradeoffs between user-driven and system-driven stages.

41


Uni- vs. Multi-faceted Most personal informatics are uni-faceted.

Some personal informatics systems have multi-faceted collection, but only support uni-faceted reflection.

42


Uni- vs. Multi-faceted Users expressed desire to see associations between different facets of their lives. •  “To understand trends in symptoms,

behaviors, and circumstances.” P26 •  “If it were easily collected, information on

food intake, calories, fat, etc., would make an interesting starting point for analysis.” P49 who tracks medication intake

43


Uni- vs. Multi-faceted

44

Active

Inactive Inactive

Location Office

Activity Shopping

People Family

M T Th F Sa Su M T


Uni- vs. Multi-faceted Most personal informatics are uni-faceted.

→ Explore support for multiple facets throughout the stages. •  I explore using contextual

information.

45


Thesis Questions

46


How do we build a PI system with context? •  Created a framework to analyze PI

systems. •  When designing, consider all the stages.


Introduction


Prototypes

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


Thesis Questions

48

Is a PI system with context better at revealing factors that affect behavior? → Deploy prototypes in field studies.

How do we build a PI system with context? → Build prototypes that explore different

ways of supporting context.


Domain: Physical Activity

49


Why Physical Activity? Lack of physical activity is a common problem that leads to obesity, diabetes, and high blood pressure.

Lack of awareness of physical activity is one reason why people are not active.

50


Physical Activity Awareness

51

Products

Research

UbiFitConsolvo et al. ʼ08

Shakra Maitland et al. ʻ06

Fish ʻn Steps Lin et al. ʻ06

want to wait for an update, he or she can manuallysynchronise via the Sync menu option.

Users specify in advance the peers they wish to shareresults with, but at any time they can change this list.Figure 2a shows the Compare Daily Activity screen thatusers can view to assess their performance in relation totheir peers. For a week’s overview of their own activity,users may use the Week’s Activity screen shown in Fig. 2b.In order to provide real time feedback to the user, ananimated representation of the user’s current mode ofactivity runs continuously on the main screen of theapplication. This is shown in Fig. 2c and d.

3.1 Sensing activity

The current activity of the user is inferred using patterns offluctuation in GSM signal strength and changes to the IDsof detected cells. This method has been demonstrated as areliable and unobtrusive way of sensing current activity [2],and has the advantage over the more traditional approach ofusing an accelerometer in that it does not require additionalsensor hardware as in Sensay [17] and the multimodalsensor board of [11]. Similarly, while the processing ofphysiological and biometric data could complement ourapproach, the benefits of encapsulating the system within amobile phone would be lost. An alternative approach wouldbe to utilise the positioning information available fromsome mobile phone networks, however this approachfrequently involves prohibitive cost, as well as dependingupon much of the same technology as our client basedmonitoring.

Rather like a traditional accelerometer, the levels ofsignal strength fluctuation change when a mobile phone ismoved. For example, Fig. 3 shows the total signal strengthfluctuation across all monitored cells during successive 30-stime periods whilst walking, remaining still and travelling

in a motor car. The figure illustrates that it is relatively easyto distinguish between moving and remaining stationary,but at times, the pattern of fluctuation whilst walking willmatch that of driving and vice versa. This is due to thestop–start nature of both walking and travelling in a motorcar in urban areas. When driving, a greater geographicaldistance will typically be covered over a given time periodwhen compared to that of running or walking. As such it ispossible to use the rate of change of neighbouring cells toinfer travel by car.

To classify these patterns we use an artificial neuralnetwork. The network inputs are the sum of signal strengthfluctuation across all monitored cells, and the number ofdistinct cells monitored over a given time interval. Thenetwork consists of a single layer of eight hidden neurons;weights are learnt using back propagation. The networkoutputs the currently sensed activity for the given inputvalues. The network is trained by repeatedly presenting datacollected during each method of movement.

The current activity of the user is conditionally depen-dent upon their previous activity. In order to provide instantfeedback to the user interface, the neural network deliber-ately does not model this behaviour. Instead, when deter-mining if any additional minutes have been earned, weapply task knowledge based upon the output from theneural network over the previous two and a half minutes.This enables noise to be filtered out and a more accuraterepresentation of the users’ activities achieved. For exam-ple, periods of low signal strength fluctuation such asstopping at traffic lights whilst driving can be ignored whenplaced between periods of high fluctuation where manydistinct neighbouring cells were monitored. It could beargued that activity would be more accurately inferred if alonger rolling filter had been applied to the GSM data.Introducing longer filters would have increased the likeli-hood of active minutes ‘disappearing’ from the users’

Figure 2 The phone interface. Images a and b show screens for examining relative and individual activity levels: compare Daily Activity andThis Week’s Activity Images. c and d show two of the screens showing the estimated current activity level: Stationary and Walking

Mobile Netw Appl (2007) 12:185–199 189

Fish’n’Steps: Encouraging Physical Activity with an Interactive Computer Game 263

!!

Fig. 1. One participant’s display after approximately two weeks into the trial in the Fish'n'Steps team-condition, also the public kiosk and pedometer platform, which rotated through each of the team fish-tanks. The components of the personal display include: 1) Fish Tank - The fish tank contains the virtual pets belong to the participant and his/her team members, 2) Virtual Pet – The participant’s own fish in a frontal view on the right side next to the fish tank, 3) Calcula-tions and feedback - improvement, burned calories, progress bar, personal and team ranking, etc., 4) Chat window for communicating with team members.

To evaluate the effect of Fish’n’Steps, we recruited 19 participants from the staff of Siemens Corporate Research to participate in a 14-week study. Two experimental conditions were designed to separately assess the impact of the virtual pet and the social influences. Application of the TTM to assess behavior that changed during the study demonstrated that Fish’n’Steps was a catalyst of a positive change for 14 out of 19 participants. This effect was evident in either an increase in their daily step count (for 4 participants), a change in their attitudes towards physical activity (for 3 partici-pants) or a combination of the two (for 7 participants). The greatest change in daily number of steps was by the participants who were at the TTM’s intermediate levels of the behavior change. For these participants the game provided just enough motivation to translate mental readiness into action.

While the overall findings were encouraging, there are a number of possibilities for future investigations. For example, selecting participants from our own research or-ganization limited representativeness of the sample to highly educated individuals relatively open to adopting new technologies. In addition, the game highlighted the importance of careful selection of incentives: unachievable or not challenging goals can fail to inspire the desired change.

2 Interventions for Behavior Change

There are a variety of techniques developed over the years to motivate behavior change. Traditional techniques are usually delivered by a trained specialist in either individual or group settings. Examples of these techniques include goal-setting, self-assessment, or monitoring of achieved progress [17] . Two particular approaches that influenced the current project include motivating behavior change by cultivating a

1 2

3

4


Physical activity is affected by lack of time, choice of activities, the environment, and social influence. (Sallis & Hovell 1990)

CDC suggests understanding of factors to circumvent barriers to physical activity.

Research on Factors

52


Physical Activity Level

Location

Activity

People


Physical Activity Level

Location

Activities

People


Research on Factors Diabetes awareness of blood sugar level and food consumption (Frost & Smith ʼ03) •  Images of food associated with blood

sugar level. •  Used in a class where people discussed

their images and blood sugar level. •  Made a prototype, but only tested with one

person.

55


Research on Factors Asthma patients videotaping daily routines found that they are in the presence of harmful allergens more often than they realized (Rich et al. ʻ00)

•  Users videotaped daily routines, but a trained observer looked at the video for assessment.

•  Matt Leeʼs embedded assessment work

56


Sedentary People and Walking Research suggests that they are less aware of their physical activity and how to become active (Sallis & Hovell 1990)

Focused on walking because it is easier to integrate into daily life. (Norman & Mills 2004)

57


Introduction


Prototypes

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


Thesis Questions

59

Is a PI system with context better at revealing factors that affect behavior? •  Run field studies with prototypes.

How do we build a PI system with context? •  Build several prototypes to try different

ways of supporting context.


Introduction


Prototypes

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


Thesis Questions

61


Diary Study How would people find factors using context? IMPACT 1.0 Would context reveal factors that affect behavior? IMPACT 2.0 What is the value of context in the long term?


Thesis Questions

62

How do we build a PI system with context? Collection Integration Reflection

DiaryPrototype

IMPACT 1.0

IMPACT 2.0

user-driven user-driven


user-driven

system-driven

system-driven

system-drivencombination


Thesis Questions

63

Time-Stamped End-of-Day

Aggregated Real-time


Date:

What? Where?Time

6a

7a

8a

9a

10a

11a

12p

1p

2p

3p

4p

5p

6p

7p

How active were you? With whom?

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

What? Where?Time How active were you? With whom?

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

Continue to the next page.!

SenseWear Booklet


No Feedback



Tracking Booklet Reflection

No Feedback

Date:

What? Where?Time

6a

7a

8a

9a

10a

11a

12p

1p

2p

3p

4p

5p

6p

7p


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:


Date:

What? Where?Time

6a

7a

8a

9a

10a

11a

12p

1p

2p

3p

4p

5p

6p

7p


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:


Date:

What? Where?Time

6a

7a

8a

9a

10a

11a

12p

1p

2p

3p

4p

5p

6p

7p


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:


:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:

:



Setup 4 female participants (A1-A4) •  Ages 25-50 •  Sedentary. Pre-screened using Stages of

Exercise Behavior Change (Marcus et al. 1998)

Audio-taped interviews every week

66


1 2 3

NoPhysical Activity

Information

Pedometer SenseWearGraph Printouts

Booklet

SenseWear Tracking


Results

68

Excellent compliance over 3 weeks •  At least one activity recorded per hour


In all phases, participants found factors that affected their physical activity.

Week 1, A3: “Writing down had an effect. I would think ʻOh good I have something active to write down.ʼ Like when I would carry my laundry to the Laundromat on foot.”

69


In all phases, participants found factors that affected their physical activity.

Week 2, A1: “It was nice to see that I walked more than I did. There was one day when I was babysitting. I walked so much with the baby. I walked all over campus.”

70


Matching SenseWear graph printouts with booklet entries.

71

8 hrs 49 m...

Duration of Vi...

Not detect...

Sleep

Not detect...

Lying Do...

11346

Step Count

1 hr 43 m...

Physical Activity (2.5 ME...

438 cal

Active ...includes off-body

estimate of 4 cal1041 calori...

Total EE

- Fri Dec 8, 2006 02:03 PMSession end

End Time

- Fri Dec 8, 2006 05:14 AM

Start Time

Start5:14 AM

End2:03 PM

FRI DEC 8, 2:03 ... THU DEC 7, 2:16 ...

cindy

8 hrs 49 m...

Duration of Vi...

Not detect...

Sleep

Not detect...

Lying Do...

11346

Step Count

1 hr 43 m...

Physical Activity (2.5 ME...

438 cal

Active ...includes off-body

estimate of 4 cal1041 calori...

Total EE

- Fri Dec 8, 2006 02:03 PMSession end

End Time

- Fri Dec 8, 2006 05:14 AM

Start Time

Start5:14 AM

End2:03 PM

FRI DEC 8, 2:03 ... THU DEC 7, 2:16 ...

cindy



Week 3, A2: “The paper feedback [SenseWear] told me when the intensity was greater than other times, so I was able to gauge my activities like if I just walk upstairs…I was calling them in the diary ʻaerobic mini burstʼ.”

72



Week 3, A1: “I had a lot of data, probably too much to decipher, but it was good. I didnʼt really compare everything to my booklet, only the peaks on the charts to see what I was doing.”

73




Detailed reflection Immediate awareness


Thesis Questions

75


•  Participants made associations between their physical activity and contextual information helping them become aware of factors that affected their physical activity.

→ Can we do this in a field study of a prototype with more people?


Thesis Questions

76

How do we build a PI system with context?

→ Help users make direct associations. → Value in aggregated real-time info and

end-of-day time-stamped info.

Collection Integration Reflection

DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



IMPACT 1.0

77


Thesis Questions

78




Thesis Questions

79


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



Pedometer Booklet


5

had an interface to label time segments with contextual information, including a participant’s activity, location, and

people the participant was with. In addition to the per-day

and per-week visualizations of step counts, they also used

the online service to see visualizations showing the

association between their daily activities and their number

of steps (Figure 3e, 3f, & 3g). Users carried diary booklets

with additional fields for contextual information.

FIELD STUDY

We used the IMPACT system to explore if associating

physical activity with context can produce a greater self-

awareness about one’s own behaviors and can then

motivate an improvement in physical activity. We had two

hypotheses:

H1: Users who recorded the context of their physical

activity would find more opportunities to be physically

active.

H2: Users who find more opportunities to be physically

active would increase the number of steps they take.

Participants and Method

We recruited 43 participants, 14 males and 29 females,

using a recruiting web site, newsgroups, and flyers. The

ages of the participants ranged between 19 and 55. All

participants owned a computer and had high-speed online

access at home or at work. Participants received $100 for

their participation.

Participants wore a pedometer (Omron HJ-112 Walking Style) (Figure 2a) all day throughout the duration of the

study. Participants also carried a pocket-sized diary booklet

to record the time and their step counts. When in the Plus-

Context phase, participants received a diary booklet with

additional fields for contextual information (Figure 2b).

Participants entered their booklet entries into the IMPACT

web site at their convenience.

Participants used the Baseline version of IMPACT during

the first week of the study. After the first week, participants

were randomly assigned to use the Steps-Only version or

the Plus-Context version for 3 weeks. After this period of 3 weeks, participants used the other version for 3 more

weeks.

During the Steps-Only phase, participants recorded their

step counts every time they changed activities or about once

an hour. During the Plus-Context phase, participants

recorded their step counts and the context of their activity

(the kind of activity, location, and the people they were

with). Participants were instructed to record every time they

changed their context, such as when they changed

activities, moved locations, or met with a different person.

The researchers met with the participants 4 times throughout the study: at the Start of the study, after

Baseline, after Steps-Only, and after Plus-Context. During

these meetings, participants received materials and

instructions for the phase they were about to begin.

Participants also answered questionnaires. Four participants

Baseline

a

Steps-Only

b

c

d

Plus-Context

e

f

g

Figure 3. a) Interface for recording steps. Steps-Only additions. b) One day of steps. c) Week of steps by day. d) Week of steps for each day. Plus-Context additions. e) Context labeler. f) Table and

chart showing association between steps and context. g) Steps by hour and period of day.

Day with "context labels

Table and charts of steps and context

Steps by hour and by period of day


Pedometer Booklet Web Site

IMPACT 1.0

Steps-Only Steps-Only Steps-Only

Steps-Only Baseline No Vis


Setup 30 participants (B1-B30) •  Sedentary. Pre-screened using Stages of


Questionnaires at the end of each phase

84


1 2 3 4 5 6 7

Ba

se

lin

e

Steps-Only IMPACT 1.0

Steps-OnlyIMPACT 1.0


Results Same level of physical activity awareness between Steps-Only and IMPACT

86


Greater awareness of factors After using IMPACT, participants self-reported greater awareness of factors that affect physical activity (5-point Likert scale)

3.93 (IMPACT) vs. 3.57 (Steps-Only) F[1,58] = 5.32, p < .05

87


Greater awareness of factors Mentioned context •  “It has been interesting to see what times

of the day I'm most active.” B17 •  “It turns out I get the most walking done to

and from work, which I can't say I wasn't expecting, but I also had no idea that walking around Squirrel Hill for just an hour or two made such a difference.” B24

88


Greater awareness of factors

89

Mentioned Context Mentioned Context

Excluding Time

Control

Steps-Only

IMPACT 1.0 18

13

11

13

7

6

out of 30

out of 30

out of 30


IMPACT was rated most useful “The [context] I used the most was the one asking who I was with during my most active periods…I hadnʼt realized that I was so sedentary most of the time I spent with my friends.” B1

90


But IMPACT was harder to use 17 of 30 participants preferred Steps-Only.

“IMPACT gave a lot of cool information, but having to input all the various factors was a hassle.” B4

91


Problem can be fixed Those who preferred Steps-Only said: •  “There were times I wanted to explain my

context.” B22 •  “IMPACT should be provided as an option.” B30

90% reported they would continue using IMPACT if collection of context was automated.

92


Thesis Questions

93


•  Context can increase awareness of factors that affect physical activity.


Thesis Questions

94


→ Users need help collecting and integrating data over a long period of time.


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven


Visualizations


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



IMPACT 2.0

95


Thesis Questions

96




Thesis Questions

97


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



IMPACT 2.0

98


IMPACT 2.0

99

Bluetooth GPS


IMPACT 2.0

100

Context Input


IMPACT 2.0

101

Bluetooth Sync


Mobile Phone Web Site

Steps-Only

IMPACT 2.0

Steps-Only

Control No Vis

Steps-Only

Steps-Only

GPS Context Input


Setup 35 participants (C1-C35) •  Sedentary. Pre-screened using Stages of



104


1 2 3 4 5 6 7 8

Control

ControlIMPACT 2.0

Steps-Only

Baseline Phase Intervention Phase


Results Complaints were not about the tedium of writing things down, but about having to carry multiple devices. •  “I would not like carrying two devices (GPS

and phone), that was too much.” C30 •  “I would use [the prototype] if I could use

the software on my own cell phone.” C17

106


Awareness of factors increased for all groups between the phases

107

F[2,32] = 3.98, p = .0547

!"

!#$"

%"

%#$"

$"

&'()*+,)" -,.)/0),12,"

!"#$%&%''()*(+#,-)$'(

32,./2*" 4.)5(67,*8" -9:;3<"=#>"


Similar awareness of factors

108

Mentioned Context Mentioned Context

Excluding Time

Control

Steps-Only

IMPACT 2.0 8

8

6

6

3

5

out of 11

out of 12

out of 12


Long-term reflection What is the value of contextual information in the long-term?

6-months later when they were more likely to have forgotten the data

109


Follow-Up Interviews Participants: •  Control (5) •  Steps-Only (6) •  IMPACT 2.0 (3)

110


Follow-Up Interviews Expressed interest in comparing over long periods of time.

Curious about the peaks in physical activity.

But only those who collected contextual information had reminders of what happened during those peaks.

111


Collection vs. Reflection

112

Short-term Reflection

Long-term Reflection

Manual Collection

GOOD NOT GOOD

Automated Collection

NOT GOOD GOOD


Thesis Questions

113


Automation has an effect: •  In the short term, reduced interaction with

context data so not difference in awareness of factors.

•  In the long term, users had data to effectively reflect on factors.


Thesis Questions

114


•  Automation is valuable in the long term, but not as useful in the short term.


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven



Thesis Questions

115


•  Diary Study and IMPACT 1.0: YESContext can increase awareness of factors that affect physical activity.

•  IMPACT 2.0: DEPENDSAutomation reduces immediate awareness, but helps with long-term reflection.


Thesis Questions

116


DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven


Need visualization Support

Need to reduce burden of Collection and Integration stages.

Consider the effect of automation on immediate awareness.


Introduction


Completed Work

Diary Study

IMPACT 1.0

IMPACT 2.0

Proposed Work

Conclusion


IMPACT 3.0?

118


Thesis Questions

119

Is a PI system with context better at revealing factors that affect behavior? •  How do people use context in the

visualizations to find factors that affect behavior? •  How do we support comparison

between different types of context? •  Infer cause-and-effect?


Thesis Questions

120


•  Reduce the cost of using personal informatics system while increasing awareness of factors that affect physical activity.


UNI-FACETED vs. MULTI-FACETED

uni-faceted multi-faceted



DiaryPrototype

IMPACT 1.0

IMPACT 2.0



user-driven

system-driven

system-driven


IMPACT 3.0

system-driven system-drivencombination

Need visualization Support

Need to reduce burden of Collection and Integration stages.

Consider the effect of automation on immediate awareness.

Design semi-automated collection without losing immediate awareness


Forced users to collect three types of contextual information. •  Some of them may not have been useful,

i.e. they are incurring a cost in Collection, but not providing a benefit during Reflection.

123

ITERATIVE



The prototypes did not support context other than activity, location, and people. •  Weather •  Mood •  Nutrition •  Other data sources: calendar, email

124

ITERATIVE



The prototypes did not count other types of physical activity. •  Swimming •  Biking •  Sports

125

ITERATIVE



•  Allow users to select factors that is important to them.

•  Support factors other than activity, location, and people.

•  Count other types of physical activity.

126

ITERATIVE



The prototypes experienced several barriers that affected their usage.

Subsequent prototypes addressed some of these barriers.

Keep these in mind!

127

BARRIERS CASCADE


IMPACT 3.0 Requirements 1.  Maintain immediate awareness in semi-

automated collection.

2.  Give choices for other types of context and physical activity.

3.  Explore visualizations of context and physical activity.

128


1. Maintain Immediate Awareness Low-cost manual collection where visualizations are shown after input. (ES+feedback, Hsieh et al. 2008)

•  Once during the day, the system asks the user to input data manually. After, the user shows visualizations to the user of other automatically collected information.

129


1. Maintain Immediate Awareness Encourage daily use of the visualizations. •  Alert the user of an interesting fact. •  “You were active while walking in the

park.” •  Make it easy for users to find what they

are looking for. •  Direct users to the graph that would

answer their question.

130


2. Other types of context and PA Integrate other sources of context •  Online weather information •  Event information from calendars •  Status updates •  Other personal informatics tools (e.g.,

MoodJam, your.flowingdata)

131


2. Other types of context and PA Allow logging of other physical activity.

Equate amount of physical activity with amount of walking •  1 hour of softball = 30 minutes of walking

132


Personal Informatics Browser

133


3. Explore Visualizations in Detail Problems occurred in stages before Reflection, now I can explore this stage in detail. •  Visualizations for comparing between

instances of physical activity •  Visualizations for making cause-and-effect

associations between context and physical activity

134


3. Explore Visualizations in Detail

135

Same Physical Activity LevelDifferent Context

Housework Shopping


3. Explore Visualizations in Detail

136

Different Physical Activity LevelSame Context

Office

Office


Setup 40 participants •  Sedentary and active


137


1 2 3 4 5 6 7 8

Control

IMPACT 3.0

Steps-Only

Baseline Phase Intervention Phase


Awareness Effects Higher awareness of factors that affect physical activity.

Maintain immediate awareness even when collection is semi-automated.

Increased interest in the visualizations

139


Psychological Effects Higher locus of control

Higher self-efficacy

Self-reports of change in physical activity

140


Thesis Questions

141

Is a PI system with context better at revealing factors that affect behavior? •  Deployment of a personal informatics

system that has minimal cost in collection without loss in benefit in reflection.

•  Exploration of how people use context in visualizations to find factors that affect behavior.


Thesis Questions

142


•  Designs that reduce the cost of using personal informatics system while increasing awareness of factors that affect physical activity.


Limitation Focus on physical activity. •  PI systems share the same stages. •  Revealing factors that affect behavior

benefits the user. •  Diabetes (Frost & Smith 2003)

•  Early work on smart meters (Betz 2010)

143


Future Work Sharing data. •  Many PI systems allow sharing. •  Explore the quality of discussions when

data is shared with context.

144


Thesis Questions

145

Is a PI system with context better at revealing factors that affect behavior? •  Showed that context can increase

awareness of factors that affect behavior. •  Automation reduces immediate

awareness, but helps with long-term reflection.

•  Deployed a system that doesnʼt sacrifice long-term for immediate, vice versa.


Thesis Questions

146

How do we build a PI system with context? •  Created the stage-based model of PI as a

framework to analyze PI systems. •  Created prototypes that support context in

field studies. •  Identified issues in supporting context and

addressed them in subsequent prototypes.


Schedule

147

Ubico

mp

CHI2 0 1 0

Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Write Dissertation

Prototype Analysis

IMPACT 3.0 Development

Deployment

Analysis


Thank you! To my advisors, Anind Dey and Jodi Forlizzi, and the rest of my committee, Niki Kittur and John Stasko.

To people who have helped discussions, pilot studies:Gary Hsieh, Scott Davidoff, Erin Walker, Karen Tang, Matt Easterday, Amy Ogan, Amy Hurst, Ruth Wylie, Moira Burke, Matt Lee, Gabi Marcu, Queenie Kravitz, Min Kyung Lee, Turadg Aleahmad, Tawanna Dillahunt, Brian Lim, Chloe Fan, Jenn Marlow, Jason Wiese, Sunyoung Kim, Aubrey Shick, Chris Harrison, Julia Schwarz, Bilge Mutlu, Andy Ko, Johnny Lee, Ido Roll, Jeff Nichols, Jeff Wong, Sara Kiesler, Laura Dabbish, Scott Hudson, Tessa Lau, Jaime Teevan, Fernanda Viegas, Jon Froehlich, UISTʼ09 and UbiCompʼ09 Symposia attendants, Alexandra Carmichael, Gary Wolf, Nathan Yau, Nicholas Felton, Ellie Harrison, Edison Thomaz

148

personal informatics and context: using context to reveal factors that affect behavior

Technology