evaluating ubicomp applications in the field

Post on 07-Jan-2016

34 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

Evaluating ubicomp applications in the field. Gregory D. Abowd, Distinguished Professor School of Interactive Computing and GVU Center, Georgia Tech. Ubicomp evaluation in the field. Weiser (CACM 93): “ Applications are of course the whole point of ubiquitous computing. ” - PowerPoint PPT Presentation

TRANSCRIPT

Evaluating ubicomp applications in the fieldGregory D. Abowd, Distinguished ProfessorSchool of Interactive Computing and GVU Center, Georgia Tech

Ubicomp evaluation in the field

• Weiser (CACM 93): “Applications are of course the whole point of ubiquitous computing.”– Applications are about the real world

• Design: finding appropriate ubicomp solutions for real-world problems.

• Evaluation: Demonstrating that a solution works for its intended purpose.

• But these activities are intertwined.Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Defining evaluation terms

• Formative and Summative

• Who is involved in evaluation

• Empirical, Quantitative, and Qualitative Evidence

• Approaches to evaluation

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Formative and Summative

Formative–assess a system being designed

–gather input to inform design

Summative–assess an existing system

–Summary judgement of success criteria

•Which to use?–Depends on

•maturity of system•how evaluation results will be used

–Same technique can be used for either

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Who is involved with evaluation?

• End users, or other stakeholders

• The designers or HCI experts

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Form of data gathered

Empirical: based on evidence from real users

Quantitative: objective measurement of behavior

Qualitative: subjective recording of experience

Mixed methods: a combination of quant and qual

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Approach

• Predictive modeling

• Controlled experiment

• Naturalistic study

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Predictive Modeling

• Try to predict usage before real users are involved • Conserve resources (quick & low cost)• Model based

– Calculate properties of interaction– Fitts’ Law, Keystroke Level Model

• Review based– HCI experts (not real users) interact with system, find

potential problems, and give prescriptive feedback– Best if they:

• Haven’t used earlier prototype• Familiar with domain or task• Understand user perspectives

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Controlled experimentation

Lab studies, quantitative results– Typically in a closed, lab setting– Manipulate independent variables to see effect

on dependent variables

– Replicable

– Expensive, requires real users and lab

– Can use follow-up interviews for qualitative results

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Naturalistic Study

Or the field study

– Observation occurs in “real life” setting– Watch process over time

– “Ecologically valid” contends with controlled and “scientific”

– What is observed can vary tremendously

We will focus on this form of evaluation

Why is this so hard?

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Historical overviewSome examples of ubicomp technologies in the field

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Examples of field studies

• Xerox PARC– PARCTab

– Liveboard

• Georgia Tech– Classroom 2000

– Digital Family Portrait

– Personal Audio Loop

– Abaris, CareLog and BabySteps

– Cellphone proximity

– SMS for asthmaEvaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Xerox PARC

• Computing at different scales– Inch

– Foot

– Yard

• They were pretty successful at inch and yard scales with two very different approaches to evaluation

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Xerox PARC inch scale

• The PARCTab– Location-aware thin

client

– Deployed in CSL building

• Built devices and programmable context-awareness

• Gave it to community to see what they would produce

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Xerox PARC yard scale

• The Liveboard– Pen-based

electronic whiteboard

– Deployed in one meeting room

• Designed solution for IP meetings– Supporting work of

one individual

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Georgia Tech Classroom 2000

• The Liveboard in an educational setting

• Establishing theme of automated capture

• Room takes notes on behalf of student

• 4-year study of impact on teaching and learning experience.

• The living laboratoryEvaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Georgia Tech Aware Home

• What value is it to have a home that knows where its occupants are and what they are doing?

• A living laboratory?– Great for feasibility

studies and focus groups

– Never anyone’s “home”

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Georgia Tech Digital Family Portrait

• Great example of a formative study done in the wild

• Sensing replaced by phone calls

• Similar to Intel Research CareNet Display

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Georgia Tech Technology and Autism• Ubicomp got very personal for me in 2002

• Improving data collection

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

From whimsical to inspirational…Dec. 1998

Aidan, 18 months

From whimsical to inspirational…July 1999

Aidan, 26 months

DetailedScoring

ManualCalculations Hand Plotting

Abaris: Embedding Capture

Leverages basic therapy protocol to minimize intrusion

Speech detection to timestamp beginning of trial

Record handwriting using Anoto digital pen to collect grades and

timestamp end of trial

Julie Kientz, Ph.D.

04/20/23

Abaris: Embedding Access Julie Kientz

Abaris: Study

• 4 month real use deployment study– Case Study: Therapy team for one child

• 52 therapy sessions (50+ hours of video)• 6 team meetings

• Data collected– Video coding and analysis of team decisions during sampled meetings

• Meetings without Abaris: 39 decision points across 3 meetings• Meetings with Abaris: 42 decision points across 3 meetings

– Interviews with team members– Software logging of Abaris

Full study details: Chapter 5

26Julie A. Kientz, Georgia Tech

Results: Easier Data Capture

• Therapists were able to learn to use the Abaris system with very quick training

• Therapists spent less time processing paperwork

Julie A. Kientz, Georgia Tech

27

Results: Access to DataPercentage of decision points in which a given artifact was used. Oftentimes, therapists used multiple artifacts.

Artifacts %Without Abaris %With Abaris

Video* 0.0 45.2

Graphs 56.0 81.9

Data sheets** 20.5 45.2

Therapy samples* 0.0 19.0

Reenactment 0.0 4.8

Memory 92.3 83.3

Ext. Observations 25.6 21.4

Therapist Notes** 5.1 19.0

For full details, see Chapter 528

* p < .01 ** p < .05

Julie A. Kientz, Georgia Tech

Results: Improving Collaboration

29

Average Participation Levels With and Without Abaris

1.00

1.25

1.50

1.75

2.00

2.25

2.50

2.75

3.00

Conditions

Par

ticp

atio

n L

evel

(1=

low

, 3=

hig

h)

Series1 2.44 1.98

With Abaris Without Abaris

Analysis of decision points in team meetings indicated an increase in collaboration

Interviews with therapists after meeting confirmed these numbers

p < .01

For full details, see Chapter 5

Julie A. Kientz, Georgia Tech

(with Gillian Hayes (GT), Juane Heflin (Georgia State), Cobb County Special Ed. and Behavior Imaging Solutions, Inc.)

Collecting rich behavioral data in the unstructured natural environment

Retroactively saving important videoConscious selection of relevant video

episodes

Gillian Hayes

After-the fact capture and annotation

Examples of field studies

• Xerox PARC– PARCTab

– Liveboard

• Georgia Tech– Classroom 2000

– Digital Family Portrait

– Personal Audio Loop

– Abaris, CareLog and BabySteps

– Cellphone proximity

– SMS for asthmaEvaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Others

• Intel Research– CareNet Display

– Reno

– UbiFit, UbiGarden

• EQUATOR– Mixed reality games

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

Lessons from our ancestors

• Evaluation takes time– The experience of ubicomp does not come

overnight

– The “abowd” unit of evaluation; nothing substitutes for real use

• Bleeding edge technology means people must keep system afloat– The users cannot be expected or bothered with

installation or maintenance (C2K)

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

More lessons

• Sometimes you want to evaluate in the field without any working system– The idea of experience prototypes (or

paratypes), or humans as sensors

Evaluation in the field, Gregory D. Abowd, Ubicomp 2011 Tutorial

3 Types of Field Studies (Ch. 4 by Brush)• Or, why do the study?

• In somewhat chronological order:– Understand current behavior

– Proof of concept

– Experience using a prototype

36

Understanding Behavior

Insight into current practice, baseline of behaviors– To inform new designs

– To use as comparison at some later date

• Brush and Inkpen (2007)

• Patel et al. (2006)

37

Proof of Concept

Bleeding-edge prototypes but not in the lab.

• Context-Aware Power Management (2007)

• TeamAwear (2007)

38

Experience

Prolonged use that is not about feasibility of the technology, but about the impact on the everyday experience.

• CareNet (2004)

• Advanced User Resource Annotation

39

Study Design

• Study Design

• Data collection techniques– Surveys, interviews, field notes,

logging/instrumentation, experience sampling, diaries

• How long should the study be?– The “abowd” as a unit of time for a field study.

40

Participants

• Ethics

• Selection of the right participants

• Number of participants

• Compensation

41

Analysis of data

• Quantitative data– Relevant statistical methods based on data

collection

• Qualitative data– Often unstructured and must be processed (or

coded) to be understood and analyzed further

42

Let’s reflect on the other readings…• Provide summary based on:

– Type of study (Brush taxonomy)

– What was its purpose?

– How study was designed• Data to be collected, conditions of study, number and

kind of participants, length of study

– How results were analyzed

– Did the study meet its purpose?

43

top related