48
Human-Centered Computing at University of Maryland, Baltimore County 1 How to run any kind of Evaluation 3/6/14 HCC 729, Human Centered Design Amy Hurst

Human-Centered Computing at University of Maryland, Baltimore County 1 How to run any kind of Evaluation 3/6/14 HCC 729, Human Centered Design Amy Hurst

Embed Size (px)

Citation preview

Human-Centered Computing at University of Maryland, Baltimore County1

How to run any kind of Evaluation

3/6/14HCC 729, Human Centered Design

Amy Hurst

Human-Centered Computing at University of Maryland, Baltimore County2

Getting started

Share inspirations, reading reflectionshttp://hcc729s2014.wordpress.com/student-blogs/

Homework check in (paper prototypes)

Human-Centered Computing at University of Maryland, Baltimore County3

Paper prototypes

Activity (10 minutes)- Pair up with another group- Pick one task from your task list- Have other group test your task with

prototype, 5 minutes- Switch

- What worked? Any changes needed to your paper prototype? Anything missing?

Human-Centered Computing at University of Maryland, Baltimore County4

Reflection on paper prototype testingWhat did you learn?

Anything important missing from your prototypes?

Any obvious changes to make?

Human-Centered Computing at University of Maryland, Baltimore County5

More about evaluation

Human-Centered Computing at University of Maryland, Baltimore County6

Why User Test?

• Any testing is better than none – even a few users!• Saves time and money in development process by

preventing errors• Hard to tell how good or bad UI is until people use it!• Examining real users gets us away from the “expert blind

spot”• It is hard to predict what actual users will do• User testing mitigates risk• Not necessary to design flawless experiment protocol to

get usability measures• Critical to evaluate the IMPORTANT aspects of your design

Human-Centered Computing at University of Maryland, Baltimore County7

Expert-based evaluation

Aren’t there experts who can look at your site and identify problems?

• Sort of… yeah. • This usually happens too late.

“We’re going live in two weeks; do you have time to look over our site?”

• Experts don’t always have the characteristics of your users, whom you studied so carefully before starting

Human-Centered Computing at University of Maryland, Baltimore County8

Risks of Late User Testing…

Sometimes in software development, users are brought in only at the beta test stageWhat are some of the risks of doing this?

– By then most of the budget has been spent– It is very much more expensive to correct an error

than if it had been caught early

Avoid this and test early and often…

Human-Centered Computing at University of Maryland, Baltimore County9

3 Types of Evaluations

1. Formative: during development (explorative)2. Summative: at completion (assessment and

validation)3. Comparison testing

Cook tastes the soup in the kitchen

Formative evaluation

Guests taste the soup at the dinner table

Summative evaluation

Couple shares two soups and chooses one they like better

Comparison testing

Human-Centered Computing at University of Maryland, Baltimore County10

Usability Methods in Chapter 77.1 Observation7.2 Questionnaires and Interviews7.3 Focus Groups7.4 Logging Actual UseCombining Logging with Follow-Up Interviews7.5 User Feedback7.6 Choosing Usability MethodsCombining Usability Methods

Human-Centered Computing at University of Maryland, Baltimore County11

Nielson’s Categories for Usability Methods, Chapter 7 Usability Engineering

Human-Centered Computing at University of Maryland, Baltimore County12

Steps for an evaluation

1. Planning & preparation– Designing the test– Choosing participants – Selecting the task

2. Running the test – During the session– Collecting the data– Debriefing the subject

3. Analyzing the data and disseminating your findings

Human-Centered Computing at University of Maryland, Baltimore County13

How to Run any Evaluation

Planning and Preparation

Running the TestAnalysis and

Dissemination

Human-Centered Computing at University of Maryland, Baltimore County14

Planning and Preparation: ParticipantsSelect the appropriate participants

– Who are the ideal participants? – Who are acceptable participants?– Aim for the actual users of the system If unavailable, aim for the closest

approximation

Target population users may have specific characteristics– Domain-specific vocabulary – Often possess particular domain knowledge – Have a history with existing systems, methods, etc.

Note: novices and experts– Why not just novices? – Why not just experts?– User mental models differ if they are novice or expert – system won’t support both

if not tested on both

Don’t forget your user analysis, and think about how your design may bias your results

Human-Centered Computing at University of Maryland, Baltimore County15

Bias…

Human-Centered Computing at University of Maryland, Baltimore County16

Who has the hardest job in the world?

I was at the post office one day, and a student came up to the woman behind the counter and asked “who has the hardest job in the world?” She answered the president of America. He wrote this down, turned to me and asked me “who has the hardest job in world?”

– What kind of results do you expect this student will get?

– What would you change about how this student is administering this survey?

Human-Centered Computing at University of Maryland, Baltimore County17

Avoid bias in your evaluations!

Always think about how you are biasing (distorting, impacting, controlling) your results

Your goal is to gather data that is reliable, and repeatable.

Human-Centered Computing at University of Maryland, Baltimore County18

3 simple Factors you can control:

Participants

Location of evaluation

Your Behavior and Actions

How can you prevent bias?

Human-Centered Computing at University of Maryland, Baltimore County19

WHO SHOULD YOU RECRUIT FOR YOUR STUDY?

“My Roommate thought the buttons were too small” “My mom really liked my color choices”

“My Girlfriend found the following Typos”

Human-Centered Computing at University of Maryland, Baltimore County20

3 Kinds of Bias

UndercoverageNonresponseVoluntary response

Human-Centered Computing at University of Maryland, Baltimore County21

3 Kinds of Bias: UndercoverageUndercoverage. Undercoverage occurs when some members of the population are inadequately represented in the sample.

– Literary Digest voter survey, which predicted that Alfred Landon would beat Franklin Roosevelt in the 1936 presidential election. The survey sample suffered from undercoverage of low-income voters, who tended to be Democrats.

Human-Centered Computing at University of Maryland, Baltimore County22

3 Kinds of Bias: Nonresponse

Nonresponse bias. Sometimes, individuals chosen for the sample are unwilling or unable to participate in the survey.

– The Literary Digest survey illustrates this problem. Respondents tended to be Landon supporters; and nonrespondents were Roosevelt supporters. Since only 25% of the sampled voters actually completed the mail-in survey, survey results overestimated voter support for Alfred Landon

Human-Centered Computing at University of Maryland, Baltimore County23

3 Kinds of Bias: Voluntary ResponseVoluntary response bias. Voluntary response bias occurs when sample members are self-selected volunteers

– Call-in radio shows that solicit audience participation in surveys on controversial topics (abortion, affirmative action, gun control, etc.). The resulting sample tends to overrepresent individuals who have strong opinions.

Human-Centered Computing at University of Maryland, Baltimore County24

Select appropriate participants

Who are the ideal versus acceptable participants? – Aim for the actual users of the system If unavailable, aim for the

closest approximation

Things to consider:– Age– Culture– Experience

• Domain-specific vocabulary • Often possess particular domain knowledge • Have a history with existing systems, methods, etc.

– Where did you find these people?– Others?

Human-Centered Computing at University of Maryland, Baltimore County25

Laboratory vs. Real World studies

Remember your environmental Analysis?

Where should you conduct your user study?

Human-Centered Computing at University of Maryland, Baltimore County26

When I say “User Study” you think….

Human-Centered Computing at University of Maryland, Baltimore County27

But, how realistic is your user study setting?

Human-Centered Computing at University of Maryland, Baltimore County28

Exploring the role of environment

What changes if your user is…

Waiting for the train at a crowded MARC stationSitting on the grass in the park on a sunny dayCurious during a movieIn an office that is quiet and dullWorking at homeWorking in a coffee shop

Human-Centered Computing at University of Maryland, Baltimore County29

Why does this matter?

Conduct an “Environmental Analysis” and control the evaluation environment.

– Understand where your interface will be used– This is usually best done through interviews or observations of real world use

A few things to consider…– Be as faithful to real situations as possible (get creative)– Consider more complicated aspects of the environment: include distractions

and stress if appropriate (noise/heat) – Consider how the environment will effect machine performance (internet lag

time, sensors not working, etc)

Does this really matter? Ex: speech recognizer achieved 98.7% word accuracy in your user study; but the real world

deployment of your system will be on an airport tarmac…

Human-Centered Computing at University of Maryland, Baltimore County30

What is the IRB?

What is a consent form?

Why do I care?

Human-Centered Computing at University of Maryland, Baltimore County31

Informed consent

Main points to include (UMBC has its own forms)– General purpose– Participation is voluntary– Results will be confidential– There is no benefit to you, other than agreed-upon payment– There is no risk to you– 18 or over– Signature and date

Human-Centered Computing at University of Maryland, Baltimore County32

IRB Slides

Institutional Review Board• http://www.umbc.edu/irb/• Human Subjects

Training modules to conduct research• If you aren’t going to publish: Researchers

conducting no more than minimal risk research• If you might publish: Social / Behavioral

Research

Human-Centered Computing at University of Maryland, Baltimore County33

How to Run any Evaluation

Planning and Preparation

Running the TestAnalysis and

Dissemination

Human-Centered Computing at University of Maryland, Baltimore County34

During the session

Write a task script– I literally write down everything I am going to say

Prepare the user – “I am testing the system and not you” – “We expect problems, that’s why we are doing this” – “You can stop at any time, for any reason” – “I need to know what you are thinking as you go” (if

appropriate)

Have the task ready – Written down – Give the same verbal instructions each time

Human-Centered Computing at University of Maryland, Baltimore County35

Choosing your actors

How many people will be in the room?• What roles will they have?• Should the greeter, facilitator and observer all be

the same person?

What kind of persona should they take on?• Manager / task master?• Student / paid worker?• Researcher?

Human-Centered Computing at University of Maryland, Baltimore County36

“Always wear blue in court”

“Wearing Green makes people think of money”

Does my behavior really matter?

Human-Centered Computing at University of Maryland, Baltimore County37

Yes, your behavior matters!

Unfortunately, there are some variables that are hard to control– Your gender, age, ethnicity– Being in a position of “power”

In order to avoid bias, you want to control for as many variables as possible.

– Make the experience the same for each user

What are some of the factors of the experimenter that could impact results?

– Clothing– Attitude (are they grumpy, or not paying attention?)– What is said to the participant

Human-Centered Computing at University of Maryland, Baltimore County38

Ensure consistency: use a task scriptGive each participant has the same experience

– Make sure they get the same instructions– Make sure you ask all questions the same way– Helps control evaluation duration– Makes it easier for you to repeat the study

Write a task script of everything that will happen in the study.

– Treat this like a script for a play– I literally include that happens from “hello” to

“goodbye”

Human-Centered Computing at University of Maryland, Baltimore County39

Collecting the data

Write down observations– Consider how this may bias the user’s behavior

Record actions– Video/ audio recording– Camtasia or other screen recording– Will have to spend time “coding” data to understand what

happened during evaluation

Take detailed notes immediately after session– Best to postpone doing anything else immediately after session– Want to capture everything that is in your head while it is fresh– Risk: you may have forgotten details

Human-Centered Computing at University of Maryland, Baltimore County40

Debrief

This is where you usually administer questionnaires

– Make sure it happens before any interview or discussion

– Ask for any comments the users might have on the system

– Ask for clarifications on areas where the participant had trouble

Thank participant and give them a method for contacting you in the future

Human-Centered Computing at University of Maryland, Baltimore County41

For next week

Assignment

Readings

Human-Centered Computing at University of Maryland, Baltimore County42

Readings

Required• Controlled experiments

Optional• Statistics in usability research• Usability Testing: current and future

Human-Centered Computing at University of Maryland, Baltimore County43

Assignment: Test Paper PrototypesUse the think aloud protocol to test your paper prototypes

KEEP YOUR PAPER PROTOTYPES (turn them in next week)

Complete appropriate Critical Incident UARs

Example paper prototype test: http://www.youtube.com/watch?v=9wQkLthhHKA&feature=related (you should probably let the user drive more)

Human-Centered Computing at University of Maryland, Baltimore County44

Assignment: Test Paper PrototypesPerform a think aloud with 3 people who represents someone from your user analysis, and have them use the prototype you created. Have your users perform the 5 tasks you created these paper prototypes for.Complete CI UARs based on what you saw

• Fill out the top part for all users first• Aggregate across all users• Then, complete the bottom half

Write 200 words about what you learned

Human-Centered Computing at University of Maryland, Baltimore County45

In-Class Activity

Verify your paper prototypes are complete– Test your other tasks with a different group– Take notes: are your prototypes complete? Any

obviously missing parts?• Fix it before you complete the assignment

At the end of your test:- testers: any obvious changes?- participants: any bias? Feedback on the procedure?

Human-Centered Computing at University of Maryland, Baltimore County46

HE Notes

For the final report

Human-Centered Computing at University of Maryland, Baltimore County47

Human-Centered Computing at University of Maryland, Baltimore County48

Notes about the HE method

Don’t forget that you (the designer) are supposed to AGGREGATE your evaluators UARs into a final set

• You search for duplicates• If your evaluators gave you severity ratings,

aggregate them• You provide overall severity ratings• You provide solution recommendations