Upload
james-wirth
View
507
Download
1
Tags:
Embed Size (px)
Citation preview
Measuring survey usability
and respondent fatigue
Survey design and usability
Erik KotoCEO
QuestionPro
• We used TryMyUI to test QuestionPro UI
• Discovered a major usability problem
• “Cool!”
• “I wonder if our customers could use this?”
Usability testing: A true story
Survey usability
• Fatigue: length, repetitiveness, or overwhelming complexity
• Clarity: poor or misleading word choice and question phrasing
• Answerability: unsatisfactory or incomplete answer options
• Poorly wording, misspellings, incorrect grammar, dumb mistakes
• Questions which don't quite mean what the researcher intended
• Questions which don't probe to find out what the respondent really meant
• ‘Double barrel’ or multipurpose questions
• Questions without an answer option suitable to the respondent
• Questions which repeat what has already been asked
• Questions which have been missed completely
Common problems
Error blindness
What is the error?
• Best practices help, but won’t
catch errors
• How does design affect data quality?
• Length
• Number of open ended questions
• Rows in matrix table
• Page breaks
• Progress bar
• Intro & instructions
• Number of required fields
• Number of answer options per question
Can we quantify usability?
Introducing
The Survey Respondent Score
Jeff SauroFounding Principal
MeasuringU
1. Create a list of items
2. Test
3. Winnow
4. Assess Reliability & Validity
Steps to Psychometric Validation
Overall UsabilityThis survey was easy to take.In terms of flow, this survey flowed well.Compared to other surveys I have taken, this survey was average.
FatigueThis survey was quick-paced.This survey took average time to complete.If I were asked, I would be likely to take this survey again.Throughout the survey, I felt engaged.
AnswerabilityAll of the questions could be answered accurately.With the answers provided, I was able to answer all of the questions.Did you have enough information to answer the questions?
ClarityThe questions were clear.I could understand the questions easily.To understand the questions, I had to work hard.
13 Candidate Items
Testing The Items
511 Participants 5 Surveys
Winnow: ITEM total Correlations
3 items with low item-total correlations were removed, leaving 10 items with strong internal reliability (Cronbach Alpha = .93)
Initial Factor analysis
One Factor Solution
A scree plot of the eigenvalues suggests a one factor solution. A parallel analysis also shows one factor with eigenvalues greater than those from randomly simulated matrices.
Refining the Items Using IRT
An examination of the items using a logit-transformation reveals some items have a similar discrimination profile, suggesting redundancy.
• Did you have enough information to answer the questions?
• I could understand the questions easily.
• The questions were clear.
Redundant Items
Room for more difficult Items
2 Factor Structure (OBLIQUE)
15
A 2 Factor structure provides at least two items per factor. All items loaded high (>.5)—although factor 2 provides only a negligible amount of variance.
Factor
Fatigue Clarity
In terms of flow, this survey flowed well. 0.84 -0.64
Compared to other surveys I have taken, this survey was average. 0.84 -0.60
Throughout the survey, I felt engaged. 0.81 -0.62
This survey was easy to take. 0.77 -0.68
If I were asked, I would be likely to take this survey again. 0.77 -0.65
I could understand the questions easily. 0.68 -0.92
The questions were clear. 0.72 -0.88
% of Variance Explained 68% 9%
Extraction Method: Principal Axis Factoring.
Rotation Method: Oblimin with Kaiser Normalization.
Reliability & Validity
1
1.5
2
2.5
3
3.5
4
4.5
5
1 2 3 4 5
SR
S
Survey Number (Hardest to Easiest)
High-reliability (Cronbach alpha = .93) and able to discriminate well across surveys
Validity
The SRS correlated highly with survey abandonment rate & total number of questions (r >.8)
Abandonment Rate
Number of Questions
Predictive Validity & Future Research
18
7 Item SRS
This survey was easy to take.In terms of flow, this survey flowed well.Compared to other surveys I have taken, this survey was average.If I were asked, I would be likely to take this survey again.Throughout the survey, I felt engaged.The questions were clear.I could understand the questions easily.
Predicting Response Rates & Continuing to Refine
3 Factor Structure (Orthogonal)
A 3 Factor structure provides at least two items per factor. All items loaded high (>.5).
Rotated Factor Matrixa
Fatigue Clarity Answerability
Compared to other surveys I have taken, this survey was average. 0.77 0.25 0.23
In terms of flow, this survey flowed well. 0.71 0.27 0.27
Throughout the survey, I felt engaged. 0.67 0.26 0.31
If I were asked, I would be likely to take this survey again. 0.61 0.34 0.33
This survey was easy to take. 0.59 0.38 0.33
I could understand the questions easily. 0.34 0.77 0.34
The questions were clear. 0.39 0.71 0.37
With the answers provided, I was able to answer all of the questions. 0.36 0.36 0.76
All of the questions could be answered accurately. 0.39 0.36 0.69
% of Variance Explained 65% 9% 6%
Forcing a 4 Factor Structure (Oblique)
Given that factors were likely to be correlated, an exploratory factor analysis using principal axis factoring with oblique rotation (direct oblimin) was then conducted and four factors were retained. All items loaded high (>.5) on at least one factor but one item cross-loaded high on two factors and could be removed.
Structure Matrix
Factor
Fatigue Clarity
Answer-
ability Usability
Compared to other surveys I have taken, this survey was average. 0.84 -0.60 -0.59 -0.59
Throughout the survey, I felt engaged. 0.82 -0.60 -0.49 -0.62
In terms of flow, this survey flowed well. 0.79 -0.60 -0.61 -0.60
If I were asked, I would be likely to take this survey again. 0.74 -0.64 -0.63 -0.65
The questions were clear. 0.68 -0.94 -0.55 -0.74
I could understand the questions easily. 0.64 -0.86 -0.55 -0.72
This survey was easy to take. 0.71 -0.65 -0.94 -0.64
With the answers provided, I was able to answer all of the questions. 0.67 -0.73 -0.54 -0.91
All of the questions could be answered accurately. 0.67 -0.72 -0.53 -0.87
Did you have enough information to answer the questions? 0.66 -0.80 -0.51 -0.81
Extraction Method: Principal Axis Factoring.
Rotation Method: Oblimin with Kaiser Normalization.
High Cross
Loading
Forced 4 Factor Structure (Orthogonal)
A varimax rotated factor analysis was also conducted showing the same factor structure as the oblique rotation. All items loaded high (>.5).
Rotated Factor Matrixa
Factor
Fatigue Clarity
Answer-
ability Usability
Compared to other surveys I have taken, this survey was average. 0.73 0.24 0.22 0.27
Throughout the survey, I felt engaged. 0.71 0.26 0.29 0.13
In terms of flow, this survey flowed well. 0.64 0.26 0.25 0.31
If I were asked, I would be likely to take this survey again. 0.52 0.31 0.32 0.36
I could understand the questions easily. 0.30 0.74 0.33 0.23
The questions were clear. 0.35 0.72 0.35 0.23
With the answers provided, I was able to answer all of the questions. 0.32 0.35 0.74 0.23
All of the questions could be answered accurately. 0.35 0.35 0.68 0.22
This survey was easy to take. 0.39 0.29 0.27 0.77
Extraction Method: Principal Axis Factoring.
Rotation Method: Varimax with Kaiser Normalization.
a. Rotation converged in 7 iterations.
QuestionPro Integration with TryMyUI
Ritvij GautamCEO
TryMyUI
• Crowdsourced, User-Interface usability testing tool.
• Traditionally when people hear UI/usability testing, they think of Websites & Mobile
Apps/Web-apps.
• ANYTHING which involves a user has a UI that needs testing.
• How do we leverage our tool to test something like a survey?
TryMyUI’s Challenge
• QuestionPro is a comprehensive survey
tool that allows its customers to ask a
variety of questions.
• Now we need to make sure you ask the
right ones.
Wait, is this REALLY a problem?
How do you get actionable feedback on a feedback mechanism? Or meta-feedback?
1. Break it down to the elemental vectors of Cognitive Stress induced by surveys.
2. Have a psychometrically validated meta-survey to assess cognitive stress along these vectors.
3. Cross-reference the data with video of a demographically screened respondent taking the survey.
4. GET THE USER’S VIEW!
The Solution
Ordering the Test
Results Received
Watch the Videos
Review the SRS
Thank you!
Erik KotoCEO
www.questionpro.com
Jeff SauroFounding Principal
MeasuringU.com
Ritvij GautamCEO
TryMyUI.com