self-administered cognitive interviewing - aapor...past research it is possible that the goals of...
TRANSCRIPT
Self-Administered CognitiveInterviewing
or
Do Respondents Really Need Us?
Jennifer Edgar
AAPORMay 16, 2013
Background
Cognitive interviews are typically face to faceinterviews used for problem identification
In-depth probing used to explore responsestrategies
Small sample sizes are common, thoughsome research suggests larger sample sizescan increase results (Blair & Conrad, 2011)
Traditional cognitive interviews are timeintensive
2
Past Research
It is possible that the goals of cognitiveinterviewing can be met using a selfadministered format
Promising findings (Edgar, 2012)Quantity and quality of data was
comparableEfficiencies in cost & data collection time
Research focused on response strategye.g. “How did you arrive at that answer?”
3
Research Questions
Can information typically obtained throughcognitive interviews be collected using a selfadministered web survey?
Are self administered web interviews anefficient way to collect data about questioncomprehension?Are substantive conclusions the same?Are potential issues identified the same?
4
Method
43 traditional cognitive interviews 34 self administered, web interviews Same protocol used when possible
5
Questions
Testing of new and revised questions for BLSproduction survey
Focused on comprehension
1. Swimsuits, warm-up and ski suits2. Delivery and installation charges3. Mortgages
6
Lab Cognitive Interviews
Traditional cognitive interviewsSemi-structured, combination of scripted
and spontaneous probing Testing purpose and method explained to
participants Conducted in three US cities Most interviews lasted 45-50 minutes
7
Self Administered Interviews TryMyUI: remote usability testing serviceVolunteer panelParticipants rated on past work
Questions administered via web survey Participants given survey question and follow-
up questions (following lab protocol)Five spaces given for examples
Video and audio recording of computerscreen and participant voices
Sessions limited to 20 minutes8
Tasks
Respond to survey question Explain what you were thinking of when
you answered the question Give examples of items in category Complete inclusion/exclusion exercise
Participants
CI WebGender
Male 51% 26%Female 49% 74%
AgeAverage 47.6 38.2Median 48 36
Same demographic requests for both modes
Analysis
Transcripts were made of all think aloudresponsesFor cognitive interviews, number of probes
and responses to follow up probes were alsorecorded
Web survey data compared with data enteredfrom cognitive interviews
Results: Sportswear
Straightforward test of question comprension,little evidence of issues
Lab participant issues were with task (“give meexamples of what a swimsuit is”) not question
Web participants gave more examples Additional probing (average 1.2) did not
encourage lab participants to add examples
Avg. # of examples% of Ps with issue First Probe Total
CI 5% 2.6 2.7Web 0% 3.6
Results: Delivery and Installation
13
More complicated question construct Lab participants had more issuesIncluding shipping or other incorrect
charges (e.g. pizza)Naming items that were delivered for free
Avg. # of examples% of Ps with issue First Probe Total
CI 44% 4.1 / 4 4.2 / 4Web 2.7% 3.7 / 4
Results: Mortgage
14
Many people showing evidence of issues Lab debriefing revealed a ‘show stopper’ issue related
to refinancing, no evidence of this found on the web Web participants made comments indicating
sensitivity issues, no evidence of this in the lab
GeneralUncertainty
ComprehensionRefinance Issues
SensitivityIssues
CI – think aloud 46% 18% 0%Web –whileanswering
21% 3% 3%
Web – thinkaloud
29% 3% 25%
Substantive Conclusions
15
SportswearCI: question works fineWeb: question works fine
DeliveryCI: question has potential comprehension
issueWeb: question works fine
MortgageCI: question has show stopping issueWeb: question is potentially sensitive
Conclusion: Comprehension
16
Minor differences found in how peopleunderstood the question, or in how theyexplained their understanding of it
Web participants had fewer issues, but alsonamed fewer examples
Could be evidence of the lab probingtechnique creating questionsParticipants working hard to give more
examples, and ended up making mistakes
Conclusion: Sensitivity
17
Different conclusions about question sensitivityCIs included more expenditure questions
preceding the mortgage questions,participants may have gotten used toanswering this type of questionsBoth groups a ‘spiel’ about the study at the
beginning, but web participants skimmed it Self-administration is often thought to be
better for sensitive questions, maybe not forpretesting?
Conclusion: Question Function
‘Show stopper’ issue found in the lab notfound online
Follow-up probing identified the issue, whichwasn’t identifiable through the question-answer process or the single probe usedonline
Additional follow-up questions could havebeen added to the web survey, if potentialissue was caught in time
18
Observed Limitations
Some answers or explanations requiredfollow-up or re-focusingComments which would have prompted
probing went un explainedSeveral participants spent time going to
their banking site to look up mortgageinformation
Technical issues held some participants upand they spent their time working them outrather than completing the survey
19
Observed Strengths
Some web participants talked aloud the wholetime, giving useful information throughoutthe process.Lab participants generally talk only when
asked a question Some web participants gave un-requested,
honest feedback about the questions
20
Method Advantages
Self administered interviewing saves time andmoney
Large amounts of data can be easily collectedMaking the potential ‘bad’ data less
important as those cases are easilyreplaced
There is no interviewer effect on the web
21
Method Disadvantages
Potential for bias, if participants systematicallydiffer by mode (web participants were younger)
Instructions must be clear and work foreveryone
Cannot provide probes to follow up onparticipants’ commentsAll probes must be scripted and therefore
must be applicable to all participants No way to bring participants ‘back on track’
22
Conclusions
Can information typically obtained throughcognitive interviews be collected using a selfadministered web survey? YES
Are self administered web interviews anefficient way to collect data about questioncomprehension and recall? YESAre conclusions about the questions the
same? Not AlwaysAre potential issues identified the same? NO
23
Potential Applications Combination of standard CI and web studies
is likely an effective approachTake advantage of the cost and time
efficiency of web collectionFocus lab testing on in-depth probing
Run studies concurrently, modifying webprobes based on lab testing results
24
Contact
Study Costs
Total Cost PerParticipant
ParticipantIncentive
Web $27.30 $10
CI $42.66 $40
26
Study TimesTask CI Web
Requestingparticipants
20 minutes; explainingcriteria to recruiter
30 minutes total;specifying test groups
and criteria
Screening 10 minutes perparticipant
0 minutes; done byTryMyUI
Scheduling &confirming
15 minutes perparticipant
0 minutes; study doneat participantconvenience
Preparing forinterviews
10 minutes perparticipant
60 minutes total;setting up Web survey
and tasksConductinginterviews
45 minutes perparticipant
0 minutes, self-administered
Total 31.7 hours 1.5 hours