improved questionnaire design yields better data: experiences from the uks annual survey of hours...
TRANSCRIPT
Improved Questionnaire Design Yields Better Data: Experiences from the UK’s Annual Survey of Hours and Earnings
Jacqui Jones, Pete Brodie, Sarah Williams & Jane CarterUK Office for National Statistics
Overview
• Data Collection Methodology at ONS
• Methods standards
• Design standards
• Case study: The Annual Survey of Hours and Earnings
Slide 1
Background to DCM at ONS
• Mid 1990s 1st QTU set-up for Census
• Then expanded to include social surveys and business surveys
Slide 2
Methods Standards
• Aim was to develop a framework of processes to be used for all DCM question(naire) development work
• Outcome was the Data Collection Methodology Improvement and Measurement Framework
(DCM IMF)
Slide 3
The DCM IMF
Identification of data requirements
+Expert Review +
Evaluation Review
Stage 2 Redesign & Development
Stage 3 Pre-field Testing
Stage 4
Stage 1
Stage 5 Field Implementation
Identify and Disseminate Best Practice
Stage 7
Stage 6
Field Testing
Post-implementation Evaluation
Slide 4
The DCM IMF
Stage 1
Stage 2
Stage 3
Stage 4
Literature review
Expert review
Evaluation review
Instrument development
Expert review
Qualitative testing
Expert review
Re-development
Quantitative analysis
Qualitative follow-up
Slide 5
Design Standards
• Aim was to develop tested standards for the design of questions and questionnaires
• Outcome is draft standards
– with constraints – colour and cross boxes
Slide 6
Example of Design Standards
Example of Design Standards
Case Study: ASHE
• ASHE collects data for earning statistics e.g. minimum wage
• ASHE questionnaires are sent out to employers to complete on behalf of selected employees
• 250,000 sample of employees
Slide 9
ASHE: Need for Change
• National Statistics Quality Review
• New data requirements e.g. Pay for other reasons Pension data
• Existing data problems e.g.
Possible inclusion of pay elements in the wrong responses
Slide 10
Overview of the ‘Old’ Questionnaire
• Very cramped 2 sided questionnaire which collected 26 data items
• 4 sides of additional notes
• Respondent coding
Slide 11
‘Old’ Questionnaire
Slide 12
‘Old’ Questionnaire
Slide 14
Overview of the ‘Old’ Questionnaire
• Unclear data requirements
• Inconsistent terminology
• Double-barrelled questions
Slide 15
‘Old’ Questionnaire
ASHE Development
Followed DCM IMF processes:
• Focus group (8 large businesses)
• Pre-field testing• 30 cognitive interviews (5 waves of testing)
Waves 1-3 concurrent interviewingWaves 4-5 retrospective interviewing
Slide 17
ASHE Development
2004 Field tested 6 sided questionnaire:
• Split sample design (98% ‘old’ , 2% ‘new’)
• Quantitative analysisNon-responseRespondent burden (actual & perceived)Processing errors
Slide 18
Results of ASHE Field Test
• Response (no statistical difference)• Errors (no statistical difference)
• Respondent feedback (positive)• Respondent burden (reduced per question)• Some problems with question wording
• Costs (increased paper, printing, postage, processing)• Time (increased processing time)
Slide 19
6 sided to 4 sided compromises
• Minimum standard – based on testing - keep question & response category wording + instructions to stay with questions
• Challenge was to promote positive respondent perceptions and clear navigational path
• No time for further testing – redesign by DCM experts & ‘quick perception’ telephone interviews
Slide 20
‘Old’ Questionnaire
Slide 21
ASHE 6 Sided Questionnaire
ASHE 4 Sided Questionnaire
2006 ASHE 4 Sided Questionnaire
Slide 24
2005 Qualitative Follow-Up
• Looked at differences between responses to the 2004 ‘old’ questionnaire and the 2005 4 sided questionnaire.
• Focussed on 5 key areas e.g. the affect on basic pay of a new ‘other pay’ question
• 31 in-depth telephone interviews
Slide 25
‘Old’ Basic Pay Question
Slide 26
‘New’ Basic Pay and Other Pay Questions
Slide 27
‘Other Pay’ Results
Qualitative study found:• Better comprehension of new questions• Improved reporting of ‘other pay’• Respondents preferred ‘new’ layout
Quantitative analysis found:• ‘Old’ questionnaire - ‘other pay’ had been excluded
by some• Correction by imputation added 0.8% growth to
basic pay at the all employee levelSlide 28
What did we learn?
Must have:
• clear & agreed data requirements• agreement on the maximum number of pages• DCM experts involved in redesign & testing• method standards• design standards • improved methods for measuring data quality
Slide 29
Thank You