SlideShare une entreprise Scribd logo
1  sur  22
Télécharger pour lire hors ligne
A subsidiary of:
Weighting a probability online panel with
multiple waves of recruitment
JSM
August 2020
www.srcentre.com.au
Authors
Dina Neiger
Chief Statistician, the Social Research Centre
Campus Visitor, ANU Centre for Social Research and Methods
Andrew Ward
Principal Statistician, the Social Research Centre
Jack Barton
Analyst, Statistical and Survey Methods, the Social Research Centre
The Social Research Centre is a for-profit wholly owned subsidiary of the
Australian National University
2
www.srcentre.com.au
Acknowledgements
Darren Pennay
Founder & Senior Adviser, the Social Research Centre
Campus Visitor, ANU Centre for Social Research and Methods
Adjunct Professor, Institute for Social Science Research, the University of Queensland
Ben Phillips
Chief Survey Methodologist, the Social Research Centre
Campus Visitor, ANU Centre for Social Research and Methods
Paul J. Lavrakas
Senior Methodological Adviser, the Social Research Centre
Charles Dove
Research Consultant, the Social Research Centre
Panel Manager, Life in Australia™
3
www.srcentre.com.au
Introducing Life in Australia™
4
www.srcentre.com.au
Australia’s only probability-based online panel
Created by the Social Research Centre in 2016
Ø Non-probability panels in Australia similar to US/Europe, relative to
probability samples, they are
o more biased
o more variable; and
o bias may be increased via traditional weighting approaches.
Ø Quality online social research
Self-funded
Ø Designed for 2,000 completed surveys per wave
Ø From 2019, expanded to provide 3,000 or more completes for a full wave
Ø Primarily split-panel waves with different clients
5
www.srcentre.com.au
Substantive areas measured
6
Australia’s place in
the world
Crime and justice Attitudes to
disability, autism
Political views Sun protection
attitudes /
behaviours
Social cohesion Communications
use
Awareness of
alcohol harms
Image-based abuse Cyber crime
Data privacy and
linkage
Attitudes about
immigration
Fertility Gambling Elections
Clients to date: government, academic and not-for-profit
www.srcentre.com.au
Operating model
Waves most months
Ø Typically whole-of-panel with occasional sub-panel selections
Ø Main client for most waves
Ø Add-on omnibus modules for secondary clients
Incentives
Ø 10 AUD base incentive for waves under 20 minutes
Ø 5 AUD additional for every extra 5 minutes
Use for primary data collection and calibration
Ø Rich profile for all new recruits including: range of primary and secondary
demographic variables, risk behaviours, health outcomes, Big5
personality inventory and social desirability indicators
7
www.srcentre.com.au
Evolution of recruitment
8
www.srcentre.com.au
Panel establishment
• Single-frame design weights where
probability of selection is
•𝑝 =
!!!""
#!!$%!!
+
!"#&'
#"#
• where 𝑆"" is number of landline
respondents, 𝐿𝐿 is 0,1 indicator of having
landline, 𝑈"" is size of landline universe,
𝐴𝐷"" is number of adults in HH, 𝑆&' is
number of mobile respondents, 𝑀𝑃 is 0,1
indicator of having mobile and 𝑈&' is size of
mobile universe
2016recruitment
• Dual-frame RDD
• Landline (n=911)
• Mobile (n=2,411)
2016designweights
9
www.srcentre.com.au
Evolution of Life in Australia™ recruitment 2018 refresh
2016recruitment
• Dual-frame RDD
• Landline (n=911)
• Mobile (n=2,411)
• Panel size n=3,322
2018
replenishment
• Mobile RDD (n=267)
• Only age 54 and
under
• Quotas on university
education
• Panel size n=2,839
10
www.srcentre.com.au
Evolution of Life in Australia™ recruitment - 2019 refresh
2016recruitment
• Dual-frame RDD
• Landline
(n=911)
• Mobile
(n=2,411)
• Panel size
n=3,322 2018replenishment
• Mobile RDD
(n=267)
• Only age 54 and
under
• Quotas on
university
education
• Panel size
n=2,839
2019expansion
• ABS (n=1,810)
• Area-level
stratification by
SES and age
• Expand to 3,000
completes/ wave
• Panel size
n=4,025
11
www.srcentre.com.au
Evolution of Life in Australia™ recruitment
2016recruitment
• Dual-frame RDD
• Landline
(n=911)
• Mobile
(n=2,411)
• Panel size
n=3,322
2018replenishment
• Mobile RDD
(n=267)
• Only age 54 and
under
• Quotas on
university
education
• Panel size
n=2,839
2019expansion
• ABS (n=1,810)
• Area-level
stratification by
SES and age
• Expand to 3,000
completes/ wave
• Panel size
n=4,025
12
non-random attrition, increasingly unknowable probabilities of selection
decreasing weighting efficiency, increasing weighting complexity
www.srcentre.com.au
Weighting approaches
13
www.srcentre.com.au
Derive exact probabilities of selection
Ø Potential to derive exact probabilities of selection either
o Assuming single frame or
o Treating as composite frames
Ø Challenges
o Future-proofing, increasingly complicated
o Large weight differentials leading to inefficiencies
o Attempted for 2018 – very low efficiency
o Not feasible for 2019 – too many non random elements
Ø References
o Kish, Leslie. 1965. Survey Sampling. New York: Wiley
o Lohr, Sharon L. 2011. ‘Alternative Survey Sample Designs: Sampling with Multiple
Overlapping Frames.’ Survey Methodology 37(2):197–213.
14
www.srcentre.com.au
Approximate probabilities of selection
Ø Propensity response model to match new cases (full coverage and
design weight applied) and continuing cases to approximate selection
weight
Ø Challenges
o Disproportionate selection of new cases (reference sample)
o Lack of common support for parts of the population, large weight differentials leading to
inefficiency
Ø References
o Rosenbaum, P.R. and D.B. Rubin (1983) The central role of the propensity score in
observational studies for causal effects. Biometrika 70(1), 41-55.
o Valliant, R., J. Dever, and F. Kreuter (2013). Practical Tools for Designing and Weighting
Survey Samples. New York: Springer.
o Bai, H., & Clark, M. (2019). Propensity score methods and Applications. Thousand Oaks,
CA: SAGE Publications, Inc.
15
www.srcentre.com.au
Propensity response model
16
A range of primary and secondary demographic variables, risk behaviours,
health outcomes, Big5 personality inventory and social desirability indicators
Propensity classes using stepwise logistic regression
Bias is measured as an absolute average error (difference) from the high
quality (official statistics) benchmarks for the modal response.
• Unweighted bias = 4.2
With propensity selection weights:
• Bias = 3.6
• Weighting Efficiency = 68.9%
www.srcentre.com.au
Doubly-robust approach
Ø Combine
o Approximate probabilities of selection
o Post-stratification (GREG)
Ø Challenges
o Calculation of probabilities of selection
o Model choice
o Further reduction in efficiency
Ø References
o Valliant, R. (2020) Comparing Alternatives for Estimation from Nonprobability Samples.
Journal of Survey Statistics and Methodology (2020) 8, 231–263
17
www.srcentre.com.au
Doubly-robust approach
Many combinations of variables (2,000 solutions for illustration):
Max weighting efficiency 69.1% for Bias < Unweighted Bias (4.1)
Minimum bias* 2.3 (Weighting efficiency 37.0%)
Ave weighting efficiency of 55.2% for solutions with bias btw 3. 5 & 3.7
*Variables used in calibration are excluded from bias measurement
18
www.srcentre.com.au
Super-population model
Ø Observed values are treated as if they have been generated by a model
o Combine all samples and
o Predict values for those units of the population that have not been sampled through the
whole panel using auxiliary information from external sources
Ø Challenges
o Model failure
o Tension between including covariates and efficiency
Ø References
o Valliant, R. and Dever, J. A. (2018) Survey Weights: A Step-by-Step Guide to Calculation.
Stata Press, 2018
19
www.srcentre.com.au
Comparison results – 2019 refresh panel weights
Approach Weighting
efficiency
Bias Measure
(unweighted
4.2)
Final variables used for
selection weight
Final variables used for
calibration
Approximate
selection weights
68.9% 3.6 Profile variables None
Doubly-robust
approach
Max 69.1%
37.0%
Ave 55.2%
4.1
Min 2.3
Btw 3.5 and 3.7
As above >2000 solutions
Super-population
model
Max 97.4%
53.9%
Ave 77.0%
Goldilocks
Solution:
75.8
4.1
Min 2.6
Btw 3.5 and 3.7
Goldilocks
Solution:
2.9
None >2000 solutions
Goldilocks Solution:
Age, Country of Birth,
Gender, Phone status,
SEIFA, State
20
www.srcentre.com.au
Lessons learnt
• Design-based weights are not suited for a long-term panel with multiple
refreshes
• Model-based approach is
• Simpler to implement
• Simpler to explain
• More efficient and less biased
• Universally applicable independent of the sample selection
mechanism
21
* PO Box 13328
Law Courts Victoria 8010
Thank you
( 03 9236 8500
A subsidiary of:
dina.neiger@srcentre.com.au
22

Contenu connexe

Similaire à Weighting a probability online panel with multiple waves of recruitment

Predicting student performance using aggregated data sources
Predicting student performance using aggregated data sourcesPredicting student performance using aggregated data sources
Predicting student performance using aggregated data sourcesOlugbenga Wilson Adejo
 
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...NAP Global Network
 
KwanCaseStudyICES2011
KwanCaseStudyICES2011KwanCaseStudyICES2011
KwanCaseStudyICES2011Stephen Kwan
 
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...The Social Research Centre
 
Workshop session 9 - Alternatives to CATI (2) probability online panels
Workshop session 9 - Alternatives to CATI (2) probability online panelsWorkshop session 9 - Alternatives to CATI (2) probability online panels
Workshop session 9 - Alternatives to CATI (2) probability online panelsThe Social Research Centre
 
Uses and misuses of quantitative indicators of impact
Uses and misuses of quantitative indicators of impactUses and misuses of quantitative indicators of impact
Uses and misuses of quantitative indicators of impactBerenika Webster
 
In metrics we trust?
In metrics we trust?In metrics we trust?
In metrics we trust?ORCID, Inc
 
UDL In Assessments
UDL In AssessmentsUDL In Assessments
UDL In AssessmentsScott Rains
 
Evaluation of Library STEM Programs: Learning from the BISE Project
Evaluation of Library STEM Programs: Learning from the BISE ProjectEvaluation of Library STEM Programs: Learning from the BISE Project
Evaluation of Library STEM Programs: Learning from the BISE ProjectNCIL - STAR_Net
 
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...NAP Global Network
 
Nursing Research Sampling Technique .pptx
Nursing Research Sampling Technique .pptxNursing Research Sampling Technique .pptx
Nursing Research Sampling Technique .pptxChinna Chadayan
 
Sampling merits & Demerits.pptx
Sampling merits & Demerits.pptxSampling merits & Demerits.pptx
Sampling merits & Demerits.pptxheencomm
 
Sampling merits & Demerits.pptx
Sampling merits & Demerits.pptxSampling merits & Demerits.pptx
Sampling merits & Demerits.pptxheencomm
 
Challenges in Software Ecosystems Research
Challenges in Software Ecosystems ResearchChallenges in Software Ecosystems Research
Challenges in Software Ecosystems ResearchAlexander Serebrenik
 
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...Big data solutions for smallholder farmers in Southeast Asia: machine learnin...
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...Sustainable Cassava Disease Solutions Asia
 
Transferring biodiversity models for conservation: Opportunities and challenges
Transferring biodiversity models for conservation: Opportunities and challengesTransferring biodiversity models for conservation: Opportunities and challenges
Transferring biodiversity models for conservation: Opportunities and challengesPhil Bouchet
 

Similaire à Weighting a probability online panel with multiple waves of recruitment (20)

Predicting student performance using aggregated data sources
Predicting student performance using aggregated data sourcesPredicting student performance using aggregated data sources
Predicting student performance using aggregated data sources
 
sampling methods in research design
sampling  methods in research designsampling  methods in research design
sampling methods in research design
 
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, data sources and assessment | Laszlo Pinter, ...
 
KwanCaseStudyICES2011
KwanCaseStudyICES2011KwanCaseStudyICES2011
KwanCaseStudyICES2011
 
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...
Workshop session 10 - Alternatives to CATI (3) address-based sampling and pus...
 
How to prepare a thesis
How to prepare a thesisHow to prepare a thesis
How to prepare a thesis
 
Workshop session 9 - Alternatives to CATI (2) probability online panels
Workshop session 9 - Alternatives to CATI (2) probability online panelsWorkshop session 9 - Alternatives to CATI (2) probability online panels
Workshop session 9 - Alternatives to CATI (2) probability online panels
 
Uses and misuses of quantitative indicators of impact
Uses and misuses of quantitative indicators of impactUses and misuses of quantitative indicators of impact
Uses and misuses of quantitative indicators of impact
 
In metrics we trust?
In metrics we trust?In metrics we trust?
In metrics we trust?
 
UDL In Assessments
UDL In AssessmentsUDL In Assessments
UDL In Assessments
 
Evaluation of Library STEM Programs: Learning from the BISE Project
Evaluation of Library STEM Programs: Learning from the BISE ProjectEvaluation of Library STEM Programs: Learning from the BISE Project
Evaluation of Library STEM Programs: Learning from the BISE Project
 
chap1.ppt
chap1.pptchap1.ppt
chap1.ppt
 
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...
Measuring Progress: Indicators, Data Sources and Assessment | Laszlo Pinter, ...
 
Nursing Research Sampling Technique .pptx
Nursing Research Sampling Technique .pptxNursing Research Sampling Technique .pptx
Nursing Research Sampling Technique .pptx
 
Sampling merits & Demerits.pptx
Sampling merits & Demerits.pptxSampling merits & Demerits.pptx
Sampling merits & Demerits.pptx
 
Sampling merits & Demerits.pptx
Sampling merits & Demerits.pptxSampling merits & Demerits.pptx
Sampling merits & Demerits.pptx
 
Challenges in Software Ecosystems Research
Challenges in Software Ecosystems ResearchChallenges in Software Ecosystems Research
Challenges in Software Ecosystems Research
 
Sampling
SamplingSampling
Sampling
 
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...Big data solutions for smallholder farmers in Southeast Asia: machine learnin...
Big data solutions for smallholder farmers in Southeast Asia: machine learnin...
 
Transferring biodiversity models for conservation: Opportunities and challenges
Transferring biodiversity models for conservation: Opportunities and challengesTransferring biodiversity models for conservation: Opportunities and challenges
Transferring biodiversity models for conservation: Opportunities and challenges
 

Plus de The Social Research Centre

Workshop session 11 - The future of CATI - including panel discussion
Workshop session 11 - The future of CATI - including panel discussionWorkshop session 11 - The future of CATI - including panel discussion
Workshop session 11 - The future of CATI - including panel discussionThe Social Research Centre
 
Workshop session 8 - Alternatives to CATI (1) non-probability online panels
Workshop session 8 - Alternatives to CATI (1) non-probability online panelsWorkshop session 8 - Alternatives to CATI (1) non-probability online panels
Workshop session 8 - Alternatives to CATI (1) non-probability online panelsThe Social Research Centre
 
Workshop session 5 - the effectiveness of standard methods to improve general...
Workshop session 5 - the effectiveness of standard methods to improve general...Workshop session 5 - the effectiveness of standard methods to improve general...
Workshop session 5 - the effectiveness of standard methods to improve general...The Social Research Centre
 
Workshop session 3 - Sampling frames for telephone surveys
Workshop session 3 - Sampling frames for telephone surveysWorkshop session 3 - Sampling frames for telephone surveys
Workshop session 3 - Sampling frames for telephone surveysThe Social Research Centre
 
Workshop session 2 - The telephone status of the population
Workshop session 2 - The telephone status of the populationWorkshop session 2 - The telephone status of the population
Workshop session 2 - The telephone status of the populationThe Social Research Centre
 
Increasing cooperation in telephone surveys with the progressive engagement t...
Increasing cooperation in telephone surveys with the progressive engagement t...Increasing cooperation in telephone surveys with the progressive engagement t...
Increasing cooperation in telephone surveys with the progressive engagement t...The Social Research Centre
 

Plus de The Social Research Centre (8)

Workshop session 12 - Closing remarks
Workshop session 12 - Closing remarksWorkshop session 12 - Closing remarks
Workshop session 12 - Closing remarks
 
Workshop session 11 - The future of CATI - including panel discussion
Workshop session 11 - The future of CATI - including panel discussionWorkshop session 11 - The future of CATI - including panel discussion
Workshop session 11 - The future of CATI - including panel discussion
 
Workshop session 8 - Alternatives to CATI (1) non-probability online panels
Workshop session 8 - Alternatives to CATI (1) non-probability online panelsWorkshop session 8 - Alternatives to CATI (1) non-probability online panels
Workshop session 8 - Alternatives to CATI (1) non-probability online panels
 
Workshop session 5 - the effectiveness of standard methods to improve general...
Workshop session 5 - the effectiveness of standard methods to improve general...Workshop session 5 - the effectiveness of standard methods to improve general...
Workshop session 5 - the effectiveness of standard methods to improve general...
 
Workshop session 3 - Sampling frames for telephone surveys
Workshop session 3 - Sampling frames for telephone surveysWorkshop session 3 - Sampling frames for telephone surveys
Workshop session 3 - Sampling frames for telephone surveys
 
Workshop session 2 - The telephone status of the population
Workshop session 2 - The telephone status of the populationWorkshop session 2 - The telephone status of the population
Workshop session 2 - The telephone status of the population
 
Workshop session 1 - How did we get here?
Workshop session 1 -  How did we get here?Workshop session 1 -  How did we get here?
Workshop session 1 - How did we get here?
 
Increasing cooperation in telephone surveys with the progressive engagement t...
Increasing cooperation in telephone surveys with the progressive engagement t...Increasing cooperation in telephone surveys with the progressive engagement t...
Increasing cooperation in telephone surveys with the progressive engagement t...
 

Dernier

proposal kumeneger edited.docx A kumeeger
proposal kumeneger edited.docx A kumeegerproposal kumeneger edited.docx A kumeeger
proposal kumeneger edited.docx A kumeegerkumenegertelayegrama
 
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptxerickamwana1
 
cse-csp batch4 review-1.1.pptx cyber security
cse-csp batch4 review-1.1.pptx cyber securitycse-csp batch4 review-1.1.pptx cyber security
cse-csp batch4 review-1.1.pptx cyber securitysandeepnani2260
 
Internship Presentation | PPT | CSE | SE
Internship Presentation | PPT | CSE | SEInternship Presentation | PPT | CSE | SE
Internship Presentation | PPT | CSE | SESaleh Ibne Omar
 
General Elections Final Press Noteas per M
General Elections Final Press Noteas per MGeneral Elections Final Press Noteas per M
General Elections Final Press Noteas per MVidyaAdsule1
 
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunity
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunityDon't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunity
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunityApp Ethena
 
Chizaram's Women Tech Makers Deck. .pptx
Chizaram's Women Tech Makers Deck.  .pptxChizaram's Women Tech Makers Deck.  .pptx
Chizaram's Women Tech Makers Deck. .pptxogubuikealex
 
GESCO SE Press and Analyst Conference on Financial Results 2024
GESCO SE Press and Analyst Conference on Financial Results 2024GESCO SE Press and Analyst Conference on Financial Results 2024
GESCO SE Press and Analyst Conference on Financial Results 2024GESCO SE
 
INDIAN GCP GUIDELINE. for Regulatory affair 1st sem CRR
INDIAN GCP GUIDELINE. for Regulatory  affair 1st sem CRRINDIAN GCP GUIDELINE. for Regulatory  affair 1st sem CRR
INDIAN GCP GUIDELINE. for Regulatory affair 1st sem CRRsarwankumar4524
 
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...漢銘 謝
 
Testing with Fewer Resources: Toward Adaptive Approaches for Cost-effective ...
Testing with Fewer Resources:  Toward Adaptive Approaches for Cost-effective ...Testing with Fewer Resources:  Toward Adaptive Approaches for Cost-effective ...
Testing with Fewer Resources: Toward Adaptive Approaches for Cost-effective ...Sebastiano Panichella
 
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATION
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATIONRACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATION
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATIONRachelAnnTenibroAmaz
 
Quality by design.. ppt for RA (1ST SEM
Quality by design.. ppt for  RA (1ST SEMQuality by design.. ppt for  RA (1ST SEM
Quality by design.. ppt for RA (1ST SEMCharmi13
 
A Guide to Choosing the Ideal Air Cooler
A Guide to Choosing the Ideal Air CoolerA Guide to Choosing the Ideal Air Cooler
A Guide to Choosing the Ideal Air Coolerenquirieskenstar
 
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...Sebastiano Panichella
 
Engaging Eid Ul Fitr Presentation for Kindergartners.pptx
Engaging Eid Ul Fitr Presentation for Kindergartners.pptxEngaging Eid Ul Fitr Presentation for Kindergartners.pptx
Engaging Eid Ul Fitr Presentation for Kindergartners.pptxAsifArshad8
 
Application of GIS in Landslide Disaster Response.pptx
Application of GIS in Landslide Disaster Response.pptxApplication of GIS in Landslide Disaster Response.pptx
Application of GIS in Landslide Disaster Response.pptxRoquia Salam
 

Dernier (17)

proposal kumeneger edited.docx A kumeeger
proposal kumeneger edited.docx A kumeegerproposal kumeneger edited.docx A kumeeger
proposal kumeneger edited.docx A kumeeger
 
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx
05.02 MMC - Assignment 4 - Image Attribution Lovepreet.pptx
 
cse-csp batch4 review-1.1.pptx cyber security
cse-csp batch4 review-1.1.pptx cyber securitycse-csp batch4 review-1.1.pptx cyber security
cse-csp batch4 review-1.1.pptx cyber security
 
Internship Presentation | PPT | CSE | SE
Internship Presentation | PPT | CSE | SEInternship Presentation | PPT | CSE | SE
Internship Presentation | PPT | CSE | SE
 
General Elections Final Press Noteas per M
General Elections Final Press Noteas per MGeneral Elections Final Press Noteas per M
General Elections Final Press Noteas per M
 
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunity
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunityDon't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunity
Don't Miss Out: Strategies for Making the Most of the Ethena DigitalOpportunity
 
Chizaram's Women Tech Makers Deck. .pptx
Chizaram's Women Tech Makers Deck.  .pptxChizaram's Women Tech Makers Deck.  .pptx
Chizaram's Women Tech Makers Deck. .pptx
 
GESCO SE Press and Analyst Conference on Financial Results 2024
GESCO SE Press and Analyst Conference on Financial Results 2024GESCO SE Press and Analyst Conference on Financial Results 2024
GESCO SE Press and Analyst Conference on Financial Results 2024
 
INDIAN GCP GUIDELINE. for Regulatory affair 1st sem CRR
INDIAN GCP GUIDELINE. for Regulatory  affair 1st sem CRRINDIAN GCP GUIDELINE. for Regulatory  affair 1st sem CRR
INDIAN GCP GUIDELINE. for Regulatory affair 1st sem CRR
 
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...
THE COUNTRY WHO SOLVED THE WORLD_HOW CHINA LAUNCHED THE CIVILIZATION REVOLUTI...
 
Testing with Fewer Resources: Toward Adaptive Approaches for Cost-effective ...
Testing with Fewer Resources:  Toward Adaptive Approaches for Cost-effective ...Testing with Fewer Resources:  Toward Adaptive Approaches for Cost-effective ...
Testing with Fewer Resources: Toward Adaptive Approaches for Cost-effective ...
 
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATION
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATIONRACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATION
RACHEL-ANN M. TENIBRO PRODUCT RESEARCH PRESENTATION
 
Quality by design.. ppt for RA (1ST SEM
Quality by design.. ppt for  RA (1ST SEMQuality by design.. ppt for  RA (1ST SEM
Quality by design.. ppt for RA (1ST SEM
 
A Guide to Choosing the Ideal Air Cooler
A Guide to Choosing the Ideal Air CoolerA Guide to Choosing the Ideal Air Cooler
A Guide to Choosing the Ideal Air Cooler
 
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...
Testing and Development Challenges for Complex Cyber-Physical Systems: Insigh...
 
Engaging Eid Ul Fitr Presentation for Kindergartners.pptx
Engaging Eid Ul Fitr Presentation for Kindergartners.pptxEngaging Eid Ul Fitr Presentation for Kindergartners.pptx
Engaging Eid Ul Fitr Presentation for Kindergartners.pptx
 
Application of GIS in Landslide Disaster Response.pptx
Application of GIS in Landslide Disaster Response.pptxApplication of GIS in Landslide Disaster Response.pptx
Application of GIS in Landslide Disaster Response.pptx
 

Weighting a probability online panel with multiple waves of recruitment

  • 1. A subsidiary of: Weighting a probability online panel with multiple waves of recruitment JSM August 2020
  • 2. www.srcentre.com.au Authors Dina Neiger Chief Statistician, the Social Research Centre Campus Visitor, ANU Centre for Social Research and Methods Andrew Ward Principal Statistician, the Social Research Centre Jack Barton Analyst, Statistical and Survey Methods, the Social Research Centre The Social Research Centre is a for-profit wholly owned subsidiary of the Australian National University 2
  • 3. www.srcentre.com.au Acknowledgements Darren Pennay Founder & Senior Adviser, the Social Research Centre Campus Visitor, ANU Centre for Social Research and Methods Adjunct Professor, Institute for Social Science Research, the University of Queensland Ben Phillips Chief Survey Methodologist, the Social Research Centre Campus Visitor, ANU Centre for Social Research and Methods Paul J. Lavrakas Senior Methodological Adviser, the Social Research Centre Charles Dove Research Consultant, the Social Research Centre Panel Manager, Life in Australia™ 3
  • 5. www.srcentre.com.au Australia’s only probability-based online panel Created by the Social Research Centre in 2016 Ø Non-probability panels in Australia similar to US/Europe, relative to probability samples, they are o more biased o more variable; and o bias may be increased via traditional weighting approaches. Ø Quality online social research Self-funded Ø Designed for 2,000 completed surveys per wave Ø From 2019, expanded to provide 3,000 or more completes for a full wave Ø Primarily split-panel waves with different clients 5
  • 6. www.srcentre.com.au Substantive areas measured 6 Australia’s place in the world Crime and justice Attitudes to disability, autism Political views Sun protection attitudes / behaviours Social cohesion Communications use Awareness of alcohol harms Image-based abuse Cyber crime Data privacy and linkage Attitudes about immigration Fertility Gambling Elections Clients to date: government, academic and not-for-profit
  • 7. www.srcentre.com.au Operating model Waves most months Ø Typically whole-of-panel with occasional sub-panel selections Ø Main client for most waves Ø Add-on omnibus modules for secondary clients Incentives Ø 10 AUD base incentive for waves under 20 minutes Ø 5 AUD additional for every extra 5 minutes Use for primary data collection and calibration Ø Rich profile for all new recruits including: range of primary and secondary demographic variables, risk behaviours, health outcomes, Big5 personality inventory and social desirability indicators 7
  • 9. www.srcentre.com.au Panel establishment • Single-frame design weights where probability of selection is •𝑝 = !!!"" #!!$%!! + !"#&' #"# • where 𝑆"" is number of landline respondents, 𝐿𝐿 is 0,1 indicator of having landline, 𝑈"" is size of landline universe, 𝐴𝐷"" is number of adults in HH, 𝑆&' is number of mobile respondents, 𝑀𝑃 is 0,1 indicator of having mobile and 𝑈&' is size of mobile universe 2016recruitment • Dual-frame RDD • Landline (n=911) • Mobile (n=2,411) 2016designweights 9
  • 10. www.srcentre.com.au Evolution of Life in Australia™ recruitment 2018 refresh 2016recruitment • Dual-frame RDD • Landline (n=911) • Mobile (n=2,411) • Panel size n=3,322 2018 replenishment • Mobile RDD (n=267) • Only age 54 and under • Quotas on university education • Panel size n=2,839 10
  • 11. www.srcentre.com.au Evolution of Life in Australia™ recruitment - 2019 refresh 2016recruitment • Dual-frame RDD • Landline (n=911) • Mobile (n=2,411) • Panel size n=3,322 2018replenishment • Mobile RDD (n=267) • Only age 54 and under • Quotas on university education • Panel size n=2,839 2019expansion • ABS (n=1,810) • Area-level stratification by SES and age • Expand to 3,000 completes/ wave • Panel size n=4,025 11
  • 12. www.srcentre.com.au Evolution of Life in Australia™ recruitment 2016recruitment • Dual-frame RDD • Landline (n=911) • Mobile (n=2,411) • Panel size n=3,322 2018replenishment • Mobile RDD (n=267) • Only age 54 and under • Quotas on university education • Panel size n=2,839 2019expansion • ABS (n=1,810) • Area-level stratification by SES and age • Expand to 3,000 completes/ wave • Panel size n=4,025 12 non-random attrition, increasingly unknowable probabilities of selection decreasing weighting efficiency, increasing weighting complexity
  • 14. www.srcentre.com.au Derive exact probabilities of selection Ø Potential to derive exact probabilities of selection either o Assuming single frame or o Treating as composite frames Ø Challenges o Future-proofing, increasingly complicated o Large weight differentials leading to inefficiencies o Attempted for 2018 – very low efficiency o Not feasible for 2019 – too many non random elements Ø References o Kish, Leslie. 1965. Survey Sampling. New York: Wiley o Lohr, Sharon L. 2011. ‘Alternative Survey Sample Designs: Sampling with Multiple Overlapping Frames.’ Survey Methodology 37(2):197–213. 14
  • 15. www.srcentre.com.au Approximate probabilities of selection Ø Propensity response model to match new cases (full coverage and design weight applied) and continuing cases to approximate selection weight Ø Challenges o Disproportionate selection of new cases (reference sample) o Lack of common support for parts of the population, large weight differentials leading to inefficiency Ø References o Rosenbaum, P.R. and D.B. Rubin (1983) The central role of the propensity score in observational studies for causal effects. Biometrika 70(1), 41-55. o Valliant, R., J. Dever, and F. Kreuter (2013). Practical Tools for Designing and Weighting Survey Samples. New York: Springer. o Bai, H., & Clark, M. (2019). Propensity score methods and Applications. Thousand Oaks, CA: SAGE Publications, Inc. 15
  • 16. www.srcentre.com.au Propensity response model 16 A range of primary and secondary demographic variables, risk behaviours, health outcomes, Big5 personality inventory and social desirability indicators Propensity classes using stepwise logistic regression Bias is measured as an absolute average error (difference) from the high quality (official statistics) benchmarks for the modal response. • Unweighted bias = 4.2 With propensity selection weights: • Bias = 3.6 • Weighting Efficiency = 68.9%
  • 17. www.srcentre.com.au Doubly-robust approach Ø Combine o Approximate probabilities of selection o Post-stratification (GREG) Ø Challenges o Calculation of probabilities of selection o Model choice o Further reduction in efficiency Ø References o Valliant, R. (2020) Comparing Alternatives for Estimation from Nonprobability Samples. Journal of Survey Statistics and Methodology (2020) 8, 231–263 17
  • 18. www.srcentre.com.au Doubly-robust approach Many combinations of variables (2,000 solutions for illustration): Max weighting efficiency 69.1% for Bias < Unweighted Bias (4.1) Minimum bias* 2.3 (Weighting efficiency 37.0%) Ave weighting efficiency of 55.2% for solutions with bias btw 3. 5 & 3.7 *Variables used in calibration are excluded from bias measurement 18
  • 19. www.srcentre.com.au Super-population model Ø Observed values are treated as if they have been generated by a model o Combine all samples and o Predict values for those units of the population that have not been sampled through the whole panel using auxiliary information from external sources Ø Challenges o Model failure o Tension between including covariates and efficiency Ø References o Valliant, R. and Dever, J. A. (2018) Survey Weights: A Step-by-Step Guide to Calculation. Stata Press, 2018 19
  • 20. www.srcentre.com.au Comparison results – 2019 refresh panel weights Approach Weighting efficiency Bias Measure (unweighted 4.2) Final variables used for selection weight Final variables used for calibration Approximate selection weights 68.9% 3.6 Profile variables None Doubly-robust approach Max 69.1% 37.0% Ave 55.2% 4.1 Min 2.3 Btw 3.5 and 3.7 As above >2000 solutions Super-population model Max 97.4% 53.9% Ave 77.0% Goldilocks Solution: 75.8 4.1 Min 2.6 Btw 3.5 and 3.7 Goldilocks Solution: 2.9 None >2000 solutions Goldilocks Solution: Age, Country of Birth, Gender, Phone status, SEIFA, State 20
  • 21. www.srcentre.com.au Lessons learnt • Design-based weights are not suited for a long-term panel with multiple refreshes • Model-based approach is • Simpler to implement • Simpler to explain • More efficient and less biased • Universally applicable independent of the sample selection mechanism 21
  • 22. * PO Box 13328 Law Courts Victoria 8010 Thank you ( 03 9236 8500 A subsidiary of: dina.neiger@srcentre.com.au 22