This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
AAPOR 2015 Mini-ConferenceReassessing Today's Survey Methods:Voters, Election and Non-Probability Samples
Meta-Analysis of Online Paneland Non-Panel Sampling:Electoral and Non-ElectoralBehavior Metrics
2
The survey research industry is in an era of transition and flux: ‘traditional’ methodologies (now defined aslive interviewer landline phone and cellular phone) are waning in prevalence – if not importance – whilenewer methodologies (online, IVR, blended) are gaining prominence. The impact of this on measurementaccuracy is still being explored, with new information emerging all the time.
To further this body of knowledge specifically within online work, we analyze behavior measurementaccuracy of the two main sampling approaches for internet-based research: panel only and non-panel(River and other approaches) sampling. Our paper will comprise a meta-analysis of nonprobablity onlinesampling, looking across over 400 data points, at election and non-election behaviors as validating points.Behavior measurement accuracy will be assessed against external benchmarks from market-accepted datasources such as the US Census, Current Population Survey (CPS), and the General Social Survey (GSS).
The analysis will focus specifically on differences in accuracy between sample sources or blendedcompositions, as well as which socio-demographic subgroups are most accurate across a range of differentvalidation measures.
1. ‘Known’ panels from which we draw targeted sample– Our own panel (iSay)– Other panels
2. ‘Ad hoc’ panelists REALLOCATED from other surveys
3. Non-panel aka River– ALL non-panel sample, including social media,
communities, gaming, reward & loyalty programs, and adnetworks
All goes into the ROUTER
10
Think of the router as a very clever funnel
Your Survey
Respondent
Respondent
Respondent
RespondentRespondentRespondent
OtherSurveys
OtherSurveys
QualifyDoes NotQualifyR
EALLOCATION
11
This approach requires a HIGH volume of survey work
o c. 50 river suppliers from over 200 sites
o Screen c. 130,000 respondents /day
o Yields c. 4 million respondents each month
65% -75% of these are unique respondents
o Over 48,000,000 respondents a year
12
We know it works(and, for many of our clients, ‘knowing it works’ is good enough…)
13
• Measuring incidence of people who received flu shot in January 2013• 40% of Americans stated that they already got a flu shot this season;
CDC figure 41.5%
Flu Shot Study (also: 2012 and 2014 elections)
14
Research Question:What are the differences from anaccuracy standpoint between thesenonprobability sample types?
15
Aggregate data from May 1st 2014 through May 1st
2015
Focus on 37 questions with 102 variables 612cases for which there is a robust market comparator(Census, CPS, other government statistics, largepublished probability-based survey, etc.)Looked at Average Absolute Difference
TM491Y15_TOT - Do you currentlyhave health insurance from a stategovernment health coverageexchange or throughHealthcare.gov? (BASED TO TOTALRESPONDENTS, ASKED OF OUTOF POCKET OR MEDICAID ONLY)