Your SlideShare is downloading. ×
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples

681

Published on

At the Advertising Research Foundation’s (ARF) 2011 annual re:think convention, a key issues forum presentation was held entitled Blend, Balance, and Stabilize Respondent Sources: Building …

At the Advertising Research Foundation’s (ARF) 2011 annual re:think convention, a key issues forum presentation was held entitled Blend, Balance, and Stabilize Respondent Sources: Building Representative Samples. Through the use of empirical examples and side by side tests, methods and solutions for building representative samples are presented. The presentation was given by Mitchell Eggers Ph.D, Chief Scientist at GMI inc. & Eli Drake Statistician at GMI inc.

Published in: Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
681
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Blend, Balance, and StabilizeRespondent SourcesBuilding Representative SamplesMitchell Eggers Ph.D. Eli DrakeChief Scientist StatisticianGMI, Inc. GMI, Inc.
  • 2. Presentation• Section 1: Industry Opportunity – Business Drivers for Multiple Sources – Methodological Drivers for Multiple Sources• Section 2: Basic Method: Matching a Standard• Section 3: Three Empirical Examples• Section 4: Questions
  • 3. Section 1:Industry OpportunityMarch 2011
  • 4. OpportunityBusinesses Need Capacity• Use multiple sources for single studies• Connect to new sources of respondents• Evolve with the connected world as it transforms• Continually embrace new sources of respondentsAnalysts Need Representative Samples• Know how sources are skewed• Control for each source’s unique skews• Stabilize a wide range control variables as respondent sources evolve or completely change• Monitor to ensure sample stability• Provide assurance to researchers their data are good and their conclusions are sound
  • 5. Areas of ImprovementQuota Cell Management• Practical limits of quota cells• Mathematical limits of quota cellsMeasurement of Skews• No Sample frame for online like RDD or address lists• Limited knowledge of bias• Limited techniques for control of those skews• Different skews among suppliersMixing Algorithms• Empirically driven• Platform enabledStabilizing Respondent Sources• Need to stabilize a wide range control variables as respondent sources evolve and wobble
  • 6. Section 2: The Basic Method:Matching an Open StandardMarch 2011
  • 7. Step 1 of Solution• Select an Open Standard – Nationally-representative measures of: • Demographics, • Psychographics, and • Behavior• General Social Survey: N.O.R.C. and the University of Chicago
  • 8. N.O.R.C.s’ General Social Survey“The GSS is the largest project funded by the Sociology Program of the National Science Foundation. Except for the U.S. Census, the GSS is the most frequently analyzed source of information in the social sciences. We know of over 14,000 research uses such as articles in academic journals, books, and Ph.D. dissertations based on the GSS and about 250,000 students annually who use it in their classes.” N.O.R.C. Home Page
  • 9. Step 2 of Solution• Create a set of respondents from various sources (panels, river, social networking) that match the nationally representative standard on all 60 dimensions• Test to see if respondents produce accurate results for market researchers when they ask a wide variety of MR questions
  • 10. The MethodSetup (Pre-Study) Run Time (During Study)Calibrate sourcesMeasure skewsEmpirically justified mixing(i.e., minimize and counter balance )Define the optimal Age-Gender mixSet up Stabilizer and Balancer on platform Order Sample – Launch Ask 3.5 Questions per respondent Cumulate stability statistic Balance if needed (screen out) Send respondents to study Close study and confirm stability
  • 11. Section 3:Three Empirical ExamplesMarch 2011
  • 12. Calibration ResultsDrake-Eggers Distance from Standard Drake-Eggers Distance from Standard Source A Source B HIV Test HH Income HH Income Age Age Dwelling Type US Born Highest Degree Pars Born US HH Size Race Grnd Pars Born US Highest Degree Work Stat Soc Class Race Grnd Pars Born US Ever Save Soul Bio Fam 16 Yrs HIV Test Migrate Post 16 Yrs Region at 16 Yrs Dwelling Type Pars Born US Marital Stat Bio Fam 16 Yrs HH Size Poly Views Work Stat Marital Stat Hood Diversity Soc Class Ever Save Soul Gender Region at 16 Yrs Hispanic Tot Child Migrate Post 16 Yrs Poly Views US Born Gender Tot Child Region Region Hispanic Hood Diversity 0 5 10 15 0 5 10 15
  • 13. Age, Gender, Region, Race,Ethnicity ControlsDrake-Eggers Distance from Standard Drake-Eggers Distance from Standard Source A vs. Regular Controls Source B vs. Regular Controls HIV Test HH Income HH Income Age Age Dwelling Type US Born Highest Degree Pars Born US HH Size Race Grnd Pars Born US Highest Degree Work Stat Soc Class Race Grnd Pars Born US Ever Save Soul Bio Fam 16 Yrs HIV Test Migrate Post 16 Yrs Region at 16 Yrs Dwelling Type Pars Born US Marital Stat Bio Fam 16 Yrs HH Size Poly Views Work Stat Marital Stat Hood Diversity Soc Class Ever Save Soul Gender Region at 16 Yrs Hispanic Tot Child Migrate Post 16 Yrs Poly Views US Born Gender Tot Child Region Source A Raw -- 6.1 Region Source B Raw -- 6.6 Hispanic Regular Controls -- 3.6 Hood Diversity Regular Controls -- 5.5 0 5 10 15 0 5 10 15
  • 14. Algorithmically Mixed ResultsDrake-Eggers Distance from Standard Drake-Eggers Distance from Standard Source A vs. Mixture Source B vs. Mixture HIV Test HH Income HH Income Age Age Dwelling Type US Born Highest Degree Pars Born US HH Size Race Grnd Pars Born US Highest Degree Work Stat Soc Class Race Grnd Pars Born US Ever Save Soul Bio Fam 16 Yrs HIV Test Migrate Post 16 Yrs Region at 16 Yrs Dwelling Type Pars Born US Marital Stat Bio Fam 16 Yrs HH Size Poly Views Work Stat Marital Stat Hood Diversity Soc Class Ever Save Soul Gender Region at 16 Yrs Hispanic Tot Child Migrate Post 16 Yrs Poly Views US Born Gender Tot Child Region Source A Raw -- 6.1 Region Source B Raw -- 6.6 Hispanic Mixed -- 3.2 Hood Diversity Mixed -- 3.2 0 5 10 15 0 5 10 15
  • 15. Mixed and Balanced ResultsDrake-Eggers Distance from Standard Drake-Eggers Distance from Standard Source A vs. Stratified Mixture Source B vs. Stratified Mixture HIV Test HH Income HH Income Age Age Dwelling Type US Born Highest Degree Pars Born US HH Size Race Grnd Pars Born US Highest Degree Work Stat Soc Class Race Grnd Pars Born US Ever Save Soul Bio Fam 16 Yrs HIV Test Migrate Post 16 Yrs Region at 16 Yrs Dwelling Type Pars Born US Marital Stat Bio Fam 16 Yrs HH Size Poly Views Work Stat Marital Stat Hood Diversity Soc Class Ever Save Soul Gender Region at 16 Yrs Hispanic Tot Child Migrate Post 16 Yrs Poly Views US Born Gender Tot Child Region Source A Raw -- 6.1 Region Source B Raw -- 6.6 Hispanic Mixed And Stratified -- 2.4 Hood Diversity Mixed And Stratified -- 2.4 0 5 10 15 0 5 10 15
  • 16. Summary Skew Reduction• No controls: 11.0• Standard controls: 7.5• Scientifically optimal mix: 4.0• Mixed and balanced: 3.0• Mixed, balanced, and weighted: 0.6• Total reduction in bias 95%
  • 17. Side-by-Side Test 1:Two SourcesMarch 2011
  • 18. Side-by-Side 1• Validation with general consumer study• Testing to see if GMI’s science and methods produce similar results when we use two alternative and skewed sources, but fix them so the respondents match the GMI panel• Does the study produce similar results?
  • 19. AwarenessAwareness of Consumer Items Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 10 Item 11 Item 12 Item 13 Item 14 Item 15 Item 16 Item 17 Item 18 Item 19 Item 20 Item 21 Item 22 Item 23 Item 24 Item 25 Item 26 Item 27 Item 28 Item 29 Item 30 Item 31 Item 32 Item 33 Item 34 Item 35 Item 36 Item 37 Item 38 Item 39 Item 40 Item 41 Item 42 Item 43 Item 44 Item 45 Item 46 GTM GTM None of the Above Pinnacle CSS 20% 40% 60% 80% % Aware
  • 20. OwnershipHome Appliance Ownership Appliance 1 Appliance 2 Appliance 3 Appliance 4 Appliance 5 Appliance 6 Appliance 7 Appliance 8 Appliance 9 Appliance 10 Appliance 11 Appliance 12 Appliance 13 Appliance 14 Appliance 15 Appliance 16 Appliance 17 Appliance 18 Appliance 19 GTM GTM None of the Above Pinnacle CSS 20% 40% 60% 80% % Own
  • 21. PurchasesRecent Past Purchases Zero One to Tw o Three to Five More than Five GTM GT Pinnacle CS 0% 5% 10% 15% 20% 25% 30% 35%
  • 22. Retail OutletPurchase by Retail Outlet Type Type 1 Type 2 Type 3 Type 4 GTM GTM Pinnacle CSS 0% 20% 40% 60% 80% 100%
  • 23. FrequencyPurchase Frequency by Retail Outlet Type Zero One to Tw o Three to Five More than Five Zero One to Tw o Three to Five More than Five Zero One to Tw o Three to Five More than Five Zero One to Tw o Three to Five GTM More than Five CSS 0% 20% 40% 60% 80%
  • 24. IntentIntent to Purchase by Retail Outlet Type Definitely w ill Probably w ill May or may not Probably w ill not Definitely w ill not Definitely w ill Probably w ill May or may not Probably w ill not Definitely w ill not Definitely w ill Probably w ill May or may not Probably w ill not Definitely w ill not Definitely w ill Probably w ill May or may not Probably w ill not GTM Definitely w ill not CSS 0% 5% 10% 15% 20% 25% 30% 35%
  • 25. ControlsDemographics 18 - 24 25 - 34 35 - 44 45 - 54 55 + Female Male Divorced/Separated/Widow ed Living w ith partner Married Single, never married African American Asian or Pacific Islander Hispanic Other White or Caucasian One person Tw o persons Three persons GTM Four or more persons CSS 0% 20% 40% 60% 80% 100%
  • 26. ResultsPercentages in GTM vs. Pinnacle Blend 100% 80% % Certified Source 60% 40% 20% 0% 0% 20% 40% 60% 80% 100% % GTM
  • 27. Side-by-Side Test 2:Three SourcesMarch 2011
  • 28. 100% 80%% Certified Source 60% 40% 20% 0% 0% 20% 40% 60% 80% 100% % GTM
  • 29. Additional Tests
  • 30. Internal Structure ResultsRelationships betweenGTM variablescompared to therelationships betweenPinnacle datasetvariables. Less than 5%of the thousands ofrelationships testedwere different, whichreveals a dataset thatnot only matchesmeans andvariances, but alsomatches in its internalcovariance structure.
  • 31. Side-by-Side Test 3:BenchmarkMarch 2011
  • 32. Out of Bounds Test• Can we match a published benchmark?• We selected the Consumer Sentiment Index – University of Michigan, Reuters
  • 33. The Results: Side-by-Side IIIConsumer Sentiment Index 77 78 76 75 GMI 74 U of M 73 72 70 71Index Value 68 69 66 64 67 62 65 60 Date
  • 34. GMI Pinnacle DeliversCapacity, Stability, and ConfidenceOur science and methods work• We can use multiple sources• We can match a standard• We include psychographic and behavioral measures with demography• We can match other panels or our panel• We can stabilize across multiple waves
  • 35. Blend, Balance, and StabilizeRespondent SourcesBuilding Representative SamplesThank YouQuestions?Mitchell Eggers Ph.D.Chief ScientistGMI, Inc.

×