Online panels for health surveys Charles DiSogra Academy Health, June 14, 2011 © 2011 Knowledge Networks, Inc. Outline Panels and types of online panels Frames and accuracy Innovations in address-based sample (ABS) recruitment Calibrating probability and non-probability samples Example of a health tracking study 1 Panels Panels are pre-recruited persons that have agreed to participate in multiple surveys over time Examples: The U.S. Current Population Survey Longitudinal surveys with repeated measures A panel can be a “true” sample of the population … or not. Depends on: how constructed who is recruited 2 Online panels The similarity (mode): Web administered questionnaires The difference (frame): How panels recruit members is not the same Key Questions: What population does this panel represent? Are survey findings generalizable? 3 Types of online panels Non-probability opt-in panels Probability-based panels 4 Non-probability opt-In Web panels Large, volunteer membership – in the millions People on the Web join through ads, aggregator Websites, pop-up invitations, e-mail marketing Used extensively by market researchers Low cost Rapid turnaround Can target defined audiences with member profile data Recruitment, sampling, weighting methods are not transparent Survey completion rates are low (2-9%) Not generalizable for prevalence estimates – but people do this all the time! Industry organizations, e.g., Advertising Research Foundation, set voluntary standards for membership management (i.e., minimize professional respondents, overlap among panels) 5 Probability-based Web Panels Recruited with probability samples (no non-sampled volunteers) Random-digit dial (formerly) RDD dual frame samples add cell phone augment Address-based sampling (now more common) Members have known selection probability (adj. in base weights) Used by government, academic and non-profit researchers and private companies where more rigor is desired Completion rates are high (65-70%) Results are generalizable, can calculate prevalence estimates with applicable confidence intervals American Association of Public Opinion Research recognizes probability based samples, ergo panels recruited as such, as a valid and reliable survey method 6 50,000 members representing America Probability-based recruitment, representative of U.S. adult population Includes: Households with no Internet access at time of recruitment – 33% of US adults have no Internet access – KN provides laptop computer, free monthly ISP Cell phone only households (28% of US) through ABS mail recruitment Spanish-language households Extensive profile data maintained on member demographics, attitudes, opinions, behaviors, health conditions, media usage, etc. • Samples from the panel are assigned to studies using e-mail invitations and a link to the online survey questionnaire 77 Population Coverage by Mode/Frame KN-ABS Frame probability based and covers non-internet HHs U.S. Adults KnowledgePanel Frame Landline Telephone Frame Unknowable clustering of non-probability opt-in Web panel samples Internet Survey Frame 74%2 72%1 97% 1 8 2 National Center for Health Statistics. Internet access from any location. October 2009 CPS. 100% non-probability online panels Accuracy of probability-based KnowledgePanel and RDD samples “Both were significantly more accurate than any of the non-probability opt-in sample Internet surveys.” Average absolute errors for probability and non-probability sample surveys across 13 secondary demographics and non-demographics, with post stratification. Non-Probability Samples Probability Samples Source: Yeager, Krosnick, et. al., Comparing the Accuracy of RDD Telephone Surveys and Internet Surveys Conducted with Probability and Non-Probability Samples. August, 2009. 99 The American Association of Public Opinion Research (AAPOR) Task Force On Online Panels Avoid non-probability online panels when objective is to accurately estimate population values Claims of “representativeness” should be avoided when using these sample sources Non-probability online panels are generally less accurate when compared to benchmark data “AAPOR Report on Online Panels,” prepared by the AAPOR Online Task Force Report, March 2010. Available at www.aapor.org. 10 10 Using an ABS Sample Frame to Recruit a Probability-Based Online Panel The Knowledge Networks Experience 11 50,000 members representing America Probability-based recruitment, representative of U.S. adult population Includes: Households with no Internet access at time of recruitment – 33% of US adults have no Internet access – KN provides laptop computer, free monthly ISP Cell phone only households (28% of US) through ABS mail recruitment Spanish-language households Extensive profile data maintained on member demographics, attitudes, opinions, behaviors, media usage, etc. • Samples from the panel are assigned to studies using e-mail invitations and a link to the online survey questionnaire 12 12 Mail Recruit Address-based Sample (ABS) U.S. Postal Service Computerized Delivery Sequence File (CDSF) ~97% coverage of physical addresses Frequently updated including status of addresses, such as, seasonal homes, vacant houses, etc. Can be matched to available telephone numbers Can be geo-coded Can attach demographic data (actual and modeled) from a variety of sources for purposes of Non-response analyses Targeted demographic mailings 13 13 Mail Materials and Schedule Day 7 Reminder PC Day 28 NR Letter Current Resident / Residente Actual 123 Your Street The City, State 99999 Initial Mailing 14 Three Response Modes for Joining Respond by: 1. Mail 2. Online 3. Telephone Toll-free number Non-Responders: Outbound Telephone Recruitment 15 15 ABS Recruitment Breaking new ground in 2011 Expanding KnowledgePanel with innovative recruitment samples Objective: Recruit more Hispanics and young adults (18-24) Predictive ancillary information is used to define targeted strata 16 Calibrating probability and non-probability samples When the finite size of the probability-based panel is unable to deliver enough sample for a given study Supplement with quota-controlled “opt-in” sample Include ancillary information among weighting variables to calibrate the opt-in cases to KN panel cases “Early adopter” behavior consistently differentiates to two sample sources Goal: minimize bias introduced by opt-in cases 17 Example of a calibrated sample Figure 1. Blended using standard variables 18 Figure 2. Blended using standard variables plus a calibration variable (early adopter Q) Health tracking study example California Tobacco Control Program 19 California Tracking Survey – Design Specific media evaluation survey General population, approx. 2,000 adults (ages 18-55) Smokers and non-smokers Supplemented with California smokers from opt-in sample Fielded immediately after a flight of media Every December and June/July Version 1.0 RDD telephone data collection July 2001 to June 2004 (5 waves) Version 2.0 KnowledgePanel online data collection December 2005 to ongoing (11 waves to date of these results) Longitudinal + fresh cross-sectional samples (combined for x-sec) California Tracking Survey – Measures Aided ad awareness Executional and message awareness Talk with others New information Attitudes and knowledge Second hand smoke (SHS), anti-tobacco industry, regulation Health effects Self-reported smoking and cessation behavior Use of tracking data Cross-sectional analysis General ad and campaign awareness Ad message awareness and/or impart new information Take action step Tracking of attitudes over time Examine the relationship between ad recall and ad placement Longitudinal analysis Is ad awareness related to attitudinal changes? Is ad awareness related to behavioral changes? Percent of Californians that agree that multi-unit housing or apartment buildings should require that at least half of their rental units be smoke-free, 2006-2010 (KnowledgePanel data) 90 80 70 Percent 60 50 Ads placed that specifically address secondhand smoke in multi‐unit housing 40 30 20 10 Dec‐10 Jun‐10 Dec‐09 Jun‐09 Dec‐08 Jun‐08 Dec‐07 Jun‐07 Dec‐06 Jun‐06 0 Percent of Californians that have seen or heard any anti-tobacco messages and media spending, 2000-2009 RDD KnowledgePanel Effectiveness of DRTV DRTV = Direct Response TV advertisement; TRP = Target Rating Points Thank you! cdisogra@knowledgenetworks.com © 2011 Knowledge Networks, Inc.