International Conference on Methods for Surveying and Enumerating Hard-to-Reach Populations H2R New Orleans, Louisiana,

1. International Conference on Methods forSurveying and Enumerating Hard-to-Reach Populations(H2R)New Orleans, Louisiana, USAOctober 31-November 3, 2012

2. Inspiration Nancy Bates at U.S. Census Bureau, co-chair Sparse survey methods literature on hard-to-reach (but growing--see special issue of Methodological Innovations Online) Only 3 conferences/workshops in past 20 years (but most recently 2010 in Paris) Help for survey practitioners

3. Sponsors American Association for Public Opinion Research American Statistical Association and its sections on government statistics, survey research methods, and social statistics World Association for Public Opinion Research Institut fur Arbeitsmarkt- und Berufsforschung

4. 2012 Schedule Info about short courses, hotel up on website end of January/early February Preliminary program April/May Registration opens August Short courses October 31 Conference sessions November 1-3

5. Publications in 2013 Invited papers: Monograph (�Interviewing the Hard-to-Reach�), to be published by Cambridge in paperback, hardcover, on-line Contributed papers: Journal of Official Statistics International Journal of Public Opinion Research

6. Book Sections Intro (Tourangeau, ed) Difficult Settings (Edwards, ed) Special Populations (Johnson, ed) Sampling Strategies (Wolter, ed) Data Collection Strategies (Edwards & Bates, eds.)

7. Section on Sampling Strategies Sampling Elusive and Mobile Populations (Kalton) Recent Developments of Sampling Hard-to-Reach Populations: An Assessment (Lee, et al) Indirect Sampling for Hard-to-Survey Populations (Levallee) Sampling Subpopulations using Partial Lists, Imperfect Census Data, and Rough Screening Questions (Clark) Network Methods for Sampling Hard-to-Survey Populations (McCormick and Zheng) Respondent-Driven Sampling and Related Techniques (Thompson)

8. Graham Visitors and travelers: best sampled and surveyed on location Standard probability methods have limitations Alternatives: location sampling, snowball sampling, respondent-driven sampling Location sampling (aka venue-based or time-space sampling) covers only those who visit specified location in survey time period Weights must compensate for unequal selection probabilities of visits and for the number of visits a sampled person makes during period (not straightforward) Variability in standard weights can cause serious loss of precision, need procedure to reduce variability Inferences from snowball and respondent-driven samples based on many assumptions that need to be critically assessed.

9. Lee Demands for data on hard-to-reach led to new sampling methods (respondent-driven sampling, community-based sampling, time-and-space sampling) not found in traditional sampling textbooks; mostly based on non-probability methods. Introduce each method, its theoretical background, critical assumptions required for inferences Examine characteristics of populations sampled with these methods, specific applications in published studies Illustrate when the assumptions are likely and unlikely to be met in data collection operations Benefits and limitations of these methods with total survey error perspective.

10. Levallee If no sampling frame available for the desired target population, may be possible to use a sampling frame that is indirectly related Populations A and B Produce an estimate for B by selecting a sample from A and using the existing links between the two populations. Estimates difficult to achieve if links between A and B not one-to-one. A solution: generalized weight share method (GWSM) Network sampling, adaptive cluster sampling, snowball sampling, multiple frames fit into context of indirect sampling Theory and developments surrounding indirect sampling, GWSM provide a unified mathematical framework for these approaches.

11. Clark Ethnic or indigenous subpopulations often lack sampling frame Multi-stage samples target areas known to have higher proportions of the subpopulations Challenges how to target these areas without creating excessive variation in estimation weights how to allow for inaccuracy or outdatedness of the census data how to use screening questions to efficiently identify subpopulations how to combine an area frame with a partial list of the subpopulation New methodology for efficiently using all these elements, based on separate training and validation samples Sampling of New Zealand's Maori, Pacific and Asian populations used as motivating example.

12. McCormick and Zheng Statistical models leverage social network data to estimate characteristics of H2R subpopulations using Aggregated Relational Data (ARD) -- questions like ``How many X's do you know?' Require no special sampling strategy Are easily incorporated into standard surveys Don�t require respondents to reveal their own group membership Bayesian hierarchical model for estimating demographic characteristics of H2R groups, or latent demographic profiles, using ARD Two estimation techniques Markov-chain Monte Carlo algorithm for existing data or cases where the full posterior distribution is of interest When new data can be collected, guidelines and a simple estimate motivated by a missing data approach

13. Section on DC Strategies Paid Media � 2010 Census (Datta, et al) EU-SILC (Nicaise, Schockaert) Mode Choice, Response Rates (Haan, Ongena) Recruiting Hispanics with R-Driven Sampling (Schonlau et al) Engaging Aboriginals (Connors) Linguistic/Cultural Minorities (Pan, Lubkemann) Mobilizing to Participate Fully (Vargas, Olson) Finding and Keeping Them Engaged (Becker et al)

14. Datta et al 2010 Census Integrated Communications Program Evaluation (CICPE) (NORC): Best Practices Eliciting confirmed awareness of advertisements Capturing receptivity to advertising messages Tracking changes in attitudes, beliefs addressed in advertisements Use of advertisement metrics to estimate exposure to paid media campaign. Actual Census participation data Data from experimental variation in paid media exposure in a sub-set of sites Key results: overall effectiveness of the campaign in increasing mail return rates Differential effectiveness across people�s initial dispositions toward the Census? Inducing changes in attitudes or beliefs targeted in advertising content? General receptivity to advertisements?

15. Nicaise and Schockaert Belgian EU-SILC data (Eurobarometer) Inventory of excluded groups (collective households, illegal residents, itinerant groups, homeless) Analysis of selective non-response and attrition � and ways to minimize or correct Complementary survey of homeless and undocumented with adapted questionnaires. Comparison of extent and depth of poverty between these two groups and the rest of the population.

16. Haan and Ongena Experiment within the European Social Survey Half, randomly approached face-to-face or by phone, could choose to be interviewed face-to-face, by phone or online. With mode choice, expect higher response rates Possible to develop a tailored approach to lower nonresponse rates in the future

17. Schonlau et al In RDS, critical to obtain estimates of R�s network size and to know who recruited whom ALP: Internet panel of 3000 respondents Currently adding a Hispanic sub-panel to the American Life panel using RDS Recruitment effort evaluated with respect to the 5 RDS assumptions, geographic spread, and socio-demographic composition.

18. Pan and Lubkemann 546 observations of Census 2010 nonresponse followup interviews in English and 7 other languages, plus debriefing interviews with respondents and enumerators Discussion of findings: Linguistic factors that affected communication Socio-cultural factors that shaped social interaction during the survey encounter Conclusions highlight several fundamental methodological implications, outline key design issues for such populations.

19. Vargas and Olsen Community engagement campaigns Leverage partnerships from community and grass roots organizations, mobilize constituents Two perspectives � nationally known community leader, statistical agency�s national field manager Examples from the United States 2010 Census pro-bono engagement of more than 257,000 organizations hiring nearly 4,000 outreach staff with skills in 146 languages coordination with messaging campaign via Spanish-lang. media Lessons learned

20. Becker et al How to find the hard to reach, how to stay in touch with them? Often no fixed address, �cell only� with precarious service/continuity of numbers, don�t appear in traditionally used search databases Methods for finding and tracking based on travel patterns, service locations, personal preferences and habits of connecting with others Databases useful in tracking these groups and how to access them Methods involve higher costs, IRBs, design that allows access while protecting privacy, safety Methods for staffing, outreach and rapport building, appropriate incentives, protection of privacy, legal issues, maintaining trusting relationships over time

21. City of New Orleans Spanish, French, Creole, Cajun, American Southern flavors, now joined by Latin Americans One of America�s top restaurant cities Birthplace of jazz, vibrant music scene today Rebuilding after Hurricane Katrina 2005 Hotel: Marriott at Convention Center, in the Warehouse District, walking distance to French Quarter

22. www.amstat.org/meetings/h2r/2012 H2R2012@amstat.org Thanks! Brad Edwards bradedwards@westat.com