Online Program

Using ancillary information in national ABS samples to recruit hard-to-reach young adults and Hispanics

Charles DiSogra, GfK - Knowledge Networks 
*J. Michael Dennis, GfK - Knowledge Networks 
Mansour Fahimi, Marketing Systems Group 

Keywords: ABS, young adults, Hispanics, Web panel

GfK-Knowledge Networks uses national address-based samples to recruit members for its probability-based online panel. In addition to the need of offsetting natural attrition, it was decided for 2011 to expand the panel among two hard to reach segments: young adults (ages 18-24) and Hispanics. In 2010, Hispanics were fairly successfully oversampled by targeting predominantly (>30% density) Hispanic census blocks. Using the 2010 experience, eight general population (GP) samples for 2011 (approx. 23,000-29,000 each) were designed by taking advantage of ancillary information attached to the sampling frame of addresses from the USPS computerized delivery sequence file. This frame could then be partitioned into the following four strata for a planned oversampling of both young adults and Hispanics: Hispanics 18-24; All Else 18-24; Hispanics 25+; All Else 25+. Additionally, nine young-adult (YA) national samples were selected (approx. 11,000-16,000 each) where a combination of ancillary information was used to maximize hit-rates for households with at least one person ages 18-24. For this purpose, a nonparametric variant of Chi-squared automatic interaction detector (CHAID) was used to identify which combinations of available ancillary variables could most effectively target households of interest. In the GP samples, the non-Hispanic young adult stratum yielded about 11% recruited households of which 30% were 18-24 year olds (vs. 13% in the CPS). In contrast, the 25+ Hispanic stratum yielded 8% recruited households of which 75% were Hispanic adults (vs. 14% CPS) and the YA samples yielded between 5% and 6% young adults (vs. a predicted 3.8% yield) of which all were 18-24 years old. The authors will show that stratification based on ancillary data can deliver a more efficient yield when targeting these segments in comparison to the earlier national ABS samples that did not use any ancillary information for stratification purposes.