Release Date: 7/10/2013
Two teams of Abt SRBI researchers will be presenting papers at the upcoming conference of the Federal Committee on Statistical Methodology (FCSM), November 4-6 in Washington, D.C.
Both papers focus on innovative methods to improve sampling efficiency when targeting populations that require screening, including rare populations:
The major goals of the Federal Committee on Statistical Methodology (FCSM) are:
“Adaptive Design Features for Using Address-based Sampling in a National CATI Survey of Households with Children” tests the use of address-based sample (ABS) incorporating adaptive design techniques, such as ancillary demographic data, to reduce screening for a nationally representative sample of households with children and under age 18.
“Adaptive Adjustment of the Multicriteria Optimal Allocation of a Hard to Reach Population Survey” uses data from previous sample surveys, geocoded establishment information (synagogues and religious education organizations), and sample vendor ethnic name prevalence estimates to develop the initial county-level small area estimates of Jewish-by-religion incidence. These estimates are updated as the survey data are being collected, and ultimately used to stratify the sample for the remaining field period.
Abt SRBI is also a conference sponsor. Both abstracts are listed below.
Communicating and disseminating information on statistical practice among all Federal statistical agencies.
Recommending the introduction of new methodologies in Federal statistical programs to improve data quality.
Providing a mechanism for statisticians in different Federal agencies to meet and exchange ideas.
Adaptive Design Features for Using Address-based Sampling in a National CATI Survey of Households with Children
Charles DiSogra (Abt SRBI), David Finkelhor (University of New Hampshire), Heather Hammer (Abt SRBI), Stanislav Kolenikov (Abt SRBI), Heather Turner (University of New Hampshire)
The primary purpose of Wave III of The National Survey of Children’s Exposure to Violence (NatSCEV III) is to document changes in the incidence and prevalence of children’s exposure to a broad array of violence, crime and abuse experiences. NatSCEV III uses an address-based sample (ABS) design to construct a nationally representative sample of households with children and adolescents age 0-17. While 34% of US households have children under age 18, CATI surveys of these households generally require costly telephone screening. The availability of ancillary demographic information that can be matched by a sample vendor to some proportion of the addresses in an ABS sample suggests the feasibility of a more efficient approach that uses an optimal allocation among matched and unmatched strata based on the expected proportion of households with children. Although presence of children is one of several indicators that can be used to stratify matched addresses in an ABS design, the accuracy of the ancillary data is imperfect and variable across indicators. NatSCEV III incorporates an adaptive design where sample vendor-provided ancillary demographic data for the matched addresses and demographic data collected from the first 1,000 completed interviews will be used to assess the accuracy of the vendor-appended ancillary demographic data and to build a logistic regression model that predicts the prevalence of households with children in each of the matched and unmatched strata. These results will then be used to develop an optimal stratified allocation for the main survey effort. Findings and conclusions about the viability of this approach for NatSCEV III and other national surveys of households with children will be presented.
Adaptive Adjustment of the Multicriteria Optimal Allocation of a Hard to Reach Population Survey
Benjamin Phillips (Abt SRBI), Stanislav Kolenikov (Abt SRBI)
The effectiveness of a survey of a rare population hinges on the information available about the prevalence of this rare population within the general population and areas where this rare population may be concentrated. Matters are further complicated when overall population and subdomain estimates within the population of interest are required, as these estimation objectives are frequently in conflict. We describe our work on a survey of American Jews, in which the estimates of the total population, as well as its demographic, religious and social characteristics are all of interest; moreover, the survey objectives encompass both estimates of the population as a whole as well as of domains of interest (Orthodox and Russian Jews). To approach the sample design, data from previous sample surveys, geocoded establishment information (synagogues and religious education organizations), and sample vendor ethnic name prevalence estimates were all used in deriving county-level small area estimates of Jewish-by-religion incidence. These estimates were used to stratify the general population into several strata with incidence varying from 0.25% to above 10%. Nonlinear programming running under readily available software (Excel solver) was used to obtain the optimal allocation that included multiple accuracy criteria as well as the required sample sizes. The availability of the data from the field allowed reviewing the accuracy of the small area estimates in retrospect. More importantly, as the field period progressed, the observed incidence across strata and landline/cell phone frames was used to update the sample allocation in the adaptive design pursuing the most efficient use of the interviewer time and samples acquired from the provider. We present the trajectories of the observed incidences throughout the field period, discuss our adaptive design decisions, and quantify the economic impact of the adaptive design on the resulting total survey cost.