Tuesday, Aug 6: 8:30 AM - 10:20 AM
1668
Topic-Contributed Paper Session
Oregon Convention Center
Room: CC-B112
Ralph Folsom Jr., former chief scientist at RTI International and ASA Fellow, passed away on December 14, 2022, in Raleigh, North Carolina. He joined RTI in 1969 as a statistician and became a chief scientist in 1998. While working at RTI, he earned his Ph.D. in biostatistics from UNC in 1984. He was a past member of the National Academy of Sciences' Panel to Evaluate the Survey of Income and Program Participation, the ASA working group to advise Census Bureau staff on the Survey of Income and Program Participation, the Board of Governors for the Panel Survey of Income Dynamics, and the Committee on National Statistics' Panel on Statistical Methods for Measuring the Group Quarter Population in the American Community Survey.
Ralph's 47-year career at RTI was filled with many innovative advancements in the field of survey data analyses. Ralph's early work on developing Taylor series standard errors for balanced effects also extended to Taylor series estimation of sampling errors for regression coefficients, which became the basis for the analysis of complex survey and other clustered data in SUDAAN® (statistical software to analyze clustered correlated data). Ralph was the first to propose using calibration weighting to adjust for unit nonresponse. He made his proposal at the 1991 ASA annual conference (ASA Proc. Soc. Statist. Sec., 197–202) before the concept of calibration weighting was formalized by Deville and Särndal (JASA, 1992). In a series of papers presented at the 2000 ASA annual conference, Ralph went on the generalize the class of calibration weighting schemes proposed by Deville and Särndal to cover their use for nonresponse and coverage-error adjustment in a more scientifically defensible manner.
In the mid-1990s, Ralph started working on developing small area estimation (SAE) methodologies to enable Substance Abuse and Mental Health Services Administration to produce reliable and cost-effective state and local area level estimates in a timely manner. He developed Survey Weighted Empirical Bayes (SWEB) (Folsom and Judkins, June 1997) SAE methodology for unit-level binary outcomes from complex survey data. Subsequently, Ralph developed the full hierarchical Bayes version of SWEB methodology and called it as the Survey Weighted Hierarchical Bayes (SWHB) methodology (Folsom, Shah, Vaish, 1999). Ralph's innovative work on SAE played a critical role in the expansion of the National Survey on Drug Use and Health (NSDUH) in 1999 from a national design to the currently implemented state stratified design. He collaborated with Babu Shah (developer of SUDAAN®) and developed a highly efficient state-of-the-art SWHB software. Since then, SWHB software is being used to produce annual state estimates and biennial substate estimates for 35+ binary NSDUH outcome variables. State and local health departments use NSDUH data to assess area substance use and mental health problems and to develop appropriate funding strategies and prevention measures.
Applied
Yes
Main Sponsor
Survey Research Methods Section
Co Sponsors
Government Statistics Section
Social Statistics Section
Presentations
Ralph Folsom made many contributions to the analysis of survey data subject to nonresponse, from a design-based perspective. Nonresponse propensities play a central role in unit nonresponse adjustments from both design and model-based perspectives, but are often not clearly defined because of lack of clarity about the variables on which the propensities are conditioned. A definition of response propensity for the purpose of nonresponse adjustments is proposed, where the conditioning is restricted to include the variables measured in the survey as well as design and auxiliary variables measured for respondents and nonrespondents. The proposed definition is justified from both design-based and model-based perspectives. The role of the missing at random assumption is discussed for cross-sectional surveys and longitudinal surveys with attrition.
Inference on overall ranking of a set of entities, such as chess players, subpopulations or hospitals, is an important problem. Estimation of ranks based on point estimates of means does not account for the uncertainty in those estimates. Treating estimated ranks without regard for uncertainty is problematic. We propose a Bayesian solution. It is competitive with recent frequentist methods, and more effective and informative, and is as easy to implement as it is to compute the posterior means and variances of the entity means. Using credible sets, we created novel credible distributions for the rank vector of the entities. We evaluate the Bayesian procedure in terms of accuracy and stability in two applications and a simulation study. Frequentist approaches cannot take account of covariates, but the Bayesian method handles them easily.
Survey researchers have increasingly used web surveys to collect information for population health research and dissemination. These data are often referred to as nonprobability samples due to the lack of a well-defined probability sampling structure. Certain statistical adjustments are therefore needed to make proper inferences using web surveys. One popular adjustment approach is to create pseudo weights that properly ``weight'' the web survey samples back to the target population using a reference survey. Li et al. (2022) showed that it is crucial to include variables that are both related to the outcome of interest and the sample selection into web surveys ("confounders" in epidemiology) in the weighting adjustment. In practice, however, it can be challenging to ensure that all important confounders are included in the data collection and subsequent weighting adjustment. Therefore, a plausible strategy for evaluation is to conduct a sensitivity analysis based on the web survey estimates when certain confounders are excluded. We illustrate this idea using public-use data from the from the Research and Development Survey and National Health Interview Survey.
Co-Author(s)
Katherine Irimata, National Center for Health Statistics
Yan Li, University of Maryland, College Park
Guangyu Zhang, National Center for Health Statistics
Speaker
Yulei He, National Center for Health Statistics
Dr. Ralph Folsom, former chief scientist at RTI International and ASA Fellow, passed away on December 14, 2022. Ralph's 47-year career at RTI was filled with many innovative advancements in the field of survey data analyses. He developed Survey Weighted hierarchical Bayes (SWHB) small are estimation (SAE) methodology for unit-level binary outcomes from complex survey data (Folsom, Shah, Vaish, 1999) to enable Substance Abuse and Mental Health Services Administration to produce reliable and cost-effective state and local area level estimates in a timely manner. Ralph's innovative work on SAE played a critical role in the expansion of the National Survey on Drug Use and Health (NSDUH) in 1999 from a national design to the current state stratified design. Since then, SWHB methodology is being used to produce annual state estimates and biennial substate estimates for over 35 binary NSDUH outcomes. State and local health departments use NSDUH data to assess area substance use and mental health problems and to develop appropriate funding strategies and prevention measures. This presentation is a dedication to Ralph, and it highlights some of the salient feature of the SWHB methodology.