How Innovations in Modeling Improve Local Area Estimates

The demand for small area estimation (SAE) has grown over the past decades spurred by policymakers’ quest for finer details about their local areas. Westat statisticians have recently developed some innovations to SAE, which are highlighted by Westat co-authors Tom Krenzke and Leyla Mohadjer, Ph.D., in their issue brief, Innovative Small Area Modeling with Multivariate Estimation (PDF).

“There is an increased demand for reliable small area estimates, and this has resulted in significant advancements in the SAE methodology and approaches in recent years,” says Leyla Mohadjer, Ph.D., a Vice President in Westat’s Statistics and Evaluation Sciences Unit. The latest enhancements to SAE approaches and methodology include improvements in the precision of input estimates, enhanced methods to select predictor variables, expansions to multivariate models, and advanced approaches to evaluate SAE models through rigorous diagnostic checks.”

Tom Krenzke, also a Vice President in Westat’s Statistics and Evaluation Sciences Unit, points to Westat’s contribution to the SAE methodology through a sophisticated modeling approach to produce model-based estimates for states and counties for the Program for the International Assessment of Adult Competencies (PIAAC). PIAAC, he explains, is an international survey that examines and assesses literacy, numeracy, and digital problem-solving skills of adults ages 16-65 across participating countries. The model-based estimates support policymakers in their planning and allocating resources and targeting education interventions for specified populations. Westat also generated PIAAC model-based estimates for New Zealand, Italy, Germany, Sweden, and Slovakia.

“The results from our approach, designed for PIAAC for the National Center for Education Statistics, enables researchers for the first time to compare and analyze state- and county-level estimates of adult literacy and numeracy proficiency with the use of an interactive Skills Map that we developed,” notes Mr. Krenzke.

The issue brief reveals the steps taken to develop the Westat model: the hierarchical Bayes linear 3-fold model involving 3 nested levels of random effects, a set of predictor variables and a thorough evaluation of the model. “Our model uses a multilevel approach to take into account the variations across the counties beyond what is explained by the predictor variables in the model,” says Dr. Mohadjer.

Although requests for SAE are primarily made by government policymakers, other decisionmakers can gain important information with its use, says Mr. Krenzke: “For example, a business considering relocating to another state will want to know about the workforce skills in literacy and numeracy in various counties.”

Westat is supporting the Centers for Disease Control and Prevention to develop a model using small area estimation methodology toward producing model-based estimates of the percentage of the population with no primary doctor within states and counties. “With the growing demand for state- and county-level data and Westat’s expertise in small area estimation, we are well positioned to support clients in meeting a range of research needs,” says Mr. Krenzke.