Brian M. Wells, PhD

Brian is the Data Quality and Survey Methodology Manager for the California Health Interview Survey (CHIS) at the UCLA Center for Health Policy Research. His primary responsibility has been to oversee and evaluate the sample and data collection redesign of the CHIS implemented in 2019. He is a doctoral graduate of the Michigan Program in Survey Methodology where his research focused on the recurrent collection of biomeasures in longitudinal surveys.

CHIS Army Knife? Examining Multiple Methods to Sample Smaller Racial/Ethnic Groups in California

Population-based surveys are often limited in their ability to report on smaller or less common racial/ethnic groups, ultimately deferring to broad racial/ethnic categories, which still may struggle with small sample sizes (e.g., American Indian and Alaska Natives, Native Hawaiian and Pacific Islanders). In health survey research (made especially relevant during the COVID-19 pandemic), the lack of disaggregation in data can limit our ability to identify and make meaningful changes in relation to within-group differences and disparities. One of the largest hurdles for these types of surveys is obtaining sufficient sample sizes for these smaller groups whom may be hard-to-sample or hard-to-find, and may require supplemental methods to obtain the desired sample.

This presentation will explore a variety of techniques used to identify and interview these small racial/ethnic groups in one of the most diverse states in the US, California, through the efforts of the California Health Interview Survey (CHIS) over its 20-year history. First, we look at the use of surname list frames to oversample Korean, Vietnamese, and Japanese Californians, and explore what happens when screening is not paired with that frame. Second, we examine to two different approaches at obtaining more American Indian and Alaska Native sample: the Indian Health Service (IHS) frame and geographic stratification. Third, we discuss the use of web-based respondent-driven sampling (RDS) to survey the population of Native Hawaiians and Pacific Islanders, and what lessons we learned about understanding your population of interest and utilizing preexisting networks. Finally, we examine the use of machine learning methods that combine auxiliary frame data and self-reported survey data to predict household attributes to use for disproportionate stratification. We explore how these models perform for different racial/ethnic groups.