Prevalence of Sexual Orientation and Gender Identity Behaviors: Model-Based State and National Estimation Derived from the Behavioral Risk Factor Surveillance System (BRFSS)

• Activity Number: 197 - Health Data Analytics
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Government Statistics Section
• Author(s): Yangyang Deng* and Ronaldo Iachan and Adam Lee and Marjorie Biel
• Companies: ICF and ICF and ICF and ICF Macro, Inc.
• Keywords: BRFSS; Sexual Orientation; Gender Identity; Multilevel Model; Machine Learning; Imputation

Abstract

The paper uses machine learning to extend model-based estimation at the state/national level when data are available only from a subset of states. We used the Sexual Orientation and Gender Identity (SOGI) optional module questions from the Behavioral Risk Factor Surveillance System (BRFSS) from 2014 to 2018, which were included in the surveys used by subsets of states. Goal was to derive state and national estimates separately for LGBT population. We also used machine learning to identify good predictors candidates from BRFSS core variables for all dependent variables. Multilevel model and random forest are used to provide estimates for LBGT prevalence at state and national level. The models include state laws and policies which reflect the states’ level of degrees on hospitality to LGBT subpopulations. Cross-validation by comparing models and direct estimates is used for model validation. The methodology supported the computation of national estimates based on the incomplete mosaic of states with module data. While developed in the context of the BRFSS data for SOGI outcomes, the approach can be used for other BRFSS topics and/or for other national surveys based on state samples.