svycdiff

svycdiff estimates controlled differences in complex survey data for health disparities research, implementing propensity score and survey weighting methods to account for covariate imbalance and generalizability when comparing outcomes between groups.

Key Features:

• Population Average Controlled Difference Estimation: Computes the average controlled difference (ACD) in outcomes between groups, such as Black and White individuals, using identification formulas that integrate propensity score weighting with complex survey weights.
• Complex Survey Integration: Properly handles survey weights that depend on the group variable under comparison, addressing sampling schemes like those in the National Health and Nutrition Examination Survey (NHANES) where weights depend on self-reported race.
• Covariate Balance Adjustment: Implements propensity score methods to achieve covariate balance across racial groups while adjusting for social determinants and socioeconomic confounders.
• Performance Validation: Methods validated via extensive simulation demonstrating superior bias, mean squared error, and coverage properties compared to traditional analytic approaches.

Scientific Applications:

• Racial Health Disparities Research: Estimates racial disparities in health outcomes, exemplified by the analysis of telomere length differences between Black and White individuals in NHANES data.
• Observational Study Analysis: Provides appropriate causal inference methods for observational data from complex surveys where randomized experiments are not feasible.
• Social Determinants of Health Investigation: Characterizes controlled effects of race on biological outcomes after accounting for confounding by socioeconomic factors and other social determinants.

Methodology:

The package builds a propensity for race to adjust for social determinants, then applies identification formulas that weight for both covariate imbalance across racial groups and generalizability using survey weights. The estimation combines propensity score methods with complex survey weighting techniques.