B0712
Title: Unconditional treatment effect with high-dimensional covariates and unmeasured confounding
Authors: Chunlin Li - Iowa State University (United States) [presenting]
Jing Zhou - University of East Anglia (United Kingdom)
Abstract: In many applications, determining whether a specific variable has a causal impact on the outcome is of major interest. For example, one might investigate the influence of a university degree on income and evaluate the income disparity between degree-holders and non-degree holders. A method is presented that investigates the marginal/unconditional effect of an indicator $D = 0, 1$ (e.g., with or without a university degree) on the outcome based on the factor-augmented sparse regression model, which allows for high correlations between predictive variables and correction of unmeasured confounders. The marginal treatment effect is derived from four measures. While mean and quantile treatment effects have been explored in existing literature, variance and Gini coefficient are underexplored in high-dimensional literature, yet they offer valuable insights into biology, finance, and economics. Furthermore, a hypothesis test is developed to assess the significance of the difference between the two groups ($D=0, 1$).