Thomas Jaki
Deparmtent of Statistics and Mathematics
Lancaster University
Direct Effects Testing: A Two-Stage method for stepwise Testing for
Effect Size and Variable Importance for Correlated 2x2 Tables
In applications such as medical statistics and genetics, we frequently
encounter situations where a large number of highly correlated predictors
explain a response. Constructing a good predictive model in
such cases is well studied while recovery of the 'true sparsity pattern',
that is finding which predictors have direct effect on the response, and
indicating the statistical significance of the results is less well
understood. Restricting attention to binary predictors and response, we
study the recovery of the true sparsity pattern using a two-stage method
that separates the process of establishing the presence of direct effects
from inferring their exact relationship with the predictors. Simulations and
a real data application demonstrate the method discriminates well between
associations and direct effects and comparisons with lasso based methods
demonstrate favorable performance of the proposed method.
Back to Colloquium Series