Quantile Correlation‐Based Sufficient Variable Screening by Controlling False Discovery Rate.

Sufficient variable screening (SVS) with the false discovery rate (FDR) controlled rapidly reduces dimensionality with high probability in high dimensional modeling. By using quantiles, this paper proposes a new SVS procedure by controlling the FDR based on two‐stage Pearson's goodness testing with Chi‐square statistics for high dimensional data, abbreviated as QC‐SVS‐FDR. Without any specified distribution of the actual model, the QC‐SVS‐FDR method screens important predictors by a series of testing procedures combined with the adaptive composite of Pearson's chi‐square statistics. The quantile correlation‐based sufficient utility is sensitive to capture the subtle correlations under different quantile levels and is easy to implement with computational efficiency. Asymptotic results and sufficient screening properties of the proposed methods are obtained under mild conditions. Numerical studies including simulation studies and real data analysis demonstrate the advantages of the proposed method in practical settings. [ABSTRACT FROM AUTHOR]