hidetify: Identify Influential Observations in High Dimension

Efficient tool for identifying influential observations in high dimensional linear regression. The tool implements two detection techniques single detection (Barry et al. (2020) <doi:10.1080/03610926.2020.1841793>) and multiple detection (Barry et al. (2021) <arXiv:2105.12286>). The single detection is an adaptation of Cook's measure for high dimensional data. The method relies on the concept of expectile to construct an influence measure based on asymmetric correlations. The multiple detection technique applies a group deletion procedure to build the algorithm on three main steps. The first stage applies an ultra conservative score to mitigate the swamping effect, the second stage uses the clean sample generated in the previous stage and applies an aggressive score to attenuate the masking phenomenon. Finally, the last step is concerned with the validation of the influential set generated by the two previous steps. The main functions take a response variable and a design matrix as input and output a set of potential influential observations.

install.packages("hidetify")

Try the hidetify package in your browser