Using Generalized Correlation to Effect Variable Selection in Very High Dimensional Problems. Hall, P. & Miller, H. J Comp Graph Stat, 18(3):533-550, 2009. bibtex @article{hal09gen,
title = {Using Generalized Correlation to Effect Variable Selection in Very High Dimensional Problems},
volume = {18},
number = {3},
journal = {J Comp Graph Stat},
author = {Hall, Peter and Miller, Hugh},
year = {2009},
keywords = {bootstrap,instrumental-variables,ranks,measurement-error,confidence-intervals-for-ranks,cubic-splines,errors-in-variables,generalized-correlation,hidden-explanatory-variables,linear-model,ranking-correlations},
pages = {533-550},
citeulike-article-id = {13265815},
posted-at = {2014-07-14 14:10:05},
priority = {0},
annote = {fitting an overall model may hide components that have potential for linearly influencing the response;group lasso or group LARS may not detect all influential variables;using a cubic spline fit for each variable separately and computing ordinary R{$^2$};"global modeling techniques generally preclude basis expansions on the grounds that they create an even larger dimensionality problem";use bootstrap percentile prediction intervals for rankings;nice dot charts;see hal09usi}
}