An approach to choosing subsets of explanatory variables in regression problems that eliminates some variables, ‘shrinks’ the regression coefficients of others (similar to what happens in ridge regression), and gives relatively stable results unlike many of the usual subset selection procedures.