On Using Linear Regressions in Welfare Economics