I. A reinterpretation of shift-share analysis