Testing for complementarities in reduced-form regressions: A note