I'd think the logic is harder to tease out than the statistical correlation; isn't the latter more-or-less intended for situations where you can't determine (let alone measure) all the variables?