Skip to content

Absence of Evidence Evaluation

Why It Matters

People misapply "absence of evidence is not evidence of absence" by treating all missing evidence as equally uninformative. But if we've thoroughly searched where evidence should be, its absence is informative. This matters for: evaluating conspiracy theories (wouldn't there be leaks?), assessing risks (shouldn't we have seen warning signs?), and updating beliefs (we predicted X would happen, it didn't, what does that tell us?).

The Key Move

When evidence E is absent, ask three questions:

  1. If the hypothesis were true, would we expect to observe E?
  2. Have we looked for E in the right places with adequate methods?
  3. How powerful was our search - could E be present but undetected? If yes to 1 and 2, and 3 indicates good detection power, then absence of E is evidence against the hypothesis. If no to any question, absence is uninformative.

Classic Application

Fraud detection. If a company's reported profits were genuine, we'd expect to see corresponding cash flows, tax payments, and legitimate business operations. When accounting shows profits but these expected correlates are missing (Enron had profits without cash flow), their absence is evidence of fraud - not mere lack of evidence for legitimacy.

Surprising Application

Social relationships. If someone "really wants to be friends" but never initiates contact, never remembers things you tell them, never makes time for you despite claiming to be busy but finding time for others - the absence of friendship behaviors is evidence of absence of friendship, not just lack of evidence for it. This distinguishes "benefit of the doubt" (appropriate when you lack information) from "ignoring clear negative evidence" (inappropriate).

Failure Modes

Expecting evidence in the wrong place or form (you looked for E₁ but H would produce E₂). Not accounting for suppression of evidence (cover-ups do happen). Failing to quantify detection power (did you look hard enough?). Treating any absence as strong evidence without considering how much you'd expect to see. Ignoring that many true hypotheses generate subtle evidence requiring expertise to detect.

Go Deeper

Sober, "Evidence and Evolution" on evidential reasoning; Hempel, "Studies in the Logic of Confirmation"