Premorbid Probability Anchoring¶

Why It Matters¶

A positive test feels definitive: "the test says they have it." But all tests have false positives. If you test a very low-probability condition, most positive results are false. Premorbid probability anchoring corrects the systematic error of base rate neglect - ignoring how common a condition is before testing.

The Key Move¶

Before ordering a test, write down: "I estimate the probability of this condition at X% before testing." When results return, use Bayes' theorem (formally or intuitively): combine your prior probability with the test's characteristics to get the posterior probability. If you started at 1% and got a positive test with 95% sensitivity and 90% specificity, you're at ~9%, not 95%. The prior anchoring prevents the test from overwhelming your clinical judgment.

Classic Application¶

D-dimer testing for pulmonary embolism. D-dimer is very sensitive (negative result rules out PE) but not specific (positive result doesn't mean much). A patient with low pre-test probability (5%) gets a positive d-dimer - naive interpretation says "clot detected, get a CT." But 5% prior with a positive d-dimer only increases probability to ~15-20%, which may still be below CT threshold. The pre-test anchor prevents over-testing.

Surprising Application¶

Hiring decisions and reference checks. A candidate seems strong (60% confidence they'll succeed). Then a reference check reveals concerning information. Naive reaction: reject the candidate. Premorbid anchoring: you were at 60% before this datapoint, the reference provides some negative evidence, but unless it's highly reliable and diagnostic, it might only move you to 40% - still worth considering. The prior prevents a single datapoint from overwhelming the accumulated evidence.

Failure Modes¶

Stubborn priors that refuse to update - if evidence is strong, you must shift significantly. The tool fails when you don't actually know the base rate (no reference class for genuinely novel situations). Also fails when applied mechanically without considering information quality - strong, reliable evidence should shift you more than weak, questionable evidence. The worst failure is using it to justify ignoring test results: "my clinical judgment says..." when the test is actually highly informative.

Go Deeper¶

Fagan, T.J. (1975). "Nomogram for Bayes' Theorem." New England Journal of Medicine, 293(5), 257; Gigerenzer, G. & Hoffrage, U. (1995). "How to Improve Bayesian Reasoning Without Instruction: Frequency Formats." Psychological Review, 102(4), 684-704.

Anchoring - Behavioral Economics: Reasoning About Systematic Deviations