Statistical1 min read
◆ The PatternJudging probability by how easily examples come to mind.
The availability heuristic (Tversky & Kahneman, 1973) — judging probability by how easily examples come to mind.
Why it distorts:
• Dramatic events (crashes, bankruptcies) are vivid and memorable
• Steady compounding is boring and forgettable
• Media amplifies dramatic events → they feel more probable
• Result: overestimate crash risk, underestimate steady-growth scenarios
• Dramatic events (crashes, bankruptcies) are vivid and memorable
• Steady compounding is boring and forgettable
• Media amplifies dramatic events → they feel more probable
• Result: overestimate crash risk, underestimate steady-growth scenarios
Counter: use base rates (actual historical frequencies) not vivid memories.
Worked example
Thirty years with two crash years: a base rate of 6.7%. Now let a crash year come to mind five times as easily as a quiet one, and let memories fade by 10% a year. The latest crash is three years old and still vivid, so the probability that feels right is 34.7% — five times the record. The weights are made up; the shape is the point.
Where it misleads
- Ease of recall is a decent cue when what you remember is a fair sample; it fails when media and memory over-select the dramatic.
- Barber & Odean (2008) found individual investors are net buyers of attention-grabbing stocks — those in the news, with high volume or extreme returns. Attention drives the trade, not the odds.
- The cure is not “ignore crashes”: tail risk is real (see Risk & Portfolio). The cure is to count them.
In code
years = 30 crash = np.zeros(years); crash[[8, 27]] = 1 # two crash years in thirty, the latest three years ago base_rate = crash.mean() vivid = np.where(crash == 1, 5.0, 1.0) # a crash year comes to mind five times as easily fresh = 0.9 ** np.arange(years)[::-1] # and memories fade by 10% a year w = vivid * fresh felt = (w * crash).sum() / w.sum() # the probability that "feels" right
Assumes import numpy as np and import pandas as pd. Each snippet carries its own example numbers; the comments say which are made up.
Sources
- A. Tversky & D. Kahneman, “Availability: A Heuristic for Judging Frequency and Probability”, Cognitive Psychology 5(2), 1973
- S. Lichtenstein, P. Slovic, B. Fischhoff, M. Layman & B. Combs, “Judged Frequency of Lethal Events”, Journal of Experimental Psychology: Human Learning and Memory 4(6), 1978
- B. M. Barber & T. Odean, “All That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors”, Review of Financial Studies 21(2), 2008
Last reviewed October 2026 · How this site checks its content · Report a mistake
Pattern bridge: Judging probability by what comes to mind is probability estimation gone wrong — confusing salience with frequency. In ML, accuracy vs. recall captures the same distortion.
Linked from 1