Conservatism Bias
🇳🇴KonservatismebiasDefinition
Conservatism bias is the tendency to update beliefs too slowly as new data arrive. Mechanistically, people underweight the likelihood carried by fresh evidence relative to their prior due to cognitive inertia, anchoring on initial estimates, and the reputational or self-justification costs of admitting error. In sequential environments this produces a systematic lag behind a changing reality.
Real-world example
In the run-up to the 2008 housing crash, rising mortgage delinquencies and widening credit spreads signaled mounting risk, yet many banks and rating agencies trimmed loss assumptions only incrementally because their models were anchored to years of stability. Capital buffers were not raised in time, exposures were maintained, and when defaults accelerated the subsequent adjustment was abrupt and costly. This slow, partial updating exemplifies conservatism bias at organizational scale.
Supplementary perspective
Partial adjustment can be adaptive when signals are noisy, transaction costs are high, or model uncertainty is substantial; optimal Bayesian filters with uncertain signal-to-noise (e.g., Kalman gain) also prescribe dampened updates. The pattern becomes a bias when the weight placed on new evidence is reliably too small given its diagnosticity. Countermeasures include pre-specifying update rules, using log-odds to combine priors and likelihoods, and independent second reviews to challenge initial anchors.
Practical advice
Recognize
- —Ask whether your beliefs have been meaningfully updated in proportion to the new evidence, or only minimally adjusted.
- —Watch for the phrase 'I still think...' when circumstances have materially changed.
Counteract
- —Explicitly quantify how much your estimate should change given the new data — use Bayesian reasoning as a benchmark.
- —Ask, 'What would I believe if I were starting from scratch today with all available evidence?'
- —Implement structured review points where assumptions are formally re-evaluated.
Ethical use
- —Present new evidence clearly with explicit comparisons to prior assumptions.
- —Create safe environments where changing one's mind is rewarded, not penalised.