Key Takeaways: Precision Without Accuracy
The Big Idea
Precise wrong numbers are more dangerous than acknowledged uncertainty — because precision feels like knowledge and prevents the questioning that uncertainty invites. The institutional demand for legible, processable numbers systematically drives the stripping of uncertainty from quantitative claims.
Core Concepts
The Archery Analogy
- Archer A (precise + accurate): The ideal — tight cluster on the bullseye
- Archer B (accurate, not precise): Right on average but variable — honest uncertainty
- Archer C (precise, not accurate): Consistently wrong — the dangerous case
- Archer D (neither): Scattered and off-target — obviously unreliable
Key Mechanisms
- Specificity heuristic: Precise claims are perceived as more credible
- Quantification bias: Numbers are treated as more objective than qualitative assessments
- Error propagation: Combining imprecise measurements compounds the imprecision
- Uncertainty stripping: As numbers move through institutions, uncertainty is removed at each step
Cross-Domain Examples
| Domain | Precise Number | Actual Uncertainty |
|---|---|---|
| Finance | VaR: $113M at 99% | Actual loss: $26B (230x error) | |
| Medicine | IQ: 117 | Actual: ~112-122 (±5 points) |
| Nutrition | Calories: 230 | Actual: 184-276 (±20%) |
| Economics | GDP growth: 2.4% | Actual uncertainty: ±1-2% |
| Polling | Clinton: 48.3% | Non-response + modeling: ±5%+ |
| Medicine | BMI: 27.8 | Doesn't distinguish muscle from fat |
| Education | GPA: 3.47 | Based on subjective grades of varying rigor |
Epistemic Audit — Chapter 12 Addition
After this chapter: identify key quantitative claims, assess precision vs. accuracy, find sharp cutoffs, map uncertainty stripping, propose honesty upgrades.
What's Coming Next
Chapter 13: The Einstellung Effect — when expertise becomes a prison.
Quick Reference:
For ANY precise number, ask:
"What is the confidence interval?"
"Is the reported precision justified by the measurement's accuracy?"
"What sources of uncertainty are hidden?"
If the answer is "I don't know" → the precision is meaningless