Estimative language is the intelligence community’s standardized vocabulary for expressing analytical uncertainty — the linguistic system through which analysts communicate how confident they are in their judgments. The development and refinement of estimative language reflects the discipline’s ongoing struggle with its central epistemological challenge: how to communicate probabilistic assessment to decision-makers who need actionable guidance.
The Kent scale
Sherman Kent produced the first systematic treatment of estimative language in his 1964 essay “Words of Estimative Probability,” which documented the problem: the same word (“probable,” “likely,” “possible”) meant different things to different analysts and consumers. Kent proposed a standardized mapping of verbal expressions to numerical probability ranges:
| Expression | Probability Range |
|---|---|
| Certain | 100% |
| Almost certain | 93% (±6%) |
| Probable / Likely | 75% (±12%) |
| Chances about even | 50% (±10%) |
| Probably not / Unlikely | 30% (±10%) |
| Almost certainly not | 7% (±5%) |
| Impossible | 0% |
Kent’s framework was elegant but faced institutional resistance: analysts resisted reducing their nuanced judgments to numbers, and consumers often ignored the probabilistic language, reading “probable” as “certain” or “possible” as “probable.”
Current IC standards
ICD 203 requires that all analytical products use standardized estimative language. The current framework uses two dimensions:
Likelihood language
Expressions of how likely something is to occur:
| Expression | Approximate Probability |
|---|---|
| Remote / Highly unlikely | < 5% |
| Unlikely / Improbable | 5–20% |
| Roughly even chance | 45–55% |
| Likely / Probable | 55–80% |
| Very likely / Highly probable | 80–95% |
| Almost certainly | > 95% |
Confidence language
Expressions of how confident the analyst is in the judgment (separate from the likelihood of the event):
- High confidence — well-corroborated information, strong analytical basis, low sensitivity to assumptions
- Moderate confidence — credibly sourced but less complete information, some analytical ambiguity, moderate sensitivity to assumptions
- Low confidence — fragmentary information, significant analytical gaps, high sensitivity to assumptions
The crucial distinction: likelihood refers to the probability of the assessed event; confidence refers to the quality of the evidence and analysis supporting the judgment. An analyst can assess with high confidence that something is unlikely — meaning the evidence strongly supports the judgment that the event probably won’t happen.
The communication problem
Estimative language addresses a structural problem of the analyst-policymaker relationship: the analyst works in probabilities; the policymaker must make binary decisions (act or don’t act). No linguistic framework fully bridges this gap.
Common failures:
- Consumers ignore hedging. “Probably” becomes “definitely” in the consumer’s reading
- Analysts hedge excessively. Every judgment is caveated until the product communicates no clear assessment
- Language without context. “High confidence” means nothing without understanding what it’s based on
- The Iraq precedent. “High confidence” in the October 2002 NIE on Iraqi WMD was spectacularly wrong — damaging the credibility of the confidence framework itself
Related concepts
- Estimative intelligence — the product type that uses this language
- ICD 203 — the directive that mandates these standards
- National Intelligence Estimate — the product format most associated with formal estimative language
- Intelligence failure — what happens when the language fails to communicate uncertainty