Level 4 · Error · εᵢ + σ + λ
When the Error Term Travels with the Evidence
Every piece of global evidence carries its own ε, σ, and λ. When you transfer it to India, all three transfer with it — and the question is whether the combined error term makes the finding usable or dangerous.
Level III Session 4 introduced India's six friction points — the structural mismatches between global evidence and India's health financing reality. This session goes deeper: it asks what happens to the ε, σ, and λ components of the error term when a finding is transferred from its source context to India.
The answer is that each component transforms. Structured absence (ε) in the source country may become structural mismatch in India — a gap not in the data but in the system being studied. Irreducible stochasticity (σ) in the source system may be categorically different from India's σ — political stability, implementation variance, and climate exposure can be higher or lower, and the direction matters for whether the finding transfers safely. And the anchor populations (λ) are almost certainly different — the populations most at risk of being failed by an average-optimised policy in Rwanda are not the same as in rural Jharkhand.
This table shows how to think through each component of the error term when assessing whether a finding from a non-India context can be used in a WHO India brief:
| Error component | Question to ask | What partial transfer looks like | What full transfer failure looks like |
|---|---|---|---|
| ε Structured missingness | Does the source study's evidence gap map onto the same India population, or onto a different one? | Source study missing rural data; India brief uses it for urban contexts only, notes rural gap separately. | Source study missing informal worker data; India brief applies it to informal sector populations without caveats. |
| σ Irreducible stochasticity | Is India's implementation variance and political environment comparable to the source context, or systematically different? | Source RCT in Kenya; India brief uses direction of effect only, notes that Kenya's more uniform health system implies lower σ than India's federal structure. | Source study in Netherlands; India brief uses effect size directly, ignoring that Netherlands σ ≈ 0 while India σ is structurally high. |
| λ Anchor constraint | Are the most-at-risk populations in the source context the same as India's anchor populations — or are they systematically different? | Source study in Rwanda shows strong benefit for lowest income quintile; India brief applies this to lowest quintile with a note that India's ST/SC populations face structural exclusion that Rwanda's study population did not. | Source study in South Korea shows universal benefit; India brief treats it as evidence of benefit for all Indian populations including ST/SC remote communities. |
The most dangerous transferability failure is not one error component but their compounding. A finding with high ε in the source (missing informal worker data), transferred to India where σ is high (implementation variance across states) and the anchor constraint differs (ST/SC populations not in source study), produces a brief whose total error term is the product of three separate failures — none visible individually, all catastrophic in combination.
This is why the transferability stress-test in Level III Session 4 is not sufficient on its own. It tests structural friction points but does not decompose the error. The full assessment requires asking: what is this finding's ε, what is its σ, and whose anchor case does it not cover — and then asking whether those components compound dangerously in the India context.
Describe a finding you want to transfer from a non-India context into a WHO India brief. The tool will decompose all three error components (ε, σ, λ) for the source finding and for its India application, then give a compound error assessment and citation language.
Global evidence does not travel alone. When a finding crosses into a WHO India brief, its ε (structured gaps), σ (stochastic environment), and λ (anchor populations) all travel with it — and each may transform in ways that compound the total error. The full error decomposition in this session is the final quality gate before a transferred finding enters a recommendation. A finding that passes the Level III transferability stress-test may still fail the Level IV error decomposition if its three components combine dangerously in the India context. Session 5 closes Level 4 by asking what it means to operate well when all three error components are simultaneously large — the rapid review under genuine uncertainty.