— this page is the wiki’s own appraisal reasoning. Like The Underivable Optimum and
The Estimate-to-Action Gap, it carries no source attributions of its own: every claim is the
fabric’s synthesis over held method rules (the telos’s streetlight and symmetric-standards sections)
and the concept pages it links, sources: [] is legitimate here. It anchors on Ioannidis’s
field-level bias model, reached through Most Published Findings Are False (PPV of a Field).
A meta-analysis has no independent access to truth — it POOLS an existing primary base, and so inherits that base’s selection. If a question was never funded or never run, no amount of rigorous synthesis at the top of the pyramid recovers it. The apex inherits the selection of its base: nothing in, nothing out. This is distinct from garbage-in-garbage-out — the inputs are not corrupted, they are absent, and absence leaves no trace in the pooled estimate to flag.
The upstream filter is a different filter from publication bias
Publication Bias and Selective Reporting is the DOWNSTREAM filter: given that a study ran, which results reach print (positive over null, the file drawer, effect-size inflation). This page is the UPSTREAM filter: which questions get studied at all. The two compose — a question must clear the upstream filter to exist, then its result clears the downstream filter to be seen — but they are not the same defect, and the downstream tools (funnel plots, p-curve, protocol-to-report comparison) are structurally blind to the upstream one: they audit the studies that exist, while the upstream filter is about the studies that never do.
Naming the lamps — the streetlight has more than one
The telos’s streetlight effect names its filter as measurability — what is easy to measure, isolate, blind, and observe quickly yields clean data, so the hard-to-measure reads as “no evidence.” That filter already covers blindability and short latency: you cannot blind a food or randomize a lifetime -> Measurement Error in Dietary Assessment, so long-latency whole-diet questions are under-served relative to a blindable, short-outcome pill. This page adds two filters the streetlight does not name, each non-random and pushing the same way, so the skew is systematic, not noise:
- Fundability / patentability — powered trials are expensive, and commercial sponsorship flows to what can be owned. An off-patent food, a behaviour, or a whole dietary pattern has no commercial sponsor for a decisive RCT, so it is expected to be under-trialled relative to a drug of equal importance. This is a named mechanism, not a measured one — held, discounted.
- Consonance with consensus — a heterodox hypothesis clears a higher bar to be funded and run than the prevailing one, so the expected result is the cheaper one to study. (This filter straddles: it also raises the bar to publish, overlapping the downstream filter -> Publication Bias and Selective Reporting.)
Each filter is non-random and points the same way, so a body of evidence that looks thin on an exposure may be dark because the exposure sits away from every lamp at once, not because the exposure does nothing.
The decision consequence — absence of evidence is CLAIM-NEUTRAL
This is the move the page exists to block. The structure
a true finding + no sponsor + a hostile establishment -> no studies -> no meta-analysis -> absence of high-tier evidence
is claim-neutral: it fits a true-but-unstudied exposure and a false-and-correctly-unstudied one equally well. The same lever that would elevate a favoured heterodox claim elevates every disfavoured one — seed-oil alarm, carnivore cures, and supplement suppression all run the identical argument. So absence-of-a-meta-analysis cannot raise the posterior of any particular claim; it is the symmetric-standards rule made concrete (a heterodox claim earns no presumption “from novelty, suppression narratives, or who advances it”).
What DOES discriminate — the constructive half
If the missing meta-analysis is claim-neutral, what separates a real-but-unstudied finding from a correctly-ignored one? Not the absence — three things that survive it:
- The expectancy test — if this were true, would we expect to have seen evidence by now? Silence from a genuinely unstudied question is not a null; silence where a cheap study would already have spoken is -> The Insufficient-Evidence Statement.
- Mechanism strength — a human-corroborated pathway carries directional weight even without a whole-intervention RCT (held, discounted, marked as mechanism, never dressed as an outcome).
- The cheap evidence that needs no sponsor — observational cohorts, natural experiments, and Mendelian randomization do not require industry money, so their absence is more informative than a missing RCT’s. Where the fundable-lamp argument is genuinely real, these channels are exactly where a true finding would still leave a trace -> Surrogate Outcomes (prediction vs causation, MR the arbiter).
So the endogenous base is a reason to weight the cheap channels, never a reason to believe the unstudied.
What this does NOT license — the gold-gate stands, and the bias runs both ways
- It does not loosen the gold-gate. The source-tier preference governs acquisition priority and confidence calibration, not a truth-verdict. Paired with the four evidence states (insufficient != no-effect), a fundable-lamp finding is held at insufficient-evidence + directional-mechanism, its gap named — NOT downgraded to “no effect.” The endogenous base silences a truth only if the gate is read naively as believe only meta-analyses, which the four-states discipline forbids.
- The bias is symmetric in direction, not in detectability. Advocate and foundation funding inflate a favoured heterodox claim through non-independence, in the opposite direction to how establishment neglect deflates it — though the inflation is at least detectable (a present study you can scrutinize for independence) where the deflation “leaves no trace” (the absent study cannot be audited). A crowd-funded study of a fashionable exposure earns the same independence scrutiny as an industry trial — symmetric standards apply in both directions, or the correction becomes its own bias .
Decision relevance
- When you meet “there is no meta-analysis, so the mainstream is merely under-researching X” — treat the absence as claim-neutral. Ask the expectancy test, the mechanism strength, and whether the cheap observational/MR evidence exists and what it says — not whether a meta-analysis is missing.
- Read a favoured-side advocate-funded study under the same independence scrutiny you would apply to a disfavoured-side establishment one.
- This is an open loop. Nothing here grades which unstudied findings later proved true; the wiki verifies only the would-form (would a well-informed advisor decline to move a posterior on an absence alone?), never the realized hit rate.