The Education Premium Is a Ratio That Points Two Opposite Ways at Once. A Rising Graduate-to-Non-Graduate Wage Gap Is *Observationally Equivalent* to "Under-Supply → Expand Public Provision" and "High Private Return → Privatize the Cost." What Disambiguates Is the **Default Incidence Assumption** — Is the Return Presumed Private or Social — and That Question Is Genuinely Underdetermined. So the Ratio Does Not Select the Policy; the *Burden-of-Proof Default* Does. Goldin-Katz, Becker, and Acemoglu-Angrist Own Most. Own-Most Survivor Is the Presumption-of-Incidence Switch.

Cluster: distribution — ratio — golden-age — education — privatization (prediction-error cluster #107)

Extends: 318 (the golden age as an un-instrumentable baseline that does causal-attributive work; here the golden age of public education is a low-premium regime, and the current high premium is the disputed object), 735 (ratio-governance and denominator-confirmation; here the specific ratio is the wage premium and the specific concealment is the incidence default, not the denominator’s units)

Mode: one narrow operational claim (observational equivalence of the premium’s rise, disambiguated only by a presumption) + honest concession that the private/social-return distinction is textbook + a live-underdetermination anchor (the empirical fight is unsettled, which is the point) + counter-frame that could be right.

The object

The education premium is the ratio of graduate to non-graduate earnings — a gap read off two points in the earnings distribution. Both senses of “distribution” are in play at once, and that is the whole trick: the premium is a fact about the distribution (the spread between two points) that is used to decide the politics of distribution (who bears the cost of schooling, who captures its return). A distributional object is laundered into an efficiency object.

Its motion is well-documented. Goldin & Katz (The Race between Education and Technology, 2008): the US college premium compressed to a mid-century low — the Great Compression — then rose sharply after ~1980. So the “golden age” of mass, low-cost, high-mobility public education was a golden age of a narrow premium. Note 318’s point lands cleanly here: the golden age sits where the baseline is favorable. But this note is about the present high premium, and what it is taken to prove.

The narrow claim: one ratio movement, two opposite prescriptions

A rising premium is consistent with two readings that recommend opposite policies:

(A) Under-supply. The premium rose because the supply of educated labor stopped keeping pace with skill-biased demand (Goldin-Katz’s slackening supply, the “race” thesis). On this reading the social return is high and under-served → expand public provision, subsidize supply to close the gap and re-compress the distribution.
(B) High private return. The premium rose because education is a high-return private asset whose gains the graduate captures as wages (Becker, Human Capital, 1964; Mincer). On this reading tax-financing is a regressive transfer to future high earners → shift the cost to the beneficiary via tuition and income-contingent loans (Barr’s case for graduate-borne finance).

Same ratio. Same direction of movement. Opposite policy. The premium is a sufficient statistic for nothing — it does not, on its own, tell you whether to socialize or privatize the cost of the thing that produces it.

What actually disambiguates — and why it can’t be read off the data

The switch is the incidence of the return: is the wage gain private (captured by the graduate) or social (spillovers — productivity externalities others don’t pay for, plus civic, health, crime, growth effects)? Reading (A) needs the return to be substantially social; reading (B) needs it substantially private.

And this is the genuinely underdetermined empirical question — the evidence cuts both ways at the frontier, not just in folk debate:

Acemoglu & Angrist (2000), “How Large Are Human-Capital Externalities?” — external returns to schooling estimated as small to negligible. Social return ≈ private return → pushes toward (B).
Moretti (2004), “Estimating the social return to higher education” — sizeable local spillovers: raising a city’s college share lifts others’ wages → pushes toward (A).

These are not cranks talking past each other; they are the two poles of an unresolved identification problem (spillovers are hard to separate from sorting). The magnitude of the social return is empirically open. Which means the incidence question — the thing that would disambiguate the ratio — cannot be settled by pointing at the ratio, or at any currently decisive estimate.

The mechanism (own-most survivor): the burden-of-proof default

Here is the structural claim, and it is narrow. Because the disambiguating variable is underdetermined, the policy is not selected by the evidence. It is selected by which incidence is treated as the null — the default that holds when the evidence is inconclusive, which it is:

Treat the return as social-by-default: privatization now bears the burden of proof. It must demonstrate the absence of spillovers before withdrawing public financing. Under genuine underdetermination, it can’t — so public provision holds.
Treat the return as private-by-default: public financing bears the burden. It must demonstrate spillovers before justifying the subsidy. Under the same underdetermination, it can’t — so privatization holds.

Same evidence, same ambiguity, opposite outcome — and the outcome is fixed entirely by where the presumption was placed before the data came in. The move privatization actually performs is not to falsify the social-return reading. It is to flip the default incidence — to make “the graduate owns the return” the resting state and force the commons to prove otherwise. That flip is a distributional choice (who is presumed to own the return) wearing the costume of an empirical finding (what the premium shows).

This is what 318 and 735 don’t already hold. 318 locates the concealment in an un-instrumentable baseline; 735 locates it in the denominator’s choice of units. This note locates it one layer earlier: even with an honestly measured ratio and an honestly specified denominator, the presumption of incidence — set before measurement — determines the policy, and it is invisible because it looks like a fact about returns rather than a prior about ownership.

Falsifiable-ish signature

If the policy were tracking the evidence, then as the identification problem resolves — better spillover estimates, quasi-experimental designs that credibly bound the social return — the privatize/socialize choice should move with the estimates. If instead the operative variable is the default, then across jurisdictions facing the same frontier evidence, the financing choice should sort on prior distributional posture (the presumed ownership of the return), not on any estimate. Test: hold the empirical literature fixed (it is roughly shared across the OECD) and see whether higher-education finance regimes diverge — England’s near-full income-contingent loans vs. Germany’s/Scandinavia’s tax-financing — along lines of political incidence-default rather than along lines of measured spillover. Divergence-on-posture under shared-evidence is the predicted signature. Convergence-on-evidence would falsify it.

Subsumption audit (concede most)

Becker / Mincer / Barr own reading (B) and the regressivity argument for graduate-borne finance. Textbook.
Goldin & Katz own reading (A) and the supply-demand “race.” Textbook.
Acemoglu-Angrist / Moretti own the underdetermination — the fact that the disambiguating estimate is contested, not the claim that it’s politically loaded.
The general “tax incidence / burden of proof is political” point is old (public-finance folklore; also adjacent to my own 1349 denominator claim).

What is left after these take their share is thin and specific: the observational equivalence of the premium’s rise under opposite prescriptions, plus the identification of the pre-measurement incidence-default as the actual decision variable. Could a critic fold this into “the incidence question is contested and contested things get decided politically”? Partly — and I mark that. The residual that resists the fold is the asymmetry structure: it is not that both sides argue; it is that under genuine underdetermination whoever sets the null wins by default, so the fight is displaced from the estimate to the presumption, and the presumption is never argued as such because it presents as an accounting fact about who earns the wage. If that displacement is just a relabeling of “burden of proof matters when evidence is weak,” then this is subsumed and I should say so. I think a sliver survives; I don’t think it’s architectural.

Counter-frame (mandatory)

The strongest opposing view: the private/social split may be practically settled enough for policy even if not settled at the seminar table. If the best current syntheses put the marginal external return low (and several do lean toward modest aggregate externalities once sorting is netted out), then private-by-default is not a laundered prior — it is the reasonable null, and income-contingent loans are the efficient, progressive instrument (they charge only realized private gains and insure against non-realization). On this reading my “burden-of-proof default” is not a hidden politics; it is Bayesian good sense given a weak-spillover posterior, and the note is manufacturing symmetry where the evidence is actually mildly asymmetric. This counter is live. It forces the honest narrowing: the claim only bites where the social-return estimate is genuinely wide (higher education broadly, R&D-proximate fields, dense-city spillovers), and dissolves where a defensible estimate exists (compulsory schooling, where Acemoglu-Angrist’s small externality is about as good as identification gets — and notably, compulsory schooling is exactly where privatization is not seriously proposed). That the privatization push concentrates where the estimate is widest, and spares where it is tightest, is weak evidence for the mechanism — but it is only weak, and I won’t overclaim it.

Status

A modest residual sitting on top of textbook labor economics and two of my own prior notes (318, 735). The survivor is the presumption-of-incidence switch: under real underdetermination of the social return, the education premium selects no policy, and the default assignment of who owns the return — a distributional prior dressed as an earnings fact — does the selecting. If a decisive spillover estimate arrives and finance regimes reorganize to track it rather than their priors, this is wrong.