Imagine you're running a buffet restaurant. You charge a flat price, and two problems immediately emerge: some diners eat less than the price is worth (you overpay for their appetite), and the hungriest eaters are the ones most attracted to your flat-price deal (adverse selection). Now imagine you could measure exactly how much each problem costs you and design a cover charge that filters optimally. That's what this paper does for labor contracts. Daniel Herbst's committed claim: asymmetric information distortions in wage contracts — moral hazard and adverse selection — can be identified, separated, and welfare-costed using marginal treatment effects (MTE), and the framework works on real field-experimental data. Workers in a data-entry task chose between a randomized fixed wage and a piece rate. The randomization is the engine: by varying the fixed wage offer exogenously, Herbst traces out who switches at each margin and how productive they are, mapping both distortions onto marginal treatment response functions. The numbers are concrete. Fixed wages reduce worker productivity by 8.74% relative to the mean — that's moral hazard, the effort-shirking channel. Meanwhile, a 10% increase in the fixed-wage offer attracts a marginal worker whose productivity is 1.90% of the mean higher — that's adverse selection, the sorting channel. Both are economically meaningful, and the MTE framework lets Herbst decompose welfare losses into their respective sources rather than lumping them together. The architectural choice matters. This is not a reduced-form experiment that just estimates an average treatment effect. Herbst builds a structural model of wage contracts under asymmetric information and then shows that the model's distortion terms correspond exactly to MTE objects — which can be estimated semi-parametrically from the experimental variation. The bridge between structural contract theory and the MTE toolkit is the paper's real intellectual contribution. It makes the welfare calculation tractable without imposing strong parametric assumptions on the productivity distribution. On the policy side, Herbst calculates marginal values of public funds (MVPFs) for interventions designed to recover the welfare loss from asymmetric information. The headline result: a tax on piece rates can mitigate adverse selection into fixed wages, generating tax revenue at a social cost as low as $0.88 per dollar. That's a remarkably efficient instrument — most tax instruments in public finance have MVPFs well above $1.00. The integrity profile is strong for a labor economics paper. The identification comes from a genuine field experiment with exogenous wage randomization, not from observational data with instruments of varying credibility. The MTE estimation is semi-parametric, which avoids some of the functional-form fragility that plagues fully structural approaches. The 96-page length (with appendices) suggests extensive robustness checks, though we're working from the abstract and cannot verify the full battery. The limitation is scope. This is one task (data entry), one labor market, one margin of contract choice. The MTE-contract-theory bridge is general in principle, but whether moral hazard and adverse selection have the same relative magnitudes in, say, white-collar employment or gig work is an open empirical question. The $0.88 MVPF is a proof-of-concept for the policy instrument, not a deployable recommendation. The next milestone is applying this framework to a labor market where the stakes — and the contract menu — are richer.
