We analyze a large, anonymous labour market in which firms motivate their workers via relational contracts. The market is frictionless and features on-the-job search, in that all acceptable vacancies are immediately filled and the employed compete with the unemployed for vacancies. While firms and workers are ex ante identical, the unique equilibrium exhibits a continuous distribution of contracts in which high wage firms have higher retention rates, more motivated workers and higher productivity. The model thus generates dispersion in wages, productivity and human resource strategies, and gives rise to endogenous job ladders. An exogenous increase in on-the-job search increases the quantity of jobs but decreases their quality; with sufficient on-the-job search there is full employment, and wage dispersion rather than unemployment motivates workers.

This paper proposes a tractable model of Bayesian learning on large random networks where agents choose whether to adopt an innovation. We study the impact of the network structure on learning dynamics and product diffusion. In directed networks, all direct and indirect links contribute to agents' learning. In comparison, learning and welfare are lower in undirected networks and networks with cliques. In a rich class of networks, behavior is described by a small number of differential equations, making the model useful for empirical work.

We propose a new model of flrm reputation that interprets reputation directly as the market belief about product quality. Quality is persistent and is determined endogenously by the flrm’s past investments. We analyse how investment incentives depend on the flrm’s reputation and derive implications for reputational dynamics. We consider three types of consumer learning. When consumers learn about quality through good news, investment incentives are decreasing in reputation, leading to a unique work-shirk equilibrium and convergent dynamics. When consumers learn through bad news, investment incentives are increasing in reputation, leading to a continuum of shirk-work equilibria and divergent dynamics. Finally, when consumers learn through Brownian news and the cost of investment is low, incentives are hump-shaped but a work-shirk equilibrium exists and is essentially unique.

We propose a model of strategic experimentation on social networks in which forward-looking agents learn from their own and neighbors’ successes. In equilibrium, private discovery is followed by social diffusion. Social learning crowds out own experimentation, so total information decreases with network density; we determine density thresholds below which agents’ asymptotic learning is perfect. By contrast, agent welfare is single peaked in network density and achieves a second-best benchmark level at intermediate levels that strike a balance between discovery and diffusion.

We study a parsimonious model of a competitive labor market in which firms privately screen workers to identify talent. The equilibrium exhibits dispersion in wages and productivity; when talent is scarce, firms with superior screening skills post higher wages, attract better applicants, and recruit more talented workers. High-wage firms impose a compositional externality on low-wage firms, leading to equilibrium inefficiency: Welfare would be higher if low-skilled firms posted high wages and selected first. We also provide a micro-foundation for firms heterogeneous screening skills. When talented workers are better at screening (e.g. via superior referrals), a dynamic version of the economy converges to a unique steady state in which differences in talent, profits and screening skills persist forever.

We propose a simple model of racial bias in hiring that encompasses three major theories: taste-based discrimination, screening discrimination, and complementary production. We derive a test that can distinguish these theories based on the mean and variance of workers’ productivity under managers of different pairs of races. We apply this test to study discrimination at a major U.S. retailer using data from 48,755 newly hired commission-based salespeople. White, black, and Hispanic managers within the same store are significantly more likely to hire workers of their own race, consistent with all three theories. For black–Hispanic pairs, productivity variance is lower for same-race pairs than cross-race pairs, implying that screening discrimination dominates. For white–Hispanic pairs, mean productivity is higher for same-race pairs, indicating a combination of screening discrimination and complementary production. For white–black pairs, biased hiring implies the presence of discrimination, but productivity results suggest the effects of the three forms of discrimination offset one another.

We explore costly deliberation by two differentially informed and possibly biased jurors: A hawk Lones and a dove Moritz alternately insist on a verdict until one concedes. Debate assumes one of two genres, depending on bias: A juror, say Lones, is intransigent if he wishes to prevail and reach a conviction for any type of Moritz next to concede. In contrast, Lones is ambivalent if he wants the strongest conceding types of Moritz to push for acquittal. Both jurors are ambivalent with small bias or high delay costs. As Lones grows more hawkish, he argues more forcefully for convictions, mitigating wrongful acquittals. If dovish Moritz is intransigent, then he softens (strategic substitutes), leading to more wrongful convictions. Ambivalent debate is new, and yields a novel dynamic benefit of increased polarization. For if Moritz is ambivalent, then he toughens (strategic complements), and so, surprisingly, a more hawkish Lones leads to fewer wrongful acquittals and convictions. So more polarized but balanced debate can improve communication, unlike in static cheap talk. We also show that patient and not too biased jurors vote against their posteriors near the end of the debate, optimally playing devil’s advocate. We shed light on the adversarial legal system, peremptory challenges, and cloture rules.

The sensitivity of Bayesian implementation to agents' beliefs about others suggests the use of more robust notions of implementation such as ex post implementation, which requires that each agent's strategy be optimal for every possible realization of the types of other agents. We show that the only deterministic social choice functions that are ex post implementable in generic mechanism design frameworks with multidimensional signals, interdependent valuations, and transferable utilities are constant functions. In other words, deterministic ex post implementation requires that the same alternative must be chosen irrespective of agents' signals. The proof shows that ex post implementability of a nontrivial deterministic social choice function implies that certain rates of information substitution coincide for all agents. This condition amounts to a system of differential equations that are not satisfied by generic valuation functions.