What the U.S. Data Reveal
The U.S. evidence asks whether young firms’ wage terms reflect the performance-contingent beliefs described by the model, rather than a simple penalty for being small or new. This chapter follows the Census study from matched employer–employee records to worker-level earnings estimates and then to industry-level uncertainty and aggregate outcomes. The useful judgment is not that one regression proves workers’ private thoughts, but that a pattern of comparisons can support—or weaken—the explanation that noisy information makes young-firm growth more difficult.
How the records turn a belief mechanism into an earnings test
The study links two administrative sources covering 1998 through 2014. The Longitudinal Business Database tracks U.S. establishments and firms, including age, employment, and productivity-related information. The Longitudinal Employer-Household Dynamics records quarterly jobs, earnings, and worker demographics from unemployment-insurance systems. Together they allow the analysis to place a worker’s earnings beside the characteristics and history of the employer. The linked data cover 60% of U.S. private-sector employment and 29 states, so they are broad but not a complete national census of the outcome being studied.
That linkage changes the question the data can answer. A firm-only data set could show that young firms grow differently, and a worker-only data set could show that younger workers earn less, but neither could easily tell whether the two facts occur in the same employment relationship. The matched records let the study follow employers and workers together, observe a worker’s prior employer or non-employment status, and compare the firm’s current and accumulated productivity. The trade-off is that administrative coverage and job definitions impose restrictions: the analysis is rich on reported employment and earnings, but it does not directly record what workers believe or every brief transition between jobs.
Young firms are establishments or firms aged five years or less, using the same practical age boundary that appears in the cross-country evidence. Firm size is total employment. Firm productivity is measured as log real revenue per worker, normalized to 2009 dollars. For workers, the main observation is a full-quarter job that is present in the quarter before and after the focal quarter. That restriction is necessary because the employment records do not report exact job start and end dates. It also means the core data are designed to describe continuing main jobs, not every short employment spell.
To connect the records to the learning mechanism, the study first estimates a productivity process. After removing industry-by-year differences and accounting for persistence, the remaining productivity variation is separated into a firm-specific component and a temporary residual. The former is used as a proxy for persistent fundamentals; the latter is treated as the noisy component of the signal. A firm’s past-average productivity then summarizes the evidence accumulated over its observed life cycle. “High-performing” means that this average is above the relevant industry benchmark, not that the firm has been assigned a permanent quality ranking.
The wage analysis is deliberately staged. First, worker earnings are adjusted for worker-specific effects, year effects, and age profiles. This removes a large part of the fact that different workers can earn different amounts even when they work at similar firms. Second, the resulting earnings residual is related to a young-firm indicator, its interaction with high performance, current and past performance, firm size, the previous employer’s firm effect or a non-employment indicator, and industry and state effects.