Reassessing Online Job Postings as a Measure of United States Labor Market Dynamics and Employer Concentration
The measurement of labor market activity in the United States has undergone a quiet revolution over the past two decades, shifting from traditional survey-based instruments toward high-frequency, big-data applications. Among these, online job postings datasets—most notably those compiled by Burning Glass Technologies (now part of Lightcast)—have emerged as indispensable resources for economists, policymakers, and industry analysts seeking to understand the granular realities of employment demand. However, a new study released by the National Bureau of Economic Research (NBER), Working Paper 35697 (DOI: 10.3386/w35697), offers a critical methodological reassessment of these digital archives. The paper investigates the fidelity with which online job postings capture true U.S. job vacancies by systematically matching near-universal Burning Glass postings against the Bureau of Labor Statistics’ representative Job Openings and Labor Turnover Survey (JOLTS).
While the research confirms that digital job boards have grown increasingly reliable over time in capturing broad macroeconomic trends, it reveals a profound distortion in how economists measure labor market concentration. According to the findings, unadjusted online posting data overstates labor market concentration—as measured by the Herfindahl-Hirschman Index (HHI)—by nearly threefold. This massive discrepancy arises because small and mid-sized establishments utilize online portals at significantly lower rates than their large corporate counterparts. By constructing a reweighting methodology that aligns online postings with JOLTS benchmarks along observable characteristics, the researchers provide a clearer picture of how big data can be accurately harnessed to evaluate modern labor markets without falling prey to systemic sampling biases.
Historical Evolution of Labor Market Measurement and the Rise of Big Data
For decades, the benchmark for understanding American labor demand relied heavily on government surveys. The Job Openings and Labor Turnover Survey, launched nationwide by the Bureau of Labor Statistics in 2000, quickly became the gold standard for tracking total job openings, hires, and separations across various sectors and geographic regions. Despite its robust statistical design and representative sampling framework, JOLTS possesses inherent limitations: it offers limited industrial and occupational granularity and operates on a monthly publication lag, making it difficult for researchers to track real-time shifts in specialized skill demands or localized geographic micro-trends.
To fill this analytical void, researchers turned to web-scraped data aggregators. Burning Glass Technologies pioneered the systematic collection of millions of digital job advertisements posted across thousands of online job boards, corporate career sites, and local classifieds. This near-universal coverage allowed labor economists to track employer demand for specific technical skills, educational prerequisites, wage offers, and software proficiencies with unprecedented speed and spatial resolution. Throughout the late 2000s and 2010s, academic literature increasingly relied on these digital footprints to study everything from the impact of automation on local labor markets to the emergence of new occupational categories.
Yet, fundamental questions regarding selection bias persisted. Unlike JOLTS, which employs a scientifically designed probability sample of business establishments designed to represent the entire U.S. economy, online job boards depend entirely on the voluntary digital recruitment habits of individual employers. Organizations that do not routinely post vacancies online—such as small retail shops, local construction contractors, independent medical practices, and traditional service providers—risk being systematically underrepresented in big-data economic studies. NBER Working Paper 35697 directly addresses this blind spot by mapping the evolution of Burning Glass data against JOLTS benchmarks across a crucial fifteen-year window spanning from 2007 to 2022.
Methodological Approach: Bridging Burning Glass and JOLTS
The research team undertook a rigorous empirical matching exercise to evaluate the structural relationship between digital job postings and official government metrics. By comparing the trajectory of Burning Glass data against the JOLTS benchmark over the 2007–2022 period, the study established a foundational baseline of convergence. The data demonstrates that over this fifteen-year span, digital recruitment practices expanded dramatically across the American economy. As internet adoption became ubiquitous and corporate human resources departments centralized recruitment through digital applicant tracking systems, Burning Glass steadily converged toward the aggregate trends captured by JOLTS.
Despite this macro-level convergence, the study reveals that the representativeness of online job postings varies significantly across establishment characteristics such as firm size, industrial sector, and geographic location. Large corporations and institutional employers, which account for a disproportionately large share of total national job openings, are exceptionally well-represented in online databases. Because these large entities routinely utilize automated job-posting mechanisms that broadcast vacancies across multiple aggregator networks, their hiring demands are captured with high fidelity.
To correct for the underrepresentation of smaller business entities, the authors developed a systematic reweighting framework. By aligning the observable characteristics of establishments in the Burning Glass database with the benchmark distributions found in JOLTS, the researchers created an adjusted data architecture designed to neutralize selection biases. This methodological correction allows economists to test whether previously published empirical findings derived from unadjusted online data hold true under rigorous statistical scrutiny.
Impact on Macroeconomic Metrics and Skill Demand Analysis
One of the most reassuring findings for empirical labor economists is the resilience of macroeconomic skill-demand analyses. When the researchers applied their reweighting adjustment to account for establishment-size discrepancies, they discovered that both the aggregate level and the cyclicality of skill demand survived the adjustment process largely intact.
Because large establishments account for the vast majority of economy-wide job openings at any given point in the business cycle, the aggregate fluctuations in hiring requirements—such as the rising demand for cognitive skills, specialized programming languages, or advanced managerial credentials—remain robustly visible in raw online posting data. When economists analyze macroeconomic trends, the heavy weight of large firms in both JOLTS and digital archives ensures that aggregate signals of labor demand accurately reflect broader economic conditions. Consequently, historical studies utilizing Burning Glass data to track the evolution of skill requirements during economic recessions and recoveries retain their fundamental validity.
The Distortion of Labor Market Concentration
While macroeconomic aggregates proved resilient to sampling biases, the study uncovers a severe distortion when researchers shift their focus from aggregate demand to market-level structural dynamics. Specifically, the paper demonstrates that unadjusted online job postings overstate labor market concentration—measured via the Herfindahl-Hirschman Index—nearly threefold.
Labor market concentration has emerged as a central focus of antitrust policy and labor economics in recent years, with researchers increasingly examining whether dominant employers possess undue wage-setting power within local geographic areas or specific occupational categories. Because smaller and mid-sized establishments—which collectively form the competitive bedrock of local service, retail, and manufacturing sectors—post significantly fewer vacancies online compared to dominant local employers, unadjusted datasets create an illusion of hyper-concentration.
When digital archives rely primarily on large corporate job postings, the competitive presence of thousands of smaller, non-digitally active local employers is effectively muted. This structural invisibility leads researchers to mistakenly conclude that local labor markets are far more monopolized or oligopolized than they actually are. By recalibrating the data to properly weight the hiring activity of smaller firms, the NBER study demonstrates that true labor market concentration is substantially lower than previously estimated through unadjusted digital records.
Economic Implications and Policy Responses
The publication of Working Paper 35697 carries profound implications for academic researchers, antitrust regulators, and economic policymakers who increasingly rely on high-frequency alternative data sources to monitor economic health. As government agencies face mounting pressure to provide faster, more localized indicators of labor market tightness, the integration of big-data scraping with traditional survey methodologies has become standard practice.
For antitrust authorities such as the Department of Justice and the Federal Trade Commission—which frequently evaluate the labor market impacts of corporate mergers using measures of employer concentration—the findings serve as an urgent cautionary tale. Regulatory decisions grounded in unadjusted online job board data risk painting an exaggerated picture of employer market power, potentially misdirecting enforcement priorities or miscalculating the competitive dynamics of regional labor pools.
Furthermore, the study highlights the indispensable value of traditional statistical benchmarks like JOLTS in an era dominated by private-sector digital archives. While web-scraped datasets offer unmatched velocity and granularity, they cannot operate in a vacuum. Without periodic calibration against representative probability samples, big-data applications remain vulnerable to hidden structural biases. The reweighting framework introduced in this research provides a vital blueprint for future empirical work, demonstrating how modern data science can successfully reconcile high-frequency digital intelligence with rigorous economic sampling theory.
Conclusion and Future Research Directions
As the landscape of American employment continues to evolve alongside advancements in artificial intelligence, automated recruiting, and remote work, the tools used to measure labor market health must evolve in tandem. NBER Working Paper 35697 successfully bridges the gap between the speed of big-data analytics and the rigorous representativeness of official government surveys.
By proving that aggregate skill demand trends remain reliable while uncovering a threefold overstatement in labor market concentration, the study recalibrates our understanding of digital recruitment data. Future economic research will undoubtedly build upon these reweighting methodologies, ensuring that policymakers and analysts can continue to harness the power of online job postings without falling prey to the systemic blind spots of digital visibility.
