Understanding How to Use Job Postings Better in Predicting Hiring Activity

The NLx Research Hub recently released an important technical report that should get the attention of labor market information practitioners and workforce analysts who rely on online job posting data to understand labor demand. The study asks a fundamental question: How closely do online job postings reflect actual hiring activity across occupations?

In her recent article for the NLx Research Hub (www.nlxresearchub.org), economist Marissa Hashizume examined approximately 11.5 million expired job postings in the National Labor Exchange (NLx) and compared their occupational distribution with roughly 183.8 million worker job changes reported through the Current Population Survey (CPS) during 2025.

The premise is straightforward. If job postings serve as a reasonable proxy for hiring activity, then the occupational distribution of postings should broadly align with the occupational distribution of actual worker movement between jobs.

The findings are encouraging because an analysis of 22 major SOC occupational groups found that the differences between postings and job changes for most groups were relatively modest. Most occupational categories fell within two percentage points of one another across the two datasets. For practitioners conducting broad occupational analysis, that is an important finding. It reinforces that job posting data remains a valuable and timely tool for understanding labor market trends, particularly when combined with traditional workforce data sources.

At the same time, the report highlights several significant occupational exceptions that practitioners should interpret carefully.

Healthcare Practitioners and Technical occupations show the largest divergence. These occupations represented 14 percent of expired job postings but only 5 percent of worker job changes. The pattern appeared consistently across nearly every state. The most likely explanation is structural. Healthcare employers tend to rely heavily on online recruitment and are more comprehensively represented in the NLx data feed. As a result, healthcare demand appears disproportionately large in posting data relative to actual hiring activity.

Construction and Extraction occupations reveal the opposite issue. These occupations accounted for 6 percent of worker job changes but only 2 percent of expired postings. This gap likely reflects the nature of construction labor markets themselves. Many construction jobs are filled through informal hiring channels, subcontracting arrangements, union halls, or short-duration assignments that never appear in online job boards. This is not simply a data collection problem. It reflects real structural limitations in what online posting data can capture.

Computer and Mathematical occupations also stand out. They represented 7 percent of expired postings but only 4 percent of worker job changes. Remote work and multi-location recruiting appear to drive much of this difference. A single software engineering position may be advertised simultaneously across dozens of states. Although the NLx applies a “shadow job” methodology to reduce duplicate postings nationally, state and regional analyses remain vulnerable to overcounting within this occupational category.

The study also identifies important seasonal differences in Educational Instruction and Library occupations. Back-to-school hiring surges appear prominently in CPS job-change data but not in online postings because many education workers return to existing employers rather than entering newly posted jobs. That distinction matters for analysts interpreting education workforce demand trends.

For practitioners, the implications are practical and immediate.

First, workforce dashboards and analyses that rely heavily on job posting data may overstate relative healthcare demand and understate demand in construction and skilled trades. Policymakers and workforce boards should understand those limitations before drawing conclusions about labor shortages or training priorities.

Second, state and regional analysts need to be especially cautious when interpreting Computer and Mathematical occupation postings. Remote hiring patterns can substantially inflate localized demand signals if duplicate and multi-location postings are not carefully addressed.

Third, the report reinforces a broader methodological point that often gets overlooked in applied analysis. Job posting data is not a neutral mirror of labor demand. It reflects the recruitment behavior of employers that advertise online and whose postings are captured within available feeds.

Those selection biases matter.

This research represents an important step forward because it focuses on understanding the measurement properties of job posting data rather than assuming its accuracy. Further research on differences between postings and hiring behavior for detailed occupations may be even more enlightening. However, this kind of transparency about the value of job postings to workforce analysis is critical for the responsible use of increasingly influential workforce data tools.

For the LMI community, the takeaway is not that job posting data should be abandoned. Quite the opposite. These data remain timely, detailed, and geographically granular in ways traditional surveys often cannot match. But effective use requires informed interpretation, methodological awareness, and a clear understanding of where the data performs well and where caution is warranted.

As the NLx continues expanding its job feed and additional representativeness studies emerge, this line of research will become increasingly important for workforce practitioners, economic developers, policymakers, and researchers alike. CREC and the LMI Institute will continue monitoring these developments and helping practitioners better understand how to apply these tools responsibly in real-world workforce analysis.