Decade-Long Analysis of Skills Across 1 Million Graduate-Level Job Advertisements
Plain Language Summary
Job advertisements are among the few continuously available records of what employers say they want, and this study reads a decade of them to ask how expectations placed on mechanical engineers holding graduate degrees have shifted. Working from more than one million postings collected between 2010 and 2022, we track 76 recurring skills and fit each one's year-by-year prevalence to a small set of interpretable growth patterns. What emerges is a discipline still anchored in its technical core, e.g., design, drafting, and physics, but one in which employers increasingly expect engineers to manage budgets, contracts, suppliers, and regulatory compliance as well, and to conduct all of it through software. Because graduate programs can revise coursework on shorter cycles than undergraduate curricula allow, these trends offer them an evidence base for deciding what to teach next.
Contribution
We contribute a discipline-specific, decade-long account of graduate mechanical engineering labor demand by pairing a Bayesian-optimized, marker-based classifier that isolates graduate-level ME postings from an O*NET-SOC-filtered corpus with a Burnham-and-Anderson model-selection procedure that assigns each of 76 skills both an interpretable trajectory and an explicit measure of how confidently that trajectory can be claimed, moving past the short-term, cross-disciplinary snapshots that characterize most prior job-posting research.
Research Questions
- How has the prevalence of employer-articulated technical and professional graduate mechanical engineering skills changed from 2010 to 2022, and what do these changes reveal about employer demand for skills?
Methods
Working from the Burning Glass Technologies (now Lightcast) database, we filtered the 372 million postings spanning January 2010 through December 2022 down to 1,142,245 records carrying the O*NET-SOC code for mechanical engineers with complete skill data, then developed a confidence-scoring classifier, i.e., weighted markers for job titles, degree language, ME software, and industry terminology, with point values and threshold fit by Tree-structured Parzen Estimation against 534 manually classified postings, to isolate the 1,014,433 postings we treat as graduate-level ME roles (F1 = 0.86 on a held-out test set). Skills came from BGT's CanonSkillCluster field, from which we retained the 76 that appeared throughout the study period at an average annual prevalence of at least 2%. For each skill we fit five theory-driven candidate models, i.e., null, linear, logarithmic, exponential, and quadratic, selecting among them by corrected AIC under the parsimony principle while controlling for job description length, and we report alongside each selection both its discriminability from the runner-up model and whether the underlying regression assumptions held under Durbin-Watson, Shapiro-Wilk, and Breusch-Pagan testing.
Key Findings
Of the 76 skills tracked, "Drafting and Engineering Design" remained by far the most prevalent at roughly half of postings each year, and 16 skills met our high-demand threshold of at least 10% average prevalence. The steepest growth, however, sat outside that technical core: "Architectural Design" (+435%), "Regulation and Law Compliance" (+184%), "Product Management" (+138%), and "Contract Management" (+98%) led the percent-change rankings, joined by computational skills such as "Mathematical Software" (+137%), "Robotics" (+107%), and "Data Analysis" (+103%). Not everything rose, with "Industrial Design" falling 22% and "Project Management Software" 13%, which we take as evidence that the pattern reflects a genuine reshuffling rather than an artifact of postings simply growing longer over the period. We interpret the two dominant directions through a pair of idealizations drawn from the trends rather than from any actual engineer, the "business-aware technical liaison" and the "proficient software executor," while noting that the dataset closes in December 2022, days after ChatGPT's launch, so the software growth documented here precedes the current generative AI landscape rather than reflecting it.
Implications
Given that program directors face finite curricular space and cannot responsibly pursue every rising skill, we sort our recommendations by where institutions actually hold leverage: skills still climbing without a visible ceiling, e.g., simulation, mathematical software, statistics, data analysis, and software development principles, are those graduate programs can address directly through course modules, capstone requirements, and electives, taking advantage of the shorter cycles and greater elective flexibility that distinguish graduate from undergraduate curricula. Business-adjacent competencies like contract management and business development sit at the boundary of what an ME department can plausibly teach, so we argue for making them accessible through cross-listed coursework, graduate certificates, and dual-degree pathways rather than absorbing them into the major, while deeply workplace-contextual skills, e.g., supplier relationship management and shipping and receiving, are better served by co-ops and internships paired with deliberate attention to students' capacity for self-directed learning. We also identify saturating and declining skills, e.g., basic productivity software and particular project management platforms, as candidates for deprioritization, though this reasoning assumes a good deal about students' background knowledge, program flexibility, and access to bridge experiences that we do not want to take for granted.