Embracing artificial intelligence in the labour market: the case of statistics

The paper examines how artificial intelligence reshapes labour markets with a focus on the role and resilience of statistics as a traditional discipline. It situates AI as an interdisciplinary, rapidly evolving technology with significant economic impact and widespread industry adoption. The authors note that while computer science has historically dominated AI, modern AI integrates multiple disciplines, including statistics, mathematics, economics, and others. Despite extensive research on AI’s labor impacts, there is a lack of empirical studies on AI roles beyond computer science. The study selects statistics due to its foundational role in AI (probability and inference), broad industry presence, and cross-disciplinary applicability, aiming to empirically map demand trends, interdisciplinary integration, skills and credentials, substitution risks, the relative value of education versus experience, and geographic/sectoral heterogeneity for statistical AI talents.

Prior work has documented AI’s economic and labor-market impacts, including productivity and job polarization effects, but tends to concentrate on AI within computer science or remains speculative regarding traditional disciplines. Studies have highlighted AI’s interdisciplinary character and sectoral adoption (healthcare, automotive, finance) and the growing demand for AI and ML skills. However, empirical labor market evidence on AI roles outside computer science, including concrete skill and credential requirements, remains limited. This paper addresses these gaps by using large-scale job postings and actual hiring data to analyze AI-related roles in statistics, quantify demand over time, map interdisciplinary clusters, and assess evolving skills and education/experience requirements.

Data sources: The study uses Emsi Burning Glass (BG) online job postings (279.87 million U.S. postings, 2010–2022, covering nearly all online and ~60–70% of offline vacancies) and Revelio Labs actual employment profiles (2010–2022) for 262,585 U.S. companies with over 2 million AI positions (reported as 2,090,658). BG provides occupations, employers, locations, required skills, certificates, and full text of postings; text is parsed to extract standardized skills and requirements. Revelio provides employee-level attributes (gender, degree, skills, tenure, seniority, salary). Representativeness of BG is supported by alignment with U.S. Census and BLS patterns in prior research. Research process: (1) Data matching and cleaning: Merge BG sub-databases by unique posting ID to attach skills, certificates, and discipline requirements; standardize synonyms (e.g., “statistics” vs “stats”); clean missing/outliers. Merge Revelio positions and skills by position ID. (2) Identify AI positions in statistics: Label postings as AI-related if skills include core AI concepts: AI, machine learning, natural language processing, or computer vision (stringent annotation following prior work). Within those, identify postings requiring statistics and co-listed disciplines. The skill taxonomy covers 10,000+ standardized skills aggregated into clusters. (3) Analysis and visualization: Aggregate by year and discipline; visualize with heatmaps, network graphs, and combined charts. Construct an interdisciplinary network of disciplines co-occurring with statistics in AI roles; identify clusters and quantify integrity via modularity. Disciplinary cluster algorithm: Build weighted network with nodes as disciplines co-listed with statistics in AI postings; edge weights reflect co-occurrence strength. Apply a heuristic modularity optimization algorithm (Louvain-style) in two iterative phases: (i) locally move nodes to maximize modularity gain ΔQ; (ii) aggregate discovered communities into super-nodes to form a new weighted network, repeating until modularity converges. Modularity Q evaluates community structure based on observed vs expected edge weights. This yields hierarchical, stable disciplinary clusters linking statistics to other fields. Outputs: Time-series of AI demand within statistics; distribution of job titles; cluster composition and evolution; skills and certifications trends; endangered jobs/skills; education and experience requirements; geographic (state-level) and industry (NAICS) heterogeneity; comparison to actual hiring data from Revelio.

- Demand magnitude and growth: 11.16% of AI postings require statistical expertise (second to computer science). From 2010–2022, statistical job postings grew 9.73× overall; non-AI statistical postings grew 3.12×; AI-related statistical postings grew 31×. Pre-2016 annual growth for AI-related statistical postings averaged 35.71% (spikes: 2013 at 50.61%, 2014 at 44.62%); post-2016 slowed to 29.64%; in 2021–2022 it re-accelerated to 47.11%. Actual AI hiring grew 2.67× (postings growth outpaced hires, indicating a talent shortfall). - Job role diversification: By 2022, 932 distinct AI-related job types within statistics. Top roles include Data Analyst (16.12%), Machine Learning Engineer (5.96%), Machine Learning Scientist (4.34%), Risk Analyst (3.25%), Business Analyst (2.48%), and Marketing Manager (2.29%). Social science roles outnumber science/IT roles in postings, with shifts from Statistician to market-oriented roles; in IT, ML Engineer/Scientist grew. Actual hiring still shows Software Engineers and Data Scientists as prominent; Historians appear highly represented among social sciences roles using AI tools. - Interdisciplinary clusters: Number of discipline clusters with statistics rose from 49 (2010) to 190 (2022). Overall shares: Computer Science (24.52%), Mathematics (16.26%), Applied Mathematics (5.65%), Economics (13.35%), Business Administration and Management (5.48%), Physics (3.53%), Engineering (3.11%). Over time: 2010–2016—CS 20.31%, Math 16.16%, Economics 14.50%, Operations Research 5.40%; 2017–2020—CS 24.7%, Math 15.68%, Economics 12.44% with Applied Math and Physics rising; 2021–2022—CS 25.79%, Math 16.83%, Economics 13.82%. Four stable clusters emerge: (1) statistics–mathematics–CS (engineering, automation, robotics); (2) economics–business management with health links; (3) finance–accounting–operations research–MIS; (4) genetics–bioinformatics–biostatistics–applied math–physics. - Skills emphasis: 3,987 skills (6,324,462 instances) in AI recruitment for statistics; strong tilt to hard skills. Among top skills, hard skills dominate (e.g., machine learning, Python [2nd], SQL [4th], SAS [8th], predictive modeling [7th], data analysis [11th]). Research is the leading soft skill (~2.25%). Trend shifts: from SAS/Excel toward Python/Tableau; SQL remains consistently important. Teamwork is the only soft skill trending upward. Actual hires show only 4/30 top skills are soft; additional hard skills (C++, C, MATLAB, HTML) are salient. - Certifications: 399 certifications cited; most demanded include Project Management Certification (8.79%) and Financial Risk Manager (7.96%); growing interest in IT certifications (e.g., CISSP, MCSA). - Endangered jobs and skills: Fastest-declining jobs include Actuaries (3.27%→1.44%), Computer Programmers (1.59%→0.40%), among others, reflecting automation of repetitive tasks and industry shifts. Declining skills include SPSS (1.16%→0.17%), CHAID (0.23%→0.01%), Microsoft Access, spreadsheets, SAP BusinessObjects, database marketing, OLAP—signaling replacement by newer AI-driven tools. - Education vs experience: Postings primarily require bachelor’s degrees (66.85%); master’s 25.68%; Ph.D. 6.35%. Average required experience ~3.9 years and trending upward. Over periods, bachelor’s share rose (59.48%, 61.76%, 74.25%) while master’s fell (31.45%→19.56%) and Ph.D. fell (8.18%→4.62%). Actual incumbents are more highly educated (bachelor 25%, master 42%, doctoral 33%), indicating postings de-emphasize degrees relative to practice, with growing weight on experience. - Geographic and industry heterogeneity: Highest state ratios of AI talent among statistics employees: Washington (32.6%), Idaho (29.3%), North Dakota (28.7%); lowest: Alaska (17.4%), Maryland (19.4%), Nebraska (19.5%). Industry AI ratios: retail (37%), information (30.8%), management of companies and enterprises (29.2%), manufacturing (28.1%); lower in agriculture/forestry/fishing/hunting (13.3%), construction (13%), educational services (10.3%). Growth rates of AI jobs exceed non-AI jobs in both high-tech and non-high-tech sectors (approx. 73.2% vs 72.7% average annual for AI vs much lower 14.7% and 18.3% for non-AI). In entertainment and arts, statistical non-AI jobs grew 9.7× (23,307→226,397) and AI jobs 22.6× (30→677), though absolute AI numbers remain small.

The findings demonstrate that statistics remains central in the AI-driven labor market, addressing all research questions: demand for AI statistical talent has surged and diversified; statistics is deeply integrated across disciplines, forming four stable clusters that span engineering, management, finance, and bioinformatics; employers increasingly prize hard, technology-oriented skills and targeted certifications; certain traditional occupations and tools are waning under AI-driven automation; practical experience is often weighted more than advanced degrees in postings; and AI demand varies substantially across states and industries, beyond mere high-tech spillovers. Collectively, the results counter pessimistic narratives about traditional disciplines’ obsolescence, showing that statistics evolves by supplying methodological frameworks to AI applications and by embedding across sectors. This underscores the importance of interdisciplinary training, continuous upskilling, and skills-based hiring to align workforce capabilities with AI’s practical deployment.

This study maps the integration of statistics into the AI labor market using large-scale postings and actual employment data. It documents a 31-fold rise in AI-related statistical postings, expansive role diversification (932 roles), and the emergence of four robust interdisciplinary clusters. It identifies a decisive shift toward hard skills and selective certifications, highlights endangered jobs and legacy tools, and clarifies that employers prioritize experience alongside bachelor-level education, even as incumbents often hold advanced degrees. The geographic and sectoral analysis shows heterogeneous demand not confined to high-tech spillovers. The paper contributes an empirical framework for understanding how traditional disciplines can adapt within AI ecosystems and offers actionable guidance for individuals, firms, and policymakers. Future research should extend beyond statistics to other traditional disciplines, include lower-cognitive-demand occupations, and explore cross-disciplinary synergies with finer-grained task-level and longitudinal analyses as AI capabilities evolve.

The analysis focuses on occupations with higher cognitive skill needs and on statistics specifically, so generalizability to lower-skill occupations (e.g., trades, drivers) may be limited. The BG and Revelio datasets, while extensive, rely on online postings and profiles and thus may not capture all labor market activity; some fields contain synonym standardization and potential annotation errors. Actual hiring data and postings differ in degree distributions, indicating potential selection or reporting biases. Raw data cannot be shared due to confidentiality, though figure-level underlying data are available upon request. Future work should broaden disciplinary scope, incorporate additional data sources, and examine causality between AI adoption and specific job/skill dynamics.