Research Article
Navigating the skill diversity frontier: How skill complexity explains worker resilience
Synthesis: Using LinkedIn data on 2.4 million U.S. workers and 16,753 distinct skills, this paper introduces three complementary measures of skill complexity — specialization, diversity, and the diversity frontier — and demonstrates that workers near the frontier are significantly more likely to acquire new skills, receive promotions, and transition into occupations with lower automation exposure. The findings distinguish productive capital (specialization) from adaptive capital (diversity) and provide a data-driven framework for understanding workforce resilience in the era of AI-driven labor market transformation.
Key Findings
- Specialization → wages: Specialization is most strongly associated with sorting into higher-wage occupations, confirming that specialized skills constitute productive capital.
- Diversity → mobility: Diversity is associated with broader skill accumulation, occupational mobility, and transitions into occupations with lower AI exposure, consistent with its interpretation as adaptive capital.
- Frontier proximity → resilience: Workers closest to the diversity frontier — those achieving the greatest attainable diversity for their level of specialization — are significantly more likely to acquire new skills, receive promotions, and transition into occupations with lower automation exposure.
- Complementarity matters: Adaptive capacity depends on combining specialized expertise with broad capabilities, not on possessing either alone; frontier position is not associated with stronger wage sorting, reinforcing the distinctness of productive and adaptive capital.
Background: Technology Reshuffles Skills
In task-based models of labor market transformation, workers are endowed with skills while jobs consist of bundles of tasks that technology can substitute, complement, or augment. Rather than automating whole occupations, artificial intelligence recombines and reconfigures tasks across the skill distribution, so disruption occurs at the level of tasks and skills rather than occupations. Worker resilience therefore depends not only on whether current tasks are automatable, but on whether workers can reallocate existing skills or acquire new ones. Research on skill relatedness, complementarity, and skill networks shows that skills are interdependent capabilities: transitions are easier when new skills are related to existing ones, while movement across unrelated domains requires more costly investment.
The paper argues that skill complexity — the structure of a worker's skill portfolio within the broader hierarchy and domain system — offers a systemic perspective on worker adaptability. It decomposes human capital into two conceptually distinct components: productive capital, the specialized expertise valuable within existing labor market structures, and adaptive capital, the diversity across skill domains that provides flexibility and mobility under changing technological conditions. Resilience may therefore depend not on maximizing specialization or diversity in isolation, but on how much diversity workers achieve relative to their degree of specialization.
Framework: Three Dimensions of Skill Complexity
The authors reconstruct skill hierarchies directly from observed co-occurrence patterns:
| Dimension | Definition | Career Outcome |
|---|---|---|
| Specialization | Productive depth in a domain | Higher-wage occupations |
| Diversity | Adaptive breadth across domains | Skill accumulation, occupational mobility |
| Diversity Frontier | Maximum attainable diversity at a given specialization level | Promotions, new skill acquisition, automation-resilient transitions |
The diversity frontier arises because potential skill diversity is bounded from both directions: highly specialized workers are constrained by the cumulative investments required to acquire deep domain expertise, while highly generalist workers are constrained because broad foundational capabilities increasingly converge across occupations. As a result, the greatest scope for skill diversity emerges at intermediate levels of specialization. The general-manager versus radiologist contrast illustrates the point: generalists face converging foundational capabilities, while specialists such as radiologists face limits from the substantial investments required to maintain deep expertise.
Methods and Data
Using longitudinal LinkedIn data on approximately 2.4 million U.S. workers who listed at least five skills prior to the diffusion of generative AI (pre-2022), the authors construct a directed skill network from observed co-occurrence patterns. Statistically meaningful skill associations are identified against a hypergeometric null model, directionality is inferred from asymmetric conditional probabilities, and 12 broad skill domains are recovered via Louvain community detection. The resulting network contains 16,753 skill nodes and 556,206 directed edges. All explanatory variables are measured between 2020 and 2022, while career outcomes are measured between 2023 and 2025 to reduce concerns about reverse causality.
Three worker-level measures are derived. Specialization captures whether a portfolio is concentrated in downstream skills, computed as the average inverse normalized local reaching centrality. Diversity captures the breadth of domains represented in a portfolio, measured with Shannon-based Hill diversity across skill domains. Frontier position captures how much diversity a worker achieves relative to others with comparable levels of specialization, ranging from 0 to 1. Outcomes span three domains of worker resilience: reward outcomes (promotion and occupation-level wage sorting), adaptability outcomes (lateral transitions into roles with lower AI exposure), and technological adaptation (AI skill adoption and total skill acquisition).
Key Results
Consistent with Hypothesis 1, specialization is strongly associated with sorting into higher-wage occupations but only weakly related to adaptive outcomes, confirming that productive depth alone is insufficient for resilience. In line with Hypothesis 2, diversity is positively associated with skill acquisition, occupational mobility, and transitions into occupations with lower AI exposure. Most importantly, Hypothesis 3 receives strong support: workers closer to the diversity frontier display the strongest adaptive outcomes — particularly skill acquisition, occupational mobility, and AI skill adoption — while frontier position is not associated with stronger sorting into higher-wage occupations. Overall, worker resilience depends not on maximizing specialization or diversity independently, but on combining specialized expertise with the greatest feasible breadth across knowledge domains.
Demographically, specialization shows clearer stratification than diversity: men, younger cohorts, and workers with higher educational attainment are shifted toward more specialized portfolios, consistent with life-cycle human capital accumulation. AI skill adoption is most prevalent near the diversity frontier, suggesting that adapting to emerging technologies requires both specialized expertise and broad capabilities spanning multiple knowledge domains.
What this means for practice
- Policymakers. Design reskilling targets around frontier-proximate skill combinations rather than occupations or job titles: workers closest to the diversity frontier were significantly more likely to acquire new skills, receive promotions, and transition into roles with lower automation exposure.
- Policymakers. Fund breadth alongside depth across the whole life course: adaptive capacity depends on combining specialized expertise with the greatest feasible diversity, and complementary or dormant skills accumulated through prior education and employment are what continuous learning can activate.
- Policymakers. Treat access to frontier-expanding skills as an equity question: specialization was stratified by gender, age, and educational attainment, with men, younger cohorts, and more educated workers shifted toward more specialized portfolios.
- Researchers. Reuse the reconstructed network — 16,753 skill nodes and 556,206 directed edges over 12 domains — and its three measures (specialization, Hill diversity, frontier position) to evaluate training programs against promotion, lateral transition, and AI-skill-adoption outcomes.
- Researchers. Map existing portfolios to surface adjacent transitions rather than recommending entirely new careers, and test whether AI literacy functions as a frontier-expanding meta-skill, since AI skill adoption concentrated near the frontier.
Limitations
The analysis relies on LinkedIn profiles that over-represent digitally engaged and highly skilled occupations; listed skills are self-reported and may capture signaling as well as actual capabilities; the skill network depends on modeling choices including pruning, community detection, and the measurement of hierarchy through local reaching centrality; and the results remain associational rather than causal despite lagged explanatory variables and extensive controls. The diversity frontier should be interpreted as a relative measure of portfolio structure rather than an independent dimension of human capital.
Citation
Carpanelli, M., Duszynski, J., & Stephany, F. (2026). Navigating the skill diversity frontier: How skill complexity explains worker resilience. v1.