Skill generality
The distinction between general and specialized skills is widely acknowledged but rarely empirically quantified in human capital theory40,41,42,43,44,45,46,47,48,49. In this study, we quantify, analyse and classify the generality of skills on the basis of their breadth of application across occupations by analysing publicly available survey data from the US Bureau of Labor Statistics (BLS). These surveys provide detailed observations on job requirements for nearly a thousand occupations, including the importance and required level of each skill, knowledge or ability necessary for workers to perform their tasks. Each is numerically rated: skill’s importance is rated on a scale from 1 to 5 to indicate ‘not important’ to extremely important, while the required level ranges from 0 to 7, indicating the low to high level of expertise needed for a given occupation. For instance, while speaking is essential for both lawyers and paralegals, lawyers who argue in court require a higher level of proficiency than paralegals, who need only an average level29. We chose skill level as our primary indicator of skill demand—measuring how many occupations require each skill at varying levels of proficiency—because advanced skills typically develop through specialized training and education rather than being immediately applied, and thus, it better captures the sequential nature of skill acquisition and development. Nevertheless, these two measures are empirically correlated50, and our results remain robust even when considering skill importance.
Figure 1a–c shows the average skill level demand, from which we infer skill generality by their distribution profiles. Some distributions are skewed (blue on the left panel), while others are more centred (red on the right panel). Based on these shapes, we group skills into three categories—specific, intermediate and general—and arrange them along a specificity gradient, represented by the arrow at the top. The insets offer examples of these categories to aid our understanding. Specific skills (for example, dynamic flexibility and programming) are characterized as skewed demand distributions that peak near zero with a long tail. This skewed distribution shape indicates that most jobs require little or no proficiency in these skills, with only a few specialized roles requiring higher levels at the distribution tail. Meanwhile, general skills (for example, English language and oral expression) are needed at elevated levels (3–4) across a wide range of jobs owing to their broad applicability.
These demand distributions lay the groundwork for empirical definitions of skill generality, independent of their broader socio-economic contexts. General skills are those required by the majority of jobs at proficiency level 3 or higher, while specific skills are needed at this level in only a limited number of specialized roles, as most jobs need almost no proficiency. In line with this approach, we introduce two additional measures of skill generality—median skill levels across occupations and later-defined network-based metrics. All three approaches demonstrate strong consistency. Median skill levels group skills into three categories: general skills (median level 3.34), intermediate skills (2.37) and specific skills (0.87). In the following section, we introduce network-based measures such as local reaching centrality and nestedness contributions for generality and dependencies, respectively51,52.
We streamline our analysis by consistently colour-coding skills according to their generality: general (red), intermediate (grey) and specific (blue). Supplementary Table 1 provides the full list of skills—31 general, 43 intermediate and 46 specific, accompanied by various grouping methods. Our findings remain robust across various group sizes, clustering methods and unit choices (Supplementary Section 1).
Skill dependencies
The uneven distributions of skill demand that we observe create the conditions for a second-order structure beyond individual job requirements: a nested skill–occupation hierarchy. In this structure, jobs requiring specific skills are often subsets of those that demand general skills, indicating that specific skills tend to complement general ones rather than stand alone, while general skills remain largely independent. These asymmetric co-occurrences suggest that the demand for specific skills often depends on the presence of general skills, but not necessarily the reverse. This nested skill–occupation hierarchy, manifested in job co-requirements, reveals a hierarchy of skill dependencies that shape career trajectories.
Most jobs require a proficiency level of at least 2 in general skills covering a broad range of roles, including entry-level positions. As a result, workers typically begin their careers in roles that rely on broadly applicable abilities before transitioning directly into more specialized positions. In these transitions, certain skills—most often general ones—serve as prerequisites for more specialized roles. For example, a software engineer may need advanced programming skills (specific) in addition to written comprehension and problem-solving and communication skills (general), whereas a customer service representative who relies on communication and problem-solving skills (general) does not necessarily need programming skills (specific). These skill hierarchies are evident in job co-requirements.
Therefore, we calculate conditional probabilities between different skills in occupation requirements: how often occupations that require one skill also demand another. As previously discussed, the relationship is not necessarily reciprocal. If one skill is a prerequisite for another—similar to how foundational courses precede advanced ones in educational curricula—it is more likely that an occupation will require the prerequisite given that the advanced skill is needed than vice versa. This asymmetric conditional probability suggests a directional dependency between the two. We operationalize our methods by comparing p(skillA∣skillB), the conditional probability of requiring skillA given the presence of another skillB, with the reverse condition, p(skillB∣skillA). Only when p(skillA∣skillB) >> p(skillB∣skillA) do we assign a hierarchical direction from skillA to skillB as a substantive conditional direction to indicate that application or acquisition of skill B is more likely to be contingent on skill A (ref. 33).
Figure 1d illustrates our inference method with a few examples. For instance, the conditional probability of requiring math skills (whose distribution peaks at 2.5), given the need for programming skills (which skews towards zero, indicating that only few jobs require programming at this level), is greater than the reverse, that is, p(skillmath∣skillprog) >> p(skillprog∣skillmath). Accordingly, we assign a direction math → programming. Note that not all specific skills necessarily depend on general ones. In some cases, such as the pair between dynamic flexibility and math skills in Fig. 1d, skills may appear independently and yet still co-occur by random chance. We apply a z-score threshold to remove the random noise and retain only meaningful skill dependencies in Fig. 2 (Methods and Supplementary Section 3).
Figure 2a,b show the resulting dependency network in its backbone and the full connections, respectively (see Supplementary Figs. 15 and 16 for labels). Nodes are coloured by generality group as in Fig. 1 and positioned on the basis of educational requirements (x axis) and the local reaching centrality (y axis), which indicates the number of other skills reachable from the focal skill51 (Methods). Reaching centrality in this network quantifies the volume of dependent nodes (skills) connected to the focal skill and, thus, can indicate its hierarchy order (skill generality). Indeed, we find reaching centrality strongly correlates with the skill’s median level measure (Ï ≈ 0.71).
Overall, the hierarchical structure in Fig. 2a aligns with our common understanding of specialization. For instance, to develop programming skills (blue node at the bottom), one must first have a general knowledge of math and systems analysis (grey in the middle), which themselves rely on deductive/inductive reasoning (red at the top). Similarly, negotiation skills (grey in the middle) are contingent on systems analysis (grey in the middle) and oral expression (red at the top). These relations constitute the nested dependency chains running from top to bottom in the hierarchy direction in Fig. 2. In the following sections, we will provide a more technical definition of the nested structure, originally developed in ecology52,53,54, and explore the implications of these nested dependency chains for wage premiums, career trajectories and skill entrapment.
It is possible that the inferred directionality does not fully account for the underlying microprocesses and mechanisms driving skill acquisition. These dependencies may arise not only from the natural progression of individuals’ learning, but also from skill requirements imposed by firms as employees advance in job seniority. Separating these factors would require further microlevel analyses or field studies. However, our additional analysis suggests that job seniority alone does not explain our findings. To test this, we excluded skills and jobs typically associated with seniority, such as social and management skills, as well as occupations with management titles, and found that our results remained robust even with these exclusions (Supplementary Section 9.1). Therefore, the observed nested structure in career trajectories is probably driven by broader acquisition dependencies. This reasoning is especially so in today’s complex economy. With its diverse range of specialized skill demands, individuals would not pursue advanced skills—often requiring non-zero effort—unless those skills were necessary. Thus, the inferred directionality from the nested structure reflects acquisition dependencies that shape career development.
Our detailed analysis further supports the manifestation of acquisition dependency in individual career progression. First, we examine the career paths of registered nurses who transitioned to nurse practitioners and find that skill and wage differences in their resume data align with nested dependency chains in the human capital structure (Supplementary Sections 3.4). Furthermore, our analysis of the career trajectories of Hispanic immigrants shows that limited proficiency in certain general skills (for example, English language) limits the development of dependent specialized skills, resulting in demographic skill entrapment (Supplementary Section 3.5). Lastly, we analyse sequential datasets, including occupational ages, survey participants’ ages and job sequences in resumes, to further support our findings in the next section.
Skills’ structural alignment with nested architecture
We now quantify the degree to which the overarching skill–occupation structure is nested and how individual skills align with this nested architecture. We first use nestedness (N), originally developed in ecology, to quantify to what extent specialists preferentially engage with generalists, which we translate into interactions between specific and general skills52,53,54,55,56,57. Basically, N measures how often specific skills are demanded in occupations that also require general counterparts more than by random chance. There are a number of different definitions of N, such as the overlap index (Nc), checkerboard score, temperature and nestedness metric based on overlap and decreasing fill (NODF), and we test them to ensure the robustness of our analysis in Supplementary Section 2.
We then assess each skill’s alignment with the observed nestedness by its contribution score (cs). The contribution score (cs) for each skill s is calculated as ((N- langle {N}_{s}^{* } rangle )/{sigma }_{{N}_{s}^{* }}) (ref. 52), where (langle {N}_{s}^{* } rangle) and ({sigma }_{{N}_{s}^{* }}) are the average and standard deviation of nestedness across an ensemble of random counterparts, respectively. For each skill s, we generated 5,000 counterparts ({N*}) in which its dependencies are randomized (that is, skills are randomly distributed across occupations as if there were no conditional probability structure) while its generality (demand distribution shape) remains constant (Methods).
Therefore, the contribution score (cs) serves to differentiate skills that align or misalign with the observed nested dependency structure. A positive cs indicates that the dependencies of skill s are aligned with and, thus, positively reinforce the overall nested hierarchy. Conversely, a negative cs suggests that the skill’s dependencies are misaligned, diminishing the nested structure (N). For example, skills such as negotiation, programming and fine arts are embedded in strongly interdependent branches of the hierarchy in Fig. 2 (represented by blue and grey nodes on the left), which explains their positive contribution scores (cs > 0) in Fig. 3. Meanwhile, skills such as mechanical, repairing and dynamic flexibility appear in loosely dependent branches (blue nodes on the right), and thus their dependency structure diminishes the overall nestedness (cs < 0). In this way, cs quantifies how much a skill’s dependencies reinforce or diminish the overall hierarchical structure.
Figure 3 and Supplementary Fig. 10 show how the structural attribute cs correlates with key socio-economic attributes such as generality, education, wages and even automation risk. First, general and foundational skills strongly align with the hierarchical structure of human capital in Fig. 3a. Second, the greater a skill’s alignment with this nested structure (that is, the higher the cs), the higher the required education level and expected wages in Fig. 3c,d and Supplementary Fig. 10. To illustrate the implications for career development and reskilling, we present skills’ arrival probabilities in Fig. 2c,d (ref. 58). For repairing (cs < 0), there are fewer nodes that reach it through nested dependent pathways, compared with skills such as programming and negotiation, which have positive cs. With significant implications for wages and education, these nested pathways, therefore, may contribute to disparities in demographics and opportunities59.
Our analysis underscores the importance of considering both skill generality and its structural alignment with the nested hierarchy of human capital (cs) when assessing socio-economic implications. As previously discussed, career paths often begin with jobs requiring general skills and transition into more specialized roles that demand additional specific skills. These specialized skills can be divided into two categories: nested-specific skills (cs > 0) that align with the nested hierarchy and typically require higher education with greater economic rewards, and un-nested-specific skills (cs < 0), which fall outside the structure and tend to miss out on increasing economic benefits as careers progress. To streamline our analysis, we have categorized skills accordingly. In the next section, we will explore how nested skills unfold along career trajectories and how they interact with factors such as occupational age, wage premiums and skill entrapment.
Skill categories in career trajectories
We analyse three datasets of occupational sequences to examine how skill levels evolve across various categories: occupation sequences by their median ages, by following synthesized birth cohorts from individual surveys, and by job transitions in resumes. Each dataset offers unique strengths and limitations, which, when combined, provide a coherent picture of both nested and unnested career trajectories.
We begin our analysis with the median ages of occupations, assuming that skill progression and development are closely correlated with the median age of workers in those occupation groups28,60,61. Figure 4a–c shows the proficiency levels of general, nested and unnested skills across occupations, segmented and arranged by the median age of workers in those roles (Methods). Consistent with previous findings, the data indicate that skill demand shifts towards nested skills as individuals advance in their careers: occupations with older workers (median age over 30 years) tend to require higher levels of both general and nested skills compared with those with younger workers. It also suggests that, as workers age and progress in their careers, these skills become increasingly essential to meet job demands. Meanwhile, the demand for unnested skills remains relatively stable across all age groups, indicating that their necessity is not influenced by worker age.
Our results hold when examining career trajectories constructed from synthetic birth cohorts of full-time respondents aged 17–55 years in the Current Population Survey (CPS) (Fig. 4d–f). Because the CPS provides yearly cross-sectional data rather than long-term longitudinal tracking, we connected birth cohorts across surveys to emulate career trajectories. For example, we created a 1967 cohort for the insets and repeated this process for all birth cohorts in the surveys between 1980 and 2022 (Methods)10,11. These results are consistent with our earlier findings with a more detailed pattern. Age 30 emerged as a marked turning point. Up to this age, there is a sharp increase in both general and nested skill levels, while unnested skills show a moderate decline. After age 30, the overall changes in skill levels tend to stabilize.
The second dataset includes additional demographic information about the respondents, allowing us to decompose our findings by gender and, later, by race. For example, analysing skill trends by gender reveals a gap in specializations that emerges around age 30, as shown in Fig. 4d–f. While men continue to grow their general and nested skills until their 50s, for women, the growth in these skills plateaus in their early 30s, which coincides with the typical age range for first-time mothers in the USA. We further analyse the influence of parenthood on male and female workers with and without children, as well as working schedules and hours in Supplementary Sections 7.1 and 7.2. Our findings are robust to controlling for yearly economic conditions (Supplementary Fig. 28) and educational attainment (Supplementary Figs. 29 and 30).
Lastly, we analyse over 70 million individual job transitions recorded in 20 million resumes. Figure 4g,h shows that the results of these direct observations are again consistent with our two earlier findings. The baselines for general and nested skill levels are shifted because the resume dataset favours more nested job roles. Once again, career trajectories (the ith job transition) show a steady accumulation of both general and nested skills (Δi > 0) during the first five transitions. In addition, Fig. 4i indicates that general skills advance faster than nested skills early in a career (({varDelta }_{i < 3}^{{mathrm{general}}}gg {varDelta }_{i < 3}^{{mathrm{nested}}})), and even later, their growth remains comparable to specific skills (({varDelta }_{i > 3}^{{mathrm{general}}}approx {varDelta }_{i > 3}^{{mathrm{nested}}})). We provide skill growth along randomized sequences (grey triangles), ({varDelta }_{i}^{{mathrm{random}}}approx 0), to confirm that the observed job sequences explain their corresponding skill growth curves (Supplementary Section 4.1).
All three empirical observations consistently depict career paths where both general and nested skills (cs > 0) grow, while unnested ones (cs < 0) remain relatively underdeveloped or even decline. Moreover, the results prompt interesting recurring patterns. First, skill advancement extends well beyond formal education, suggesting that nested specialization pathways persist throughout a career, beyond the schooling phase (Supplementary Figs. 29 and 30). This observation challenges the traditional view that education is the primary driver of human capital development11,40,62,63,64.
Second, the findings question the simple linear progression model, which assumes that basic general skills precede advanced specialized ones, as is often portrayed in economics, sociology and psychology and even in our study’s initial assumptions28,65. Instead, we observe that general skills continue to develop alongside specific ones, even after workers enter specialized roles. For example, critical thinking is a prerequisite for transitioning into roles requiring new skills such as negotiation, as indicated by the asymmetric conditional probabilities in Fig. 2. However, even within a negotiation role, further advancement depends on honing higher levels of critical thinking. Thus, career trajectories unfold through nested specializations, where the ongoing development of general skills supports their dependent, nested counterparts, thereby echoing the cumulative and sequential nature of skill development.
Next, we explore the connection between each specialization pathway and wage premiums. Figure 5a,b shows that both educational requirements and average annual wages rise with proficiency in nested specific skills. However, as shown in Fig. 5c, these observed wage premiums associated with nested skills almost fully disappear when controlling for an occupation’s general skill requirements. This finding underscores the previous argument that specific skills complement general ones rather than stand alone, while general skills remain largely independent. Therefore, even in specialized roles, a strong foundation in general skills remains essential. That said, the wage penalties associated with unnested skills (cs < 0) in Fig. 5 turn into wage premiums of a magnitude comparable to those of nested skills once general skill requirements are factored in. These findings suggest that, although unnested skills do hold value in the labour market, their potential wage premiums are diminished by a lack of dependency structure, almost like a penalty for missing that foundation. This finding provides structural insights into career development pathways and the broader evolution of workforce skills.
We present a series of tests to demonstrate that our findings remain robust even when controlling for education, training and workplace experience, as well as holding across subsamples of major occupational groups (Supplementary Section 5, Supplementary Table 5 and Supplementary Figs. 32–34 and 39). We also confirm our findings hold even when excluding managerial roles and social skills (Supplementary Sections 9.1 and 9.2 and Supplementary Figs. 62 and 64).
Disparity, skill entrapment and long-run wage penalties across demographic groups
We now examine how our skill taxonomy helps in understanding salient features of the labour market. Specifically, we analyse skill categories across demographic groups to determine whether the nested architecture in human capital explains socio-demographic inequalities, skill entrapment and persistent wage penalties. Figure 6a–e compares skill, education and wage differences between racial/ethnic groups and their white peers. The results show, first of all, notable wage gaps between Black and Hispanic workers on the one hand, and Asian workers and the baseline of white workers, on the other hand. These wage gaps are accompanied by employment in jobs with lower requirements for nested skills for Black and Hispanic workers. Notably, for Hispanic workers in particular, another critical factor emerges: an elevated requirement for unnested skills, which may contribute to skill entrapment.
To understand this finding in more detail, we examine a case study of how language-skill requirements may limit access to jobs requiring certain nested skills (Supplementary Section 3.5). We first distinguish between nested skills that depend on language skills (general) and those that do not, and find substantial gaps in language-dependent nested skills among Hispanic workers, particularly those who recently arrived in the USA. This disparity often leads to the development of unnested skills, creating skill traps associated with long-term wage penalties (Supplementary Fig. 31)23. Taken together, these findings suggest that closing wage gaps for Black workers and Hispanic workers may require tailored solutions.
Next, Fig. 6f–j shows skill gaps between men and women across social groups. The most pronounced disparities occur for specific skills: Except in the Asian subsample, women generally work in occupations requiring more education and higher levels of general skills than men do. However, women’s jobs do not require similar levels of nested skills. These disparities probably contribute to the well-known gender wage gap observed in the right-most panel. This gap has narrowed over time, as shown in Supplementary Fig. 52, but the disconnect between education and general skills on the one hand and wages and nested skills on the other remains puzzling.
To understand this in more detail, we examine another case study on parenthood and women’s tendency to work in jobs with more regular and predictable schedules that impact both wages and skill development in Supplementary Sections 7.1 and 7.2 (refs. 66,67,68). Our analysis shows that having children is associated with reduced general and nested skills for women. By contrast, men with children tend to have higher levels of these skills compared with men without children. A key factor appears to be work schedules: when we control for irregular hours and overtime, the gender gap in nested skill requirements decreases by more than a third.
Finally, we examine the geographic distribution of skills across categories. Our analysis shows that general skills tend to concentrate in densely populated urban areas (Supplementary Section 6), a finding consistent with previous research on the diversity and complexity of economic activity in large urban economies4,36,69,70,71,72,73. Moreover, differences in general skills account for about one-third of the well-documented urban wage premium74, which is associated with employment in large cities.
Increasing nested structure in human capital
Figure 7 shows the historical changes in the hierarchical skill structure between 2005 and 2019. Over this period, the skill structure has become increasingly nested, as indicated by a decrease in the checkerboard score (from 438.67 to 356.4) and temperature (from 40.07 to 31.89), alongside increases in NODF (from 39.06 to 41.72) and Nc (from 573,873 to 651,030)55,56. This trend towards a more nested structure is attributed to increasingly uneven job requirements, with growth in high-dependency branches and a decline in low-dependency branches. Workers with broad skill sets have seen higher wage premiums, probably driven by demands for nested specialization skills (general and nested specific skills with cs > 0), while demand for unnested skills has declined (Supplementary Fig. 40). These trends are perhaps attributed to the growing complexity and interdependence of the economy (Supplementary Fig. 57)7,75.
The growing divide between nested and unnested specializations raises concerns, echoing ongoing discussions on job polarization20,22,76,77,78 (Fig. 7c). Given the important role of nested skills in career progression and wage expectations, the growing divide potentially further reinforces demographic and regional disparities. Workers with insufficient foundational general skills often find themselves stuck in unnested specialization paths with limited opportunities for upward mobility (Supplementary Figs. 20 and 31)21,22,23,77. Thus, policymakers and educators must prioritize fundamental skill development across all demographic groups and regions to mitigate these structural disparities. In other words, investing in education and training programmes that cultivate general skills and unlock specialized pathways is essential to addressing this growing divide and improving economic mobility.