A new data product for occupational skills: methodology, analysis, and a guide to using the employment projections skills data

This article provides an overview of the new data product and the methodology used to create it. In addition, the article provides information on how to use the product, the potential uses of its data, and definitions of skills categories that can assist data users with understanding the types of information captured in each category. The article concludes with analyses and highlights of the data for the 2023–33 projections cycle.

Methodology

To prepare the framework for the skills data, EP analyzed various data sources, conducted factor analysis, and reviewed feedback gathered through cognitive testing with potential data users. The resulting framework includes 17 skills categories, with data from the Occupational Information Network (O*NET) database (sponsored by the U.S. Department of Labor) serving as the foundation for determining the importance of skills by detailed occupation.⁴

This section discusses the EP-developed definition of skill; the data sources and process used to develop the EP skills categories; the approach used to map detailed O*NET elements to the skills categories; and the methodology used to create skills scores and ratings, including the approach used to provide a score for occupations for which data were not available.

Definition of skill

After analyzing various data sources and definitions of skills, EP developed a definition of skill for the skills data released alongside the employment projections. EP defines a skill as a general human capacity that is developed (learned over time through education, training, and/or experience), general (applicable across many occupations), applied (involving action beyond cognitive knowledge of a subject), and work related (limited to performance in employment).

Framework of EP skills categories and mapping of O*NET elements

To provide data users with information on the importance of skills needed for various occupations, EP created a framework of skills categories. The initial framework was developed by analyzing several data sources, including the O*NET database, the OOH, the BLS Occupational Requirements Survey (ORS), the Program for the International Assessment of Adult Competencies (PIAAC), Multipurpose Occupational Systems Analysis Inventory—Close-Ended (MOSAIC) competency studies, and Lightcast data for job postings. (More information about these sources is available in the appendix.)

EP evaluated the information from these sources to determine the level of detail for the skills categories, taking into account the terminology used in other sources. EP determined that a framework with 15–20 categories would provide sufficient detail, without overburdening potential data users. Initially, the framework included 15 skills categories, but after further testing and analysis, it was expanded to a total of 17 categories.

EP used professional judgment, factor analysis, and cognitive testing to map detailed O*NET elements to the EP skills categories. To create the mapping, EP first created an initial mapping between the detailed O*NET elements and the initial skills categories, using professional judgment and the EP definition of skill. O*NET’s Content Model provides a framework that identifies the most important types of information about work, integrating them into a theoretically and empirically sound system. The model provides information about worker characteristics, worker requirements, experience requirements, occupational requirements, workforce characteristics, and occupation specifics. EP examined information from the model to determine which of its elements met the EP definition of skill and could serve as inputs to the initial mapping. Using this analysis, EP selected O*NET elements from the following element types: abilities, knowledge, skills, work activities, work context, and work styles.

Next, EP evaluated and refined the initial mapping by using factor analysis. Factor analysis is a statistical method that reduces the number of observed variables to a smaller number of unobserved factors. This method was used to explore whether each of the skills categories potentially contained more than one skill and to confirm that the skills category was capturing a single underlying concept. In three cases, factor analysis strongly indicated two distinct subcategories, and the initial skills framework was expanded to reflect this feature of the O*NET data.⁵ EP also mapped some of the less clear connections between O*NET elements and skills categories by examining the correlations between a given element and all the factors.

After developing the initial mapping, EP analyzed the results of cognitive testing to evaluate customer understanding of skills concepts and categories, including the mapping of detailed O*NET elements to summary-level categories. The cognitive testing was conducted through personal interviews, card sorting, and feedback from the public and experts in the field. Testing participants included career counselors, career services professionals, and potential skills-data users who were not familiar with EP data, the OOH, or O*NET.⁶ For the card-sorting test, participants were provided with a list of detailed skills and the summary-level skills categories and were asked to map the detailed skills to one summary-level category. The results of the cognitive testing, together with those obtained from the factor analysis and professional judgment, informed the final terminology used for the skills categories and the EP mapping.⁷

For the 2023–33 projections, the final skills framework is composed of 17 skills categories, with 104 O*NET elements mapped to them. For each skills category, EP developed a definition (how each category is defined) and mapping of O*NET elements (how the detailed O*NET elements are mapped to the framework). Table 1 outlines the 17 skills categories, their definitions, and their mapping to O*NET elements.

Calculating a skills score

EP used the O*NET database as the primary source for calculating skills scores, providing a comprehensive coverage of EP occupations. Across the O*NET element types included in the skills framework, the domains for abilities, knowledge, skills, and work activities all contain ratings for both importance and level of skill. The domain for work styles contains only an importance rating. The domain for work context contains scores for various categories, including importance, frequency, and, for some elements, unique scales. For the analysis of the EP data product, the O*NET importance score was selected as the core data feature for several reasons. First, it had the most coverage across O*NET elements. Second, it was consistent with all O*NET elements that use it, which means that a skill with an importance score of 4 (on a 5-point scale) would translate across occupations. Finally, it was conceptually more straightforward than other O*NET concepts, such as level.

For most of the O*NET elements mapped to the EP skills framework, an importance score was available and used to calculate a skills score. However, 14 of the 104 O*NET elements were in the work-context domain, which includes elements of physical and social factors that affect the nature of work. Although an importance score was not available for every work-context element, all work-context scores were on a scale from 1 to 5, and EP determined that the meanings of the scores across that scale were similar to the meanings of the scores across the importance scale. After analyzing the work-context values, EP determined that, for the work-context elements mapped to the EP skills framework, the work-context score would be used to create a skills score.

Using the importance score (or the work-context score for the work-context elements), EP created a skills score on a 1.0–5.0 scale for the EP skills categories of each occupation with published projections. The average score across the detailed O*NET elements was used to calculate a skills score for each EP skills category. Then, the average skills score for the detailed occupations, weighted by employment, was used to calculate the skills score at the occupational group level.

The skills data tables provide a numeric skills score, which translates to a rating. EP used O*NET’s questionnaire to develop the rating scheme for the 5-point scale. According to this scheme, for a given occupation, skills scores greater than or equal to 4.5 correspond to a skills rating of “extremely important,” whereas skills scores of less than 1.5 correspond to a skills rating of “not important.” (See table 2.) These values are based on responses to the O*NET survey questions about how important an O*NET element is to a respondent’s current job. The options for response fall on a scale from 1.0 (not important) to 5.0 (extremely important).

Treatment of occupations for which skills data are not available

The Standard Occupational Classification (SOC) system contains residual occupations that include a variety of jobs that do not fall within defined occupations. These residual occupations, which usually include “all other” in their titles, are not subject to O*NET data collection, so O*NET data for them are not available. Because EP produces employment projections for residual occupations, it imputes the skills scores for them. The underlying assumption behind this imputation is that the residual occupations are similar to other occupations located in a similar position (broad or minor group) in the SOC structure. The skills scores for the residual occupations were calculated by using an employment-weighted average of the skills scores of similar occupations in the corresponding broad or minor occupational group.⁸

In addition, there were 38 nonresidual occupations for which O*NET had some data, but these data were not in the relevant domains required to create a skills score. For these occupations, EP imputed missing data by using information for the highest ranked O*NET-SOC occupation in the O*NET dataset for related occupations.⁹ O*NET’s list of related occupations is created on the basis of similarity of job duties, similarity of possible job titles, and expert review.¹⁰ In the EP skills data tables, these occupations are flagged with the same footnote as residual occupations.¹¹

Data product for skills analysis

For the release of the 2023–33 employment projections, EP produced a suite of data tables for skills analysis by using the 17 skills categories mentioned earlier. EP also produced detailed tables with skills scores by detailed skills element, as well as crosswalks of data from O*NET. This section outlines information about the skills data tables, the detailed tables and crosswalks, and their potential users and uses.

Skills data tables

EP released a suite of spreadsheet tables (tables 6.1 through 6.5) to provide data users with skills information.¹² These tables, described in table 3, provide various types of skills data, including skills scores, for each of the 17 skills categories. The tables allow different data views, including by occupational group; by detailed occupation; for top skills categories, by fastest growing detailed occupation; by typical education needed for entry; and for percentile rank of detailed occupations, by category. In addition to these tables by skills category, EP released detailed data tables (for detailed skills and projections data) and crosswalks (for O*NET elements and occupations), both described in table 4.¹³ (More information about the variables that appear in the skills data tables and the detailed tables is provided in the appendix.)

Potential data users and uses

The skills information can be used by a variety of data users, including jobseekers, career counselors, students, workers, employers, training professionals, and researchers.

Jobseekers can use the skills data to either find suitable occupations by occupational skill or, conversely, learn about the skills needed to perform a particular occupation of interest. For example, jobseekers interested in mathematics may find that the skills score for that category is very high for occupations such as data scientists, actuaries, and statisticians and decide to explore those occupations further. Conversely, jobseekers interested in becoming technical writers may use the skills data to find out what skills are required for this occupation. In addition, jobseekers may benefit from information about how employment opportunities may change over time.

Career counselors can use the skills data to assist jobseekers and students. Likewise, students interested in certain occupations can use the skills information, along with other elements of the OOH, to plan their future education and training needs (for example, to inform their choices of classes, electives, camps, or clubs). Career planners, including people looking to switch careers or pursue further education, can examine which occupations are growing and connect training or educational opportunities to the skills required for these occupations.

Workers can use the skills data to obtain information about the skills needed for their occupation or for an occupation they hope to enter in the future, informing their decisions about enhancing certain skillsets. For their part, employers can use these data to determine what skills they may want to list on their job postings. This information can also assist workers in assessing training opportunities and help employers and others in creating training programs.

Finally, researchers may find the skills data helpful for their research projects. The detailed tables provided with the new product allow researchers and others to conduct detailed occupational analyses.

Analysis and highlights of 2023–33 employment projections and skills data

In informing decisions about career-planning activities, data users can analyze the skills data released with the 2023–33 employment projections to identify important skills by occupational group or detailed occupation. In addition, they can examine relationships between the skills data and other occupational characteristics, such as education or wages. Although EP does not project how skills will change within individual occupations over the projections decade, data users can view skills scores alongside the occupational projections to conduct their own analyses of which skills may be in future demand.¹⁴ This section offers several analyses illustrating the relationships between the skills data and projected employment and between the skills data and wages.

Relationships between skills and projected changes in employment

One way to examine how the need for different occupational skills may change in the future is to compare the average score of each skills category across all detailed occupations (weighted by base-year employment versus projected employment). Because EP is not projecting changes in skills scores over the projections period, the comparison is between identical skills scores averaged with different weights that reflect projected changes in employment. Chart 1 presents the results of this comparison, showing that the average score for science skills has the highest expected increase across all skills categories. Only two skills, customer service and fine motor, are expected to have lower overall skills scores because of occupational shifts in employment. Chart 1 also shows that the changes in skills scores are relatively small across all categories. This finding is consistent with the relatively slow structural changes in the economy that underpin the employment projections.

Another way to examine how skills importance could potentially change over time is to look at the relationship between skills scores and projected percent changes in employment at the occupation level. This analysis can be performed by plotting the two variables to see whether they are related. This approach is illustrated in charts 2a through 2d, which present binned scatter plots for four skills categories: problem solving and decision making, interpersonal, computers and information technology, and mechanical. In these plots, occupations are aggregated into 20 groups according to a skill’s importance and computed average skills score, with the projected percent change in employment within each group plotted on the vertical axis.

As shown in the plots, for three of the four skills categories (problem solving and decision making, interpersonal, and computers and information technology), the relationship between skills scores and projected percent changes in employment is positive, meaning that occupations with higher skills scores have higher projected percent changes in employment than occupations with lower skills scores. Conversely, the relationship between mechanical skills and projected percent changes in employment is negative, indicating that occupations requiring high values of mechanical skills are projected to have lower projected percent changes in employment than occupations requiring lower values of mechanical skills. (As shown in chart 1, however, projections still indicate growth in the employment-weighted average level of mechanical skills.)

Chart 3 summarizes the results of single-variable regressions of projected percent employment change on each of the 17 skills categories. The estimated slope in these regressions can be interpreted identically to the regression lines shown in the binned scatter plots in charts 2a through 2d. A positive (negative) slope indicates that as the required importance of a skill increases, the projected percent change in employment increases (decreases). The chart indicates a positive relationship for 14 of the 17 skills categories, with that relationship being particularly strong for the categories of problem solving and decision making, adaptability, critical and analytical thinking, and speaking and listening. Three skills categories—physical strength and stamina, mechanical, and fine motor—exhibit a negative relationship.

Relationships between skills and projected changes in employment, by education

Besides shedding light on the overall relationships between skills and projected changes in employment, the skills data allow analyses of those relationships by various occupational characteristics, including educational requirements. For each occupation for which EP publishes projections data, the program also provides information about the education typically needed for entry.¹⁵ For the 2023–33 projections cycle, employment shares are projected to increase for occupations typically requiring a postsecondary nondegree award or higher and to decrease for occupations typically requiring some college (no degree) or less.

Chart 4 breaks down the total percent change in skills scores (see chart 1) into changes due to projected shifts in employment between education groups and changes due to projected shifts in employment within an education group. The component measuring between-group changes holds each skills score constant and multiplies that value by the projected change in the share of occupations requiring a certain level of education for entry. The component measuring within-group changes holds the share of occupations requiring a certain level of education for entry constant and multiplies that share by the projected skills-score change due to projected occupational employment shifts. These components are then summed across the eight education groups identified by EP, to arrive at total percent changes by skills category.

As shown in chart 4, much of the percent change in skills scores comes from changes due to projected shifts in employment between education groups. This means that changes in skills scores are largely driven by projected shifts in employment toward occupations requiring higher levels of education and away from occupations requiring lower levels of education. However, employment changes within education groups also play an important role in driving changes in average skills scores. The within–group component outweighs the between–group component for the science, mechanical, and physical strength and stamina skills categories. For science, the within- and between-group components both move in a positive direction, making science the skills category with the largest percent change in skills score. By contrast, the category of physical strength and stamina exhibits a small increase in its skills score (0.07 percent) because of projected employment changes, but underneath that small increase is a large decrease due to between-group employment changes and a large increase due to within-group employment changes.

Table 5 aggregates the detailed education categories into two higher level categories by using projected employment levels as weights. The table shows that most skills categories have higher skills scores when the education needed for occupational entry is a bachelor’s degree or higher. The categories of writing and reading, computers and information technology, and critical and analytical thinking exhibit especially large differences in skills scores between the two higher level education categories. However, the categories for physical strength and stamina, fine-motor, and mechanical skills all have higher skills scores for occupations that typically require less than a bachelor’s degree. The skills scores for customer service and detail-oriented skills are similar across education categories.

Chart 5 breaks down the skills-score percent change that is due to projected employment shifts into changes for two broad occupational groups defined in terms of two aggregate education categories: occupations whose typical education for entry is a bachelor’s degree or higher and occupations whose typical education for entry is less than a bachelor’s degree. For occupations in the latter group, the categories for science, physical strength and stamina, and mechanical skills have the highest percent gains in skills scores, whereas the skills categories of computers and information technology and mathematics have the highest percent losses. For occupations typically requiring a bachelor’s degree or higher, the categories of science, critical and analytical thinking, and computers and information technology have the highest percent changes. Because the scale of these changes is relatively small, they should be evaluated in light of skills-score magnitudes. For example, although the science category has the largest percent change, its average skills score for occupations typically requiring less than a bachelor’s degree is only 1.57 in the target year of 2033.

Relationships between skills and wages

The analysis in this section seeks to answer the question whether the EP skills categories contain important empirical content. Although this question may be approached from many different directions, the relationship between occupational skills and wages emerges as a natural focal point of interest given the connection between EP and Occupational Employment and Wage Statistics data.¹⁶ This section presents the results of a test performed to see whether the new skills variables developed by EP jointly explain a significant amount of across-occupation wage variation.

Table 6 shows linear regression results for the amount of variation, measured by the adjusted R-squared statistic, in occupational median annual wages regressed on readily available independent variables. The linear regressions include all possible combinations of 22 two-digit occupations, 8 education categories, and EP’s 17 skills categories. There are seven possible specifications. As shown in the table, skills categories (adjusted R-squared of 0.69) explain significantly more wage variation than either two-digit occupations (adjusted R-squared of 0.50) or education categories (adjusted R-squared of 0.55). In fact, skills categories explain nearly as much wage variation as these two sets of variables combined (adjusted R-squared of 0.72). Moreover, combining skills categories with two-digit occupations or education categories indicates that skills categories jointly explain nearly half of the residual wage variation from occupations or education alone. For example, using only dummy variables for two-digit occupations results in an adjusted R-squared of 0.50, a figure that increases to 0.75 when skills categories are added. Therefore, skills categories explain roughly half of the residual wage variation (after controlling for two-digit occupations). These results strongly indicate that skills contain substantial empirical content, at least in explaining occupation-specific wage variation.

Data limitations

In preparing and analyzing the skills data, EP identified several data limitations. Some of these limitations, such as the effect of data source on the definition of skill, the lack of skills data for certain occupations, and the variation of skills and skill levels within an occupation, have already been noted. This section outlines a few additional limitations.

Classification of skills

Skills categories can be subjective. Some categories, such as mathematics and science, are generally easily understood, and classifying subcategories into them is relatively unambiguous. This may not be the case for other categories, however, because certain subcategories may reasonably fall under multiple categories. For example, some elements mapped to the interpersonal category, such as assisting and caring for others, could also be mapped to the customer service category. In mapping O*NET elements to skills categories, EP made every effort to produce a reliable classification by using the results of cognitive testing, factor analysis, and, in some cases, professional judgment. In addition, EP has made its mapping publicly available, allowing data users to produce their own mappings and arrive at a different number of skills categories.

Aggregation of skills scores

As explained earlier, creating an aggregate skills score by EP skills category involves taking the average importance score for each O*NET element. A potential limitation of this approach is that the skills-score aggregation depends on which O*NET elements are selected for each skills category, especially because some elements are more related than others. This limitation is particularly relevant in the case of categories with a wide range of O*NET scores for detailed elements. For example, the skills category for science combines scores for four detailed O*NET elements: biology, chemistry, physics, and science. Although chemists may have higher scores for the chemistry and science elements, their relatively lower scores for biology and physics would affect the aggregate score for the overall science category. Despite this limitation of skills-score aggregation, EP has made its scores by detailed element fully available to data users.¹⁷

Within-occupation differences

The importance of various skills within an occupation can vary by level of skills needed, location, and industry. To capture different levels of skills needed for a work activity, O*NET assigns level scores on a scale from 1 to 7, with 7 being the highest level of skill. Because EP decided to develop a skills score by occupation with the use of importance scores (or context scores for work-context elements), level information was not used for this initial data release—a potential data limitation. For example, within the “working with computers” work-activity element, the levels of skills needed can vary widely. A lower level score for this element may be associated with entering information into a database, whereas a very high score may involve setting up a new computer system for a large company. In both examples, however, the importance score would be similar, because working with computers is an important skill for both.

In addition, the type and importance of skills may vary by industry and location. An occupation in one industry may have work requirements that differ from those in another industry. Because the present data product aggregates information to the occupation level, it does not capture within-occupation differences across industries. Additionally, given that EP develops employment projections only at the national level, skills data are not produced at the state or local level.

Conclusion

The new product for skills data will be updated annually as new O*NET data are released alongside the BLS employment projections. The goal of this new product is to provide data users with additional information about occupational skills and assist them with making informed career-planning decisions. The product provides skills information that is easy to use and linked to the BLS occupational projections. EP plans to gather feedback from data users and use it to enhance future iterations of the skills data.¹⁸ EP also plans to integrate the skills data into the OOH. In addition, EP will conduct and publish additional analyses based on the skills data.

Appendix: Data sources and variable definitions

This appendix provides additional information about the data sources used to develop the EP skills framework (see table A-1) and definitions of the variables underlying the skills data tables (see table A-2).