angle-up angle-right angle-down angle-left close user menu open menu closed search globe bars phone store

Skills and Earnings in the Full-Time Labor Market
Neeta Fogg, Paul Harrington, and Ishwar Khatiwada

Effects of Literacy and Numeracy Proficiencies, Educational Attainment, and Other Explanatory Variables on Earnings

In addition to literacy and numeracy proficiencies, the human capital earnings regressions that we have estimated for 25- to 54-year-old full-time employed workers included a number of other explanatory variables that are known to affect earnings. In the preceding section we have presented findings from a series of regressions with a particular focus on the regression-adjusted connection between earnings and skills. In this section, we present the complete regression findings for the full model (Model 6) that contains covariates of human capital traits of workers, job traits, employment-related traits of workers, and demographic traits of workers. The full earnings regression model (Model 6) was estimated separately four times, each with a different specification of skills.49 In keeping with the numbering system used for the discussion of regression findings in this report, tables 5 through 8 contain findings from the following four earnings regressions: Set A-Model 6 (standardized literacy test score), Set B-Model 6 (standardized numeracy test score), Set C-Model 6 (literacy proficiency levels), and Set D-Model 6 (numeracy proficiency levels).50 The R-squared for each of these four models ranges from .500 to .502, meaning that these models explain half of the variation in the monthly earnings of 25- to 54-year-old full-time employed workers in 2012-2014.

Findings presented in these tables contain the percent effect of a one unit change in each explanatory variable on the monthly earnings of 25- to 54-year-old full-time workers. The first column (column A) contains findings from earnings regressions in Set A-Model 6 in which the skills of workers was specified as the standardized literacy proficiency test score. Columns B, C, and D present findings from the same earnings regressions (Model 6) except with different measures of proficiency: standardized numeracy proficiency test score in Set B-Model 6 (Col. B), literacy proficiency levels in Set C-Model 6 (Col. C), and numeracy proficiency levels in Set D-Model 6 (Col. D).

Human Capital Traits of Workers

The discussion of literacy and numeracy proficiencies of prime-age, full-time workers found that these basic skills are closely related to their earnings. Even in the regression models that control for a wide variety of worker characteristics and job traits, literacy/numeracy proficiencies of workers are found to have a strong effect on their earnings (figures 11 to 14). The full regression models also find that educational attainment exerts a strong and positive impact on the earnings of prime-age, full-time employed workers at the bachelor's degree level or higher. Earning a bachelor's degree had the effect of increasing monthly earnings by about 30 percent relative to high school graduates in each of the full regression models (Model 6) of sets A, B, C, and D. Similarly, full-time workers with a master's degree are expected to earn between 43 to 45 percent more than high school graduates. And, the regression-adjusted monthly earnings premium of workers with a professional degree relative to the base group of high school graduates is estimated to be 60 percent in each of the four full regression models (Table 5).

High school dropouts had regression-adjusted earnings that were about 16 to 17 percent below those of high school graduates. It is of special importance to note that among prime-age, full-time employed workers, the regression-adjusted earnings of those with some college, but no degree or certificate award, were not significantly different from the earnings of their high school graduate counterparts. Furthermore, three out of the four full regression models found no statistically significant regression-adjusted earnings advantage relative to high school graduates among workers with an associate's degree, and the fourth model measured only a marginally significant regression-adjusted earnings advantage for workers with an associate's degree relative to those with just a high school diploma (Table 5). These findings are of considerable importance since a very large share of the nation's workforce falls in one of these two educational categories.

Paid work experience is considered to be another form of human capital. As workers gain labor market experience, they learn new skills and knowledge and improve the skills and knowledge that they already possess. Additional work experience also provides workers with seniority that is frequently accompanied with higher pay. Paid work experience is expected to have a strong positive effect on the earnings of workers. Descriptive analysis of PIAAC data shows a strong and positive connection between paid work experience and monthly earnings of workers (Appendix F). Compared to workers with less than 10 years of paid work experience, the earnings premiums of workers with higher levels of paid work experience ranged from 17 percent for 10-19 years, to 40 percent for 20-29 years, to 43 percent for 30-plus years. Mean monthly earnings for all subgroups of workers included in the full earnings regressions models are presented in Appendix F.

Human capital earnings functions include work experience as an explanatory variable with a quadratic specification to capture the relationship between work experience and earnings postulated by Mincer, that is, earnings rise with additional work experience but at a diminishing rate.51 Earnings increase with work experience: sharply for the first few years, followed by a more gradual rate of increase. Indeed, findings from our earnings regression analysis support this relationship between work experience and earnings. An additional year of work experience is expected to raise monthly earnings by 3.1 to 3.3 percent, holding everything else constant (Table 5). The negative and statistically significant coefficient on the experience-squared variable indicates that the earnings of workers grow with additional work experience, but the rate of earnings growth slows down as the years of work experience increases (diminishing returns to additional years of work experience), which means that at a certain level of work experience, monthly earnings will be maximized (about 15.7 years based on estimated regression coefficients of the two work experience variables in the full regression models). Regression findings for paid work experience were similar across all four earnings regressions presented in Table 5.

English language proficiency is vital in the U.S. labor market. English language skill is closely linked with earnings, employment, and other labor market outcomes of workers, particularly foreign-born workers. The English speaking ability of immigrants was found to be closely connected with their employment and earnings outcomes as well as other social outcomes such as attainment of U.S. citizenship.52 In PIAAC 2012-2014 surveys, all respondents (foreign-born as well as native-born) were asked questions about their English language proficiency.53 Respondents were asked to self-assess their ability to understand spoken English, read English, speak English, and write English. They were asked to select one of the following four options to rate their English language proficiency: very well, well, not well, and not at all. We found that the mean monthly earnings of workers were strongly linked to their English writing ability.

Table 5: Percent Change in Expected Monthly Earnings from One-Unit Increase In Predictor Variables in Full Regression Models for 25- to 54-Year-Old Full-Time Employed Workers (Findings for a Subset of Explanatory Variables Measuring Human Capital)

Explanatory Variables

(A)

Set A-Full Model 6: Standardized Literacy Proficiency Score

(B)

Set B-Full Model 6: With Standardized Numeracy Proficiency Score

(C)

Set C-Full Model 6: With Literacy Proficiency Levels

(D)

Set D-Full Model 6: With Numeracy Proficiency Levels

Explanatory variables in full regression model (Model 6): Literacy/Numeracy proficiencies; educational attainment; paid work experience, English writing ability sector of employment, occupation, weekly hours of work, school enrollment status, region of residence, gender, race-ethnicity, foreign-born status, and disability status.

Source: 2012 and 2014 PIAAC surveys, restricted use files, tabulations by authors.

Standardized Proficiency Score-Plausible Values (PVs)

PVLiteracy or PVNumeracy

8.4 8.3 --- ---

Proficiency Levels (base group is Level 2)

Level 1

--- --- -3.8 -4.8

Level 3

--- --- 6.8 8.0

Levels 4/5

--- --- 21.0 21.8

Educational attainment (base group is high school graduates)

No H.S. Diploma

-15.6 -16.1 -17.5 -17.5

Some College

0.7 0.7 1.3 1.2

Certification

1.9 1.3 2.6 1.8

Associate's Degree

6.2 5.8 6.8 6.3

Bachelor's Degree

30.0 28.9 30.5 28.9

Master's Degree

44.9 43.5 45.2 43.0

Professional Degree

60.3 59.7 60.5 59.4

Doctorate Degree

56.5 54.8 56.0 53.0

Years of work experience (continuous variable, range: 0-47)

Experience

3.1 3.3 3.3 3.3

Experience squared

-0.1 -0.1 -0.1 -0.1

English writing ability (base group is English writing ability "well")

Very well

0.1 0.7 0.4 0.8

Not well or not at all

-8.2 -8.8 -11.2 -11.0

R-Squared

0.500 0.500 0.501 0.502

In our full earnings regression model, we included variables representing self-reported English writing ability of workers for "very well," "not well," and "not at all." Workers who reported their English writing ability as "well" were included in the base group. Findings presented in Table 5 (Col. A) reveal that among prime working age (25- to 54-year-old), full-time employed workers, the regression-adjusted earnings of workers in the base group ("well" English writing ability) were not expected to be statistically different from that of their counterparts with the best English writing ability ("very well") or with poor English writing ability ("not well" or "not at all"). Although the mean monthly earnings of these workers varied by their English writing ability (see Appendix F), after controlling for all the other covariates included in the full earnings regression model, coefficients of both measures of the English writing ability of workers were not statistically significant (Table 5, Col. B). Similar findings were noted in the remaining three earnings regressions presented in Table 5, columns B, C, and D.

Job Traits: Characteristics of the Jobs of Workers

Also included in the full earnings regression models are variables that represent characteristics of the jobs of workers. One of these variables is the economic sector of employment of workers. Workers' wages vary by the economic sector in which they were employed. A recent report by the Congressional Budget Office shows that workers without a college degree employed in federal government jobs had 21 percent higher wages than their comparable peers employed in the private sector, whereas workers with a Ph.D. or professional degree who were employed in the federal government sector earned 23 percent less than peers in the private sector.54

In the PIAAC survey, workers were asked to report their economic sector of employment. Responses were grouped into the following three sectors in the PIAAC data file: the private for-profit sector, the public sector, and the nonprofit sector. We have added two explanatory variables to the full earnings regression model: the nonprofit sector and the public sector. The base group includes workers employed in the private for-profit sector.

Regression analysis (see Table 6) found that even after controlling for the skills, education, and other job traits—employment-related traits of workers, and demographic traits of workers—the monthly earnings of 25- to 54-year-old workers employed full-time in the nonprofit sector were expected to be nearly 11 percent less than the earnings of those employed in the private for-profit sector (statistically significant at the .05 level), and those working in the public sector were expected to earn 5 percent less than the base group (private for-profit sector workers; significant at the .10 level). Similar findings were noted in the remaining three regressions in Table 7 that used the standardized numeracy proficiency score, literacy proficiency levels, or numeracy proficiency levels as explanatory variables representing worker skills.

Another important job trait that is closely related to the earnings of workers is occupation. Occupations represent what workers do on the job and are closely related to the knowledge, ability, and skills of workers. Occupations that require high level of skills can be staffed with only highly skilled and educated workers who can perform the required tasks. These occupations pay high wages that are required to attract and adequately compensate workers with high levels of human capital in the form of skills and educational attainment. In fact, the earnings premiums associated with a college education are closely connected to access to jobs in high-level occupations—sometimes known as college labor market occupations—because these occupations require the skills, knowledge, and abilities that are typically acquired with a college education.55

Using the skill-based classification of occupations in PIAAC data, we have divided 25- to 54-year-old full-time employed workers into the following four groups: workers employed in skilled occupations, semi-skilled white-collar occupations, semi-skilled blue-collar occupations, and elementary occupations. A few examples of occupations in each group are presented below:56

  • skilled occupations—professional, technical, managerial, and high level sales occupations such as executives, managers, engineers, scientists, health practitioners, IT professionals, teaching professionals/educators, lawyers and judges, insurance/finance/real estate sales 
  • semi-skilled white collar occupationsadministrative support and clerical occupations, personal services, protective services occupations
  • semi-skilled blue-collar occupations—construction workers, machine assemblers/operators/repairers, vehicle operators
  • elementary occupations—laborers, helpers, handlers

Prime-age, full-time employed workers were distributed across the four major occupational groups as follows: skilled occupations (54.5 percent), semi-skilled white-collar occupations (23 percent), semi-skilled blue-collar occupations (17.5 percent), and elementary occupations (5 percent). The occupational distribution of all employed workers (aged 16 years or older) had somewhat higher shares of workers in lower level occupations: skilled occupations (48 percent), semi-skilled white-collar occupations (26 percent), semi-skilled blue-collar occupations (16 percent), and elementary occupations (8 percent).

Earnings of workers varied widely by occupation. The mean monthly earnings of 25- to 54-year-old full-time employed workers were highest among workers in skilled occupations ($5,968), followed by $3,589 among semi-skilled blue-collar workers (40 percent lower than skilled occupations), $3,136 among semi-skilled white-collar workers (47 percent lower than skilled occupations), and just $2,448 among elementary occupation workers (59 percent lower than skilled occupations).

As mentioned above, occupations are included in the earnings regressions as a set of explanatory variables representing job traits and consisting of three dummy variables representing semi-skilled white-collar, semi-skilled blue-collar, and elementary occupations. Skilled occupations are used to represent the base group. Earnings differentials between workers in each of the four occupational categories remained large even after statistically controlling for human capital and other job traits, job-related worker traits, and demographic traits of workers. Findings in Table 6 (col. B) indicate that compared to 25- to 54-year-old full-time employed workers employed in skilled occupations (the base group), their counterparts employed in other occupations were expected to earn 23 percent less in semi-skilled blue-collar occupations, 27 percent less in semi-skilled white-collar occupations, and 35 percent less in elementary occupations. These regression-adjusted earnings differentials by occupation were the same across all four regression models presented in Table 7.

Table 6: Percent Change in Expected Monthly Earnings from One-Unit Increase In Predictor Variables in Full Regression Models for 25- to 54-Year-Old Full-Time Employed Workers (Findings for a Subset of Explanatory Variables Measuring Job Characteristics)

Explanatory Variables

(A)

Set A-Full Model 6: Standardized Literacy Proficiency Score

(B)

Set B-Full Model 6: With Standardized Numeracy Proficiency Score

(C)

Set C-Full Model 6: With Literacy Proficiency Levels

(D)

Set D-Full Model 6: With Numeracy Proficiency Levels

Explanatory variables in full regression model (Model 6): Literacy/Numeracy proficiencies; educational attainment; paid work experience, English writing ability sector of employment, occupation, weekly hours of work, school enrollment status, region of residence, gender, race-ethnicity, foreign-born status, and disability status.

Source: 2012 and 2014 PIAAC surveys, restricted use files, tabulations by authors.

Economic sector of employment (base group is for-profit private sector)

Nonprofit sector

-10.6 -10.9 -10.7 -10.7

Public sector

-5.1 -4.6 -5.0 -4.5

Occupation (base group is skilled occupations)

Semi-skilled white-collar occupations

-26.6 -26.8 -26.6 -26.9

Semi-skilled blue-collar occupations

-23.0 -23.7 -22.7 -23.2

Elementary occupations

-35.1 -35.7 -35.3 -35.5

Occupations missing

-11.3 -12.5 -11.1 -12.3

R-Squared

0.500 0.500 0.501 0.502

Employment-Related Traits of Workers

The next three groups of explanatory variables in the earnings regressions presented in Table 7 represent employment-related worker traits. These traits include their school enrollment status, weekly hours of work, and the region in which they reside. The first of these three employment-related traits of workers is their school enrollment status. Workers who are enrolled in school are expected to have lower earnings for a number of reasons. School-enrolled workers are generally younger and are still in the process of securing the labor market work experience that will raise their earnings in the future. Also, these workers are engaged in their education, which will result in higher earnings in the future. Furthermore school-enrolled workers are less likely to be fully engaged in the labor market as part of their time is spent on schooling activities—even when working a full-time weekly work schedule. The full earnings regression models presented in Table 7 found that workers who were enrolled in school during the PIAAC surveys were expected to earn between 8.3 and 8.9 percent less than their counterparts who were not enrolled in school. The school enrollment coefficient was statistically significant at the .01 level in all four full regression models (Table 8).

The monthly earnings of workers are closely related to the wages that workers earn per hour and the number of hours of work that they perform per week during the month. PIAAC data provide information on the number of hours that workers worked per week at the time of the PIAAC survey. This variable, representing job traits, is included in the full earnings regression model as a continuous variable between 35 and 60 hours. Findings show a strong and positive connection between weekly hours of work and monthly earnings of 25- to 54-year-old full-time employed workers; each additional hour of work is expected to increase monthly earnings of workers by 1.8 percent, holding all other explanatory variables constant (Table 8).

Earnings of workers also vary by location. Variations in labor supply and demand, cost of living, local institutions, policies and regulations, and other labor market variations result in geographic variations in earnings. The PIAAC data file provides data on the residence of respondents. We have used the region of residence of workers as explanatory variables in the earnings regressions to capture the geographic variation in the monthly earnings of workers. Descriptive measures of the regional differences in earnings (Appendix F) found that the mean monthly earnings of workers in the Northeast ($5,228) and the West ($5,055) exceeded earnings of their counterparts in the South ($4,413) by 18 percent and 15 percent, respectively, while the earning of workers in the Midwest ($4,530) were only 3 percent higher than that of their counterparts residing in the South.

Earnings regressions found statistically significant differences in the earnings of workers by region. Workers in the Northeast region and the West region were expected to earn 10.6 percent and 12.6 percent, respectively, higher monthly earnings than workers residing in the South region, holding constant human capital traits, job traits, employment-related worker traits, and demographic traits of workers. These two coefficients were statistically significant at the .05 level in each of the four full regression models (Table 7). The coefficient on the Midwest region was positive, but not statistically significant.

Table 7: Percent Change in Expected Monthly Earnings from One-Unit Increase In Predictor Variables in Full Regression Models for 25- to 54-Year-Old Full-Time Employed Workers (Findings for a Subset of Explanatory Variables Measuring Employment-Related Traits of Workers)

Explanatory Variables

(A)

Set A-Full Model 6: Standardized Literacy Proficiency Score

(B)

Set B-Full Model 6: With Standardized Numeracy Proficiency Score

(C)

Set C-Full Model 6: With Literacy Proficiency Levels

(D)

Set D-Full Model 6: With Numeracy Proficiency Levels

Explanatory variables in full regression model (Model 6): Literacy/Numeracy proficiencies; educational attainment; paid work experience, English writing ability sector of employment, occupation, weekly hours of work, school enrollment status, region of residence, gender, race-ethnicity, foreign-born status, and disability status.

Source: 2012 and 2014 PIAAC surveys, restricted use files, tabulations by authors.

School enrollment status (base group is not enrolled in school)

Enrolled in school

-8.6 -8.9 -8.3 -8.8

Weekly hours of work (continuous variable, range: 35-60)

Weekly Hours

1.8 1.7 1.8 1.8

Region of residence (base group is South region)

Northeast

10.6 10.7 10.6 10.6

Midwest

5.1 5.2 5.1 5.1

West

12.6 12.6 12.6 12.6

R-Squared

0.500 0.500 0.501 0.502

Demographic Traits of Workers

The last set of covariates in the full earnings regressions model consists of variables representing demographic traits of workers. The following four demographic traits of workers are included as explanatory variables in the earnings regressions: gender, race-ethnicity, foreign-born status, and disability status. Mean monthly earnings of 25- to 54-year-old full-time employed workers included in this study varied widely by these four demographic characteristics (Appendix F).

The four models presented in Table 8 estimated the value of the male coefficient at between .22 and .24, meaning that, with all else remaining the same, the monthly earnings of men were expected to be between 25 and 27 percent higher than the monthly earnings of women. Even after statistically controlling for skills, education, and work experience, and job traits including occupation, employment-related workers traits, and other demographic traits of workers, these earnings regressions estimated a sizable earnings gender gap in favor of male workers. Further research is needed to understand the factors that underlie this sizable regression-adjusted gender gap in earnings.

Although there were differences in the mean earnings of workers by race-ethnicity, after controlling for literacy/numeracy proficiency, educational attainment, and all other covariates included in the regressions in Table 8, the earnings of Black, Hispanic, Asian, and other race-group workers were not expected to be statistically different from the earnings of White workers. The same findings prevailed in all four full earnings regression models presented in Table 8. These findings indicate that the race-ethnicity-based differences in the monthly earnings of these workers were attributable to differences in their human capital and employment-related traits, the types of jobs to which they have access, and demographic traits of gender and disability status. After statistically controlling for these traits (by including these traits as covariates in these earnings regressions), the regression-adjusted differences in earnings by race-ethnicity were not statistically significant.

A comparison of the mean monthly earnings of workers included in this study found a difference by the foreign-born status of workers; the earnings of foreign-born workers were 10 percent lower than those of native-born workers. However, each of the four full earnings regression models (presented in Table 8) found no statistically significant difference between the regression-adjusted earnings of native- and foreign-born workers (after statistically controlling for literacy and numeracy proficiency, educational attainment, and all other covariates in these regression models). The coefficient for this variable in all four full earnings regression models was not statistically significant.

Table 8: Percent Change in Expected Monthly Earnings from One-Unit Increase In Predictor Variables in Full Regression Models for 25- to 54-Year-Old Full-Time Employed Workers (Findings for a Subset of Explanatory Variables Measuring Demographic Traits of Workers)

Explanatory Variables

(A)

Set A-Full Model 6: Standardized Literacy Proficiency Score

(B)

Set B-Full Model 6: With Standardized Numeracy Proficiency Score

(C)

Set C-Full Model 6: With Literacy Proficiency Levels

(D)

Set D-Full Model 6: With Numeracy Proficiency Levels

Explanatory variables in full regression model (Model 6): Literacy/Numeracy proficiencies; educational attainment; paid work experience, English writing ability sector of employment, occupation, weekly hours of work, school enrollment status, region of residence, gender, race-ethnicity, foreign-born status, and disability status.

Source: 2012 and 2014 PIAAC surveys, restricted use files, tabulations by authors.

Gender (base group is female)

Male

27.4 25.4 27.3 24.7

Race-Ethnicity (base group is White)

Hispanic

-3.6 -3.3 -4.1 -3.6

Black

-1.7 -0.1 -2.3 -0.8

Asian, Pacific Islander

7.6 7.3 7.7 7.7

Other Race

-0.9 -0.2 -0.8 0.0

Nativity status (base group is native-born)

Foreign-Born

0.3 -0.9 -0.3 -1.4

Disability status (base group is workers without disabilities)

With disability

-8.1 -8.5 -8.5 -8.7

R-Squared

0.500 0.500 0.501 0.502

The disability status of workers is closely related to their labor market outcomes.57 Workers with disabilities are less likely to participate in the labor market compared to workers without disabilities. And, when they participate in the labor market, workers with disabilities are less likely to find a job and more likely to remain unemployed than workers without disabilities. Individuals with disabilities have lower labor force participation rates, lower employment rates, and higher unemployment rates than individuals without disabilities. Even when employed, workers with disabilities work fewer hours per week and fewer weeks per year than workers without disabilities. Our descriptive analysis of PIAAC data found that the mean monthly earnings of 25- to 54-year-old full-time employed workers with disabilities were 21 percent lower than the earnings of workers without disabilities. Even after controlling for skills, education, occupation, and all other covariates included in the four full models of earnings regressions presented in Table 5, the earnings of workers with disabilities were expected to be between 8.1 and 8.7 percent lower than the earnings of workers without disabilities. The coefficient was statistically significant at the .01 level in all four regression models (Table 8).

Notes

49 See Box 1 for details.

50 Tables 5 through 8 present (separately) findings for subsets of all explanatory variables included in the full regression Model 6. Findings for all explanatory variables included in the full regression Model 6 are presented (together) in Appendix E.

51Mincer, 1974, Schooling, Experience, and Earning.

52 See Andrew Sum, Irwin Kirsch, and Kentaro Yamamoto, A Human Capital Concern: The Literacy Proficiency of U.S. Immigrants (Princeton, NJ: Educational Testing Service, 2004), https://www.ets.org/Media/Research/pdf/PICHUMAN.pdf.

53 In U.S. decennial censuses and American Community Surveys (ACS), English-speaking abilities questions were asked only to foreign-born persons 5 years and older.

54 See Justin D. Falk, Comparing the Compensation of Federal and Private-Sector Employees (Washington, DC: Congressional Budget Office, January 2012), http://cbo.gov/doc.cfm?index=12696.

55 Neeta P. Fogg and Paul E. Harrington, "Rising Mal-Employment and the Great Recession: The Growing Disconnection between Recent College Graduates and the College Labor Market," Continuing Higher Education Review 75 (2011): 51-65.

56 A list of detailed occupations in each of the four skill-based categories is presented in Appendix G.

57 Neeta P. Fogg, Paul E. Harrington, and Brian T. McMahon, "The Impact of the Great Recession upon the Unemployment of Americans with Disabilities," Journal of Vocational Rehabilitation 33 (2010): 193-202; Neeta P. Fogg, Paul E. Harrington, and Brian T. McMahon, "The Underemployment of Persons with Disabilities During the Great Recession," The Rehabilitation Professional 19, no. 1 (2011): 3-10. For current labor force statistics of persons with disabilities, see Bureau of Labor Statistics, "Persons with a Disability: Labor Force Characteristics – 2015," news release, June 21, 2017, http://www.bls.gov/news.release/pdf/disabl.pdf.