Growing inequality of incomes is one of the most important economic issues of our time. When wages are not rising over time for a large fraction of American workers, these workers do not fully share in economic growth. Moreover, inequality of incomes becomes inequality in household resources. As household resources become more important in equipping new workers for jobs in which they can earn higher incomes (for example, funding from parents and other relatives to pay for postsecondary education), inequality today has the potential to further increase inequality in succeeding generations. Such ever-widening inequality will mean growing gaps in opportunities between Americans of different household income levels.
For more than four decades, inequality of incomes has been growing nearly continuously and has been the subject of a tremendous amount of empirical research. Lawrence F. Katz and David Autor provide a thorough survey of this literature as it stood 20 years ago.1 Since then, the works of David S. Lee and of David Card and John E. DiNardo have been particularly influential in emphasizing the potential role of declining real minimum wages and of unionization in explaining the growth of wage inequality.2 The work of David H. Autor, Lawrence F. Katz, and Melissa S. Kearney has also been particularly influential in emphasizing the importance of technological change in explaining wage inequality growth, because automation reduced demand for what had been routine work done by middle-income employees.3 A related literature using tax data to study income inequality has followed the work of Thomas Piketty and Emmanuel Saez,4 who used these tax data to show dramatic increases in incomes for people at the top of the income distribution.
All these studies focus on understanding details in the patterns of income inequality growth to better understand the underlying causes of inequality growth. However, recent research articles based on data from the Current Population Survey (CPS) show some evidence of a change from the trend of ever-increasing income and wage inequality in the years immediately preceding the 2020 recession. In their recent survey of the wage inequality literature, Florian Hoffmann, David S. Lee, and Thomas Lemieux include figures showing small declines in labor income inequality for men beginning around 2015.5 Jay C. Shambaugh and Michael R. Strain also present figures showing that wages at the lower percentiles of the wage distribution had greater growth than median wages from 2017 to 2019.6
In this article, we examine recent trends in wage inequality using the confidential microdata collected from the Occupational Employment and Wage Statistics (OEWS) survey, which relies on a large representative sample of employers in the United States. We also examine the wage questions in the confidential version of the CPS data, collected from a large representative sample of households in the United States. The OEWS and CPS data are completely independent surveys that include wage information. James R. Spletzer and Elizabeth Weber Handwerker show very similar patterns of wage variances between the CPS and the OEWS survey from 1998 to 2010, overall, by sector, industry, and occupation, and in the fraction of wage variance explained by these factors.7 Here, we examine what happened to wage inequality in these two surveys from late 2002 or 2003 through 2019. We further corroborate our findings from these two surveys with annual wage and salary earnings data from income tax filings.
We find that wages were growing more unequal from 2003 to 2013 in both the CPS and OEWS data, because wages increased more among high-wage workers than among low-wage workers. However, from 2013 to 2019, both surveys show substantial wage gains for workers in the second and third deciles of the wage distribution. In the OEWS data, wage growth among lower wage workers was great enough to reduce overall wage inequality from 2013 to 2019, while in the CPS data, wage inequality plateaued. In tax data, we find substantial increases in annual earnings among workers with lower annual earnings, which were great enough to reduce overall earnings inequality from 2010 through 2018 (the most recent year available). Occupational information in the CPS and OEWS data allows us to identify the occupations most important for wage growth among lower wage workers, because of their wage gains and employment levels. These occupations included health aides and material moving workers. We also examine which characteristics of workers and their employers were particularly important for wage inequality in the CPS and OEWS data and highlight the importance of occupations in wage inequality.
We begin by describing the wage or wage and salary data available from each of our sources and the differences between the data available from each one.
In either May or November each year, the OEWS survey asks employers about their occupational and wage patterns, with approximately 200,000 establishments surveyed at each date. Administrative data on the occupations and wages of employees are also collected from state and federal governments. The OEWS data do not include much of the agricultural sector, the unincorporated self-employed, or private household employers. Employers are asked to report hourly wage rates for part-time or hourly workers and annual rates for salaried workers and for workers in occupations that are generally paid an annual salary but work less than full-time hours over the course of a year, such as pilots, flight attendants, and teachers. Wages in this survey include base pay, guaranteed pay, incentive pay, commissions, bonuses, and tips. They do not include overtime pay, severance pay, or employer costs for employee benefits. Before 2020, the OEWS survey collected wage information in 12 wage intervals, defined in terms of either hourly wages or annual wages equivalent to hourly wages multiplied by 2,080 hours.
The OEWS survey has a complex sample design. A full sample is selected over a 3-year period, with establishments generally not selected more than once every 3 years. All OEWS estimates rely on 3 years of data, as well as data from other U.S. Bureau of Labor Statistics (BLS) programs. Within each wage band, a mean hourly wage is estimated by using wage data collected by the BLS National Compensation Survey.8 To adjust wage estimates collected at different dates, the OEWS program uses the BLS Employment Cost Index for each occupational division. Employers that do not respond to the survey have occupational employment and wage values imputed on the basis of responses from employers that are similar in location, industry, and size (in Quarterly Census of Employment and Wages [QCEW] data). All OEWS estimates are benchmarked to employment totals from the BLS QCEW, by location, industry, and size, in the final year of data collection. More information about the OEWS survey can be found in the BLS Handbook of Methods.9
We use the OEWS microdata beginning in November 2002, when that survey began sampling 200,000 establishments every November and May and had completed training all staff in using the Standard Occupational Classification (SOC) system. We convert the resulting nominal wage estimates into constant 2016 dollars using the CPI for All Urban Consumers (CPI-U). Because OEWS estimates are based on 3 years of data, OEWS estimates cannot be compared from one year to the next. We have calculated estimates for every 3-year aggregation of data (November 2002 to May 2005, November 2003 to May 2006, etc.) but present only estimates for the 3-year aggregations ending in May 2005, May 2010, May 2013, May 2016, and May 2019. The choice of years we have chosen for presentation does not substantially change our findings, although the overall log wage variance measured in the OEWS is highest for November 2011 to May 2014 (this period has a standard deviation of 0.618, compared with the standard deviation of 0.616 for November 2010 to May 2013 shown in the charts discussed later in this article).
The CPS data are generally used in studies of wage and income inequality, because microdata from this survey have been publicly available to researchers for several decades. This survey uses a rotational sample design, in which each household in the sample is interviewed for 4 consecutive months and then, 8 months later, interviewed for another 4 consecutive months. For the best comparison to the OEWS, we focus on the CPS wage questions, which are asked in the fourth and eighth interviews. Since these are the months when respondents are rotating out of the sample, they are referred to as Outgoing Rotational Groups (ORGs). We use CPS data from the ORG beginning in 2003, when the CPS began using the same SOC system used in the OEWS as the basis for its occupational codes, allowing us to compare wage trends by occupation between the CPS and the OEWS. Authors who focus more on overall income inequality (rather than wages) use questions about income from the previous year, which the CPS asks only in the Annual Social and Economic Supplement.
The wage questions in the CPS are only asked of respondents about their main jobs and are not asked of self-employed people. We add overtime pay, tips, and commissions to hourly wages for those who report these types of income. We use hourly wages for jobs reported as such and otherwise convert annual, monthly, or weekly earnings to hourly wages using usual hours reports, following research by Anne E. Polivka and by John Schmitt.10 About 58 percent of respondents report hourly wages. Depending on the year, 2.5 percent to 3.8 percent of respondents report annual, monthly, or weekly earnings but not an hourly wage and report that their usual hours vary, making it difficult to calculate their hourly wages. For these people, we use a regression-based imputation method to model weekly hours separately for men and women working full- or part-time, based on age, race, education, marital status, and immigration status. All wage data in the CPS are reported in nominal terms and converted into constant 2016 dollars using the CPI-U. To avoid the top coding of incomes applied to the public-use version of the CPS data to protect the confidentiality of respondents, we use the confidential CPS microdata not available to researchers outside BLS or the U.S. Census Bureau. However, in the appendix (to this article), we show that the public-use version of these data yield very similar trends in wage inequality over time.
The greatest hourly wage reportable in the CPS data is $99.99, but this affects only 0.05 percent of hourly wage reports; most high earners report their earnings annually, monthly, or weekly, not hourly, and the CPS allows respondents to report weekly earnings up to $99,999.99 per week. Following Card and DiNardo, Thomas Lemiuex, and Sarah A. Donovan and David H. Bradley,11 we remove observations with an hourly wage of less than $1 or more than $100 in 1979 dollars—less than $3.50 or more than $350 in 2016 dollars. This censoring of particularly high- and low-wage values means dropping 0.8 percent to 0.9 percent of observations in the confidential data, varying very slightly by year.
The most important differences in wage data collection between the OEWS and the CPS are that wage information is collected from employers in the OEWS and from workers in the CPS and that wage questions in the CPS ask respondents only about their main job.
We additionally examine annual person-level wage and salary income reported on Form W-2 Wage and Tax Statement to the Internal Revenue Service (IRS) by employers for each wage earner in each job in each year. The U.S. Census Bureau has received an extract of these data from tax years 2005 through 2018 and is authorized to use these data to improve measurements of income and produce new income statistics. Form W-2 data are annual, but they reflect earnings on each job, and so many W-2 observations have low values representing short duration jobs. We aggregate wage income across all W-2 forms for each person to capture total annual wage earnings in these data.
There are several ways this annual wage and salary earnings measure differs from the CPS and OEWS wage measures. The W-2 data include wage and salary income across all jobs over a calendar year, rather than the current “main jobs” for which we measure wages in the CPS. They do not include any measure of hours or weeks worked, rather than the hourly wage report in the CPS or the hourly wage or equivalent annual salary reported in the OEWS. This wage and salary earnings measure will be affected not only by wages but also by hours worked per week and by weeks worked per year. The W-2 data can also include overtime pay (like the CPS but unlike the OEWS), severance pay (unlike either the CPS or the OEWS), and potentially other forms of compensation, such as stock options and bonuses. Nonetheless, we include these annual earnings here to show how trends in the wage inequality measured with the OEWS and the CPS compare with trends in inequality in overall wage and salary incomes measured with income tax data.
We begin by showing trends in the standard deviation of log wage income in the CPS-ORG, OEWS data, and W-2 data. We examine trends in the CPS-ORG data in two ways. First, to compare our results with authors in the economic literature, we examine trends in the CPS for full-time workers and weight these workers by the number of hours they work. (We cannot further restrict the data to full-year work, because these questions are only asked in the CPS-ORG data for workers who report wages on an annual basis.) Second, to better compare wage variation in the CPS with the OEWS and the W-2 data, we examine trends in the CPS for all workers, without weighting workers by the number of hours worked, since the OEWS and W-2 data include part-time and seasonal workers and cannot be weighted by the number of hours worked.
Chart 1 thus compares trends in the standard deviation of log hourly wage income in the OEWS with two versions of overall trends of the standard deviation of log hourly wage income in the CPS-ORG and with trends in the standard deviation of log annual wage and salary income in the W-2 data. Note that the standard deviation of log annual wage and salary income in the W-2 data is much greater than the standard deviation of hourly wages in the OEWS and CPS data, and so it is plotted on a different scale.
Comparing worker restrictions and weightings within versions of the CPS data, we find greater dispersion of wages among all workers than among full-time workers only. However, weighting full-time workers by the number of hours they work increases the dispersion in their wages, since higher earning full-time workers work more hours. Combining both worker restrictions and hours weighting, we find a greater dispersion of wages overall among all workers who are not weighted by hours worked (estimates more comparable to the OEWS) than among full-time workers only who are weighted by hours worked (estimates more comparable to the wage variation literature). However, the two versions of the CPS wage inequality series show similar time trends in wage inequality. In both CPS series, wage inequality was clearly rising from 2003 to 2013 and then stopped rising, with perhaps some modest decline from 2016 through 2019. Versions of these series estimated using public-use CPS data are shown in the appendix.
Comparing the CPS data to the OEWS and W-2 data, we see that all three data sources show growing inequality from the beginning of the series through 2010. OEWS data show increased inequality of wages through 2013 and declining inequality of wages thereafter, particularly from 2016 to 2019. CPS data show a similar increase in wage inequality to the OEWS through 2016, but they do not show as much decline in inequality thereafter. The CPS series for all workers not weighted by hours is much more similar to the OEWS data in its measured level of wage inequality than the CPS series for full-time workers weighted by hours worked. The two CPS series, however, show similar time trends in wage inequality. The W-2 data show earnings inequality that peaked in 2010 and has been declining thereafter.
The rest of this article examines the distribution of wages further by using only the OEWS and the confidential version of the CPS from 2003 through 2019 that is more comparable to the OEWS (all workers, not weighted by hours) as well as the distribution of annual wage and salary income in the W-2 data.
Another common measure of overall wage inequality is the ratio of wages at the 90th percentile of the wage distribution to wages at the 10th percentile of the wage distribution. We show this measure for all three data sources in chart 2 and find broadly similar patterns to those shown in chart 1. In the OEWS data, this measure is increasing from 2005 to 2013, flat from 2013 to 2016, and decreasing from 2016 to 2019. In the CPS data, this measure is increasing from 2003 to 2013 and flat or declining from 2013 to 2019. In the W-2 data, this measure is increasing from 2005 to 2010 and declining afterward. We note that the OEWS is the last series to show a decline, perhaps because each OEWS estimate is based on 3 years of data; an OEWS estimate for May 2016 uses data collected in November 2013 through May 2016.
The overall pattern of declining wage/earnings inequality after 2013 in all three data sources could be driven either by slower wage growth for high earners or by particularly strong wage growth for lower earners. Thus, chart 3 shows a common measure of wage/earnings inequality for the top half of the wage distribution, the ratio of wages at the 90th percentile of the wage distribution to wages at the 50th percentile. In the OEWS data, this measure is increasing from 2005 to 2016 and decreasing from 2016 to 2019. In the CPS data, this measure is generally increasing from 2003 to 2016, with unclear trends from 2016 to 2019. In the W-2 data, this measure is increasing from 2005 to 2012, flat from 2012 to 2013, and declining afterward.
Chart 4 shows a similar measure of wage inequality for the bottom half of the wage distribution, the ratio of wages at the 50th percentile to wages at the 10th percentile of the wage distribution. In the OEWS data, this measure is flat or declining over the whole period (except for a small increase from 2010 to 2013), with a particularly strong decline from 2016 to 2019. In the CPS data, this measure is flat or declining over the full period, with a clear decline from 2012 to 2017. In the W-2 data, this measure is flat or slightly increasing from 2005 to 2009 and declining strongly thereafter. Again, the declining wage inequality in OEWS data is similar to a pattern of declining earnings inequality in the W-2 data but with a lag.
For a more complete understanding of how overall wage inequality has been declining in recent years in the OEWS and W-2 data (but not as clearly in the CPS data), we show the percentage changes in wages or annual earnings from 2013 to 2019 for selected percentiles of the wage/earnings distribution in all three data sources in table 1. This table shows an overall greater growth in hourly wages in the CPS than in the OEWS survey, particularly for wage earners at the 75th, 90th, and 95th percentiles of the wage distribution. The W-2 data have greater growth in annual earnings than either the OEWS or the CPS at the 10th, 25th, and 50th percentiles of the wage distribution. The earnings growth levels in the W-2 data are between those of the OEWS and the CPS at the 75th, 90th, and 95th percentiles of the wage/earnings distribution (estimates of earnings growth in the W-2 data for the 5th percentile are omitted for disclosure avoidance purposes).
 Estimates of earnings growth in the W-2 data for the 5th percentile are omitted for disclosure avoidance purposes.
Note: W-2 wages are from the Form W-2 Wage and Tax Statement, reported to the Internal Revenue Service. CPS = Current Population Survey, and OEWS = Occupational Employment and Wage Statistics. Numbers in bold highlight percentiles of the wage distribution with particularly high real log wage growth from 2013 to 2019 in each of the three data sources. CPS estimates are based on the confidential version of the CPS microdata available to U.S. Bureau of Labor Statistics employees, not the public-use version of the microdata.
Source: U.S. Bureau of Labor Statistics and U.S. Census Bureau.
This greater wage growth for high-wage workers in the CPS than in the OEWS explains why overall wage inequality declined in the OEWS but plateaued in the CPS from 2013 to 2019. Both surveys show particularly strong wage growth during this period for wages at the 10th and 25th percentiles of the wage distribution.
The even stronger wage growth for low-wage and salary workers in the W-2 data drives the difference in overall wage inequality growth patterns between the W-2 and the OEWS. Perhaps the particularly strong annual earnings growth for low earners in the W-2 data reflects increases in their hours worked per week or weeks worked per year.
The analysis just discussed shows that wage growth at the 10th and 25th percentiles of the log wage distribution was particularly strong during the 2013–19 period in both the OEWS survey and CPS data and in the W-2 data. We can say more about the occupations of the worker who saw this wage growth in the OEWS survey and CPS data, because both these surveys ask workers and employers about occupations. In each survey, these responses are then assigned detailed occupational codes by statistical agency staff or their state partner agencies. A similar analysis of the occupations experiencing strong earnings growth is not currently possible in the W-2 data, because although U.S. individual tax returns ask each tax filer to report an occupation, the U.S. Census Bureau does not currently receive occupational responses from tax filers.
We start by examining workers into 10 equally sized deciles of the wage distribution in each time in each dataset. For each decile d at time t, overall wages w are the sum of wages earned by people working in each occupation j, , multiplied by the share s of people in that decile who work in each occupation j, . Thus, the overall wage level for each decile in each time is . The change in the average wage in decile between time t0 and time t1 is the change in wages for each occupation within that decile, , multiplied by the share of decile employment for each occupation at time t1, plus the change in the share of employment in the decile accounted for by each occupation, , multiplied by the initial wage level for each occupation:
We conduct this analysis at the minor occupational category (three-digit SOC code) level, slightly adjusting the SOC codes to make them consistent over time and consistent between the CPS and the OEWS. After codes are adjusted, there are 92 occupational categories.
During the 2005–19 period overall, both the CPS and the OEWS data show that the largest increases in wages were for the top three deciles of the wage distribution. Within each decile, changes in wages were overwhelmingly due to occupation-specific wage changes, rather than to changes in the distribution of occupations within the decile. However, the CPS shows larger increases in wages in the top three deciles of the wage distribution than the OEWS shows, and the CPS shows even less of this change was due to changes in the distribution of occupations within each decile than the OEWS.
From 2005 to 2013, wage inequality was increasing in both the CPS and OEWS data. In both surveys, increasing wage inequality was due to large increases in wages for the top decile of wage earners, accompanied by wage declines for most of the seven lowest deciles of wages. Again, in both surveys (but even more in the CPS than in the OEWS), these patterns were driven more by wage changes within occupations than by shifts in the occupational composition.
By contrast, from 2013 to 2019, wage inequality was declining in the OEWS and flat in the CPS data. Both surveys showed rising wages in every decile, with particularly substantial increases in wages for the second and third deciles of the wage distribution, in both the CPS and the OEWS. The decile-level decomposition of these wage changes, summed across all occupations, is shown in table 2. This table shows that in both the OEWS and the CPS, these wage increases were due almost entirely to within-occupation wage increases rather than to changes in the shares of each occupation. It also shows that the difference in overall wage inequality trends in the CPS and the OEWS from 2013 to 2019 comes from greater increases in wages for the top deciles of the wage distribution in the CPS than in the OEWS. These greater increases in wages at the top in the CPS meant that overall wage inequality in the CPS held steady, while smaller increases in wages at the top in the OEWS meant that overall wage inequality in the OEWS declined during this period.
|2013 real hourly wage||2019 real hourly wage||Difference||Change due to occupation-specific wages||Change due to occupation shares||2013 real hourly wage||2019 real hourly wage||Difference||Change due to occupation-specific wages||Change due to occupation shares|
Note: CPS = Current Population Survey, and OEWS = Occupational Employment and Wage Statistics. CPS estimates are based on the confidential version of the CPS microdata available to U.S. Bureau of Labor Statistics employees, not the public-use version of the microdata.
Source: U.S. Bureau of Labor Statistics.
We delve into the individual occupations aggregated in table 2 to answer two questions about the 2013–19 period:
1. Both the OEWS and the CPS show remarkable wage growth in the second and third deciles of the wage distribution during this period. Which occupations benefited from this wage growth?
2. The CPS showed much greater wage increases than the OEWS in the top deciles of the wage distribution during this period. Which occupations drove this difference in measured wage growth?
What occupations accrued these wage increases in the second and third deciles of the wage distribution during this period? The answers differ somewhat between the OEWS and the CPS; the correlation of occupation-specific contributions to these wage increases between the two surveys is 0.28 in the second decile and 0.22 in the third decile.
Table 3 presents the 10 minor occupational groups that contributed most to wage growth in OEWS data in the second and third deciles of the wage distribution. In the OEWS data, health aides (SOC code 31-1) were the most important minor occupational category in explaining wage growth within the second decile from 2013 to 2019, with large increases in both employment and wages in this decile of the wage distribution. This occupational group includes such occupations as home health and personal care aides (SOC code 31-1120) and nursing assistants (SOC code 31-1131). Much of the employment growth in this occupational category was the result of including about half a million additional workers in the OEWS beginning in 2017 who were providing nonmedical home care for the elderly or people with disabilities. These additional workers were added to the OEWS as the result of an effort in some states to more consistently classify such work as part of North American Industry Classification System (NAICS) code 624120, services for the elderly and disabled (covered by the OEWS) instead of NAICS code 814110, private households (not covered by the OEWS).12 The same occupational group had a smaller contribution to wage growth in the third decile of the OEWS, because it had wage growth but did not increase its share of employment in this decile of the wage distribution.
|Decile||SOC||Share of employment||Average wage||Contribution to wage growth|
|Building cleaning and pest control workers||0.044||0.054||9.46||10.58||0.15|
|Information and record clerks||0.032||0.042||9.41||10.54||0.15|
|Material recording, scheduling, dispatching, and distributing workers||0.036||0.042||9.21||10.57||0.11|
|Cooks and food preparation workers||0.051||0.055||9.45||10.61||0.09|
|39-9000||Other personal care and service workers||0.018||0.021||9.44||10.61||0.06|
|Other protective service workers||0.016||0.020||9.60||10.59||0.05|
|Supervisors of food preparation and serving workers||0.005||0.010||10.08||10.55||0.05|
|Other production occupations||0.018||0.019||9.38||10.54||0.03|
|Assemblers and fabricators||0.013||0.014||9.35||10.52||0.03|
|Food and beverage serving workers||0.057||0.073||10.90||12.16||0.26|
|Material moving workers||0.048||0.059||11.04||12.39||0.21|
|Health technologists and technicians||0.014||0.022||11.03||12.58||0.12|
|Construction trades workers||0.015||0.022||10.98||12.56||0.11|
|Other installation, maintenance, and repair occupations||0.015||0.022||11.01||12.57||0.11|
|Other educational instruction and library occupations||0.021||0.024||10.99||12.52||0.07|
|Business operations specialists||0.004||0.009||11.14||12.60||0.07|
|Counselors, social workers, and other community and social service specialists||0.009||0.014||11.01||12.56||0.07|
|Material recording, scheduling, dispatching, and distributing workers||0.043||0.043||11.02||12.45||0.06|
Note: OEWS = Occupational Employment and Wage Statistics, and SOC = Standard Occupational Classification.
Source: U.S. Bureau of Labor Statistics.
In the third decile of the OEWS wage distribution, food and beverage serving workers contributed most to wage growth, with a rising share of employment and rising wages. This occupational group consists of workers in occupations such as bartenders (SOC code 35-3011), fast food and counter workers (SOC code 35-3023), and waiters and waitresses (SOC code 35-3031). However, this occupational group did not contribute to wage growth in the second decile of the wage distribution because its employment fell sharply in this wage group. Overall, this occupational group saw little change in total employment across all wage deciles from 2013 to 2019, but strong wage growth meant that its employment shifted from the first two deciles to higher deciles of the overall wage distribution.
Similarly, material moving workers were the second most important occupational group contributing to wage growth in the third decile of the OEWS wage distribution, with rising employment and rising wages, but this occupational group also had a negative impact on wage growth in the second decile of the wage distribution because its employment fell in this decile of the wage distribution. Although the material moving occupational group grew overall from 2013 to 2019, wage growth for this occupational group meant that it made up an increasing share of employment in the third, fourth, and fifth deciles of the overall wage distribution and a decreasing share of employment in the first and second deciles.
Table 4 presents the 10 minor occupational groups that contributed most to wage growth in CPS data in the second and third deciles of the wage distribution. In both deciles, the minor occupational group that contributed most is material moving workers, which grew in employment in both deciles and had large increases in average wages. This group includes such occupations as crane and tower operators (SOC code 53-7021), industrial truck and tractor operators (SOC code 53-7050), and stockers and order fillers (SOC code 53-7065),13 an occupation that would include many e-commerce order fulfillment workers. The second most important contributing occupational group in the second decile of the wage distribution is retail sales workers, who had declining employment but sharply rising wages during 2013 to 2019. Another important contributing occupational group in both the second and third deciles of the wage distribution is health technologists and technicians, such as lab technicians (SOC code 29-2010), paramedics (SOC code 29-2040), and licensed practical nurses (SOC code 29-2061), who had growing employment and rising wages. Yet another such group is information and record clerks, such as customer service representatives (SOC code 43-4050), file clerks (SOC code 43-4071), and order clerks (SOC code 43-4150), who also had growing employment and rising wages.14
|Decile||SOC||Share of employment||Average wage||Contribution to wage growth|
|Material moving workers||0.048||0.050||$9.63||$10.99||$0.09|
|Retail sales workers||0.113||0.108||9.55||10.84||0.09|
|Health technologists and technicians||0.009||0.015||9.78||11.00||0.08|
|Information and record clerks||0.053||0.053||9.69||10.97||0.08|
|Cooks and food preparation workers||0.055||0.056||9.61||10.91||0.07|
|Other production occupations||0.024||0.027||9.53||10.99||0.07|
|Material recording, scheduling, dispatching, and distributing workers||0.040||0.041||9.59||10.86||0.06|
|Other management occupations||0.017||0.020||9.74||11.05||0.06|
|Other office and administrative support workers||0.019||0.022||9.75||10.97||0.06|
|Material moving workers||0.044||0.049||11.36||13.02||0.13|
|Information and record clerks||0.055||0.056||11.34||12.93||0.10|
|Other production occupations||0.029||0.034||11.42||12.93||0.10|
|Health technologists and technicians||0.015||0.021||11.42||12.98||0.09|
|Other office and administrative support workers||0.026||0.029||11.38||13.07||0.09|
|Cooks and food preparation workers||0.040||0.041||11.24||12.88||0.08|
|Construction trades workers||0.040||0.041||11.41||13.08||0.08|
|Other healthcare support occupations||0.015||0.020||11.35||12.98||0.08|
|Other management occupations||0.023||0.026||11.35||13.11||0.08|
|Retail sales workers||0.066||0.064||11.22||12.84||0.07|
Note: CPS = Current Population Survey, and SOC = Standard Occupational Classification.
Source: U.S. Bureau of Labor Statistics.
One possible reason we might find different occupational groups contributing the most to this wage growth in the OEWS and the CPS could be that the CPS measures wages only in workers’ main jobs, whereas the OEWS survey measures wages in all jobs. However, this reason does not appear to explain the difference in which occupations have particularly strong wage growth. Although the CPS does not record wages for workers’ second jobs, it does record the occupations worked in these jobs. The most common second jobs in the CPS are in construction trades and teaching, and these are not the jobs with particularly strong wage growth in the OEWS survey.
While both the CPS and OEWS data show strong wage growth in the second and third deciles of the wage distribution from 2013 to 2019, the CPS data show much greater wage growth in the top deciles than the OEWS data during this same period. Using the same methodology as just previously described, we can identify the occupational groups that contribute most to this difference in wage growth between the surveys. We find that the pattern of which occupations contributed most to decile-level wage increases is remarkably similar between the two surveys. The correlation of occupation-specific contributions to wage increases between surveys is 0.51 in the eighth decile, 0.71 in the ninth decile, and 0.51 in the tenth decile.
Table 5 lists the occupation-wage-decile combinations for which the difference in contributions to wage growth in the OEWS and the CPS during this period is greater than or equal to 50 cents. “Other management occupations” appears in this table in all three of the uppermost wage deciles. This occupational group includes such occupations as farm managers, construction managers, education administrators, engineering managers, food service managers, medical managers, postmasters, funeral home managers, and others. Both surveys show an increasing share of employment in the top deciles between 2013 and 2019, as well as increasing wages in both surveys but somewhat greater share growth and substantially greater wage growth in the CPS than in the OEWS. “Computer occupations” appears in this table in the top two wage deciles. This occupational category includes such occupations as computer systems analysts, computer and information research scientists, computer network support specialists, network administrators, computer programmers, and others. Again, both surveys show an increasing share of employment for this occupational category in the top deciles between 2013 and 2019, as well as increasing wages in both surveys, but somewhat greater share growth and substantially greater wage growth in the CPS than in the OEWS. The last two occupational groups in table 5 are “Health diagnosing or treating practitioners” and “Sales representatives, wholesale and manufacturing,” which appear only in the top decile. The occupational group of health diagnosing or treating practitioners includes such occupations as dentists, dieticians, optometrists, pharmacists, physician assistants, nurses, physicians, surgeons, and others. This group made up a growing share of employment in the top decile in the CPS but a falling share of employment in the OEWS during this period, and it showed greater wage growth in the CPS than in the OEWS (overall and within the top decile). Sales representatives, wholesale and manufacturing, made up a falling share of employment in the top decile in both surveys and saw an increase in wages of more than $13.00 an hour in the CPS but a decrease in wages in the OEWS of $0.23 during this period.
|Decile||SOC||Contribution to the change in wages in OEWS survey data||Contribution to the change in wages in CPS data||Difference in contributions|
|Other management occupations||$0.08||$0.58||$0.50|
|Other management occupations||0.37||0.90||0.52|
|Other management occupations||0.48||1.69||1.22|
|Healthcare diagnosing or treating practitioners||–0.00||1.05||1.06|
|Sales representative, wholesale and manufacturing||–0.36||0.17||0.53|
Note: CPS = Current Population Survey, OEWS = Occupational Employment and Wage Statistics, and SOC = Standard Occupational Classification. CPS estimates are based on the confidential version of the CPS microdata available to U.S. Bureau of Labor Statistics employees, not the public-use version of the microdata.
Source: U.S. Bureau of Labor Statistics.
The richness of the survey data from the OEWS and the CPS allow us to show how much wage inequality can be explained by various characteristics of workers and their employers and how this explanatory power has been changing over time. In the previous section, we emphasized the role of occupations in identifying which workers have seen notable changes in wages, because occupations are collected in both the OEWS and the CPS. In contrast, Hoffman and coauthors emphasize the important role of education in driving wage inequality growth in the United States in recent decades by showing that after accounting for education, occupations explain very little more of wage variation among workers.15 However, education and occupation are closely related; one mechanism for the strong relationship between education and wage inequality is the way that education determines occupations of workers. For example, Daron Acemoglu and Pascual Restrepo find that much of the divergence in wage growth from 1980 to 2016 for people of different education levels can be explained by the different exposure of people with different education levels to automation technologies.16
In this section, we closely follow Hoffman and coauthors, regressing wages on categorical dummy variables and showing how much variation in wages can be explained by each variable individually or how much additional variation can be explained in regressions that include multiple variables.17 These figures plot R2 values (and increases in R2 values from wage regressions including additional variables), multiplied by the wage variance in each period. The categorical variables used here are broad age, education, occupation, industry, and metropolitan statistical area (MSA) groups. Categories are eight 5-year age groups, five broad education categories, ten broad occupational groups using the current CPS broad occupational recoding,18 thirteen broad industry groups using the current CPS broad industry recoding, and three geographic area groups (the largest 15 metropolitan areas, all other metropolitan areas, and the balance of the United States).
In chart 5, we reproduce results from Hoffman, Lee, and Lemieux (their figure 4) using the confidential CPS-ORG data, combining data on men and women.19 Chart 5 replicates the result in Hoffman and coauthors that most of the impact of growing wage inequality comes through their “baseline,” which includes the impact only of the interaction between age/experience and schooling levels, with smaller additional amounts explained by occupation, industry, and interactions with geographic areas. However, when we reverse the order of this decomposition to show the impact of the interaction between age/experience and occupation first and then add education later, in chart 6, this pattern is reversed, showing an increasing amount of wage variation explained by occupation (with a smaller additional amount explained by education).
We cannot observe either the education or age of workers in the OEWS data, but we can observe their occupations and the industries and locations of their employers. Chart 7 shows the amount of growing wage inequality in the confidential CPS wage data that can be explained by only these three variable categories, while chart 8 shows the same results in the OEWS data.
Three notable differences are apparent between charts 7 and 8. First, as shown earlier in this article, the total amount of wage variation in the OEWS data (but not in the CPS data) is declining from the 2010–13 period to the 2016–19 period. Second, the amount of wage variation because of occupation—even the extremely broad occupational groups used in Hoffman and colleagues—is much greater in the OEWS than in the CPS. Third, the amount of wage variance that can be explained by broad occupational groups continues to increase in the CPS throughout this period, but the amount of wage variance that can be explained by occupation in the OEWS declines from the 2010–13 period to the 2016–19 period.
This article is not the first to show that more wage variation can be explained by occupation in the OEWS data than in the CPS data. Spletzer and Handwerker found a very similar result for major occupational groups in the 1998–2010 period.20 Katharine G. Abraham and James R. Spletzer found a similar result for detailed occupations in the 2003 and 2004 data and attributed it to more accurate occupational reporting by employers in the OEWS than by employees in the CPS.21
The declining amount of wage variation that can be explained by occupation in the OEWS (but not in the CPS) is consistent with the evidence shown earlier in this article. As table 4 shows, in the OEWS, several lower wage occupations had substantial wage gains from 2013 to 2019, while table 5 gives examples of high-wage occupations with much more substantial wage growth in the CPS than in the OEWS. This combination of increasing wages for lower wage occupations and relatively less wage growth for higher wage occupations in the OEWS generates this pattern of a declining amount of wage variation explained by occupation.
The explanatory power of industry or MSA groups on wage inequality in the OEWS is much less than the impact of occupational groups, even when we do not first condition on occupation. While occupational groups alone explain more than one-third of log wage variation in each OEWS period, industry groups alone can explain less than one-seventh, and MSA status alone can explain only one-fortieth.
This article documents the expansion and compression of wage and earnings inequality using three data sources: wage data collected in the OEWS employer survey, wage data collected in the CPS household survey, and W-2 earnings reports to the IRS. All three show expanded inequality of wages or earnings from 2005 to 2010. However, in the CPS, we find that wage inequality has been flat since 2013 (for all workers) or 2015 (for full-time workers), and in the OEWS, we find that wage inequality was basically flat from 2013 to 2016 and decreased from 2016 to 2019. This finding—that wage inequality did not continue rising through the 2010s—is consistent with evidence that workers’ annual wage and salary earnings in the W-2 tax data are compressing from 2010 to 2018.
All three of these very different data sources show that wages/earnings were becoming more equal among workers in the bottom half of the wage/earnings distribution from about 2013 to 2018 or 2019, because of particularly strong wage/earnings growth for workers near the 10th and 25th percentiles of the wage distribution. In the CPS and OEWS data, we show particularly strong wage growth in the second and third deciles among workers who worked in occupational groups such as material moving workers, health technologists and technicians, and information and record clerks (in the CPS) and health aides, food and beverage serving workers, and material moving workers (in the OEWS).
These results are consistent with those of Jay C. Shambaugh and Michael R. Strain, who show particularly strong wage growth at the lower percentiles of the wage distribution from 2017 to 2019.22 They are also consistent with other research suggesting a change in the trend of ever-increasing income and wage inequality in recent years. Using data from millions of households with checking accounts, Fiona Greig, Chris Wheat, George Eckerd, Melissa O’Brien, and Shantanu Banerjee examine income growth over 2-year periods, grouping households into quartiles by average incomes in their ZIP Code of residence.23 They find income growth rates for households in the lowest income quartile that exceeded the income growth rates for households in the highest income quartile beginning with the 2-year period from 2016 to 2018. Ellora Derenoncourt, Clemens Noelke, and David Weil use online job postings to study wage spillovers for low-wage workers in the same labor markets as the workers of Walmart, Target, Costco, and Amazon/Whole Foods.24 These large employers announced company policies of paying all U.S. workers wages of no less than $9–$15 per hour between 2014 and 2019. Derenoncourt and coauthors find that the wage policies of these employers not only affected wages for their own workers but also raised wage offers for lower wage workers hired by other employers in the same geographic areas.
Our examination of which occupations have played substantial roles in wage convergence during this period is accompanied by evidence that occupations play a large and growing role in wage inequality in CPS data—and even more so in wage inequality in OEWS data. However, the role of occupations in wage inequality in OEWS data was lessened from 2013 to 2019, while low-wage occupations saw large wage gains and high-wage occupations saw small wage gains in these data. Other authors, such as Hoffman, Lee, and Lemieux, have emphasized the role of education in wage inequality.25 We suggest that readers looking to reconcile these findings consider that occupation is often a mechanism by which education affects wages.
We also caution our readers that wage inequality—despite its importance—is only one piece of overall inequality in labor markets. The Organisation for Economic Cooperation and Development suggests indicators of job quality that include not only earnings but also working time arrangements (such as the share of workers with “unsocial” hours of work), job security, workplace relationships, work-related access to programs for health insurance, pensions, unemployment insurance, and family-related paid leave, as well as workplace safety.26 As an example of how workplace safety may affect the overall “quality” of an occupation, we note that despite the increasing wages we have found in multiple datasets for material moving workers, the BLS Injuries, Illnesses, and Fatalities program found that workers in the broader category of transportation and material moving occupations have among the highest rates of workplace injuries and illnesses of any broad occupational group in 2018.27
Nonetheless, wage inequality is a very important part of overall inequality in labor markets. Thus, it is important that wage inequality either plateaued or began to reverse because of high wage growth for lower wage workers in the later years of the last economic expansion.
ACKNOWLEDGMENT: Any opinions and conclusions expressed herein are those of the authors and not the U.S. Census Bureau or the U.S. Bureau of Labor Statistics. The U.S. Census Bureau reviewed the results derived from Title 13 protected data for unauthorized disclosure of confidential information and approved the disclosure avoidance practices applied to this release (approval ID: CBDRB-FY21-CES014-040). The U.S. Bureau of Labor Statistics reviewed results derived from OEWS and CPS data for unauthorized disclosure.
In our main results, we use the confidential CPS microdata not available to researchers outside BLS or the U.S. Census Bureau. The difference between the public-use and confidential versions of the CPS data is whether actual wages are reported for people earning more than $2,884.61 a week. The CPS public-use data are top coded at $2,884.61 a week to protect the confidentiality of the respondents.28 This top coding affects a growing share of respondents. In 2003, it affected 0.9 percent of respondents; in 2019, it affected 4.1 percent of respondents. Researchers working with public-use data have approached the problem of modeling earnings above this top-coding threshold in two different ways: assuming a Pareto distribution and assuming wages of 1.4 multiplied by $2,884.61 ($4,038.45). We follow two approaches to these top-coded public-use data in the published literature. Following Thomas Lemieux, we apply a uniform factor of 1.4 multiplied by the top-coded value to all these observations.29 Following Sandra A. West, Anne E. Polivka, and Barry T. Hirsch and David A. Macpherson,30 we fit a Pareto distribution to the non-top-coded observations to impute values for the top-coded observations by using 300 bins below the top-coded wage value. We note that the Pareto distribution parameters we estimate using the confidential microdata are quite similar to those published by MacPherson and Hirsch.31 Neither the 1.4 uniform factor nor the Pareto distribution approach to top-coded data is necessary in the confidential CPS microdata, because this version of the data is not top coded the same way.
Following Card and DiNardo, Lemiuex, and Donovan and Bradley,32 we remove observations with an hourly wage of less than $1 or more than $100 in 1979 dollars—less than $3.50 or more than $350 in 2016 dollars. This means dropping 0.8 percent to 0.9 percent of observations in the confidential data, varying very slightly by year, and 0.6 percent to 0.8 percent of observations in the public-use CPS data.
In appendix chart 1, we compare six variations of overall trends of the standard deviation of log hourly wage income in the CPS-ORG. Trends in
1. the confidential CPS data for all workers (the main specification shown in the article);
2. the confidential CPS data for full-time workers only, weighted by their weekly hours;
3. the public-use data for all workers, assuming a Pareto distribution of wages for those who earn more than $2,884.61 a week;
4. the public-use CPS data for full-time workers only, weighted by their weekly hours, and assuming a Pareto distribution of wages for those who earn more than $2,884.61 a week,
5. the public-use CPS data for all workers, assuming weekly earnings of $4,038.45 for those who earn more than $2,884.61 a week; and
6. the public-use CPS data for full-time workers only, weighted by their weekly hours, and assuming weekly earnings of $4,038.45 for those who earn more than $2,884.61 a week.
Comparing worker restrictions and weightings within each version of the CPS data (comparing 1 with 2, 3 with 4, and 5 with 6), we find that without weighting by the number of hours worked, a greater dispersion of wages is found among all workers than among full-time workers. Weighting full-time workers by the number of hours they work increases the dispersion in their wages, since higher earning full-time workers work more hours. In all versions of the CPS data, a greater dispersion of wages is found overall among all workers, not weighted by hours worked (estimate more comparable to the OEWS) than among full-time workers only, weighted by hours worked (estimates more comparable to the wage variation literature).
Comparing the various versions of the CPS data (comparing 1 with 3 and 5 and 2 with 4 and 6), we find much greater variation in overall wages when wages above $2,884.61 a week are modeled using a Pareto distribution fit to the shape of the earning distribution below $2,884.61 a week than if all wages above $2,885.00 are assumed to be $4,038.45 ($2,884.61 multiplied by 1.4). We also find that the assumption that all these top-coded earners earn $4,038.45 a week better fits the trend in the confidential version of these data than modeling the top-coded data using a Pareto distribution each period. The use of the Pareto distribution to model wages above the top code introduces more variation in wages in recent years than exists in the confidential data. Both the public-use version of the CPS wage data—using the assumption that top-coded wages are $4,038.45 a week—and the confidential version of the CPS data show that wage inequality plateaued after 2012.
Matthew Dey, Elizabeth Weber Handwerker, David S. Piccone Jr, and John Voorheis, "Were wages converging during the 2010s expansion?," Monthly Labor Review, U.S. Bureau of Labor Statistics, June 2022, https://doi.org/10.21916/mlr.2022.19
1 Lawrence F. Katz and David Autor, “Changes in the wage structure and earnings inequality,” Handbook of Labor Economics, vol. 3, part A, 1999, pp. 1463–1555.
2 David S. Lee, “Wage inequality in the United States during the 1980s: rising dispersion or falling minimum wage?” The Quarterly Journal of Economics, vol. 114, no. 3, August 1999, pp. 977–1,023, https://doi.org/10.1162/003355399556197; and David Card and John E. DiNardo, “Skill-biased technological change and rising wage inequality: some problems and puzzles,” Journal of Labor Economics, vol. 20, no. 4, 2002, pp. 733–783, https://www.journals.uchicago.edu/doi/abs/10.1086/342055.
3 David H. Autor, Lawrence F. Katz, and Melissa S. Kearney, “Trends in U.S. wage inequality: revising the revisionists,” The Review of Economics and Statistics, vol. 90, no. 2, May 2008, pp. 300–323.
4 Thomas Piketty and Emmanuel Saez, “Income inequality in the United States, 1913–1998,” The Quarterly Journal of Economics, vol. 118, no. 1, February 2003, pp. 1–41, https://doi.org/10.1162/00335530360535135.
5 Florian Hoffmann, David S. Lee, and Thomas Lemieux, “Growing income inequality in the United States and other advanced economies,” Journal of Economic Perspectives, vol. 34, no. 4, 2020, pp. 52–78, https://doi.org/10.1257/jep.34.4.52.
6 Jay C. Shambaugh and Michael R. Strain, “The recovery from the Great Recession: a long, evolving expansion,” The ANNALS of the American Academy of Political and Social Science, vol. 695, no. 1, August 2021, pp. 28-48, https://doi.org/10.1177/00027162211022305.
7 James R. Spletzer and Elizabeth Weber Handwerker, “Measuring the distribution of wages in the United States from 1996 through 2010 using the Occupational Employment Survey,” Monthly Labor Review, May 2014, https://doi.org/10.21916/mlr.2014.18.
8 Ibid. Spletzer and Handwerker show that collecting wage data in intervals in the Occupational Employment and Wage Statistics (OEWS) survey has a minimal impact on overall wage inequality measures; when they apply the OEWS wage intervals to Current Population Survey (CPS) data, the reduction in measured wage variance is very small.
9 For detailed information on OEWS, see “Occupational Employment and Wage Statistics: overview,” Handbook of Methods (U.S. Bureau of Labor Statistics, last modified March 31, 2021), https://www.bls.gov/opub/hom/oews/home.htm.
10 Anne E. Polivka, “Data watch: The redesigned current population survey,” Journal of Economic Perspectives, vol. 10, no. 3, 1996, pp. 169–180; and John Schmitt, “Creating a consistent hourly wage series from the Current Population Survey’s Outgoing Rotation Group, 1979–2002,” version 0.9 (Washington, DC: Center for Economic and Policy Research, August 2003).
11 Card and DiNardo, “Skill-biased technological change and rising wage inequality”; Thomas Lemieux, “Increasing residual wage inequality: composition effects, noisy data, or rising demand for skill?” American Economic Review, vol. 96, no. 3, June 2006, pp. 461–498, https://doi.org/10.1257/aer.96.3.461; and Sarah A. Donovan and David H. Bradley, “Real wage trends, 1979 to 2019,”Report no. R45090 (Washington, DC: Congressional Research Service, December 2020), https://crsreports.congress.gov/product/details?prodcode=R45090.
12 For more information about this change, see Occupational Employment and Wages: May 2017, USDL-18-0486 (U.S. Department of Labor, March 30, 2018), p. 9, https://www.bls.gov/news.release/archives/ocwage_03302018.pdf.
13 Stock and order fillers were part of the clerical occupational group in the 2000 and 2010 Standard Occupational Classification systems and were moved into the transportation and material moving occupational group in the 2018 Occupational Classification System revisions. For consistency, we consider this occupation part of the 53-7000 material moving workers occupational group in all years.
14 The within-decile employment and wage trends hold for these occupations as a whole in the CPS data, not just within the second and third deciles of the wage distribution. During this period, material moving workers, grew in employment and had large increases in average wages overall; retail sales workers, had declining employment but rising wages overall, and health technologists and technicians as well as information and record clerks had growing employment and wages overall.
15 Hoffmann et al., “Growing income inequality in the United States and other advanced economies.”
16 Daron Acemoglu and Pascual Restrepo, “Tasks, automation, and the rise in US wage inequality,” Working Paper No. 28920 (Cambridge, MA: National Bureau of Economic Research, June 2021), https://www.nber.org/papers/w28920.
17 Hoffmann et al., “Growing income inequality in the United States and other advanced economies.”
18 Ibid. Hoffmann and colleagues use a 9-category broad occupational grouping consistent back to 1980, but we prefer the modern 10-category broad occupational grouping because the older grouping conflates higher paying computer occupations with somewhat lower paying technician occupations. We estimated these results in CPS public-use data using both the occupational grouping used by Hoffmann et al. and the modern broad occupational grouping and found the choice of occupational grouping had minimal impact on the results.
19 Hoffmann et al., “Growing income inequality in the United States and other advanced economies.”
20 Spletzer and Handwerker, “Measuring the distribution of wages in the United States from 1996 through 2010 using the Occupational Employment Survey.”
21 Katharine G. Abraham and James R. Spletzer, “New evidence on the returns to job skills,” American Economic Review, vol. 99, no. 2, May 2009, pp. 52–57, .
22 Shambaugh and Strain, “The recovery from the Great Recession: a long, evolving expansion.”
23 Fiona Greig, Chris Wheat, George Eckerd, Melissa O’Brien, and Shantanu Banerjee, “How did the distribution of income growth change alongside the hot pre-pandemic labor market and recent fiscal stimulus?” (JP Morgan Chase Research Institute, September 2021), https://www.jpmorganchase.com/institute/research/household-income-spending/how-did-the-distribution-of-income-growth-change-alongside-the-hot-pre-pandemic-labor-market-and-recent-fiscal-stimulus/
24 Ellora Derenoncourt, Clemens Noelke, David Weil, and Bledi Taska, “Spillover effects from voluntary employer minimum wages” Working Paper 29425 (Cambridge, MA: National Bureau of Economic Research, October 2021), https://doi.org/10.3386/w29425.
25 Hoffmann et al., “Growing income inequality in the United States and other advanced economies.”
26 Organisation for Economic Cooperation and Development, “Well-being in the workplace: measuring job quality,” in How’s Life? 2013: Measuring Well-being (Paris: OECD Publishing, 2013), https://doi.org/10.1787/9789264201392-en.
27 “Transportation and material moving workers experienced 184,470 injuries and illnesses in 2018,” TED: The Economics Daily (U.S. Bureau of Labor Statistics, December 6, 2019), https://www.bls.gov/opub/ted/2019/transportation-and-material-moving-workers-experienced-184470-injuries-and-illnesses-in-2018.htm.
28 “Topcoding of usual hourly earnings,” Current Population Survey (U.S. Census Bureau, last revised October 8, 2021), https://www.census.gov/programs-surveys/cps/technical-documentation/methodology/topcoding-of-usual-hourly-earnings.html.
29 Lemieux, “Increasing residual wage inequality.”
30 Sandra A. West, “Estimation of the mean from censored income data,” unpublished working paper, U.S. Bureau of Labor Statistics, 1985; Anne E. Polivka, “Data watch: the redesigned Current Population Survey,” Journal of Economic Perspectives, vol. 10, no. 3, 1996, pp. 169–180, https://doi.org/10.1257/jep.10.3.169; and Barry T. Hirsch and David A. MacPherson, “Union membership and coverage database from the Current Population Survey: note,” ILR Review, vol. 56, no. 2, January 2003, pp. 349–354, https://doi.org/10.1177/001979390305600208.
31 David A. MacPherson and Barry T. Hirsch, “Five decades of union wages, nonunion wages, and union wage gaps at unionstats.com,” IZA Discussion Paper 14398 (Bonn, Germany: IZA—Institute of Labor Economics, May 2021), http://ftp.iza.org/dp14398.pdf and https://www.unionstats.com/Pareto%20Top-Code%20Earnings.xls
32 Card and DiNardo, “Skill-biased technological change and rising wage inequality”; Lemieux, “Increasing residual wage inequality”; and Donovan and Bradley, “Real wage trends, 1979–2019.”