Stagnant top-code thresholds threaten data reliability for the highest earners and make inequality difficult to accurately measure
Measuring wage growth, particularly at high wage levels, has become a difficult task. The most useful, publicly-available data for measuring trends in hourly wages is the Current Population Survey (CPS). Analyses from CPS, for example, (like this one here) are a key reason why we know that except for brief periods of decent wage growth for middle- or low-wage workers between 1979 and today, wage growth for most workers has been slow while wage growth for top earners has been far more rapid.
Unfortunately, it is increasingly difficult to report accurate data on top-end wages or wage changes using the CPS because of growing inequality and measurement issues related to top-coding, an earnings reporting method that hasn’t kept up with rising wages for top earners and thus limits measuring of earnings above a certain threshold (and therefore makes it difficult to accurately measure average wages as well). “Top codes” used in the CPS assign observations that report wages over some threshold the identical “top-coded” value in the data (the current top code is $2,884.61 in weekly wages). The top code is not updated annually for inflation or anything else. Consequently, it suppresses a larger and larger fraction of the CPS data over time. If wage-growth for workers who make more than the top code has been systematically faster than for workers beneath the top code, we will miss out on just how much wage inequality has risen.
This creates a real problem for assessing overall trends in inequality, because so much income (including wage income) has been concentrated at the very top of distributions. For example, data on annual wage growth from the Social Security Administration that is not top coded show substantially more dramatic growth in inequality than what is apparent in CPS data (see an analysis of this SSA data here). Given that the overwhelming majority of high-wage workers work full-time, it is surely the case that the SSA data for the highest-paid workers is mostly representative of very rapid growth in hourly wages, not hours worked. Yet we cannot really validate this with direct measures of hourly wages available to use in the CPS.
Aside from the problems it presents for interpreting overall trends in inequality, the CPS top code has radically different implications for analyses of smaller demographic groups. At the Economic Policy Institute (EPI), we provide labor market and wage analysis on our State of Working America Data Library page based on the best available data. When we attempt to analyze wages among demographic groups, by gender, race/ethnicity, and/or education, the parts of the wage distribution we are able to measure is even smaller because more and more observations’ data are suppressed by the top code.
The failure to adjust the top code to better capture wages at the high end isn’t a new phenomenon. Between 1973 and 1988, the top code for weekly earnings was constant in nominal dollars at $999 per week, even as inflation ran in double-digits over some of the intervening years. Then, it stayed at $1,923 per week between 1989 and 1997. The latest change in the top code was made back in 1998; the current top code has now sat at $2,884.61 per week in nominal value for the last twenty four years. Even if wages did not grow in real terms and only kept up with inflation, the ability to measure high end wages would be compromised. But, in most years, high end wages grew far faster than inflation, increasing the pace at which wages could not be reliably measured.
The Census Bureau has announced upcoming changes to the top-coding procedure for usual weekly earnings and usual hourly earnings data, which will certainly improve data analysis moving forward. But, there has been no indication that data will be changed historically, which does not solve the problem of trying to uncover high end earnings trends over the last few decades.
We start our display of the top-code issue with an examination of the shares of each demographic group that have been top-coded over time. Figure A displays top-coding shares overall as well as by gender from 1979 to 2021. Figure B displays the same for White, Black, Hispanic, and Asian American and Pacific Islander (AAPI) workers. And, Figure C displays the same trends by educational attainment.
Figure A illustrates the difficulties in measuring between and within group inequality at the high end of the wage distribution resulting from a significant portion of men being top-coded. This is particularly noticeable during 1985–1988 and 2018–2021. The share of men who were top-coded increased sharply through the 1980s, exceeding 5% of workers between 1986 and 1988, and then reached an all-time high of 7.7% in 2021. The higher shares of men top-coded compared with women is not surprising given the fact that they are far more likely to be found in higher paying jobs in the U.S. economy while women continue to face a significant gender pay gap, particularly at the middle and upper portion of the wage distribution, even among women with higher levels of educational attainment.
The top 5% of men’s wages have been topcoded in 8 of the last 43 years: Topcode shares by overall and gender, 1979–2021
| Year | All | Men | Women |
|---|---|---|---|
| 1979 | 0.6% | 1.0% | 0.1% |
| 1980 | 0.7% | 1.2% | 0.1% |
| 1981 | 1.0% | 1.8% | 0.1% |
| 1982 | 1.4% | 2.5% | 0.2% |
| 1983 | 1.8% | 3.3% | 0.3% |
| 1984 | 2.3% | 3.9% | 0.4% |
| 1985 | 2.7% | 4.7% | 0.6% |
| 1986 | 3.2% | 5.5% | 0.7% |
| 1987 | 3.7% | 6.4% | 0.9% |
| 1988 | 4.5% | 7.4% | 1.1% |
| 1989 | 0.5% | 0.8% | 0.1% |
| 1990 | 0.6% | 1.0% | 0.1% |
| 1991 | 0.7% | 1.2% | 0.2% |
| 1992 | 0.7% | 1.3% | 0.2% |
| 1993 | 0.8% | 1.4% | 0.2% |
| 1994 | 1.2% | 2.0% | 0.4% |
| 1995 | 1.3% | 2.2% | 0.4% |
| 1996 | 1.4% | 2.3% | 0.4% |
| 1997 | 1.7% | 2.6% | 0.6% |
| 1998 | 0.6% | 1.0% | 0.2% |
| 1999 | 0.7% | 1.1% | 0.2% |
| 2000 | 0.8% | 1.3% | 0.3% |
| 2001 | 0.9% | 1.4% | 0.3% |
| 2002 | 1.0% | 1.6% | 0.4% |
| 2003 | 1.1% | 1.6% | 0.4% |
| 2004 | 1.2% | 1.9% | 0.4% |
| 2005 | 1.3% | 2.0% | 0.5% |
| 2006 | 1.5% | 2.3% | 0.6% |
| 2007 | 1.7% | 2.6% | 0.7% |
| 2008 | 1.9% | 3.0% | 0.9% |
| 2009 | 2.1% | 3.2% | 0.9% |
| 2010 | 2.3% | 3.4% | 1.0% |
| 2011 | 2.3% | 3.4% | 1.1% |
| 2012 | 2.6% | 3.9% | 1.2% |
| 2013 | 2.8% | 4.2% | 1.4% |
| 2014 | 2.9% | 4.2% | 1.5% |
| 2015 | 3.2% | 4.6% | 1.6% |
| 2016 | 3.5% | 4.9% | 1.7% |
| 2017 | 3.7% | 5.1% | 2.0% |
| 2018 | 4.2% | 6.0% | 2.3% |
| 2019 | 4.6% | 6.3% | 2.6% |
| 2020 | 5.4% | 7.4% | 3.2% |
| 2021 | 5.8% | 7.7% | 3.5% |
Notes: The topcode nominal dollar thresholds for weekly earnings have been updated three times since 1973. The topcode value was $999 from 1973–1988, $1923 from 1989–1997, and has currently sat at $2,884.61 since 1998.
Source: Authors’ analysis of EPI Microdata Extracts.
Figure B makes apparent how the historical and current discrimination against Black and Hispanic workers has meant that they are less likely to be in higher paying jobs, thus leading them to have significantly lower shares of top-coded workers than their AAPI and white counterparts. Between 1979 and 1997, the share of Black and Hispanic workers who were top-coded did not exceed 1.6% for either group, while “other” (largely AAPI) workers and white workers’ shares hit 4.8% and 5.1%, respectively. After the top-code threshold was reset in 1998, Black and Hispanic shares remained low until the last couple of years, hitting 2.5% and 2.6%, respectively. At the same time, AAPI and white workers experienced significant increases in the shares of their workforce hitting the top code, 10.9% and 6.8%, respectively. Ultimately, the significant increase in top-code shares among AAPI and white workers makes measuring the true high-end wage inequality within and across racial and ethnic groups nearly impossible; in fact, it also makes calculating average wages more challenging—relying more heavily on imputation assumptions—if the upper end is increasingly top-coded.
Earnings at the top for AAPI and white workers have been impossible to accurately measure since the mid 2010s: Topcode shares by race/ethnicity, 1979–2021
| Year | White | Black | Hispanic | AAPI | Other |
|---|---|---|---|---|---|
| 1979 | 0.6% | 0.1% | 0.1% | 0.4% | |
| 1980 | 0.8% | 0.1% | 0.3% | 0.7% | |
| 1981 | 1.2% | 0.2% | 0.4% | 1.0% | |
| 1982 | 1.6% | 0.3% | 0.4% | 1.9% | |
| 1983 | 2.1% | 0.5% | 0.6% | 2.0% | |
| 1984 | 2.6% | 0.7% | 0.7% | 2.0% | |
| 1985 | 3.1% | 0.8% | 0.9% | 2.5% | |
| 1986 | 3.7% | 0.8% | 1.1% | 3.8% | |
| 1987 | 4.3% | 1.2% | 1.2% | 4.3% | |
| 1988 | 5.1% | 1.6% | 1.4% | 4.8% | |
| 1989 | 0.6% | 0.0% | 0.2% | 0.4% | 0.4% |
| 1990 | 0.7% | 0.1% | 0.2% | 0.8% | |
| 1991 | 0.8% | 0.1% | 0.2% | 0.9% | |
| 1992 | 0.9% | 0.1% | 0.1% | 1.1% | |
| 1993 | 1.0% | 0.2% | 0.2% | 0.9% | |
| 1994 | 1.4% | 0.5% | 0.4% | 1.4% | |
| 1995 | 1.6% | 0.3% | 0.4% | 1.6% | |
| 1996 | 1.7% | 0.4% | 0.5% | 1.8% | |
| 1997 | 2.0% | 0.5% | 0.5% | 2.4% | |
| 1998 | 0.8% | 0.2% | 0.2% | 0.6% | |
| 1999 | 0.9% | 0.2% | 0.2% | 0.7% | |
| 2000 | 1.0% | 0.3% | 0.3% | 1.1% | |
| 2001 | 1.1% | 0.3% | 0.2% | 1.3% | |
| 2002 | 1.3% | 0.2% | 0.3% | 1.4% | |
| 2003 | 1.3% | 0.3% | 0.3% | 1.3% | |
| 2004 | 1.4% | 0.3% | 0.5% | 1.7% | |
| 2005 | 1.6% | 0.3% | 0.4% | 1.6% | |
| 2006 | 1.9% | 0.4% | 0.4% | 2.1% | |
| 2007 | 2.1% | 0.6% | 0.5% | 2.2% | |
| 2008 | 2.4% | 0.7% | 0.7% | 2.8% | |
| 2009 | 2.5% | 0.8% | 0.8% | 2.6% | |
| 2010 | 2.8% | 0.8% | 0.9% | 3.0% | |
| 2011 | 2.8% | 1.0% | 0.8% | 3.2% | |
| 2012 | 3.2% | 1.1% | 0.9% | 3.9% | |
| 2013 | 3.4% | 1.1% | 1.0% | 4.1% | |
| 2014 | 3.5% | 1.2% | 1.0% | 4.6% | |
| 2015 | 3.9% | 1.4% | 1.3% | 5.3% | |
| 2016 | 4.2% | 1.6% | 1.3% | 5.9% | |
| 2017 | 4.5% | 1.6% | 1.4% | 6.2% | |
| 2018 | 5.2% | 1.8% | 1.6% | 6.6% | |
| 2019 | 5.5% | 2.1% | 1.9% | 8.3% | |
| 2020 | 6.4% | 2.4% | 2.3% | 10.3% | |
| 2021 | 6.8% | 2.6% | 2.5% | 10.9% | |
Notes: The topcode nominal dollar thresholds for weekly earnings have been updated three times since 1973. The topcode value was $999 from 1973–1988, $1923 from 1989–1997, and has currently sat at $2,884.61 since 1998. AAPI refers to Asian American and Pacific Islander. Race/ethnicity categories are mutually exclusive (i.e., white non-Hispanic, Black non-Hispanic, AAPI non-Hispanic, and Hispanic any race). Prior to 1989, AAPI was not reported separately and included in “Other.”
Source: Authors’ analysis of EPI Microdata Extracts.
Figure C demonstrates how higher levels of educational attainment are related to higher earnings. There is a consistent and significant gap between the top-code share of workers with an advanced or college degree and those with the next highest level of educational attainment (some college experience). In 1988, 20.1% of workers with an advanced degree were top-coded, while the same was true for 11.0% of workers with a college degree. The share of workers with some college experience who were top-coded was only 2.9%. This large gap persisted in 2021, with the top-code shares being 17.0% and 9.3% for workers with an advanced and college degree respectively while workers with some college experience had only 1.9% top-coded. The significant shares of top-coded workers within the advanced degree and college educational attainment groups increases the difficulty of measuring their wages both alone, and in comparison, to other groups.
Advanced degree holders' wages exceed 5% topcode share in all but 10 years since 1979: Topcode shares by education, 1979–2021
| Year | Less than high school | High school | Some college | College | Advanced degree |
|---|---|---|---|---|---|
| 1979 | 0.1% | 0.2% | 0.4% | 1.6% | 3.3% |
| 1980 | 0.1% | 0.2% | 0.5% | 2.1% | 3.8% |
| 1981 | 0.2% | 0.4% | 0.7% | 3.1% | 5.1% |
| 1982 | 0.2% | 0.4% | 0.8% | 4.0% | 7.3% |
| 1983 | 0.3% | 0.5% | 1.1% | 5.1% | 9.1% |
| 1984 | 0.3% | 0.7% | 1.5% | 6.1% | 11.2% |
| 1985 | 0.4% | 0.9% | 1.9% | 6.8% | 13.0% |
| 1986 | 0.3% | 1.0% | 2.2% | 8.3% | 15.3% |
| 1987 | 0.4% | 1.1% | 2.6% | 9.6% | 17.6% |
| 1988 | 0.6% | 1.4% | 2.9% | 11.0% | 20.1% |
| 1989 | 0.0% | 0.1% | 0.3% | 1.1% | 2.7% |
| 1990 | 0.1% | 0.1% | 0.3% | 1.3% | 3.4% |
| 1991 | 0.0% | 0.1% | 0.3% | 1.5% | 4.3% |
| 1992 | 0.0% | 0.1% | 0.3% | 1.7% | 4.5% |
| 1993 | 0.0% | 0.1% | 0.3% | 1.8% | 5.1% |
| 1994 | 0.2% | 0.3% | 0.5% | 2.7% | 6.8% |
| 1995 | 0.1% | 0.2% | 0.5% | 2.9% | 7.7% |
| 1996 | 0.1% | 0.2% | 0.6% | 3.0% | 7.8% |
| 1997 | 0.1% | 0.4% | 0.7% | 3.6% | 8.9% |
| 1998 | 0.1% | 0.1% | 0.3% | 1.3% | 3.3% |
| 1999 | 0.1% | 0.1% | 0.2% | 1.6% | 3.7% |
| 2000 | 0.0% | 0.1% | 0.3% | 1.8% | 4.1% |
| 2001 | 0.1% | 0.1% | 0.3% | 2.1% | 4.2% |
| 2002 | 0.1% | 0.2% | 0.4% | 2.2% | 4.9% |
| 2003 | 0.1% | 0.2% | 0.4% | 2.3% | 4.8% |
| 2004 | 0.1% | 0.3% | 0.4% | 2.3% | 5.7% |
| 2005 | 0.1% | 0.2% | 0.5% | 2.6% | 5.9% |
| 2006 | 0.1% | 0.3% | 0.5% | 2.9% | 6.9% |
| 2007 | 0.1% | 0.3% | 0.6% | 3.3% | 7.3% |
| 2008 | 0.1% | 0.5% | 0.7% | 3.6% | 8.0% |
| 2009 | 0.2% | 0.4% | 0.7% | 3.7% | 8.9% |
| 2010 | 0.2% | 0.5% | 0.7% | 4.2% | 8.9% |
| 2011 | 0.2% | 0.5% | 0.8% | 4.1% | 8.9% |
| 2012 | 0.2% | 0.5% | 0.7% | 4.6% | 10.5% |
| 2013 | 0.2% | 0.5% | 0.8% | 5.0% | 10.6% |
| 2014 | 0.2% | 0.6% | 0.9% | 5.0% | 10.6% |
| 2015 | 0.3% | 0.8% | 1.2% | 5.5% | 10.9% |
| 2016 | 0.2% | 0.7% | 1.0% | 6.1% | 11.9% |
| 2017 | 0.4% | 0.8% | 1.0% | 6.4% | 12.3% |
| 2018 | 0.4% | 1.0% | 1.3% | 6.8% | 14.1% |
| 2019 | 0.4% | 1.0% | 1.4% | 7.5% | 14.8% |
| 2020 | 0.4% | 1.2% | 1.8% | 8.4% | 16.3% |
| 2021 | 0.5% | 1.2% | 1.9% | 9.3% | 17.0% |
Notes: The topcode nominal dollar thresholds for weekly earnings have been updated three times since 1973. The topcode value was $999 from 1973–1988, $1923 from 1989–1997, and has currently sat at $2,884.61 since 1998.
Source: Authors’ analysis of EPI Microdata Extracts.
Of course, the difficulties of measuring high end earnings are even more acute when we look at demographic groups that cut across gender, race/ethnicity, and education.
Figure D makes the power of intersectionality evident. Workers who identify with the groups with the largest top-code shares across gender, race, and education have earnings which are by far the most difficult to measure. Men (7.7%), AAPI workers (10.9%), and those with advanced degrees (17.0%) have the largest top-code shares of their respective demographic groups. However, when you examine these demographics in combination, AAPI men with advanced degrees, the share of workers who are top-coded is even more significant (29.2%). The group with the second largest top-code shares across all combinations of gender, race, and education are white men with advanced degrees (25.6%). For these groups, and all those displayed in Figure D, high-end earnings are exceedingly difficult to measure due to large top-code shares.
For some demographic groups, even 85th percentile wages are not measurable: Topcode shares for specific groups, 2021
| Share for specific groups | |
|---|---|
| All | 5.8% |
| White women, advanced degree | 10.1% |
| AAPI men, college degree | 13.1% |
| White men, college degree | 14.9% |
| AAPI women, advanced degree | 15.1% |
| Black men, advanced degree | 15.6% |
| Hispanic men, advanced degree | 18.8% |
| White men, advanced degree | 25.6% |
| AAPI men, advanced degree | 29.2% |
Notes: The topcode nominal dollar thresholds for weekly earnings have been updated three times since 1973. The topcode value was $999 from 1973–1988, $1923 from 1989–1997, and has currently sat at $2,884.61 since 1998. AAPI refers to Asian American and Pacific Islander. Race/ethnicity categories are mutually exclusive (i.e., white non-Hispanic, Black non-Hispanic, AAPI non-Hispanic, and Hispanic any race).
Source: Authors’ analysis of EPI Microdata Extracts.
See our State of Working America Data Library page for wages by percentile, including NA’s for percentile values that can’t be reliably measured.
It’s unfortunate that policy has not made a significant contribution to reining in rising wage inequality over these years. But allowing rising inequality to mechanically obscure even its own measurement seems truly absurd, and yet extremely easy to fix. The BLS should commit to higher and far more regularly updated top codes, or find some other way to allow researchers to get a clearer sense of what is happening to wage inequality than what the CPS currently allows.
Enjoyed this post?
Sign up for EPI's newsletter so you never miss our research and insights on ways to make the economy work better for everyone.