Income inequality is rising so fast, data can’t keep up

Wages at the top of the U.S. income distribution continue to rise much more rapidly than wages for everyone else, according to an analysis of the latest federal data by the Economic Policy Institute, a progressive think tank.

But the data are just as notable for what they don’t say, according to the report by EPI economist Elise Gould. Increases in wages at the top are outpacing economists’ ability to measure them because the federal survey tracking the wage data “top-codes” the highest earnings amounts: For confidentiality reasons, wages are fully recorded only up to a certain threshold. The Bureau of Labor Statistics and the Census Bureau, which jointly administer the survey, haven’t changed that threshold in 20 years, even as top incomes have skyrocketed. As a result, the survey is capturing less information on top pay than it used to.

“All workers who report weekly earnings above $2,884.61 (annual earnings for full-year workers above $150,000) are recorded as having weekly earnings of exactly $2,884.61, to preserve the anonymity of respondents,” Gould writes. That top-code threshold hasn’t been updated since 1998. As a result, the survey is becoming less useful for tracking top incomes at a time when public concern over inequality is growing. Representatives from the Census Bureau did not immediately respond to a request for comment.

“In the overall wage distribution, over the period analyzed in this report, the share of workers reporting weekly earnings at or above the top-code rose from 0.8% in 2000 to 4.2% in 2018,” the report says. “In 2018, the share of white workers with weekly earnings hitting the top-code was 5.2%. For working men, that share was 5.9% in 2018.”

Economists can use a variety of statistical methods to adjust for top-coding, but that process becomes much less reliable as more workers’ wages are masked by the code. “Now that the top-code hits over 5% of the wage distribution of men and white workers,” Gould writes, “our 95th-percentile wage estimate has essentially become the weekly earnings top-code divided by a measure of usual hours. This makes our measure of 95th-percentile wages for working men and white workers unreliable.”

Nevertheless, the available data paint a clear picture of broadening disparities between top earners and everyone else. Adjusted for inflation, wages for the top 5% of earners rose from $50.46 an hour in 2000 to $63.10 in 2018, an increase of 25%. The median worker’s hourly wage, meanwhile, rose by 7% over that period, to $18.80.

The divide was particularly stark among men: The wages of the top 5% of male workers rose by 42% from 2000 to 2018, while those of the median male worker rose by 0.8%.

One way to capture that dispersion is to look at the ratio between wages at the top and wages at the middle. In 2000, for instance, the top 5% of workers earned 2.87 times as much, per hour, as the typical, or median, worker. By 2018, that ratio had increased to 3.36. Among men only, the ratio between the top 5% and the median increased from 2.83 to 3.98.

However, there was little change in wage variation at the bottom of the distribution: The median worker earned roughly twice as much as the bottom 10% of workers in 2000 and 2018, among all workers and among men only.

The report notes that bottom wages grew the fastest in the states that saw some form of minimum wage increase. From 2013 to 2018, for instance, 10th-percentile wages grew by 13% in states with a minimum wage increase and by 8.4% in states with no minimum wage hike. That finding adds to existing evidence that raising the minimum wage is an effective way to bring up earnings at the bottom of the income distribution.

But the data also show that rising inequality is primarily a function of what’s happening at the top of the wage distribution, rather than the bottom of it. Unless there’s a change in how the federal government collects wage data on top earners, we’ll continue to know less and less about what’s happening among some of the country’s richest people.