The “labor share” refers to the share of income in an economy paid to labor. Back in the dark ages of the late 1970s and early 1980s when I first started studying economics, it was often assumed that the labor share was more-or-less a constant term over decades. But in the last to decades, the labor share has dropped, both for the US economy and across most high-income countries.
For some perspective over time, the graph shows a time series of the labor share for the nonfarm business sector of US economy going back to the late 1940s. The vertical axis is set equal to 100 in 2017, which is not super-convenient for my purposes, but for eyeballing the chart, it’s roughly correct to say that 112.5 on the vertical axis is a labor share of about 64% of output produced by this part of the economy.
Thus, you can see a fall in the labor share in the 1960s, and then a bounceback in the late ’60s; a fall in the early 1970s, and a bounceback in the late ’70s. When there was a fall in the early 1980s, looking at this data, I found myself thinking “wait a few yeas to see what happens,” and there’s something of a bounceback in the late 1980s. There’s again a fall in the early 1990s, but a hearty bounceback in the late 1990s and early 2000s. With this history, when the labor share fell in the early 2000s, I again found myself saying “wait a few years to see what happens.” But as you can see on the figure, the prevailing level in the last 15 years or so (with some fluctuation around the pandemic) has been lower–more like 57% of total output than the earlier level of 64%.
There are (at least) two big questions here: measurement and causes. Loukas Karabarbounis takes on both topics in the Spring 2024 issue of the Journal of Economic Perspectives in “Perspectives on the Labor Share.” (Full disclosure: I’ve been Managing Editor of JEP for 38 years. All articles back to the first issue are available free of charge, compliments of the American Economic Association, which publishes the journal.)
The hyperattentive reader will have noticed that the figure above is the “nonfarm business sector,” which leaves out both agriculture (not a huge part of the US economy in recent decades) and also government and nonprofits, as well as the output of services received by those living in owner-occupied housing. This has been the traditional method of calculating labor share, but what if one looks instead at labor share of the entire GDP?
As intro econ students are taught, the GDP of a country can be measured in several ways. The best-known way is as a measure of output. But in the economic statistics, output only counts when it is sold, which means that output is also a form of income for someone, somewhere–workers, owners of firms, others. Thus, you can also measure GDP by adding up categories of income (along with allowing for income that just offsets depreciation of existing capital). Here are the main categories of income for 2019 from the official national income and product accounts:
Looking at the rows, it’s clear that “compensation to employees” is labor income. It’s clear that corporate profits, rental income, and interest are not labor income. The conceptual problem arises in two categories: taxes (less subsidies) on production/imports, and “proprietor’s income,” which refers to businesses owned by the person (or people) who runs them, like certain car dealerships, franchises, or health care practices. For the taxes/subsidies category, standard practice seems to be just to divide them proportionally between capital and labor; in this way, it doesn’t influence the labor share. But how to think about “proprietor’s income” is hard.
Consider someone who owns a car dealership or a fast-food franchise. Say the firm makes money, and so the owner get that money at the end of the year. Should that money be treated as labor income? Or is it really a business profit that should be treated as a return to capital? Or should it be divided in some way? The question is hard, and rather than try to resolve it, Karabarbounis calculates the labor share several different ways. For example, you can treat “proprietor’s income” as divided equally between labor and non-labor income, or you can can treat the “proprietors’s income” that matches employee income as labor, and the rest as capital income. Karabarbounis also suggests looking at the labor share just for output and income from corporations, which is about 50-60% of GDP, but avoids the question of how to divide up other kinds of income.
When he compares these various measures (and there are more measures in the appendix of the paper for those who can’t get enough of this stuff), they all show a decline in the labor share in the last couple of decades. The method closest to the official Bureau of Labor Statistics data shown above shows the biggest declines. The decline in labor share also holds across almost all US industries.
In international data, the decline in labor share appears across most large economies as well. He writes:
[Consider} the labor share of the 16 largest economies of the world based on 2015 GDP. Out of these, we can see 13 economies whose labor share has declined. The pattern of these declines is not related in any obvious way to geography, level of the labor share, or level of development. We observe labor-share declines in advanced Anglo-Saxon economies (Australia, Canada, and United States), advanced European economies (France, Germany, Italy, and Spain), advanced Asian economies (Japan and Korea), and emerging markets (China, India, Mexico, and Thailand). The only countries with increases are Brazil, the United Kingdom, and Russia.
This breadth of the declining labor share across industries and countries in the last couple of decades means that, when looking for explanations, you need to throw a broad net–that is, you need an explanation that works across countries and across sectors. For example, trying to trace the decline in labor share to the decline in US unionization rates is not much help in explaining the decline in labor share in Italy or India) over several decades.
As Karabarbounis points out, “[I]t is fair to say that there is no strong
consensus yet about the deeper causes of the labor share decline.” He goes through a list of plausible alternatives, and suggest two as being more likely than others. One is what economists call “capital-augmenting technology” and humans call “automation.” A common model is to think of work as made up of a number of separate “tasks.” “The key assumption is that some types of services can be produced with either capital or labor, whereas other services are produced only with labor. Automation decreases the number of tasks which are produced only with labor, enabling capital to substitute for labor in a larger share of tasks.” Indeed, some researchers have used the declining labor share as a measure of the effects of automation.
The other plausible alternative mentioned by Karabarbounis are based in changes in product markets. For example: “The evidence shows that an increasing share of economic activity has been concentrated at larger firms with lower labor shares.” In addition, these large firms may be able to charge higher mark-ups over the cost of production. Along similar lines, I find myself wondering if we live increasingly in an economy where the digital goods and services make up a larger share of what we consume, both directly via entertainment, but also indirectly as producers use these digital services behind the scenes. For digital goods and services, the marginal cost of an additional user can be very close to zero, but the sellers do need to charge enough to recover the fixed costs of providing these goods and service. The resulting outcomes in terms of prices and mark-ups are an evolving story.
