Believe those who are seeking the truth. Doubt those who find it. Andre Gide

Saturday, April 15, 2017

Sectoral and Occupational Trends in the U.S. Labor Market

The labor market is back to normal. Or so we are told. Here's the prime-age unemployment rate for the United States beginning in 1960.
Above we see the familiar cyclical asymmetry, in which the unemployment rate spikes up sharply at the onset of a recession and declines gradually during the recovery. As of today, the prime-age unemployment rate is at 4%, close to its recent historical norm. 

Other measures of labor market activity, however, tell a slightly different story. Here is the prime-age employment-to-population ratio. Employment took a big tumble during the recession--especially for males--and its taken a lot longer to recover than unemployment. In fact, we're not quite back to pre-recession average of about 80%. 
The reason employment has recovered more slowly than unemployment is because significant numbers of prime-age workers left the labor force in the aftermath of the Great Recession. This pattern has reversed only recently.
By these standard measures of aggregate labor market activity, the U.S. labor market appears close to "full employment." Nevertheless, there is still much anxiety concerning the present state and future course of the U.S. labor market. Much of this anxiety stems from the perennial concerns regarding the impact of international trade and technology on future employment opportunities. 

To assess the nature of these concerns, we have to move beyond the standard aggregate measures of labor market activity, which have remained relatively stable since 1990 in the face of growing trade deficits and technological progress. So if trade and technology are going to have an impact on employment opportunities, it's likely going to happen at the sectoral and/or occupational level. 

Of course, a changing allocation of human labor across sectors of the U.S. economy is nothing new. Here is the share of employment in agriculture and manufacturing since 1815. 

In 1815, 80% of employment was devoted toward the production of food. And for those of us who have never worked on a farm, here's how Charles H. Smith describes the venture in 1892:
“It’s about the meanest business I have ever experienced. It’s all fact—solemn fact – no romance, no poetry, no joke. It does seem to me that all this sort of work ought to be done by machinery or not to be done at all.” Source: Farm Life at the Turn of the Century.
As we can see, Smith got his wish. Over the centuries, machines have replaced most of human labor in the production of food. Was this a welcome development? Most of us today would likely answer in the affirmative. However, sectoral transformations like this also come at a cost, borne by those who are compelled to do the adjusting. A shovel that helps ease a manual burden is one thing. A combine harvester that renders a certain kind of labor redundant is another. When the demand for a certain type of labor in a given sector is diminished, what are workers to do? 

One thing they can do is employ their labor in the production of the machines that replaced their labor on the farm. And so youngsters left the family farm and immigrants flowed increasingly into the communities that offered employment in manufacturing and other sectors. I'm not sure how much of an improvement it was to substitute the drudgery of farm work with the drudgery of factory work, but it was probably an improvement in net terms (however small). The relatively high levels of comfort most Americans enjoy today did not come about until the latter half of the 20th century. And by that time, manufacturing employment (as a share of total employment) began its long secular decline. 

So, what accounts for the decline in manufacturing sector employment? According to Dean Baker, the U.S. trade deficit has a lot to do with it. 

Since 2000, employment in the U.S. manufacturing sector has dropped from about 17 million to 12 million workers. The sharp decline started with the "China shock" in 2001 (also a recession year). On the other hand, Germany has also experienced a decline in manufacturing sector employment while running trade surpluses. Arguably, the trade surpluses muted the secular decline in manufacturing employment in Germany, while the U.S. trade deficits exacerbated the process of structural readjustment. The trade deficit has almost surely had some impact on U.S. manufacturing employment, but it's hard for me to escape the conclusion that most of the sectoral reallocation of labor has been driven by technology. The next diagram tells the story. 
But it hardly matters to an individual worker whether they are displaced by a foreigner or an automaton. Where are the new jobs to be found and what are they paying? Are all of our good, high-paying manufacturing jobs being replaced by lousy low-paying service sector jobs? There is an element of truth to this. As Daniel Alpert points out, since the cyclical peak in 2007, the U.S. economy has added close to 7 million jobs. About 60% of these jobs appeared in the "low-wage, low-hours" sectors that account for about 36% of all private sector jobs; see table below. 
The picture is not quite as bleak as Alpert paints it. Andrew Spewak (my trusty RA) took it upon himself to produce this table. 
What this decomposition shows is that there is in fact substantial net job creation in high-wage/high-hour jobs in the service sector. The bleakness is heavily concentrated in manufacturing (which we know is in secular decline) and in construction (which is not in secular decline, but which hit a peak just prior to the housing crisis). 

Let's dig a little deeper and explore the task-based view of the labor market developed by Daron Acemoglu and David Autor (see here). According to this view, goods and services are produced by a set of tasks that can be performed by various inputs, like capital and labor. Tasks differ along two important dimensions. The first dimension measures the relative importance of "brains v. brawn" in performing a task. This is labeled "manual" v. "cognitive" in their analysis. The second dimension measures the extent to which a task can be described by a set of well-defined instructions and procedures. If it is, then it is possible to have an automaton execute a program to perform the task. Acemoglu and Autor label these "routine" tasks. If instead the job requires flexibility, creativity, on-the-fly problem-solving, or human interaction skills, the occupation is labeled "non-routine." 

Here's how some broad occupational classes fall into the four implied categories:
Of course, no classification scheme is perfect, but you get the idea. And here's how the employment shares for each category behave since 1983:
This graph could be labeled "The Decline of Routine Work." We have seen how the machines have substituted for manual labor. Robots are increasingly doing the same thing for many cognitive tasks. Here's a male/female breakdown of the phenomenon:

The decline of routine labor has been associated with "polarization"--that is, a "hollowing out" of America's middle class. This is because routine jobs, whether manual or cognitive, generally pay "middle class" wages.
The share of employment allocated to these middle class jobs is declining. Where is this middle class going? Largely to jobs in higher wage category, associated with non-routine/cognitive occupations. But there's also a moderate increase in the share of employment in the lowest wage category, in those jobs associated with non-routine/manual occupations.

This is already a long post, so I don't really have space to ponder the question of Whither Human Labor? It seems likely that an increasing number of tasks presently labeled non-routine will become routine (still lot's of room in manufacturing it seems, see here). We have seen this with robo-advisors in the financial sector and the emergence of artificially-intelligent doctor apps (see here) among many other places. How might the future unfold? Here's one take, by Andrew McAfee: Are Droids Taking Our Jobs?


  1. David,

    Where does that data for "routine manual" labor come from? I ask because it looks like a perfect example of a model I've been working on. I originally put together the model as a possible explanation of the cyclical asymmetry. Here's a short presentation on it (images or pdf at the link):

    There's a 'dynamic equilibrium' where jobs are essentially filled at a steady (relative) rate. As a side note, this relative rate can be related to a matching model in the absence of shocks. Unemployment rises via shocks (which can be seen as a mis-match in the matching model), but then starts to fall as the normal dynamic (matching) equilibrium is restored.

    One thing I did notice is that employment among men does not seem to recover to its previous peak before the next shock hits, meaning that each shock throws men out of the workforce who seem to never return (because of retirement or lack of skills for the new kinds of jobs).

    I thought you might find it interesting.

    1. Hi Jason, it comes from the BLS. I can send you the excel spreadsheet on Monday if you'd like. Thanks for the link to your post.

    2. Thanks David. I realized I can get the data from FRED aggregating the individual sector data per your organization above in fig10. Cheers.

  2. A follow-up; here's the post relating the dynamic equilibrium to the matching model:

  3. Interesting post. Labor re-allocation should fuel productivity growth, which would dampen the suffering of those stuck in declining-share jobs. The growth in non-routine cognitive jobs seems like it would be a big boost to productivity. They require more education after all, and we all know education is a high-return project.

    Yet, there are three major areas of the non-routine cognitive economy where productivity is probably stagnant: education, healthcare and government. These three are rife with perverse incentive dynamics (education signaling, for instance, or expensive treatment of preventable diseases) that generate large dollar expenditures with questionable value. Together, the three constitute about 25% of employment. Throw in FIRE (again, perverse incentives, questionable value, hard-to-measure productivity), and you get to about a third of the economy and the lion's share of high wage jobs growth.

    Those same four industries are the ones that blockchain has the biggest potential to disrupt. Productivity growth picture could significantly improve when that happens, and with it the labor share of income (as blockchain will arguably destroy economic rent, whereas the Internet produced it). Stay tuned!

  4. David:
    I appreciate the cite very much. But let's look at the composition of the "high-wage/high-hours" jobs as a counterpoint to my LWLH job. The greatest growth has been in management and in professional and technical services. Now it's tempting to think that the latter cohort is a bunch of software coders and hardware geeks. But it's not. Look deeper and you will find sales people. Lots of sales people in fact. And a smaller cohort of roduct designers. Most of whom are involved in the design and distribution of exogenous manufactures.
    What I am saying is that even though there is a respectable level of growth in HWHH jobs (which offset, to an extent, LWLH jobs) they are connected largely with the domestic manifestation of exogenous production, in no less a way as the LWLH retail sales jobs.
    Bottom line, look carefully at the service cohort that is a necessary component of offshoring. And consider that the growth of the latter proceeds from the growth of the former and is thus a "commissioned" cost to foreign production, that rises and falls in proportion to same, but is not linked to the multipliers that would be attendant to a domestic manufacturing industry.

  5. Hi Daniel. Not sure I fully understand what you're getting at here. You seem to be suggesting that a large component of the growth of service HWHH jobs is associated with activities designed to send goods-producing HWHH jobs abroad. And that this is somehow bad for the economy. Yes?