Don’t Blame the Robots for Slow Job Growth In 2000s

There is a lot of talk about robots these days, as if technological change has led to weak job creation, caused wage inequality, and even caused the profit share of the economy to increase as workers’ share (compensation) falls. We have definitely had problems with employment growth and growing wage inequality, alongside a profit boom not just during the Great Recession and its aftermath but since 2000—and for wage inequality, for two decades prior. Is technology driving all this? I think not.

This post tackles the issue of whether robots slowed job growth in the 2000s (my colleague Josh Bivens has addressed this previously, but for the recovery). In the near future I will be addressing whether robots are responsible for our current wage inequality. Spoiler alert: they aren’t.

MIT professors Andrew McAfee and Erik Brynjolfsson wrote the influential book Race Against The Machine, which has driven much of this conversation. They label the disconnect between employment growth and productivity growth in the 2000s the “Great Decoupling.” Moreover, they argue that there has been an “acceleration of technology” that has “hurt wages and jobs for millions of people” even as “digital progress grows the overall economic pie.” Brynjolfsson and McAfee know a lot more about technology and its impact on firms and work than I do, but attributing the job slowdown in the 2000s to robots or digitalization overlooks a simple alternative answer: slow aggregate output growth caused by the bursting of asset bubbles. In fact, they do not offer much evidence to connect the technological trends (robots and digitalization) they document to the aggregate job, wage and inequality trends they, and I, care about.

My former colleague Jared Bernstein offered up a chart to show the decoupling, which EPI’s Will Kimball reproduced using economy-wide productivity and hours worked, below. Hours worked (red line) used to grow alongside productivity growth (blue line), but in the 2000s there was a divergence. This decoupling, by the way, mirrors the post-1979 divergence between productivity and wage growth for the typical worker, something Jared and I introduced in 1994 in the State of Working America. An astute reader of graphs would get a clue about causes if they noticed that the hours worked line identifiably flattens at certain times prior to the 2000s: the mid-70s, early 1980s and early 1990s. Those are recessions, periods of falling or slow output growth.

prody-hours

Brynjolfsson and McAfee may be correct in how digital technologies have affected jobs in particular firms, occupations, and industries, but I’m not persuaded that technology slowed aggregate job growth in the 2000s. A simpler explanation is just as plausible, and seems to have eluded much of the commentary: slow growth in the 2000s was a result of the collapse of two asset bubbles (tech stock and housing), followed by an inadequate policy response. Brynjolfsson and McAfee, in fact, acknowledge the “macro problem” in their book’s introduction, writing, “Sluggish demand is in large part to blame for today’s lack of jobs. But cyclical weak demand is not the whole story… deeper trends are also at work”

The journalistic treatment of their work clearly overlooks the macro issue, which is not only true for the Great Recession and afterwards but, as I will show, persisted throughout the 2000s.This could be a case where microeconomists—those focused on particular markets and not the aggregate economy—and the journalists who write about their work are missing the big picture and mistaking anecdotes for data (as Jared Bernstein also suggests).

Let’s start from some basic macroeconomics mathematics:

Q = (Q/L)* L

Where Q is GDP, productivity is (Q/L) and L is total hours worked.

We know that, in log annual changes:

L growth = Q growth – productivity growth

That is, the growth of total hours worked (i.e. jobs) is basically set by the degree to which the overall economy (GDP) grows faster than productivity. For example, if productivity grows 2.5 percent each year (producing 2.5 percent more goods and services per hour worked) then the economy can grow 2.5 percent and do so without adding any additional employment. If we are generating employment it is because, in the first instance, output growth exceeds productivity growth.

It turns out that this dynamic readily explains the “great decoupling” between productivity and employment in the 2000s. The bar chart below shows the growth of output, productivity and hours in each of the last three business cycles, and since 2007. All the data are for the economy as a whole, not for any particular subset of the economy. Output grew faster than productivity in the 1980s and 1990s and, accordingly, there was a 1.7 and 1.5 percent annual growth in total hours worked, respectively. In the 2000-07 business cycle, however, output and productivity growth were pretty much the same—so there was minimal growth of hours worked, just 0.3 percent each year. The collapse of the tech bubble, in other words, led to slow output growth and, correspondingly, slow growth of hours worked and jobs. During the Great Recession and the slow recovery, 2007-12, output growth was less than productivity growth, leaving us with fewer jobs in 2012 than we had in 2007.gpd prody hours

It is true that productivity grew faster in the 2000-07 business cycle (productivity actually accelerated in the mid-90s), a fact which lends plausibility to the robot story. However, faster productivity growth need not have led to subdued employment growth, and it certainly did not do so in the late 1990s when productivity was even faster (2.4 percent) than in the 2000s and hours employed growth exceeded that of any other period (1.9 percent). The difference in the 2000s was that a bubble burst and monetary and fiscal policies did not increase GDP growth enough to maintain steady employment growth.

It is also noteworthy that productivity in the 2007-12 period has been relatively slow, so the prima facie case that robots and digitalization are hampering the jobs recovery is absent. In fact, slower productivity growth since 2007 has assisted job growth.

By the way, the rate of growth of capital (including software) investment in the business sector, which should correspond to the introduction of new technologies, was about half as fast in the 2000s as in the 1980s or 1990s. If the robots did it, there’s not much evidence. They didn’t even leave footprints at the crime scene.

It is certainly the case that the character and location (industries) of productivity growth could affect jobs growth (it depends on the elasticity of demand for the products involved. See work by Eileen Appelbaum and Ronald Schettkat here and here). Nevertheless, it seems a stretch to argue that robots slowed job growth to the extent it occurred, and a better suspect is available: slow output was at the scene of the crime and seen with a weapon in its hand.

The great decoupling, in this view, is simply the result of slow GDP growth relative to productivity, and has everything to do with the collapse of various asset bubbles—stock and housing—and the policy response. Journalists and policymakers who think they’re getting at the deeper story by focusing on robots are actually missing the big picture. Robots may not have been even near the site of the crime, so don’t blame them so quickly.


  • microsrfr

    It seems to me that you are overlooking a deeper question. Why did we create the asset bubbles? I submit that it was a result of the need to supplement flagging wage-based demand by accelerating growth of personal debt. Is it merely an accident that the Federal Reserve, witnessing the growing slack demand in the economy after the dot com bubble burst and after 9/11, quickly lowered interest rates and looked the other way while subprime, variable rate, negative amortization and undocumented mortgages were issued? This process continues today with the run up in student loan debt.

    Sometimes the simplest explanations are the most valid. The rapid drop in the labor content of goods and services driven by Moore’s Law technology which has enabled automation and outsourcing combined with the diversion of all the resultant corporate “savings” to the wealthy by CEO’s singularly focused on short-term earnings improvements.

    It would be reasonable for one to ask at this point why competition has not reduced the price of goods and services in this scenario, and sustained wage-based demand? The answer may lie in the fact that those companies who have most effectively leveraged the exponential growth of computing power have become near monopolies — Microsoft, Apple, Google,

  • benleet

    I take this away: “in the late 1990s . . . hours employed growth exceeded that of any other period (1.9 percent). ” The late ’90s also were exceptional especially the gains in lower-income household incomes. It seems that the macro argument of income distribution has been missed in this analysis. Also the relative growth of the civilian noninstitutional population which varied over the decades. Productivity gains should decrease hours worked, that is the purpose of productivity efficiencies, or is it? It isn’t in a one-sided capitalist economy that has an incurable exigency for ever-growing profit. Private sector employment was a negative growth between Dec. 2000 and July 2012, 11 and a half years, on net no new jobs after the growth of 31 million of the “working age population.” which was a 14% growth. Robots didn’t do it.

  • Jed Harris

    I wonder if your business investment numbers for the 2000s are adjusted for the exponential improvement in digital systems price / performance. I also wonder if those numbers are adjusted for the rapid growth of zero cost software packages (often Open Source) — these packages are market leaders in a number of key areas such as server operating systems, web related software, development tools, etc.

    If the investment numbers are not adjusted for those two factors they could be completely wrong. It would be interesting to look at non-monetary metrics, such as the number of megabytes of server output bandwidth installed (I don’t have these numbers but they probably exist). I would guess they tell a very different story.

    More generally, if we can get an exponentially larger job done for the same investment, why should we increase investment? And if we can get an exponentially larger job done with the same employees…

    • Jed Harris

      IDC does collect numbers like these, unfortunately behind a paywall. I found a very relevant presentation that uses the IDC numbers and clearly shows the exponential trend in capacity over the 2000-2010 period — see slide 13 in this presentation.

      This pretty clearly illustrates the divergence between economic metrics and real capability metrics. If capability is increasing exponentially while investment is flat, the argument in this post does not hold water.