“Refried confusion” on automation and jobs in manufacturing

    Jared Bernstein Posted by Jared Bernstein.

    From 2009 to 2011, Bernstein was the Chief Economist and Economic Adviser to Vice President Joe Biden, executive director of the White House Task Force on the Middle Class, and a member of President Obama’s economic team.



    “Refried confusion” on automation and jobs in manufacturing

    Source: BLS, my trends (HP filter)

    In the words of that great Ph.D. of New Orleans funk, Dr. John, “Refried confusion is making itself clear” in this NYT piece about the impact of automation on manufacturing employment. The piece makes the common claim that robots are wiping out jobs in the sector, but if there was any evidence in there, I missed it. And I know of some pretty compelling evidence to the contrary, along with some key variables left out of this analysis.

    It’s critical to be clear about what’s under the microscope re this question of jobs and automation. It is not whether or not machines replace humans. That of course has gone on since the Luddites of the early 1800s and well before. The claim today, whether its purveyors make it explicitly or not, is that the pace at which labor-displacing technology is entering the workforce is accelerating. Automation must be driving jobs out of a sector—in this case, manufacturing—at a pace that’s demonstrably faster than demand for manufactured goods can offset.

    That is, if 10 workers could make 100 widgets in a week, and now 5 workers with a robot can generate that same output, the other five lose their jobs unless demand for widgets doubles such that the producer needs to keep the 10 workers to produce the 200 widgets the market will bear.

    But you will note that I’ve embedded two key intervening variables within this example: productivity growth and demand.

    First, if the pace of automation is accelerating, it should show up as faster productivity growth—more output per labor hour. Thus, any article making this claim must either show this trend or at least explain what’s wrong with my logic.

    Now, we know that productivity growth has slowed economy-wide, and not just here in the US; it’s one of growth economists’ greatest contemporary concerns. But what about manufacturing productivity? The Times story—and I don’t mean to pick on this one; this is a very commonly held view—implies that manufacturing productivity must be speeding up. Is it?

    Nope.

    The figure shows the annual growth rate in overall manufacturing and durable manufacturing, like making cars, steel, big machines, etc. where you might imagine robots hanging out. Because these numbers are both very bouncy and even spikey around recessions, I’ve added a smooth trend. But whether you look at the actual rates or the smoothed ones, recent years show productivity growth on the factory floor to be low, flat, and perhaps even falling in the trend.

    What about employment? The piece cites one expert as arguing that “because of advances in technology, manufacturing simply doesn’t employ as many people as it once did.” But that’s certainly not what the history of employment in the sector shows. The figure below shows monthly employment in manufacturing since the beginning of time (as per BLS: 1939). Factory employment was remarkably stable in terms of headcounts for decades from the 1970s through 2000. It was of course falling as a share of total employment over those years, a trend in all advanced economies, which has as much to do with trade as technology, as manufacturing employment shares were growing in developing countries that were and remain our trade partners.

    Source: BLS

    Source: BLS

    Obviously, technological advances occurred all throughout that period while employment counts were stable. But the intervening variable—demand—absorbed productivity gains and held employment roughly constant at around 17 million.

    What happened around 2000? Well, China entered the WTO and there’s research that shows US/China competition was a strong negative for American manufacturing employment and wages in those years. I’m definitely not claiming that’s the whole story, but it is the other big piece that’s missing from this analysis. Global competition has of course been a highly significant factor in US manufacturing outcomes, as US consumers’ demand for manufactured goods has been increasingly met from abroad. Some of that is “market forces,” but I’ve argued that some of it is a function of currency management that has disadvantaged our manufacturing sector relative to those in countries that sustain large trade surpluses (see Chapter 5 here).

    Bringing in these other variables—productivity, employment trends (actual vs. imagined), currency, who’s sustaining large deficits vs. surpluses, how demand is being met—is, in my view, essential, as it drives the policy debate. In fact, the Times story is framed around the presidential debate, suggesting that candidates who talk about reviving manufacturing are all ignoring robot-driven realities.

    If what’s happening is simply accelerated labor-displacing technology, that’s not something I nor most others would want to stop. We need faster productivity growth! But many of these other issues, especially currency and our persistent trade deficits in manufactured goods, should be thought of as policy variables with relevance to trade negotiations, for example. I strongly doubt the declining trend of factory jobs as a share of total employment will reverse course. But the number-of-jobs story is a lot more up for grabs. I want the candidates to tell me what they’re going to do about that!

    Even the previous figure shows employment growing out of the recent recessionary trough. And while you can’t see it in the figure, in 2015 manufacturing employment was largely flat. Technology at work? Of course not—automation doesn’t enter and leave the economy like that; it’s a more gradual force. What hurt our manufacturers last year is the strengthening dollar and its impact of the goods trade deficit (not currency manipulation, I’d argue; more relative growth rates and relative central bank actions).

    So there’s a much more nuanced story in play here than a simple automation one. But I believe it’s the right one.