Before you blame the robots, look to the policy (and the data)

February 21st, 2017 at 11:32 am

This very incisive bit of work from the NYT editorial page makes two critical points:

  1. The data do not support the claim that there’s been an acceleration in labor-replacing technology displacing US workers. To the contrary, measures of capital investment and especially and most persuasively, productivity growth, have slowed, trends that point in the opposite direction.
  2. The adjustment to technological change (and trade, and every other structural shift) takes place in a policy context that can either help those hurt by the change, or ignore them. US labor policy used to be a lot more helpful.

Re point #1, here’s the relevant figure. If automation were increasingly displacing workers, we’d be seeing more output produced in fewer labor hours, aka, faster productivity growth. But we see the opposite. I know that measurement issues have been raised–the idea that we’re under-counting output related to IT–but if anything, the evidence pushes the other way: we’re now doing a better job of accounting for the productive aspects of new technologies.

Source: NYT

Re point #2, the editorial provides a useful, quick sweep through the kinds of labor standards and “guardrails” that have historically been in place to facilitate the adjustment to tech changes:

When automation on the farm resulted in the mass migration of Americans from rural to urban areas in the early decades of the 20th century, agricultural states led the way in instituting universal public high school education to prepare for the future. At the dawn of the modern technological age at the end of World War II, the G.I. Bill turned a generation of veterans into college graduates.

When productivity led to vast profits in America’s auto industry, unions ensured that pay rose accordingly.

Corporate efforts to keep profits high by keeping pay low were countered by a robust federal minimum wage and time-and-a-half for overtime.

Fair taxation of corporations and the wealthy ensured the public a fair share of profits from companies enriched by government investments in science and technology.

Productivity and pay rose in tandem for decades after World War II, until labor and wage protections began to be eroded. Public education has been given short shrift, unions have been weakened, tax overhauls have benefited the rich and basic labor standards have not been updated.

To be clear, none of this denies the ongoing infusion and defusion of digital technologies into our work and our lives. Obviously, you see robots in today’s factories that weren’t there years ago, and productivity is growing, albeit too slowly.

But I think a lot of people miss a really simple, fundamental point about all this: productivity is pretty much always growing. We’re almost always creating and adding new machines that complement the production processes, and no question, there are workers who get displaced in the process. But throughout it all, we’ve had periods of full employment and equitably distributed growth, and periods of slack and inequality.

The questions are thus: is there enough labor demand in the economy to provide those displaced workers with other opportunities, and are the policy measures in place to help them handily find their footing again? In recent decades, the answers to those questions has been a resounding “no.”

FTR, I made all these same points years ago, but if anything, the automation story is getting louder.

People, if you’d just let me rule the world, we wouldn’t have to go through all of this!

Print Friendly, PDF & Email

25 comments in reply to "Before you blame the robots, look to the policy (and the data)"

  1. Kevin says:

    Dear Jared, this is interesting, but I’m not sure i concur that these measures are the best at determining automation uptake, especially against those time frames. I would argue that customers expect more for free and more customer service time for their purchase. I would argue automation via internet price comparison (and, for that matter Amazon) has eliminated profitability for the businesses. Productivity can be eliminated by all the competition, which would hide that it is really robots eliminating that profitability. The robots making themselves cheaper would then also agree with lower capitalization, not necessarily higher. In short, I wonder if the data can’t support the exact opposite narrative. Interested in your thoughts!

  2. Brian says:

    I agree with the point that policies could much more helpful, but I am not sure the numbers reported really show that labor-replacing technology hasn’t displaced a large number of workers. Average productivity growth could fall for at least two reasons. First, as is suggested by the article, productivity growth across all industries could fall. But also there could be something like a composition effect. Productivity growth could fall if workers are displaced from industries that have high productivity growth to industries that have low productivity growth. For instance, from manufacturing jobs to service sector jobs. The dates don’t line up exactly with those in the article, but manufacturing productivity has been pretty impressive. Averaging over 4% from 1990-2007. You can blame the robots, if they are what is causing workers to shift industries.

    Productivity change in the manufacturing sector, 1987-2016:

    Average annual
    percent change
    1987-1990 1.5
    1990-2000 4.1
    2000-2007 4.7
    2007-2015 1.7
    Source: BLS:

    • Jared Bernstein says:

      Good thoughts, though I think the automation story goes way past manufacturing. Also, note the deceleration of even manufacturing productivity in your chart. That’s a challenge for the automationists.

  3. Gerald Scorse says:

    “Also, note the deceleration of even manufacturing productivity in your chart. That’s a challenge for the automationists.” A challenge? I’d say it totally destroys that argument.

  4. Robert Salzberg says:

    I’m a physical therapist, and since my notes went from being hand written to computerized, it now takes more than twice as long because administrators throw in the kitchen sink. I estimate that I used to spend 10-15% of my time on documentation and now it’s easily over 30%. I’ve worked on 3 different programs with similar results. I’ve also spoken at length with physicians who also said that medical record technology has slowed down their productivity a lot. Since the medical field is a big chuck of GDP, computerization of electronic medical records is causing part of the productivity slowdown.

    • Blissex says:

      Unfortunately that waste of time in computerization is *boosting* reported productivity.
      Because for services it is not clear what the output is, and what the value of it should be recorded.
      What do you produce as a physical therapist? Ideally “well being”, but that cannot be easily measured.
      So the GDP estimators *assume* that what services sell for is also the value of their output.
      Your wasting more time of each hour on record keeping means that you need to charge more per hour worked, and that becomes in GDP an *increase* in productivity.

  5. Sandwichman says:

    I don’t blame the robots, either. But “the data” do not make the case for or against because the 1948-1973 data is not comparable to the 1974-1995 data, which is not comparable to the 1996-2002 data, which is not comparable to the 2003-2007 data, which is not comparable to the 2008-2015 data. I have had this argument with Dean and he is not persuaded so perhaps you won’t be either but my disagreement stems from the artificiality of the “labor productivity” measurement.

    As you know, labor productivity is a ratio between real GDP and labor hours. There is a lot going on in each of those components and unless you know what that is, the noise could easily be overwhelming the signal. 1973 was the end of an era of sustained cheap oil. Period. End of story.

    DO NOT compare post-1973 productivity stats with pre-1973 stats without explicitly addressing that monumental “correction.” Productivity is easy if you can just throw nearly-free petroleum at it. Otherwise, not so much.

    There is a lot going on policy-wise in each of the periods being compared to obliterate any meaningful comparison of “productivity.” Next, I would like to mention FINANCIALIZATION. What does financialization have to do with productivity? A lot. On the one hand, an increasing amount of “real” GDP consists of financial services. But on the other hand…

    From 1948 to 1973, the ratio of household and non-profit net worth to GDP doubled, from about 1:2 to 1:1. From 1974 to 1995, that ratio nearly tripled. From 1996 to 2015, the ratio nearly doubled again. To sum up, over the 67 years in question, net worth of household and non-profits increased by an order of magnitude relative to annual gross product. Need I remind readers of Jared’s blog that about 75% of that wealth is concentrated among the top ten percent? That’s up from about 2/3’s, twenty-five years ago.

    Does it take a non-economist to see what is happening in front of our eyes?

    After a certain point, it makes no sense for the wealthy to take their gains as INCOME because income gets taxed. Capital gains are not taxed until the assets are sold, so why pay taxes? One may object to this condition on moral or policy rationality grounds but that is not my point. My point is that the ASSET INFLATION makes the standard “labor productivity” measure obsolete. Whatever is happening is obscured by the all the creative accounting and the policy regime that enables it.

    • Jared Bernstein says:

      I take your point re asset inflation and that cap gains, either realized or unrealized, are not counted in output. But you seem to be contemplating a productivity measure which is the value of goods+services+CAP GAINS (both realized and non, I think)/hours. That last term in the numerator isn’t really analogous, to my mind, but it would be interesting to see what someone comes up with this way.

      One problem you’d have is productivity would tank when bubbles burst. But that’s not right–we don’t forget how to produce when wealth evaporates!

      • Sandwichman says:

        I’m certainly not suggesting goods+services+cap gains would be a better measure. There is more noise in the net worth stat than there is in the GDP. Consider 2008, the fall in net worth was equivalent to 71% of GDP!

        All I am saying is the huge disjuncture between output and wealth indicates that comparison of GDP between periods is dubious. In theory, how is wealth created? It is presumably produced. But in practice net worth bubbles up out of nowhere and then evaporates.

        There seems to be a reasonably steady relationship between GDP and net worth accumulation between 1948 and 1996. But after 1996 it is utter chaos. I don’t have an explanation for why this is the case. But I take it as a very loud alarm that something is seriously, seriously out of whack.

    • SPENCER says:

      I agree with much of what you say about comparing productivity with cheap energy and with expensive energy. But there may be another explanation why they differ. In a cheap energy environment, energy production does not absorb much of the new investment in the economy. But in an expensive energy environment more and more of the annual investment is diverted to producing energy rather than improving labor productivity. Moreover, investment in energy has a lower return in an expensive energy economy. If oil goes from $50 to $100 it means it now takes twice as much investment to produce the marginal barrel of oil so the productivity is cut in half. Moreover, oil users will diver more of their capital spending to maximizing their energy efficiency rather than improving their labor productivity. So their productivity in terms of labor will suffer.

    • Sandwichman says:

      Longer reply, with chart at my blog, EconoSpeak. “Nineteen Ninety-Six: The Robot/Productivity Paradox”

  6. Kaleberg says:

    Does productivity fall when prices are falling? If a worker produces 1,000 $2 splodges an hour, then the price falls to $1 does that mean the worker is half as efficient? The prices of a lot of manufactured goods have been falling a lot as they were in the late 19th and early 20th century. How do economists handle this?

    Also, the price of robots has been falling. There was an article on manufacturing a few years back, and one manager explained that as soon as the price of the robot to do the job dropped below $75K, they’d buy it and fire the human. Robot prices, for a given ability, have been falling, so the investment needed to eliminate a job has been dropping. How does this appear in productivity figures?

    • Sandwichman says:

      Yes, I was going to mention this. I paid $3000 (1990 dollars) in 1990 for a desktop computer with 50 megabytes of hard drive memory and a black and white monitor. In 2012 I paid $300 2012 dollars for a laptop with 15 gigabytes of memory, wifi, etc. etc. etc. Does this mean that if it took, say, one-tenth as much time to build my new computer than it took to build the old one that productivity of computer workers has fallen?

      Might it not even be one of the impacts of computerization to make the way we have been measuring productivity increasingly obsolete?

  7. jonny bakho says:

    The key is investing to help displaced workers. There was much resistance among politicians to passing the GI Bill.
    ManyRepublicans had to be shamed into voting for it.

  8. dwb says:

    This comment was written by a robot.

    No, not really.

    Economists are obsessed with robots these days. Lower labor demand is not being caused by robots. People can actually observe industries that have closed since 2008 (like coal mines) and see that they are not being manned by robots. I do not see any robots making my fast food yet. Nor do I see any robots making houses, buildings. So we cannot really blame lower construction labor demand on robots. Robots are not speeding up knee or hip replacement surgery, nor are they putting stents in clogged arteries.

    I have a robots in my house now, named Alexa and Roomba. Alexa does not do much more than play music, and Roomba does not mop the floors, do laundry, or anything other than occasionally sweep. Worse, Roomba is kinda dumb and when she encounters a sticky mess (e.g. if the kids spilt ketchup on the floor), she spreads it all over the place.

    I do however see a lot of politicians, pundits, and lobbyists advocating the same old nonsense over and over. Many journalists do the same thing Roomba does: Spread a sticky mess all over the place instead of mopping it up.

    Maybe journalists, pundits, politicians, and lobbyists are the robots, they (we?) just don’t know it.

  9. Ed Brown says:

    “People, if you’d just let me rule the world, we wouldn’t have to go through all of this!”

    I think it would be unwise of us to do this until you prove yourself. It is like bringing a kid up out of high school and having him start in the major leagues. The GM would prefer to see him perform in the minors for a while first.

    Hence I recommend we have you rule something small, like Wyoming. You run that for four years, appoint their congressmen and senators, do a good job, reduce pollution, improve education, etc., and we’ll talk. Then you can add Texas to your portfolio, and so on.

    This sounds more reasonable. Good luck! 🙂

  10. Longtooth says:

    Economists look at the macro economy to for evidence of whatever might be proposed as factors in some economic measure or potential measure. What they don’t do walk the mfg’ing and services floors and ask simple questions like — how did you do (x, y, z) 10 years ago? or 20 years ago? or 5 years ago? If they did this would be forced to adjust how they view the macro economic “evidence” since it wouldn’t reconcile with the obvious at the fundamental levels of empirical facts…

    One of the things I’ve always found interesting is that if more widgets are produced by automation at lower costs of automating, then there’s a clear cut productivity increase, both for capital and labor productivity. But that’s not how productivity is measured. What we measure is the price at the end user level, and then try to back out how those prices reflect quantities produced by making all kinds of assumptions that may have been roughly correct when they were developed, but were then still only estimates based on some proportionality constants. When those assumptions are no longer valid, neither are the proportionality constants, as has been the case since the computer revolution.

    In services the productivity gains are less obvious. For example, look at checkout (cashiers)today v just 10 years ago, and 20, and etc. They can process 20x more units/hour through he checkout stands today with cashiers who don’t even have to know how to add or subtract. Now add the automated checkout stands… one person monitors 6 checkout stands… the consumer now spends their time and effort to pass their goods over the automated price reader with the universal whatever code which is pre-printed by the producer/packager so that the central business computer simple does the price look-up for that code that which then applies at that store at that moment in time.

    So if any mary-jane or joe-blow consumer can act as a checkout cashier, then how is this massive increase in services productivity showing up in macro that the economist uses to “measure productivity”? Similarly in inventory management and accounting. A computer measures number of units “in”, and a computer measures number of units “out” at the checkout or when merchandize is transferred to another store or scrapped. The retail business knows within tiny margins of error how many units of every single item in their business they have at any moment and can track consumption per unit time automatically, place orders to wholesalers and figure optimal pricing by store or by geographic location including competitive pricing which is then instantaneously passed to the computer than reads the universal code and looks up the new price. How many labor hours at each skill level in this process are no longer required per unit sold per hour of labor? .

  11. Longtooth says:

    Another thing that is more philosophical in nature regards automation / robotics.

    If automation isn’t an issue from a macro measured point of view in employment rates of gain or loss over time hence productivity, then why do people like Jared and other economists even bother trying to tell the public it isn’t an issue? Why would academic economists or business economists bother with time and effort to publish to the uninformed consumer (specifically targeted to these uninformed audiences) and public on their “rumor mills” about “robotics” as it might affect employment if there is absolutely no merit to such “wild rumors” in the first place?.

    I’ll answer my own question.

    Because they’re scared to death that the other ‘experts’ who show by empirical facts that automation and robotics is in fact a serious and major economic issue in modern economies, unlike perhaps any in prior history since the industrial revolution, have a real and valid point but for which policy makers and macro economists have no real solutions that will translate to an “acceptable” one under our existing paradigm related to government and capitalism.

    If they (macro economists, like Jared for just one of many examples) acknowledged that the automation rate is and will continue to take jobs at middle and lower income levels at increasing rates and lower capital costs, providing greater profit margins to producers then they would in fact have to acknowledge that we need to make some major and serious changes in political economics — not just in the US, but globally in most or all major economies. Since they can’t do this… policy makers are not willing to make those kinds of changes since the would be highly disruptive to the status quo, it’s far better to tell the public there’s nothing there, there.

    That’ s my considered opinion.

  12. Longtooth says:

    An interesting productivity measure at the macro level might be the domestic airline passenger miles traveled divided by the total number of domestic airline employees, and the same divided by population of the U.S.. I would include contracted services like baggage handlers, etc. since those activities used to be done by actual full-time airline employees.

    I haven’t done this analysis myself (and don’t plan to) but it would be interesting to see how much productivity in the airline services industry as grown of the past 40 years, no?

    A similar measure might be air-cargo productivity airline weight miles per full-time equivalence employee in air cargo airlines.

    I assume these would also have affected lower prices, so consumers spend less also, but that isn’t the question or issue. The question & issue is how productivity gains reduce demand for employment. and/or shift it to lower paying slower skill employment — even to part-time employment or seasonal employment. As it does so then supply of labor to lower paying lower skill jobs increases relative to demand .. which reduces lower skill wages/salaries to increase employment to equilibrium with supply/ demand. But if productivity growth continues then the lower skill services continue to increase in supply available which further reduces services wages or increases overall proportion of adults employed (employment population ratio).

    Either or both of these reduce consumers’ ability to consume (down to subsistence) which then reduces demand hence production of goods and services hence forces more business consolidations which use economies of scale and thus further reduce employment (improved productivity). The more consolidations the fewer competitors, then the lower the incentives for improving productivity so productivity growth slows. This reduces international competiveness. That in turn gives policy makers incentive to increase import duties to prevent further job losses to foreign competitors and keep employment losses from increasing even more. That erodes US domestic producers competitiveness in goods and services even more, reducing U.S. standards of living relative to other nation’s increases in standard of living…. the U.S. then falls further and further behind.

    It’s easy to stop this train but not without offending capital owners interests — and since capital owners fund elected representatives then those representatives must necessarily cater to capital owners best interests to get elected or re-elected. Thus propaganda then blames foreign competitors — currency manipulation, unfair use of gov’t funds to aid their manufacturing, unfair past trade deals the US made, etc., and “illegal immigrant” taking jobs from “legitimate citizens and residents”, and capital owners taxes being too high, as the public “enemies” to blame.

    This is played out time and time again in 3rd world oligopolies and “ostensible democracies” run by dictatorships of the “elites”. It’s standard practice and nothing new.

  13. Devin says:

    Correct me if I’m wrong, but the way these calculations work is that productivity = (the price of everything produced) / number of hours worked.

    If either a.) the savings due to the robots is passed on to the consumers or b.) human labor is shifted towards low value-add work in unrelated fields like retail and food services (both, explanations that I believe are generally consistent with the data), then this productivity measure wouldn’t increase, even if robots were taking over productive work.