24 December 2012

The Work of a Nation

Readers of this blog who have followed my discussions of innovation, productivity and economic growth will recognize two competing perspectives among economists on the role of innovation as a source of our current economic difficulties, neither of which I think adequately captures the innovation challenge facing the US economy.

One of these competing views, represented by the work of Robert Gordon and most recently expressed in this weekend's WSJ, is that we have reached the end of innovation and productivity growth, and thus economic growth will slow to a crawl. Gordon writes that "the future of American economic growth is dismal" following from what he says is a decelerating trend in rates of economic growth and innovation over the past half century. As I have shown in a series of posts, the data do not support Gordon's argument - US per capita economic growth rates are largely unchanged since 1870.

At the opposite pole from Gordon are those economists who argue that rather than reaching the end of innovation and productivity growth, we in fact have far too much [UPDATE: See below for a response from Brynjolfsson]. Erik Brynjolfsson and Andrew McAfee argue that we "need to start preparing for a technology-fueled economy that’s ever-more productive, but that just might not need a great deal of human labor." Similar to Gordon they base their argument on an analysis of recent trends in labor productivity as compared to employment, finding a "great divergence." Here as well, I don't think that the data actually supports their argument.
The graph above provides further evidence that the productivity hawks and doves are both reading into the data signs that just are not there. The graph shows data on GDP per hour worked in the economies of the the US, Germany and Japan (data from BLS). There is no indication of a rapid acceleration or deceleration in the output per hour worked in any of these three economies. (The one-time effect of the reunification of Germany in 1991 does stand out.) What this graph suggests is that changes in productivity are not the source of more fundamental issues associated with employment, income, equality and other economic outcomes that we care about.
The graph above, also from the BLS data, tells an interesting story. It shows the total hours worked annually across the entire economies of each of the US, Germany and Japan. The graph shows that Japanese workers were performing the same total amount of work in 2011 as in 1970. We can deduct from this that all gains in Japan's GDP since 1970 are a result of productivity gains. Germany has seen its total work increase by 11% since 1970, but with essentially no change since reunification. Since 1970 Germany has seen an almost 30% reduction in hours worked per employed person (Japan saw a 15% decline over the same period). Germany has policies focused on maintaining high employment, which leads to lower rates of change in GDP per employed person, but which has not penalized GDP growth per hour worked, as shown in the first graph above. Germany provides good evidence that productivity gains and employment need not be closely related, and employment policies matter a great deal for aggregate levels of employment.

The United States, interestingly, did the same total amount of work in 2011 as it did in 1998, despite seeing overall population grow by 36 million and employment grow by 8.5 million. Since 1970 the average US worker is working only 8% less than in 1970, but most of this decrease has occurred since 1998. With global GDP increasing by 133% since 1998 the world obviously has not been suffering for a lack of work to be done. Indeed, according to the McKinsey Global Institute the world added 1.1 billion new jobs from 1980 to 2010, with 164 million in so-called "advanced economies."  Unlike Japan and Germany, the US does not have a population that is either decreasing (in the case of Japan) or slowly increasing (Germany).

It seems that at the root of US economic problems lies a challenge of stagnant work, a condition that cannot be attributed to too much or too little productivity gains -- the work that is being done is improving its productivity as it always has. Further, the stagnation in work appears to have pre-dated the recent economic crisis, but was no doubt exacerbated by it.

In this context, reducing unemployment would mean sharing work more broadly (e.g., the average German worker works 20% less hours per year than the average American worker) or simply creating or finding more work to be done. Sharing work makes a lot of sense in a slowly growing or shrinking workforce, as in Japan or Germany. Such policies may make sense in the US, but they are not long-term solutions to the challenges of work in an economy that is seeing significant population growth. In that case more work is needed. Creating or finding more work means innovating -- and specifically, by competing more effectively in existing markets or creating new markets to compete in.

In an era of globalization, we have learned that competing more effectively in existing markets can be difficult and, as has been the case in manufacturing, may lead to continued economic growth unmatched by growth in employment. By contrast, the prospect of creating and competing in new markets or even leveling the playing field in old markets offers potential for creating more work. More fundamentally, as the world has become wealthier there is increasing demand for economic necessities such as infrastructure, energy and food. The United States is well-positioned on these fronts to put the nation to work. However, doing so requires a forward looking approach to innovation and a renewed commitment to globalization and economic growth around the world.

UPDATE 12/25: Erik Brynjolfsson sends this comment:
One correction: Andy and I don't think there is "far too much" innovation and productivity growth.  We are huge fans of both and say so repeatedly in our book and our other writings.  In fact, we'd like to see more innovation in work practices and organizations to keep up with the technological advances, aka learn how to better race with machines.   Some people mistakenly lump us with the luddites, but that's about 180 degrees from where we are, even if we don't think technology always and everywhere helps all people evenly.


  1. Lack of investment by companies may explain some of the stagnant U.S. figures (although companies worldwide are hanging on to their cash).

  2. Within the last 2-3 years, I read of one of the big up-and-coming tech companies (it might have been Twitter) boasting to a room full of investment banker types of how few employees they had. This is now the mindset within the private economy movers-and-shakers. Fewer employees means more profits for us. Lenin talked about the Imperialists searching for super-profits overseas, when their own markets matured and could no longer grow. These people have turned this idea inside out. Unable to get rich selling to China - the long dream of Western businesses - they now see their own internal market and their own citizens as subject to their blood-from-a-rock squeezing.

    Regarding infrastructure, energy and food. Agriculture will never be a big employer again. I've watched three combines work a field that went on for as far as I could see in Iowa. And more restaurant won't do it either - even Hooters. ;-) Energy, in the sense of new natural gas, will bump us up locally, but the national effects will probably be trivial. And please, for the love 'o God, don't mention green jobs. Infrastructure? We're overwhelmed with work to be done - bridges are falling apart nationally. But of course, that kind of work only comes from government spending, and we could create the same jobs by hiring men to dig holes and fill them in again.

    Count me pessimistic. Innovation has created the need for software engineers. I saw recently that the CEO of the largest app producer said that he hired employees out of Ivy League school and the big tech schools - MIT and Caltech. So in other words, graduates that were guaranteed jobs and the good life in any case now work for him rather than for someone else. Not exactly what the nation needs. The children of Scarsdale and Shaker Heights aren't the ones hurting.

  3. Maybe, just maybe you are applying the optimum to a phenomena that doesn’t run optimally, That is, the economy is not an economy as much as it’s a political economy. However, a political economy will run somewhat optimal until it becomes political only, as in, government privilege economies e.g. the many government privilege based economies prior to the year 1700.

    Hence a third rail exists as explained by public choice theory. That in fact many “advanced” economies have been purposely morphed into government privilege economies by politicos through the mechanism of government over the last seven or so decades.

    Therefore, notional propositions of shared work, productivity gain graphs, innovation or non-innovation observations don’t get at the base problem: purposely constructed government privilege economies by politicos and for politicos as political constituency building exercises and schemes.

    One might say the old expression of ‘money is the root of all evil’ is off target and the correct expression is ‘other people’s money is the root of all evil’.

  4. Reasons for unemployment include many that are simply there to take power from employees.

    Creating a reserve pool of labour .

    Keeping wages down

    Keeping the employed in fear of being sacked and ending up on welfare.

    In 1950's America, doctors, engineers and store workers lived in similar houses. High inequality was deliberately created. It isn't a nasty side effect of capitalism. Just a side effect of nasty capitalism.

    The power of repetition and the desire to conform are enormous. A UK physics head of dept in the Guardian forums stated he didn't believe that computers can model the climate. I asked him if that was the case, why was engaging in end of the world fantasies with hysterics.

    Yes, he said, but you can extrapolate the CO2 output into the future and come up with a temperature figure. I suggested he could afford expensive cruises for the whole department because with a man as clever as him, they could sell their computers. I said it in an inoffensive way. Sensitivity numbers roughly come from models which he claims don't work.

  5. Roger
    I am curious how the German data was compiled as there is no distinction made between East/West. Does the dataset simply add the former GDR stats to to previous West German?

  6. -5-Reiner

    Yes, prior to 1991 is West Germany is my understanding -- hence the discontinuity.

    Merry Christmas!

  7. Hi,

    If I draw a regression line on this curve, it looks like the number of instructions per second per $1000 goes up by "only" a factor of a little more than 100 per decade. (I'd thought the value was more like a factor of 1000 per decade.)


    The computing power passed 10^3 MIPS per $1000 in 2002. So that's 1 billion instructions per second per $1000 in 2002. If the pace continues at a factor of 100 per decade, it would be 10^5 in 2012, 10^7 in 2022, and 10^9 in 2032.

    10^9 instructions per second is 1 quadrillion instuctions per second...roughly the power of a human brain (which has been estimated at between 500 trillion instructions per second and 20 quadrillion instuctions per second).

    So sometime between now and 2032 the effective human population is going to skyrocket. The Solow-Swan economic growth model predicts that population growth actually slows down economic growth. I think the exact opposite will happen. Economic growth models need to account for expected future growth in artificial intelligence.

  8. Hi,

    I guess I've linked to and commented on this before, but I think it's worth repeating. If Robin Hanson and Arnold Kling (and I!) are even close to correct, it's not an exaggeration to say it will be the most important thing ever to happen in economics. In fact, it's not an exaggeration to say that it might well be the last thing to happen in economics.

    Specifically...in my last comments, I wrote, "Economic growth models need to account for expected future growth in artificial intelligence." I'd forgotten/misremembered the fact that Robin Hanson has already used the Solow-Swan economic growth model to predict that, circa 2025, annual world GDP growth should be approximately 50 percent. (!!!!!!!!!)


    From page 6 of that paper: "Without machine intelligence, world product grows at a familiar rate of 4.3% per year, doubling every 16 years, with about 40% of technological progress coming
    from ordinary computers. With machine intelligence, the (instantaneous) annual growth
    rate would be 45%, ten times higher, making world product double every 18 months!"

    Let's say he's way, way wrong, and the world GDP growth rate is "only" 7-10% per year. That still means world GDP doubles in less than a decade.

    World per-capita GDP is currently about $10,000. So even doubling every 7-10 years starting in 2025 would mean that world per-capita GDP would be higher than the Luxembourg and Qatar (currently the countries with the highest per-capita GDP in the world) by approximately 2050.

    That would change everything. The problems with Social Security, Medicare, and the National Debt would be erased. Global warming as a problem would be erased. (That's because it would become a trivially small percentage of global GDP to lower CO2 levels back to the pre-industrial level of 280 ppm, via ambient air carbon capture and sequestration, within a few decades.)

    All current diseases of poverty (e.g. malaria, dengue fever, intestinal parasites, tuberculosis, etc.) would be eradicated:


    In fact, it would be difficult to name a human problem that would not be potentially solved by such wealth. For example (and apologies to those affected by the tragedy) imagine a gun that "knows" that it's at a school (via GPS coordinates) or that it's pointed at a child (via recognition) and refuses to fire.

  9. I tend to agree with Brynjolfsson (and not just because I envy him is surname). In particularly, i think there's enormous productivity increase to be wrung out of second-stage integration. At home and at work we have a lot of very clever gadgets and software, which now usually talk to the internet and sometimes the cloud, but not very effectively to the other gadgets and software. At work, I have one program for managing my class, another for entering my grades, a third for managing my employment information. Word processing and research data management is another hodge-podge. I also run my own mail and web servers because I don't much like the way the university does its mail and web. My office has a little device that detects when I'm present and adjusts heating and airflow accordingly. But it could do far more. Imagine if it could interact with a public calendar to let people know when I'm physically present and available. Calendaring itself is crude and inefficient; we use doodle to set up meetings. An intelligent agent could do so much better.

    At home, I manually (when I'm home), open and close blinds to optimize solar capture. This should be done by a computer. TVs, for all their bells and whistles, are still largely 20th century technology, though perhaps Apple will change that. I have a little vacuum cleaning robot which is wonderful, but it has a remote; why can't I operate it via an iPhone app over bluetooth? Why can't I operate my TV, dishwasher, coffeemaker, hot water system and heat pump via bluetooth (or in most cases, have a program do it).