Innovation Files has moved! For ITIF's quick takes, quips, and commentary on the latest in tech policy, go to

Robert Gordon’s Cassandra World

The Federal Trade Commission has rules about unfair and deceptive advertising.   Too bad they don’t apply to academic papers, because if they did Robert Gordon would be facing an FTC inquiry.   His new Cassandra-like paper, “The Demise of U.S. Economic Growth”, has little to do with U.S. economic growth.   Rather it is focused on other factors like transfer payments, taxes, and income inequality.  He should have titled his missive “The Demise of Robust After-Tax Income Growth for Low and Moderate Income U.S. Workers.”   But that’s nowhere near as catchy as his chosen title.

Gordon’s new NBER paper restates his slow-growth forecasts from two years ago, which come in turn from his long tradition of dismissing the potential of technology to drive productivity. This time he is careful to label his more controversial “growth headwinds” (slower innovation and continued globalization) as “speculative”.  Still, he fails to make a more convincing argument for an overall growth slowdown. This is partly due to his reliance on assumptions about education, inequality, and globalization, coupled with a fundamental lack of understanding of the nature of 21st century innovation. But it also stems from his varying usage of growth and productivity: although he dedicates an entire section of his paper to “distinguishing between per-capita output and productivity,” he fails to do exactly that when making his arguments or selecting his title.

The first of Gordon’s four headwinds is demography, which is indeed likely to impact GDP  growth. But while it may affect growth (measured in economic output or output per person), demography is irrelevant to productivity (measured in economic output per hour worked). Because it better relates to per-worker income, productivity is the more accurate measure of economic welfare.  In contrast, GDP includes work hours, so we could raise GDP if we allowed 100 million immigrants move the United States.  But this would do nothing to affect per-worker income of the Americans who area here now. Moreover, since our free time is also valuable, less work can be a good thing even if it detracts from measured GDP.

There are two issues here, of course: people reduce hours because they want to work less (reduced labor supply; e.g. because they get older and want to retire), and people reduce hours because they can’t find a job (reduced labor demand). Gordon argues that the recession is a significant cause of reduced work hours, and this is clear from the data as well—the drop in the 2000s was too sharp for any demographic cause. For this reason he reduces his predictions of labor force contraction by one-third relative to the current trend.  But he still counts the reduction in work hours due to long-term demographics as a negative—which is like saying that having more old people around means we will be poorer. This is not an indicator of productivity and therefore not worth worrying about (except where per-capita statistics are specifically relevant, like with regard to the federal budget).

Gordon’s second headwind is education, and is based on the idea (detailed most prominently in Goldin and Katz, 2008; summary here) that education contributed significantly to our previous economic growth: as the average levels of education increased up through the 1970s, so did our productivity. Gordon’s view of the future, though, is that educational gains are going to be harder to come by, because the average years of schooling has been stagnant since the 1970s (although it hasn’t, really: Gordon notes that it has increased since 2000). One problem with this argument is that education is only tenuously linked to growth in many growth models, particularly “new growth” models. Innovation and technology are much more important to growth than adding a few more months of schooling to the human capital stock. Another problem is that educational attainment is a fairly bad proxy for actual quality of learning —there are still large gains to be had in terms of bringing all students, including college students, up to higher levels of educational competence, even holding constant years of schooling.

The third headwind – inequality – is misleading because it’s not about concerned strictly about economic growth as normally understood by economists. It’s concerned instead about growth for the bottom 99 percent of income earners. Because of rising inequality over the past 40 years, Gordon estimates households in the bottom 99 percent of the population will continue to fall behind the top 1 percent. If the trends for the top and for everyone else continue diverging at the same rate as they did in 1993-2012, growth will slow by 0.5 percent per year for the vast majority of the population.

While inequality is not the same thing as growth, is it still a problem to be concerned about? Perhaps less than Gordon’s data might indicate. Previous work on inequality by ITIF and others has shown that when you control for the rising cost of non-wage compensation as well as reductions in household size, inequality growth, while still real, is less than scholars like Gordon acknowledge. Moreover, as shown by the 1990s, fast growth in labor productivity and a tight labor market contribute significantly to rising median wages. Wages for the bottom 99 percent would in fact grow more rapidly if  inequality declines in the future.

Gordon’s final headwind is public debt: the need to bring the federal government budget into balance after a long period of surplus spending. This contraction in government spending relative to taxes will eventually affect after-tax income and thus consumption, and thus could impact growth. However, this has little or nothing to do with either GDP growth or productivity growth and is a relatively short and medium-term phenomenon: our current debt in the year 2077 (the year Gordon uses for long-term predictions) may well be tiny if GDP growth rates do increase somewhat and if retirement ages are raised modestly.

Gordon distances these “four headwinds”, which he claims are fairly uncontroversial, from what he admits are his two more speculative challenges to growth (he included all six of them as headwinds in his previous paper). He claims that globalization and innovation are both going to slow down growth of the United States’ economy going forward. However, his worries about globalization are centered around globalization’s possible role in increasing inequality—which, again, is not a growth issue but a distribution issue. (Gordon is again talking about growth for the 99 percent.) The notion that low wages in the U.S. lead to reduced standards of living in the future is simply illogical – low wages in other nations may affect the distribution of wages in the United States, they don’t affect overall economic output. Furthermore, if trade balances turn positive due to a depreciating dollar, globalization could reduce the pressure on wages for workers in traded sectors.  And if the trade balance turns positive because the U.S. finally enacts a national competitiveness policy (e.g. see here) then globalization would be a tailwind for growth for the bottom 99 percent.

Despite his attempts to downplay the importance of his innovation claims to his pessimistic forecast, his predictions about innovation predictions still take up a significant part of the paper and remain a central argument. Unfortunately, his claims about innovation and industry are where he goes most wrong. In his continued attempt to downplay the future potential for innovation, Gordon misunderstandsor fails to see the significance of a number of major new technologies.

For instance: Gordon gives Baxter, the new (as well as cheap, adaptable, and human-scale) industrial robot, a “big yawn” because he claims that industrial robots have been around for a long time. Gordon fails to appreciate the way that decreasing cost and increasing ease of use can transform a technology’s potential—which is odd, because he acknowledges that electricity continued to have major impacts on productivity as it declined in price and improved in functionality and began to play a new role in people’s home and work lives. Why is this not true of robots as they get cheaper and better? Certainly, Baxter is a product of technologies that have existed before. But these technologies are now accessible for a whole new range businesses and applications, and Baxter is only another step in a continuing evolution.

Gordon’s take on big data is similarly one-dimensional: he argues that it is merely a weapon in a zero-sum game of marketers competing for customer surplus with better targeted advertising. Leaving aside the fact that better marketing knowledge can mean large efficiency gains for resource allocation by both businesses and government, big data has enormous potential in productive contexts as well. Machine learning and the automation of work hold enormous potential for productivity gains, both for physical robots in terms of machine vision and interaction within environments, and for wholly information-based tasks like document and textual analysis.

But this naïve view and simplistic understanding of innovation should come as no surprise.  Gordon has a long history of not understanding digital technology innovation.  Indeed, as late as the year 2000 he stated that the only two major IT innovations in the US economy to that point were the automatic teller machine and the Sabre airline reservation system.  Gordon falls into the trap that unless it is an atom innovation (e.g. steel, airplanes, cars, highways), it’s not real.

Looking at the paper as a whole, “The Demise of U.S. Growth” makes three big errors. While Gordon ostensibly tries to be clear about the differences between growth, inequality, and productivity, in effect he ends up conflating them through his misleading title and by grouping unlike things together. Second, he suffers from a crippling lack of understanding of current technology.  I wonder if he has ever visited a technology company or a company that is adopting advanced technologies?  Ultimately, growth and productivity gains are based in eliminating costs and creating new value. We believe that the digital revolution is still in its early stages—with no more than half its productivity gains achieved. In other words, more growth is certainly possible and even likely.

Worst of all, in his dissection of growth and productivity trends Gordon also completely misses important causes of innovation, productivity, and growth. He treats innovation as a deterministic process based on current technology, instead of the product of innovation ecosystems embedded in government and private sector organizations and their policies and incentives. Innovation certainly depends on existing technology, but it also depends on the incentives and resources people have to improve that technology. Growth is not something that falls out of the sky, or as economist Robert Solow once called it, “manna from heaven.”  It comes from intentional human action, so we need to make sure that we are doing what we can to understand and promote productivity and innovation—the most important sources of growth.

(photo credit to Timo Newton-Sims)

Print Friendly, PDF & Email