The great decoupling

Tyler Cowen’s e-book The Great Stagnation offers a novel explanation of the slowdown in U.S. median income growth since the 1970s. Here’s his causal model:

Innovation —> economic growth —> median income growth

In this model there are three potential sources of the reduction in median income growth:

1. Innovation has slowed.

2. The degree to which innovation boosts economic growth has declined.

3. The degree to which economic growth boosts median income growth has declined.

Cowen argues for hypothesis #1. He cites an estimate by Jonathan Huebner, a Pentagon physicist, that the rate of global innovation per capita peaked in the late 1800s, remained high to the mid-1950s, and then steadily declined. And he suggests that whereas “The period from 1880 to 1940 brought numerous major technological advances into our lives…. Today … apart from the seemingly magical internet, life in broad material terms isn’t so different from what it was in 1953.” The high rate of innovation through the mid-1950s enabled rapid economic growth for a few additional decades. But beginning in the 1970s economic growth slowed, and along with it median income growth.

The book is well worth reading. (At four dollars it’s also a good deal — less than a large latte, a Sunday New York Times, or a newsstand copy of The Atlantic.) But I’m skeptical on two counts.

First, I’m not convinced that innovation has in fact slowed significantly. Cowen discusses the internet but not computers more generally. Computers are the engine of the postindustrial economy; they are the modern counterpart to steel, railroads, and the assembly line. Advances in computer hardware and software, their widespread dissemination, and their application to myriad tasks — automation and coordination of supply chains in manufacturing, record keeping and scheduling in services, and much much more — surely represent a massive improvement.

Second, the data point to hypothesis #3. A key difference between the WW2-1973 period and the decades since then is that median income growth has become decoupled from economic growth. (Mark Thoma makes this point too.) The rate of economic growth has been lower in the recent era, but it’s nevertheless been decent. Yet median income growth has been very slow. This contrasts sharply with the prior period.

Here’s one way to see this (others here):

Between 1947 and 1973, GDP per family increased at a rate of 2.6% per year and median family income grew at 2.7% per year. From 1973 to 2007, GDP per family increased at 1.7% per year, but median family income grew at just 0.7% per year.

And note the absolute numbers: GDP per family rose by $52,000 during 1947-73 and then by $82,000 during 1973-2007. Median family income increased by $26,000 during 1947-73 but then by just $13,000 in 1973-2007.

Median family income was $64,000 in 2007. Had it kept pace with GDP per family since the mid-1970s, it instead would have been around $90,000.

I’m all for helping to accelerate the rate of innovation. But the big change in recent decades lies in the degree to which economic growth lifts middle-class incomes. If we want to understand slow income growth, that should be our focus.

10 thoughts on “The great decoupling

  1. Just a guess, but it seems likely that innovations in information technology allowed first movers who recognized its potential (for example, financial experts) to profit from information asymmetries. That seems like just the sort of thing that could account for enormous rents in the face of comparatively slow overall prductivity growth. It also seems possible that recent productivity improvements might be more in the nature of public goods and therefore less well represented in conventional measures of growth.

  2. The first time I heard “innovation” measured as “inventions per capita” I said “what?”

    I still don’t believe it. It may be a convenient number, but it is a nonsense number (esp. when you substitute “patents” for “innovation.”)

    My main argument is that with greater connectivity and greater information flow, each invention matters more, not less, than it did with smaller and more isolated populations.

    That seems obvious.

  3. Several easy ‘leads’ in this mystery. First is that while population increased faster during the later measurement periods, America since the late 60s was no longer the exclusive workshop of the world and able to easily absorb unskilled newcomers (native or immigrant) into a middle class factory wage situation–in fact globalisation has shrunk that capacity. Computerization and robotics further reduced the entry paths for the unskilled. Meanwhile, educational services have not kept up on average and deteriorated for many. But production and profits, including those for offshore activities, have continued to accelerate thanks to technology.

    More difficult to see is the solution. Given a time machine the U.S. might have been wise to seal itself off at the pinnacle of its advantage, allowing in only raw materials and the most high-aptitude immigrants and allowing out only last generation finished goods. Obviously the Cold War and corporate growth aspirations made such a move inconceivable at the time.

    But from here where to go? Get used to sharing, improve education and R&D, facilitate inbound brain drain from the rest of the world, and invest in public infrastructure as a way to ameliorate shrinking real median incomes and improve competitiveness without sacrificing so much on the labor compensation side. Those are some of the things I hope President Obama was referring to in his latest report.

  4. That curve bending illustrates the rise in income inequality that reflects a US-centric measurement in a world where businesses no longer recognize national boundaries. Decades ago discoveries by American scientists got developed into American technology and became mass produced in American factories that employed American labor. We’re still doing the front end, and the shrinkage to the middle has only just begun, but most of the back end has been offshored for some time. And the back end is where the benefits accrue to the median income.

  5. “Today … apart from the seemingly magical internet, life in broad material terms isn’t so different from what it was in 1953.”

    Cowen’s hypothesis is silly beyond belief. Why would anybody take him seriously?

  6. Lane,

    You appear to be using the data from the Census Bureau’s official definition of income (“money income excluding capital gains”). But as the CB acknowledges, that measure has serious problems. In particular, it does not account for the effects of taxes and transfers, including non-cash benefits such as Medicaid. The CB provides historical income data for a number of alternative income measures. The most sophisticated of these is income definition 15, which makes adjustments for taxes, transfers, health coverage, capital gains and other factors. The data is here:

    Using this alternative income definition, the growth rate is much higher. The CB historical data only covers the period 1979-2003, but that is enough to show a large difference.

    Change in median real household income, 1979-2003, using income definition 1 (Lane Kenworthy method):

    1979 income: $38,649
    2003 income: $43,318
    Growth: +12%

    Change in median real household income, 1979-2003, using income definition 15 (accounts for taxes and transfers, capital gains, health insurance, etc):

    1979 income: $37,776
    2003 income: $45,154
    Growth: +20%

    12% vs. 20%. That’s quite a difference. But we should also adjust for the change in household size. You say this made little difference to your analysis, but you don’t show your calculations. The CB reports that between 1979 and 2003, average household size declined from 2.76 to 2.57. When I use these numbers with income definition 15 to calculate the change in median real household income per household member, I get the following:

    Change in median real household income per household member, 1979-2003

    1979 income: $13,787
    2003 income: $17,570
    Growth: +28%

    So making these two changes — using a more sophisticated measure of household income and adjusting for the change in household size — increases the growth rate dramatically, from 12% to 28%.

    This is why I don’t trust your analysis.

    Historical income data by income definition is here (Table RDI-4):
    Historical average household size data is here (Table H-11, All Races):

  7. Miller,

    Your first adjustment makes sense and, if correct, largely accounts for the discrepancy between Lane’s $78K and his counterfactual $90K. But the second doesn’t since both the GDP figures were expressed in $ per family. One is a mean and the other a median and the growing discrepancy between those two measures undoubtedly reflects increases in measured income inequality.

  8. I have no expertise whatsoever on the subject, but am wondering about the concept and definition of “family.” Is it the case that what constitutes a family has been statistically adjusted over time to represent roughly the same “thing”?
    Just wondering…

Leave a Reply