Automation and US Employment: New Research

new abstract

The popular press is full of articles about how automation destroy jobs.  It seems so intuitive, capital, in the form of technological advancement, replaces labor.  In response to claims that low wage economies, like Mexico and China behind the collapse of American manufacturing jobs, it is sometimes noted the real culprit is not trade as much as technology. 

However, the impact of automation on employment is still hotly debated by investors, economists, and policymakers. A new study posted the Center for Economic and Policy Research suggests automation has created more jobs than it has destroyed in the US.  The study, conducted by two PhD students at the University of Bonn (Katja Mann and Lukas Puttmann) looks at the period between 1976 and 2014.  Over that period, US non-farm payrolls rose an average of 133k a month.

The economists focus on patents. 
Patents are often used as proxies for innovation, but they can also be used as a proxy for technology available to be exploited by businesses.  The study uses a machine learning algorithm to all five million patents the US granted in that 1976-2014 period, and identify the patents related to automation.  Automation is defined broadly in terms of hardware (e.g., robots) and software (e.g., process, programs and cognitive inventions).

The algo identified 40% or 2 mln of the 5 mln patents were for automation.
  There number of patents for automation accelerated from 70k in 1976 to 180k in 2014.  The share of automation patents of overall patents rose from a quarter to two thirds over the period.

Next, the economists link the patents to industries that are likely to use them. 
They recognize that the industry that uses the patent may be different than the industry of origin.  As one would suspect, there is a link between industries that have a high share or routine tasks in 1960.  However, over time the link has gotten weaker.  There are at least two possible explanations for this development.  First, automation has already made routine jobs redundant. Alternatively, it could also be that newer automation technology can replace some non-routine tasks.

The study identifies “commuting zones” of industries that can use the automation technology.
  The economists calculate the number of new patents that individual workers in a commuting zone could use.  Two distinct patterns are evident.  In the 1970s and 1980s, the automation technology was concentrated in the Great Lakes area where had a high share of manufacturing.  However, by 2014, automation technology had become more dispersed, which speaks to the changing nature of automation, and change in local industry structure.  For example, auto production is no longer concentrated in the Great Lakes area.  Mississippi, Alabama and Kentucky, South Carolina and Tennessee have important auto plants.

The economists then examine the effect of automation technology on employment across the more than 700 commuting zones over nearly 40 years.
  The economists then regress five-year period of employment-to-population ratio with the automation indicator that they create from the data.  They control for several variables, including non-automation patents and demographic variables.

Contrary to what many intuitively seem to think, the research found that more patents in a commuting zone translates into an increase in the employment-to-population ratio.
  Specifically, the economists found that a one-standard deviation increase in their automation measure predicts a 0.2% in the employment-to-population ratio per five-year period.

The study found important differences between manufacturing and services.
  “Manufacturing employment falls and service sector employment grows in response to new automation technology.”  On a net basis, routine jobs benefit less from automation and manufacturing jobs tend to do worse, but the rise in service sector jobs more than offsets it.

The research is about the quantity of jobs and the relationship with automation technology.
  It does not address compensation issues.  Nevertheless, the conclusion that automation technology is spurring a shift in the allocation of labor and not undermining it is important.  In terms of public policy, more effort may be needed to help people make the transition through skill-based training. A new generation of Luddites can be avoided.  From a macroeconomic point of view, the research suggests that facilitating innovation and patent developments does not have to undermine labor force participation rates.