A major cause of jobless recoveries is recession-related switches to automation

9 Nov 2012 by Jim Fickett.

Henry Siu and Nir Jaimovich, in an article on VoxEU, classify all US jobs as one of (1) non-routine jobs with high wages, (2) routine/repetitive jobs with middle income, or (3) low-paying service jobs. They show that in each of the last three recessions, the number of routine/repetitive jobs fell drastically and did not recover, while the high- and low-paying jobs recovered more normally. They interpret this to mean that the recent jobless recoveries are mostly due to automation, and that each stage of switching from humans to machines occurs mainly with the spur of a recession.

The fact that employment is recovering much slower than GDP is a relatively new phenomenon; jobless recoveries have only really occurred after the recessions of 1991 and 2001. These last three recoveries represent a distinct break from previous postwar episodes of recession when both GDP and employment would vigorously rebound following recessions …

Our current research indicates that a jobless recovery is not simply an ‘economy-wide‘ delay in firms hiring again. Instead, it can be traced to a lack of recovery in a subset of occupations; those that focus on “routine” or repetitive tasks that are increasingly being performed by machines …

As recently as the mid-1980s, about one in three Americans over the age of 16 was employed in a routine occupation. Currently, that figure stands at one in four. The fact that polarisation is occurring should not surprise anyone who understands the influence of robotics and automation on machinists and machine operators in manufacturing. Indeed, the influence of robotics is increasingly being felt on routine occupations in transportation and warehousing. Of equal importance is the disappearance of routine employment in ‘white-collar’ occupations - think bank tellers being replaced by ATMs, or secretarial work being replaced by personal computers and Siri, Apple’s iPhone-integrated ‘intelligent personal assistant’. Thus, all of the per capita employment growth of the past 30 years has either been in ‘non-routine’ occupations located at the high-end of the wage distribution, such as software engineers and economists, or in low-paying jobs, such as service occupations like restaurant waiters and janitors. For this last set of occupations, this has been especially true in the past decade. …

What is surprising is the link between job polarisation and the business cycle. Figure 1 highlights our simple point; it plots per capita employment in routine occupations (in log levels) from 1967 to the end of 2011. Since about 1990, there is an obvious 28 log point decline in routine employment. What is equally clear is that this fall has not happened gradually over time but that the decline is concentrated in economic downturns. 92% of the 28 log point fall occurred within a 12 month window of NBER-dated recessions.

… jobless recoveries cannot be traced to the business cycle behaviour of ‘non-routine’ jobs: employment in these occupations experience only mild contractions, if at all, during recessions, and have experienced essentially no change in the nature of their recoveries over the past half century. …

A recent report by the National Employment Law Project (2012) indicates that the recovery from the Great Recession has been particularly lopsided, with the majority of jobs added being low-paying jobs.

The pace of job polarisation was greatly accelerated in this last recession, and the pace of automation and progress in robotics and computing technology is not slowing down either (Brynjolfsson and McAfee 2011). If the past 30 years is any guide, we should expect future recessions to continue to spur job polarisation. Jobless recoveries may be the new norm.


I was led to this paper, indirectly, by a link on Early Warning. The VoxEU post cited above is only a summary of the research and does not give, for example, the details of how the authors define routine/repetitive jobs. The full paper is available for purchase from NBER.