Job polarization in the 2000s?

18 Feb

by John Schmitt

Real World Economics Review Blog, January 19, 2013 [Shortened by Sviluppo Felice]

In a recent post Dylan Matthews[1] takes a dim view of a new paper[2] on the role of technology in wage inequality. For us, proponents of the “job polarization” view of technological change provide no evidence that the framework actually works in the 2000s.

The lack of evidence for the 2000s supports our view that other policy-related factors are what is really driving inequality. We also think that if this purportedly unified framework doesn’t work well for the 2000s, that it is likely not helpful for earlier periods either.

But, even if you still think technology is the main or even an important culprit, we would argue that you need a new theory of technology that actually fits the facts of the 2000s.

For about ten years, a group of economists (most prominently, MIT’s David Autor) has argued that a particular type of technological change has been driving much of rising wage inequality.

In their view, the economy has three broad types of jobs: abstract cognitive jobs (think lawyers, doctors, managers), routine jobs (think manufacturing or many clerical jobs), and non-routine manual jobs (think restaurant workers or office cleaners).

Historically, a lot of good middle-class jobs were in the “routine” category: manufacturing or clerical and administrative jobs that offered decent pay and benefits to workers with a high school degree and sometimes even less than that. Computers and robots etc., the argument goes, are an excellent substitute for workers of routine manual or cognitive tasks. As a result, the advance of technology has dried up the stock of middle-wage jobs and put downward pressure on their wages.

These same technological developments have made abstract workers more productive. Lawyers, doctors, managers and other professionals now have vast processing power at their disposal. These changes in work should, the argument goes, raise the employment and wages of workers in high-skilled occupations.

Meanwhile –these economists argue– computers and robots have made few inroads into non-routine, manual jobs, such as food service and cleaning –the kinds of services that are typically provided in-person (these workers are about 15 percent of the workforce). But technology isn’t undercutting them either. With technology driving down employment in the middle and having no important impact on employment at the bottom, this framework predicts that the wage gap between workers in the middle and the bottom should be falling.

The end result of these forces –strongly rising demand for jobs at the top, falling demand for jobs in the middle, and rising demand for jobs at the bottom– is, in this view, a “polarized” occupational employment structure with expanding high and low-wage occupations and a shrinking middle.

This is a reasonably close approximation of what appears to have happened to wages in the 1990s. The wage gap between workers at the top and the middle continued to grow rapidly, as it had in the 1980s. But, the wage gap between workers in the middle and the bottom shrank, reversing the pattern of widening inequality between the middle and the bottom that had prevailed in the 1980s. Economists call this “employment polarization” and “wage polarization.”

The figure below, taken from a paper by Autor and Dorn,[3] shows (on the vertical axis) changes between 1980 and 2005 in the employment share of occupations arranged by the “skill level” of the occupation (the horizontal axis).


The line has a well-defined U-shape. This suggests that employment in high-skilled and in less-skilled occupations was rising more rapidly than in middle-skilled occupations. The data combine trends for the 1980s, the 1990s, and the 2000s (through 2005).

When the data are broken down by decade (which roughly correspond to three separate business cycles), however, we see three very different stories. Here is a version of the figure presented by Autor in a 2010 paper prepared for the Hamilton Project and the Center for American Progress.


The data for the 1990s (the light green line) trace a U-shaped pattern that looks a lot like the iconic graph for the longer period from 1980 through 2005. But the data for the 2000s (the dark green line) do not show a U-shaped pattern, rather something closer to an L-shape, with expanded employment at the bottom and small declines in employment at the middle and the top.

The data actually show no signs of job polarization in the 2000s. Employment expanded in low-wage occupations, but was flat or falling in middle- and high-wage occupations. It is hard to appeal to “job polarization” as an explanation for widening wage inequality in recent years when the data used to establish occupational employment polarization for the 1990s also show that polarization had stopped by the 2000s.

We would have expected that: (1) wages at the bottom would have grown sharply relative to those at the middle; and (2) wages at the top would have grown at roughly the same pace as wages at the middle.

However, between 2000 and 2007, inflation-adjusted wages of workers at the bottom (the 10th percentile) grew 2.6 percent, wages at the middle grew 2.5 percent, and wages of workers at the top (the 90th percentile) grew 7.6 percent. So, contrary to what the job polarization view would suggest: (1) wages at the bottom did not grow faster than wages in the middle and (2) wages in the middle fell further behind wages at the top.

Both of these patterns are awkward for the job polarization view of technology. Low-wage occupations saw big employment gains, but these did not translate into rising relative wages for low-wage workers. High-wage occupations saw no change in their employment share, but high-wage workers experienced much more rapid wage growth than those at the middle (and the bottom). As was the case at the low-end of the wage distribution, something other than occupational employment changes appears to be driving overall wages at the top of the distribution.

Technology-driven job polarization is, at best, a story of the 1990s.

[1] Dylan Matthews, “Inequality is rising. Should we blame robots or the government? Or both?”, Wonkblog, January 12, 2013.

[2] Larry Mishel, John Schmitt, Heidi Shierholz, “Assessing the job polarization explanation of growing wage inequality, Economic Policy Institute, January 11, 2013.

[3] David Autor and David Dorn, “The Growth of Low Skill Service Jobs …”, IZA, Discussion Paper No. 7068, Dec. 2012.


Una Risposta to “Job polarization in the 2000s?”

  1. valeria cirillo 18 febbraio 2013 a 15:20 #

    I’m really glad to read this post. I’m working on this topic and I know Mishel et al. (2013)’ paper. It will be nice to receive comments on it and maybe share ideas with other researchers.


Inserisci i tuoi dati qui sotto o clicca su un'icona per effettuare l'accesso:


Stai commentando usando il tuo account Chiudi sessione / Modifica )

Foto Twitter

Stai commentando usando il tuo account Twitter. Chiudi sessione / Modifica )

Foto di Facebook

Stai commentando usando il tuo account Facebook. Chiudi sessione / Modifica )

Google+ photo

Stai commentando usando il tuo account Google+. Chiudi sessione / Modifica )

Connessione a %s...

%d blogger hanno fatto clic su Mi Piace per questo: