Policy-making relies on narratives, and narratives often come from data. Or claim that they do. One story often told by economists—by everyone from Dani Rodrik to Erik Brynjolfsson and Andrew McAfee to James Kynge to Laurence Summers—is that China’s manufacturing sector has been shedding workers since the mid-1990s. This story leads us to believe that something like this is happening:
This argument ends up as a morality tale with serious policy implications: if even China, manufacturing powerhouse with wages developed countries cannot hope to compete with, is losing manufacturing jobs, then surely manufacturing jobs are obsolete and the U.S. is foolish to try to maintain them—let alone get them back.
Unfortunately, this story is based on a gross misreading of inaccurate evidence. There are three major problems. First, even based on a simplistic look at the data, it’s flat out wrong. Take a look at this chart that shows the actual manufacturing employment in China. (You may note that this chart only goes back to 1998, and that the peak of employment underlying most claims was in 1996—more on that in a bit.)
Strangely enough this graph looks nothing like the first one. As you can see, China now has more people working in the manufacturing sector—in raw numbers and nearly as a percentage of the total workforce—than at any earlier point. In fact, the numbers keep rising even in the midst of the 2009 recession.
To put the data in context: China has added more manufacturing workers since 2002 than the entire manufacturing workforce of the United States—and more than double what the U.S. lost over the same period. These graphs might help you get a sense of scale:
If Chinese manufacturing employment has been increasing since 2002, where does the China-losing-manufacturing-jobs story come from, then? It comes from the data in the 1990s that shows that China had a surge of manufacturing employment up through 1996. Back in 2005 (and in fact until updated data became available) it may have been excusable to claim that China was losing manufacturing employment. Here is the manufacturing employment series from Banister’s 2005 paper:
But this data only looks like a big decrease if you don’t read the surrounding article very carefully. As interesting as it may be, this chart tells us basically nothing about manufacturing employment trends. This is because of the second and third problems with our technology-kills-manufacturing story, which come from the data itself. Problem #2 is that the manufacturing industry underwent massive shifts in ownership and institutional structure, rendering any technology-centric narratives meaningless. Problem #3 is that there are significant measurement changes leading to discontinuities in the measurements themselves.
The data counts both private manufacturers and state-owned enterprises (SOEs). Technologists like McAfee and Brynjolfsson ignore the critical importance of institutional shifts (placing all the blame on robots), but the shift of ownership from public to private has played a critical role in manufacturing employment. China’s economy has gone through massive restructuring over the past 30 years, and the mid-1990s saw manufacturing transition from SOEs to far more private employment. (Richard B. Freeman’s recent paper provides a good overview.) This transition meant that the surplus labor that had built up in inefficient SOEs in the 80s and early 90s was forced into the private job market and more efficient private manufacturing companies. In the words of Judith Banister (former director of International Programs at BLS and author of the definitive papers on the subject):
The genuine declines in PRC manufacturing employment in the late 1990s were caused by restructuring and privatization of state-owned and urban collective-owned factories in the cities, which brought about massive layoffs of urban manufacturing workers and sharp increases in manufacturing labor productivity. Private sector manufacturing has thrived in both urban and rural areas in the late 1990s and the early 21st century.
Note the word “genuine” in that paragraph: Banister is referring to the fact that some of the decline shown by the data may not have occurred at all. This is the third and final reason the 1990s employment peak is irrelevant: part of the drop between 1996 and 1998 appears to be partly a statistical artifact.
Chinese manufacturing statistics are collected by two different agencies—the Ministry of Labor collects stats on urban manufacturing, while the Ministry of Agriculture collects stats on manufacturing everywhere else. This means that Agriculture is in charge of statistics about Town and Village Enterprises (TVEs), special enterprise zones, and all other manufacturing outside of official cities. There was no effort made in the 1990s to coordinate the statistics between the two agencies, and as a consequence trying to reconcile these two statistical series and get an accurate picture of overall employment is not an easy task.
Moreover, the agencies themselves made a number of changes that resulted in discontinuities in the data, and have never fully explained them. For example: part of the large drop from 1996-1998 can be explained by the reclassification of manufacturers in urban areas; meanwhile, in 1997 TVE employment dropped by 20 percent, then almost fully recovered in 1998 (see Banister 2005). Depending on the way you count, employment sums can vary by as much as 20 million workers—more manufacturing workers than the United States has ever had at one time.
Clearly, there are both large statistical anomalies in the 1990s data and large, real shifts due to structural changes. Until the Chinese government agencies release their methodology we may never know which is which, but either of these reasons renders the declining manufacturing employment argument moot.
We need to be careful when using data to create stories that shape our policies—and to respect the limits of our own knowledge. Particularly since 2013when the newest BLS data was released, there is no excuse to continue making the same faulty arguments based on oversimplified logic and speculative data. The ostensible decline of Chinese manufacturing employment is nothing of the sort, and it is time to retire the myth. U.S. manufacturing has been decimated over the past 15 years and there is no room here for fatalistic attitudes–particularly not inaccurately fatalistic ones.
(Zombie photo credit: Thierry Ehrmann)
1. Banister, Judith, “Manufacturing Employment and Compensation in China,” Bureau of Labor Statistics, 2005, http://www.bls.gov/opub/mlr/2005/07/art2full.pdf.
2. Banister, Judith, “China’s Manufacturing Employment and Hourly Labor Compensation, 2002-2009,” International Labor Comparisons, June 7, 2013, http://www.bls.gov/fls/china_method.pdf.
3. St. Louis Federal Reserve Bank FRED, All employees: Manufacturing (seasonally adjusted), http://research.stlouisfed.org/fred2/series/MANEMP.
4. World DataBank, World Development Indicators: China: Population, http://databank.worldbank.org/data/Popular_countries/id/556d8fa6.
5. International Labor Organization, ILOSTAT: China: Labor Force, http://www.ilo.org/ilostat/faces/home/statisticaldata.