The Bureau of Labor Statistics (BLS) announced on February 5, 2016 that the unemployment rate fell to 4.9% in January 2016. The unemployment rate a year ago in January 2015 was 5.7%.  Although the decline should bode well for the economy, conflicting data with other unemployment statistics have analysts and the financial news media offering different interpretations of this announcement.  For example the long-term unemployed (those without jobs for over 27 weeks) remained unchanged at 2.1 million, accounting for 26.9% of the unemployed. In addition, the category of discouraged workers, defined as those not currently looking for work because they believe there are no suitable jobs available, was 623,000. Both these numbers have been used to downplay the declining unemployment rates even though a recent study by the Federal Reserve Bank of St. Louis showed most discouraged workers are transitory and return back to employment soon.1

One way to mitigate the conflicting interpretation of data is for the BLS to provide more granular data with alternative categories of employment. The incorporation of technology and automation at all levels of the workforce make the current figures inadequate to properly assess the impact of unemployment. The additional data should not be based on the traditional divisions of the labor force, manufacturing, services and government sectors. Instead, the categories should incorporate technology and skill sets required for completion of the job.

One problem with the selection of the appropriate categories is that the impact of technology on employment is yet to be fully understood. Earlier analysis predicted job losses whenever and wherever automation replaced workers. Yet recent studies have shown that this relationship may not hold for a range of job categories. James Bessen, Professor of Law at Boston University, has provided evidence that even routine jobs that have been computerized have seen increases in employment.2 One example cited is the increase in the number of bank tellers between 2000 and 2013 even as the number of ATMs have proliferated. He reasons that the introduction of ATMs has allowed banks to open more branches at lower cost and thereby increase bank tellers. Similarly, barcode scanners that were introduced in the 1980s have reduced checkout times at most retail outlets but the number of cashiers continue to show strong growth. In addition, both bank tellers and cashiers are using advanced equipment (computers and scanners) indicating partial automation of their jobs.

Another factor compounding the analysis of jobs data is the outsourcing of work overseas, making it difficult to determine if automation or outsourcing is responsible for job losses. Both factors have been blamed for “labor polarization” in the U.S., a term coined recently to indicate a loss of middle skills jobs. Automation has a direct impact on jobs that are simple and routine while outsourcing impacts both low and middle level jobs. If labor polarization persists, it would be helpful for unemployment statistics to contain functional and technological components in each job category.

Recent academic studies have begun to analyze labor trends with different categorizations. A recent study by the Federal Reserve Bank of St. Louis examined the growth rates of employment using the following categories.3

  1. Non-routine cognitive : includes management and professional jobs
  2. Routine cognitive : includes sales and problem-solving administrative work
  3. Non-routine manual : includes bookkeepers and mail sorters
  4. Routine manual : includes welders, fitters and forklift operators

As the chart from the Fed study below shows, employment in both non-routine cognitive and non-routine manual sectors increased steadily from 1983 to the present. In contrast, employment in routine cognitive and routine manual sectors remained sluggish, if not declined, throughout the 30-year period (see also Siu and Jaimovich for more discussion on the employment categories).4

Time for new Unemployment data

In addition, the graph below shows that the rate of non-routine cognitive unemployment has also been low throughout this period while routine manual unemployment remained the highest of the four categories.

Time for new Unemployment data


In sum, the reporting of employment statistics has to expand in scope and granularity. The BLS has an excellent data gathering system in place for their monthly forecasts. Their sampling techniques and questionnaires has been modified and finessed over many decades to provide consistent and reliable unemployment statistics. It is time for them to tweak their questionnaire to provide “cognitive” data that can be interpreted and utilized more effectively by economists and investment managers.

1. See our blog, “Unemployment Statistics, Fact or Fiction” October 22, 2014. The Fed study on discouraged workers is B. Ravikumar and Lin Shao, “Discouraged Workers: What Do We Know?” Economic Synopses, Federal Reserve Bank of St. Louis, March 14, 2014.

2. James, Besson, “How Computer Automation Affects Occupations: Technology, Jobs, And Skills”, Boston University School of Law, Law & Economics Working Paper No. 1, January 16, 2016.

3. Maxmiliano, Dvodkin, “Jobs Involving Routine Tasks Aren’t Growing, On the Economy, Federal Reserve Bank of St. Louis, January 4, 2016.

4. Sui, Henry and Nir Jaimovich, “The Trend is the Cycle: Job Polarization And Jobless Recoveries,” Working Paper 18334,, August 2012.