Methodology for "Transitioning to a non-routine economy: it’s not happening, but it has to."

The job “scores” I refer to were pulled from the O*NET jobs database. The O*NET jobs database is a federally maintained repository of detailed descriptions of over 900 occupations which fairly comprehensively describes the US workforce. O*NET has performed surveys to create an exhaustive list of every skill, knowledge category, work activity, and other job characteristics that could possibly be required for job performance. They’ve then given a numeric score for each of these variables assessing the level at which they need to be performed for every single US occupation.

Each O*NET occupation has a Standard Occupational Classification code: a 6-digit number used by the Bureau of Labor Statistics to classify occupations. The Occupational Employment Survey is conducted each May by the BLS and measures occupation-level employment totals. I used the SOC code to pair occupations and their job characteristics as measured by O*NET in 2013 with employment levels from the OES survey for each year going back to 1999 -- the earliest year for which the BLS used the current 6-digit SOC format. BLS did not have any occupation-level employment data available for years before 1997. I gathered data for all 702 occupations used by Osborne & Frey in their analysis.

I looked at two sets of jobs to view their share of the total US workforce. First, I looked at all jobs that Osborne & Frey had found to be 40% or less likely to be automated in the next 20 years, a total of 270 jobs. These jobs tended to have high O*NET ratings for originality or fine arts knowledge (grouped under “creativity”), assisting and caring for others, persuasion, negotiation, and social perceptiveness (grouped under “social intelligence”), and finger and manual dexterity (grouped under “perception and manipulation). For each year, I divided the total number of these jobs by the total number of employed persons it he US according to the OES surveys to arrive at a total percentage of the workforce deemed by Osborne & Frey to be less than 40% likely to be automated in the next 20 years.

The second subset of jobs I looked at in hopes of measuring the relative size of the creative workforce, leaving out jobs that were deemed unlikely to be automated primarily due to their social intelligence or perceptive/manipulative characteristics. To do this, I summed the originality and fine arts knowledge scores for each occupation. I then charted this combined “creative score” in a scatter plot against their probability of automation as calculated by Osborne & Frey. This chart showed a significant increase in probability of automation for occupations with creative scores under 4.5.

In addition, I ranked the 270 jobs deemed less than 40% likely to be automated in terms of their combined “creative score.” I looked at the top 90 most creative jobs of those 270, because 90 is one-third of 270 and because creativity is one of three areas of cognition deemed unlikely by Osborne & Frey to be computerized in the near future due to engineering bottlenecks. Of these 90 most creative jobs, the vast majority had a creative score above 4.5.  

To measure the share of the workforce comprising creative jobs only, I therefore looked at every job with a creativity score over 4.5 as a percentage of the total number of employed persons in the US for a given year according to the OES surveys.