Worker training and skill upgrading programs are a major focus in impact evaluation work. The design of such training programs implicitly involves the identification of the activities that a worker needs to accomplish in a job. Only then can the program offer training in the set of skills required to complete these identified tasks. Economists have long been interested in measures of worker skills – starting with old school measures of human capital such as education and experience up to more nuanced and comprehensive measures such as those in the Program for the International Assessment of Adult Competencies at the OECD. However economists have paid little attempt to measure the task content of jobs. At least until recently.
A new literature tries to empirically untangle the skill requirements of different jobs through a “task framework”. Behind this approach is the concept of the job as a “bundle of task demands”. An early example of the task framework is the 2003 paper by Autor, Levy, and Murnane that measures the average task requirements of different occupations in order to understand the role of computerization for changes in the U.S. wage structure.
While this early work is highly influential we know that task demands vary within an occupation as well across occupations, yet there has been little attempt to extend the task measurement of jobs until now. Autor and Handel analyze data from a first attempt to assess task measurement at the job level. What they find is useful for understanding the functioning of labor markets as well as suggestive for measurement approaches that can be incorporated into impact evaluation.
Autor and Handel collect data on the job activities across a variety of task domains for a representative sample of US workers. To render this initial analysis tractable they group the task domains into three broad areas of activity – cognitive, routine, and manual. These domains are defined and measured by the following:
Cognitive or Abstract Tasks include abstract problem-solving and organizational and managerial activities. These are measured by such tasks as (1) the length of the longest document typically read as part of the job, (2) frequency of mathematics tasks at the high-school level or higher, (3) frequency of problem solving of at least 30 minutes duration, and (4) proportion of work day managing or supervising others.
Routine Tasks are defined as cognitive and manual tasks that follow explicit procedures and are measured by (1) the percent of the work day spent performing short repetitive tasks, as well as the absence of face-to-face interactions with (2) customers or clients, (3) suppliers or contractors, (4) students or trainees.
Manual Tasks involve non-routine manual activity that require physical adaptability and are measured by (1) the percent of workday spent performing physical tasks such as standing, operating machinery or vehicles, or making or fixing things by hand.
This measurement effort results in a simple snapshot of US labor tasks interesting in its own right: for example they find that 24 percent of wage or salary workers use any kind of higher level math in their job, 37 percent read documents longer than 6 pages on a regular basis for their job, and 29 percent manage or supervise others at least half of their time. As expected, tasks vary quite a bit by occupation. In fact broad occupation categories are more greatly correlated with the frequency of tasks than is the education of the worker.
There are also clear differences in average tasks for workers with differing characteristics, largely reflecting differential occupational allocation. Women are more likely than men to spend at least half their time on repetitive tasks. And, as expected, education plays a strong role in determining the task content of jobs: the gap between workers with high school and those with post-college is around 1 standard deviation each for abstract tasks, routine tasks, and manual (higher levels of education lead to more abstract tasks and less routine or manual one). Tasks also remain significant predictors of wages, even after controlling for human capital and demographic measures.
A separate effort to measure task is a text based analysis of 12000 occupation descriptions from the Dictionary of Occupational Titles over the period 1880-2000. The authors Michaels, Rauch, and Redding focus on 3000 verbs from repeated versions of this dictionary to establish their metric of tasks per occupation. They then match these tasks with the relative importance of each occupation based on employment micro-data over the same historical period.
As one might expect, systematic change in the task-based content of common occupations becomes readily apparent. Here are some of the verbs most commonly associated with urban occupations at 60-year intervals:
1880 - Thread, Stretch, Sew, Braid
1940 - File, Bill, Compile, Distribute
2000 - Develop, Determine, Analyze, Review
The authors also identify a change towards interactive activities described by verbs associated with thought, communication, and social activity. This work is a fascinating window into the consequences of technological change and economic development for labor demand.
The task-based approach can reshape how we think about changes in labor demand as changes in the prices paid for specific tasks, which in turn affect the earnings of different skill and demographic groups. There is clearly a long-run measurement agenda here, but it is worth flagging the current efforts as they can enrich questions explored by the impact evaluation of labor and training programs.