Nice work

Today I am happy with Equitable Growth, they done fundamental and good research on the labor market. 

In the process they solved a bunch of economic problems, for real.  The made the labor matching smart machine, a Watson style machine to manage labor searches. They want to prove the usual talking points, but ignore that bias and focus on how they match in the labor markets.

First, teach the bot to read:

In this paper we show that, in fact, substantial changes in task composition did occur within occupations since 1960.2 Moreover, we find that these within-occupation changes in task content account for much of the observed increase in earnings inequality. We start by constructing a new data set using the text content of approximately 4.2 million job ads appearing in three major metropolitan newspapers — the New York Times, the Wall Street Journal, and the Boston Globe. We then map the words contained in job ad text to different classifications of task content. Our main strategy relies on Spitz-Oener’s (2006) classification of words into routine (cognitive and manual) and nonroutine (analytic, interactive, and manual) tasks. We complement this approach using the classifications in Deming and Kahn (2017), Firpo, Fortin, and Lemieux (2014), and the Occupational Information Network (O*NET). This last mapping is useful, since O*NET is at present the de facto standard in the literature for measuring the content of occupations

Then read the want ads

In our first quantification, we incorporate our occupational measures into Fortin, Lemieux, and Firpo (2011)’s methodology for decomposing changes in the wage distribution across points in time. Using these methods, we break down changes in the distribution of earnings over time into changes in the attributes of workers and their occupations (the “composition” effect), as well as changes in the implicit prices of those observable characteristics (the “wage structure” effect). Next, we further break down changes in the distribution of earnings into the contributions of observable characteristics (e.g., the contribution of changes in educational attainment and of changes in the returns to education). 

Do some sequence matching

In our second quantification, which we view as complementary to the earnings distribution decomposition, we construct a general equilibrium model of occupational choice. In our model, individual occupations are represented as a bundle of tasks. Workers’ skills govern their abilities to perform each of the individual tasks in their occupation, and give rise to comparative advantage. These skill levels are functions of workers’ observable characteristics — like gender, education, and experience — but also contain an idiosyncratic component. Based on their skill levels and the demand for tasks within each occupation, workers select into the occupation with the highest payoff.

They find market uncertainty

Using data from CareerBuilder, Marinescu and Wolthoff (2016) document substantial variation in job ads’ skill requirements and stated salaries within narrowly-defined occupation codes.

They find that within occupations, routine tasks have been eliminated. 

Anyway, hire these guys, en masse, including half the folks they reference. Their research is also the solution, something that matches hiring managers and workers. This is where WeWork needs to go, evolve a job language, a grammar both hired and hiring can understand.