The rise of the machines (?)
The analysis of the impact of technological growth in the labor market has gained momentum in the last years. Nowadays, almost every economist agrees with the statement that technological progress affects the labour market; however, there is much less consensus on the how. Opinions in the topic had proliferated: from the early approaches of Katz and Murphy (where the adoption of new technologies leads to an increase in the productivity of skilled workers) to the offshorability literature pioneered by Blinder (where technology allows for a delocalization of the production of goods and services, thus increasing competition), including the seminal contributions of Autor (computers and robots can replace only some of the tasks done by workers).
As a part of this ongoing debate, the Instytut Badań Sturkturalnych (IBS) organized a conference that attracted presenters from various countries . Among the papers presented, one of the most interesting (Graetz and Michaels – their presentation will be soon available in the IBS channel) analyzed the impact of the use of robots on labor productivity and employment in a number of industries and countries. Unfortunately, the robots they analyzed resembled more robotic arms than the robots we observe in the movies.
In any case, their approach is quite straightforward: they employ an input database with information on the value added by country and industry (EU-KLEMS) and combined it with a database on the amount of robots bought by the same industry. They calculate the effect using a dif-in-dif approach. The results suggest that the introduction of robots had a large impact on productivity on those industries. The authors claim that the productivity growth that ensue robot adoption was of the same order magnitude than the development of the steam engine. With respect to employment, the authors argue that robots did reduce employment, but only for a specific group of workers, namely those in so-called low skill occupations. Their paper then would somehow fit in the tradition inaugurated by Katz and Murphy.
An easy critique of their analysis concerns the issue of reverse causality, i.e. that productivity growth allowed firms in those industries to finance the introduction of robots. Thus, even if the outcome is the same, causality might have run in the opposite direction. The authors proposed an instrumental variable (IV) approach to solve this issue. The construction of the instrument is quite interesting and deserves some additional thought. They created an index of replaceability, which combines the extent to which workers can be replaced by machines (close to Autor, but not quite) and the proportion of these workers in US manufacturing industries in 1980.
The construction of this instrument has some weak points that the authors might be interested in addressing in the future. The first and most obvious comes from the use of US data, both in the description of occupations and their shares in the industry. This approach hides two assumptions: a) that workers in the same occupation perform the same tasks in all countries and b) that the occupational structure within each industry is the same in all countries. Secondly, by using US data for the remaining countries, it is not clear how the IV differs from an industry dummy.
The second concern with the construction of the IV comes from the theory and concerns to what extent machines can replace workers. Graetz and Michaels solve this problem quite straightforwardly. They analyze the task description of different occupations and the tasks that are conducted by robots. If a task appears in both lists, they consider the occupation to be replaceable. Now, if we were to take Autor's insight seriously, we have to recognize that the criterion for replaceability is insufficient. For once, these tasks might represent only a small part of the task content of the job. Second, the routine content of occupations refers also to a controlled environment; where there is little need to adjust to the conditions. An example should clarify the problem, while robots can be used to paint cars in a factory, they might have problems to paint a room.
In short, the paper offers new opportunities to explore the role of other technological changes beyond those occurring in the ICT and allows for the coexistence of several types of shocks to the labor market. However, authors should consider improving both their empirical strategy.