Why We Won't Even Know If A Robot Takes Your Job

If it's not in the data, did it really happen?JL

Robert Seamans comments in Forbes:

If a robot takes your job, we won’t know. Alas, the U.S. government does not (yet) collect information about businesses’ use of robots. Robots are everywhere, except, as it turns out, in the data.
Robots and automation have received lots of attention over the past year, with much of the interest ranging from alarmist to curious. Elon Musk has said that robots will take your job. The Economist detailed the long history of human anxiety over automation-induced job loss.  And, at the recently concluded 50^th annual Consumer Electronics Show, companies rolled out robots to monitor your child and brew your coffee and tea. Robots are everywhere, except, as it turns out, in the data.
To be clear, I don’t mean there’s no data about robots. There is some, and existing data sources show large increases in the use of robots. According to the 2016 Economic Report of the President, worldwide robot shipments nearly doubled between 2010 and 2014, and there has also been an increase in robot-related patents. But the data used in the report, and in two recent academic studies that I discuss below, are all from the same data source. And we need additional, more disaggregated, data to better understand the disparate effects of robots on the economy, and to design appropriate policy responses. Reasons for Cautious Optimism
Using national level data on worldwide robot shipments across 17 countries, George Graetz and Guy Michaels show that robots may have been responsible for about a tenth of the increases in those countries’ gross domestic product between 1993 and 2007, and may have increased labor productivity growth by over 15%. This might sound like it’s a small number, but it’s not. According to the authors, this number is comparable to the impact of steam engines on British labor productivity growth in the 19^th Century.
Does this growth come at the expense of labor? Graetz and Michaels find some evidence that wages go up on average as robot use increases. But they also find some evidence that hours worked drops for low-skilled and middle-skilled workers. A paper by Daron Acemoglu and Pascual Restrepo focuses on the effect of robots on the U.S. labor market and estimates that each additional robot reduces employment by seven workers and that one new robot per thousand workers reduces wages by 1.2 to 1.6%.
Both of these studies are carefully done by well-respected economists, and provide important findings that will be useful to policymakers and future researchers. But it is important to provide the caveat that both of these studies rely on the same data set provided by the International Federation of Robotics (IFR), which has aggregated robotics shipments to a relatively “macro” level (that is, for country-industry pairs). The IFR data appear to be the best (and maybe only) data currently available. While the data are useful for some purposes, its aggregated nature obscures underlying differences across firms and regions. This makes is hard to uncover the economic conditions under which robots might be substitutes for or complements to labor.
How Can Robots Complement Labor?
Consider the case of Amazon, which has reportedly been rapidly increasing the number of robots in its fulfillment centers. Amazon had 15,000 robots in 2014, 30,000 in 2015, and 45,000 in 2016. But a review of Amazon’s annual reports shows that Amazon has grown its workforce over the same time frame: Amazon employed 117,000 workers in 2013, 154,000 in 2014 and 230,000 in 2015 (2016 numbers are not yet available). This is just one example, but it suggests that robots may be complementary to labor in some cases. By investing in robots, Amazon is able to grow quicker and hire more workers. Anecdotally, Amazon is able to accomplish this by re-organizing its fulfillment centers so that the robots bring shelving units to centrally located (human) employees who use their judgment and manual dexterity to rapidly sort and fill orders—tasks at which humans continue to have a sizable comparative advantage over robots.
So Will A Robot Take Your Job?
If a robot takes your job, we won’t know. Alas, the U.S. government does not (yet) collect information about businesses’ use of robots. The Census Bureau routinely surveys businesses about a range of other issues including total sales, sales by e-commerce, expenses, wage bill, and others. Economists have used data from these surveys to conduct many useful studies on business dynamics, industrial organization, firm strategy, entrepreneurship, and innovation. And the government uses findings from these systematic surveys to inform itself, and the public, about how well its policies are performing (think of the importance placed on unemployment numbers, GDP, and trade deficits).
How might data on robots be useful? For starters, researchers could use the data to address a host of questions regarding the impact of robots (and automation more generally) on the economy, including (1) how robots affect firm-level productivity; (2) which types of firms are more likely to invest in robots; (3) the extent to which robots are complementing or substituting for labor; and (4) how market structure affects a firm’s incentives to invest in robots, among other questions.
Good Policy Requires Good Data
Understanding the conditions under which robots serve as a complement to or substitute for labor has important policy implications—and can help avoid costly policy mistakes. The European Union has reportedly considered taxing firms that use robots. This is a risky policy, premised on the assumption that robots substitute for labor. Robot taxes would disincentivize firms from investing in robots, which would lower economic growth, and, to the extent that robots complement labor in some cases, would lead to less hiring and lower wage growth.
While a tax on robots is unlikely to pass in the EU (let alone in the United States), government data on robots could be used to undertake targeted policy responses. For example, the data could be used to identify areas that need more resources for job retraining or other assistance. The data could also be used to “stress test” existing safety nets. For example, unemployment insurance varies by state, and the unemployment insurance reserve in some states has been slow to recover from the recent financial crisis. Data on regional adoption of robots could be used to simulate the extent to which future robot adoption will increase unemployment, and whether unemployment insurance reserves are adequately funded.
It is important for the government to conduct more systematic data collection on the use of robots in our economy. At a minimum, government data can be used to replicate the existing studies that rely on the IFR data. But the disaggregated firm-level data can also help us understand the conditions under which robots complement or substitute for labor. In so doing, the data can help policymakers design and assess the appropriate policy responses. And of course, the data will help us predict where and when any robot apocalypse might begin.