In the Automation Debate, Don't Forget the Job Multiplier Effect


In his 1950s satire Player Piano, author Kurt Vonnegut describes a dark dystopia where automation has led to a world of meager consumption and desperate idleness. The vision of workers displaced by machines predates this though, and is perhaps most associated with the 19th century Luddite movement where workers sabotaged machinery for fear of losing jobs. In economic thought, the prospect of labor-replacing technology has a still much longer history.

The opinion of most economists has been that “Luddite fears” are misplaced. New technology is economically synonymous with increased efficiency, new and cheaper products, expanded national income and demand for goods, and ultimately an expanded demand for labor and higher wages. With recent technology, however, most notably robots and artificial intelligence, a growing number of economists are sounding alarm.

Will the future be one where capital in the form of robots and other machines make remunerative work increasingly obsolete? Will it be one where smart policy aims at the maintenance and fostering of labor-intensive processes, while shunning automation and capital intensity? These two questions increasingly dominate the economic debate. In this brief exploratory paper we highlight an important element in that debate.

Don’t Forget Multiplier Effects

The jobs supported by a given industry extend beyond those specifically employed in that industry. A more or less wide variety of produced inputs are needed, and jobs are created as well in the industries supplying these inputs. And the suppliers themselves need inputs, as do their suppliers, and so on, creating an often long and complex chain of input supply and job creation. Importantly, some industries have deeper supply chains than others, and a deep supply chain means higher off-site job effects.

Turning to economic models, the off-site job effects of a given industry are captured by the employment multiplier of an input-output model. Employment multipliers measure total jobs divided by on-site jobs—a multiplier of 3 means for every job on-site two more are created off-site through supply chain multiplier effects. Now it may be that industry A offers fewer jobs on-site than industry B yet offers more jobs in total when multiplier effects are included. In framing a jobs policy, failure to include multiplier effects could lead to the erroneous choice of B over A.

Off-site job creation extends beyond the chain of industrial inputs. An industry with a given number of workers and high wages will create more jobs through the effects of personal consumption spending than one paying low wages. Likewise an industry with much capital (buildings, machines, etc.) creates more property income than one with little capital, and this means greater personal consumption spending. More importantly, though, in the case of high-capital industries, considerable annual expenditures will be required to maintain, repair, and periodically replace the capital stock, and this creates jobs in the broad collection of industries that provide these essential capital goods and services.

Multipliers and Automation

So we ask the question: Do industries characterized by automation have greater off-site employment effects, i.e., multiplier effects, than other industries? If we had some definitive index of automation by industry we could simply compare industry I-O employment multipliers to this index and determine the answer. Unfortunately, to our knowledge, no such index exists. Is there a suitable surrogate?

To begin with note that any tool or machine, even simple and inexpensive ones, contribute to the abridgement of labor and thus to some degree of automation. At the same time, a thoroughly automated factory, with its robots and advanced technology, is a very expensive factory, and thus a factory with a high ratio of capital stock to labor. So as a tentative exploration of the relation between automation and employment multipliers, let us compare industry capital-labor ratios and multipliers (see italicized footnote for how we estimate the value of an industry’s stock of capital).

The multiplier effect we wish to examine includes particularly the action of personal income and induced investment spending. These are derived from our input-output model “closed” with respect to household spending and investment. Such models are strictly intended to portray the economic base of regional economies. When constructed at the national level, they tend to overstate multipliers, the result of assuming, in effect, that all economic activity is explained by national exports. However, absolute size notwithstanding, industry-by-industry comparisons provide an entirely reliable indication of relative multiplier magnitudes.

Drawing an overall comparison of multipliers and capital-labor ratios, across all of the approximately 1,000 North American Industrial Classification System (NAICS) industries, provides a less than perfect yet solidly positive correlation. Figure 1 shows indicative findings.

Leading the collection is petroleum refineries (NAICS 324110), with nearly $21 million in plant and machinery per employee and an employment multiplier of nearly 100. Think of the great investment in building a refinery, all the moving parts, the ongoing investment needed to maintain it, all the many inputs per worker and an employment multiplier near 100 is perhaps not surprising, especially as it is derived from a national-level model. Other industries with large capital investments (per worker) and employment multipliers include light truck and utility vehicle manufacturing (336112), petrochemical manufacturing (325110), and tobacco manufacturing (312230).

At the other extreme, low multiplier-low capital investment, we find fine arts schools (611610). With a modest building, capital, and equipment investment of less than $9,000 per employee, art schools appear with an employment multiplier of barely 1.5. Other sectors at the low end, mainly service sectors, include child day care services (624410), mobile food services (722330), and nail salons (812113). It is easy to see how modest wages and minimal capital investment results in shallow multiplier effects.

Implications for Policy

While more research on the particulars of consumer spending and investment effects is warranted, and a more explicit measure of automation than simply the ratio of capital-to-labor would be helpful, our findings are nonetheless indicative of a need to consider multiplier effects in framing policy. As automation proceeds, employment multipliers will, of mathematical necessity, increase: A theoretical factory, fully automated, with no jobs at all, would have an employment multiplier approaching infinity. So in judging which industries fit better with a jobs and industry policy, consider where the inputs come from, including especially the investment goods and services needed to maintain plant and equipment. A factory full of domestically made and serviced robots may employ more workers than it appears.

* Measuring the value of an industry’s capital stock:

Among the annual data included in Emsi’s I-O model are figures on the flow of property income by industry. Property income is the return on invested capital. Assuming a uniform rate of return across all industries (we assume 4%, a rough but for our purposes acceptable assumption), the total value of capital in a given industry is computed as the industry’s flow of property income divided by the assumed uniform rate of return. Finally, dividing the value of an industry’s capital stock by the number of its employees provides that industry’s capital-labor ratio: the economist’s standard measure of industry capital-intensity.

Dr. Robison is EMSI’s co-founder and senior economist with 30 years of international and domestic experience. He is recognized for theoretical work blending regional input-output and spatial trade theory and for development of community-level input-output modeling. Dr. Robison specializes in economic impact analysis, regional data development, and custom crafted community and broader area input-output models. Contact Josh Wright with questions about this analysis.

Photo credit: Flickr/Steve Jurvetson