New Perspectives on the Decline of U.S. Manufacturing Employment

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. New Perspectives on the Decline of U.S. Manufacturing Employment Teresa C. Fort, Justin R. Pierce, and Peter K. Schott 2018-023 Please cite this paper as: Fort, Teresa C., Justin R. Pierce, and Peter K. Schott (2018). “New Perspectives on the Decline of U.S. Manufacturing Employment,” Finance and Economics Discussion Series 2018-023. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2018.023. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

New Perspectives on the Decline of U.S. Manufacturing Employment∗ Teresa C. Fort† Tuck School at Dartmouth & NBER Justin R. Pierce‡ Board of Governors of the Federal Reserve System Peter K. Schott§ Yale School of Management & NBER First Draft: October 2017 This Draft: March 2018 Abstract We use relatively unexplored dimensions of US microdata to examine how US manufacturing employment has evolved across industries, (cid:28)rms, establishments, and regions from 1977 to 2012. We show that these data provide support for both trade- and technology-based explanations of the overall decline of employment over this period, while also highlighting the di(cid:30)culties of estimating an overall contribution for each mechanism. Toward that end, we discuss how further analysis of these trends might yield sharper insights. ∗Any opinions and conclusions expressed herein are those of the authors and do not necessarily representtheviewsoftheU.S.CensusBureau,theBoardofGovernorsoritsresearchsta(cid:27). Allresults have been reviewed to ensure that no con(cid:28)dential information is disclosed. Part of this research was conducted while Teresa Fort was a Peter B. Kenen Fellow in the International Economics Section at Princeton University. She thanks the IES for (cid:28)nancial support. We thank the editors of the JEP for helpful comments and Jim Davis for his exceptional aid in the disclosure review process. †100 Tuck Hall, Hanover, NH 03755, tel: (603) 646-8963, email: teresa.fort@tuck.dartmouth.edu ‡20th&CStreetsNW,Washington,DC20551,tel: (202)452-2980,email:justin.r.pierce@frb.gov. §165 Whitney Avenue, New Haven, CT 06511, tel: (203) 436-4260, email: peter.schott@yale.edu. 1

1 Introduction US manufacturing since World War II exhibits three notable trends, illustrated in the two panels of Figure 1. First, the manufacturing employment has diverged from nonmanufacturing employment, as shown on di(cid:27)erent axes in Figure 1A. While both series move upward until the late 1970s, manufacturing employment then begins to decline, even as other non-farm employment continues a steady rise. As a result, there is a continual decline in manufacturing employment’s share of total US non-farm employment, from 32 percent in 1948 to 8 percent in 2017. Second, while US manufacturing employment fell just 12 percent over the 21 years between the post-war peak in 1979 and 2000, it then dropped by more than twice as much (cid:21) 25 percent (cid:21) from 2000 to 2012. Third, despite the relative (cid:29)atness and subsequent sharp decline in US manufacturing employment, the right-hand panel of Figure 1 shows a steady rise in manufacturing real value added at more or less the same rate as non-manufacturing GDP over the same period, at least between the late 1970s and the Great Recession. The combination of relatively steady and then declining employment, and rising output, indicates that, over the long term, labor productivity has risen faster in the manufacturing sector than in the broader economy. For a variety of reasons, including the perception that workers in manufacturing receive comparatively high wages conditional on education (Langdon and Lehrman (2012); Ebenstein et al. (2014)), these trends have stirred intense discussion among both policy makers and academics. This debate can broadly be summarized as a dispute between views that emphasize the relative importance of trade versus technology. The trade-based explanation contends that import competition has reduced US manufacturing employment by inducing labor-intensive, low-labor-productivity industries to move abroad. The technology view argues that the decline in manufacturing employment stems from innovations in production techniques, such as automation, that have dramatically increased output per worker. If, as implied by Figure 1, consumers spend a constant share of their expenditure on manufactured goods, then an increase in labor productivity means fewer workers are needed to meet demand for those goods. Discussions about the decline in US manufacturing employment often culminate in a request to decompose the decrease into the part that is due to trade and the part that is due to technology. Our view is that providing a de(cid:28)nitive accounting of the amount of employment change attributable to either factor is extraordinarily di(cid:30)cult for two reasons. First, identifying the numerous changes in tari(cid:27) and non-tari(cid:27) barriers that have occurred over the last few decades, let alone the wide range of technologies 1 that have been adopted, is a daunting task. More importantly, even if one could 1For example, even while ad valorem tari(cid:27) rates have trended downward over time, and regional trade agreements have proliferated, implementation and repeal of contingent protection measures like antidumping and countervailing duties remains frequent and widespread (Bown (2016)). These temporary barriers have been linked to relative declines in physical productivity and increased prices among protected manufacturing plants (Pierce (2011)). Identi(cid:28)cation of the numerous technological innovations introduced during this period, including computerization, electronic communication, computer-aided design and manufacturing, just-in-time inventory management, and enterprise resource planning, is similarly di(cid:30)cult. 2

Figure 1: U.S. Manufacturing Employment and Real Value Added identify all of these changes, it is di(cid:30)cult to see how their intertwined impacts on employment could be teased apart. As an example, consider an anecdote from a recent Wall Street Journal article (reported in Michaels (2017)), which takes place around the time of an important US trade liberalization with China discussed below: (cid:16)When Drew Greenblatt bought Marlin Steel Wire Products LLC, a small Baltimore maker of wire baskets for bagel shops, he knew nothing about robotics. That was 1998, and workers made products manually using 1950s equipment. ... Pushed near insolvency by Chinese competition in 2001, he started investing in automation. Since then, Marlin has spent $5.5 million on modern equipment. Its revenue, sta(cid:27) and wages have surged and it now exports to China and Mexico.(cid:17) Are changes in Marlin’s employment and output driven by the availability of robots or increased Chinese competition? What about employment and output at other producers of steel wire products, who face increased competition from both China and from Marlin? These questions are even more di(cid:30)cult to answer if robots’ availability itself is in(cid:29)uenced by trade liberalization, for example, by robot manufacturers’ ability to source intermediate inputs from China. In this paper, we provide a brief overview of recent e(cid:27)orts to answer such questions before turning to relatively unexplored dimensions of US microdata for further input. These data allow us to examine changes in US manufacturing employment across industries, (cid:28)rms, and regions, and thereby o(cid:27)er four new perspectives on how US manufacturing has evolved over the last several decades. We (cid:28)nd that while employment changes along these dimensions provide support for both trade- and technology-based explanations, they also highlight the di(cid:30)culties of cleanly separating one force from the other. Toward that end, we discuss how further analysis of the data we use might yield sharper insights. Our (cid:28)rst perspective examines how the overall growth of US manufacturing employment and value added varies by sector. We (cid:28)nd that some sectors (cid:21) such as transportation equipment (cid:21) exhibit increases in output even as employment is falling, a potentially clear indication of technology adoption. On the other hand, it is not hard to (cid:28)nd examples of sectors, such as apparel, characterized by simultaneous increases 3

in import penetration and reductions in both employment and output. Furthermore, the set of sectors experiencing declines in both employment and output increases after 2000. Our second perspective analyzes employment loss along (cid:28)rm and establishment margins of adjustment. One of our more striking (cid:28)ndings (cid:21) given conventional expectations about how creative destruction due to trade and technology likely manifest (cid:21) is that net (cid:28)rm death accounts for just 25 percent of the overall decline in US manufacturing employment between 1977 and 2012. On the other hand, we (cid:28)nd a large role for net plant exit within incumbent (cid:28)rms, perhaps because adopting new technologies or adapting to import competition entails high (cid:28)xed costs that continuing (cid:28)rms are better able to absorb, and which are easier to implement by opening new plants. Our third perspective breaks down the aggregate change in US manufacturing employment between 1977 and 2012 along regional margins of adjustment. We (cid:28)nd a steady reallocation of manufacturing employment away from the north and east towards the south and west until 2000, when employment starts falling in all regions. The earlier transition may re(cid:29)ect (cid:16)domestic o(cid:27)shoring,(cid:17) that is, a movement from higher- to lower-wage US regions in an era before foreign o(cid:27)shoring was cost-e(cid:27)ective. Our (cid:28)nal perspective takes a wider view of manufacturing (cid:28)rms by examining their non-manufacturing activities. We (cid:28)nd that manufacturing (cid:28)rms’ non-manufacturing employmentincreasesuntil2000(cid:21)primarilyviatheadditionofnewnon-manufacturing establishments (cid:21) before leveling o(cid:27). About a third of this growth is in professional services,atrendthatmayrepresentanevolutionofUSmanufacturing(cid:28)rmsinto(cid:16)neurofacturers(cid:17) that increasingly provide intellectual services rather than physical goods (Leamer (2009)). Prominent examples include Pitney Bowes, which has abandoned the production of postage meters to o(cid:27)er logistics services, IBM, which increasingly o(cid:27)ers data solutions rather than mainframes, and Apple, which designs the iPhone in the US but uses o(cid:27)shore contractors for assembly . 2 Some of the Evidence Thus Far The last three decades have witnessed dramatic changes in both trade and technology. WeprovideasenseofsomeofthesechangesinFigure2,whichplotsU.S.manufacturing (cid:28)rms’ use of two speci(cid:28)c forms of technology (cid:21) computers and electronic networks (cid:21) at (cid:28)ve-year intervals from 1977 to 2012. As indicated in the (cid:28)gure, the share of (cid:28)rms purchasingcomputersinthenotedyearsincreasesthroughthe1990s, withalargejump in the early 2000s. Data tracking use of electronic networks to control or coordinate shipments are available starting in 2002, and exhibit an analogous increase in adoption 2 during the 2000s. Figure 2 also reports several dimensions of trade activity. First, starting in 1992, we report the share of manufacturing (cid:28)rms that imports from any country as well as 2As discussed in Fort (2017), plants’ use of electronic networks to control or coordinate shipments involves not just using the internet or other networks, but also integrating electronic communication in the production process. Computer purchase data are not available in 1997, so we supplement the Census of Manufactures data with information from the 2000 Annual Survey of Manufactures. 4

Figure 2: Technology and Trade Trends the share that imports from China, by census year. Here, as with our indicators of technologyuse, weseeincreasesintheearly2000s. Second, wedisplayannualmeasures of import penetration and import penetration from China. These series are de(cid:28)ned as manufacturing imports (or manufacturing imports from China) divided by the sum of domestic manufacturing shipments plus manufacturing imports less manufacturing exports, allinrealterms. Importpenetrationfromallsourcesisrisingovertime, witha pronouncedupwardshiftafterthe1981recessionandrelativelyrapidgrowthduringthe 1990s. Chinese import penetration rises relatively slowly in the 1990s before picking 3 up in the 2000s. A key message of Figure 2 is that both technology adoption and importing, including by USproducers, generallyrise over thesampleperiod, sometimes simultaneously. Researchers have adopted several approaches to identify e(cid:27)ects of trade (cid:16)shocks(cid:17) on employment. Perhaps the narrowest de(cid:28)nition of a trade shock is a change in trade policy, such as a reduction in import tari(cid:27)s, that leads to increased trade (cid:29)ows. Broader de(cid:28)nitions include the impact of other factors, such as transport or communication 3Appendix Figure A.1 displays the levels of overall U.S. imports, exports, manufacturing value added, and manufacturing absorption (value added plus imports minus exports). 5

costs, or foreign capital accumulation, that alter comparative advantage and the terms of trade. A complication associated with identifying such shocks is that they can be induced by technology shocks. For example, a trading partner’s productivity growth may be driven by its adoption of new technologies or production techniques. Investigating the US steel industry, Oster (1982) shows that large US producers were relatively slow in adopting new blast-furnace technologies during the 1970s, a factor which may have contributed to the rise in steel imports from their faster-adopting Japanese rivals. Agrowingempiricalliteratureusesspeci(cid:28)ctradeliberalizationstoexaminewhether US manufacturing employment or wages drop disproportionately in industries with greater exposure to changes in policy. Hakobyan and McLaren (2016), for example, use industry variation in US tari(cid:27) reductions due to the North American Free Trade Agreement (NAFTA) to document a negative wage e(cid:27)ect of NAFTA on less-educated workers between 1990 and 2000. Focusing on the next decade, Pierce and Schott (2016a), show that the post-2000 decline in US manufacturing employment is relatively larger for industries exposed to the granting of Permanent Normal Trade Relations to China in October, 2000. This non-traditional trade liberalization eliminated the possibility of sudden, substantial spikes in US tari(cid:27)s on many Chinese imports, thereby removing a signi(cid:28)cant deterrent to greater integration of the two economies. Research into the broader set of shocks that might alter US terms of trade makes use of changes in imports to identify reallocation. These papers devote considerable e(cid:27)ort to excluding variation in imports driven by non-trade factors, such as secular declines in demand or common technology shocks. Bernard et al. (2006), for example, (cid:28)nd that US manufacturing plant survival and employment between 1977 and 1997 are negatively associated with increasing import penetration from low-wage countries. To identify a causal e(cid:27)ect of trade, they use changes in US import tari(cid:27)s and ad valorem trade costs over their sample period as instruments for import penetration. Autor et al. (2014) and Acemoglu et al. (2016), show that workers in industries with higher growth in Chinese imports experience increased unemployment between 1992 and 2007. In these papers, Chinese import growth in other countries is used as an instrument for its growth in the United States. The identifying assumption is that Chinese exports to these other countries are driven by productivity growth in China, and not by changes in demand or technology outside of China that might also a(cid:27)ect US manufacturing employment. A related body of work exploits spatial variation in the distribution of manufacturing industries across the United States. Autor et al. (2013) demonstrate that regions with higher initial shares of employment in industries with greater exposure to imports from China experience relatively larger declines in employment and labor force participation. Regions with higher initial shares of employment in exposed industries also exhibit relative declines in the provision of public goods (Feler and Senses (2017)) and marriage rates (Autor et al. (2017)), as well as relative increases in household debt (Barrot et al. (2017)) and crime (Che et al. (2017)). These consequences carry over to health: Pierce and Schott (2016b) show that regions more exposed to US trade liberalization with China exhibit relative increases in (cid:16)deaths of despair,(cid:17) including drug overdoses. This connection is reminiscent of the spike in mortality rates among high-tenure workers laid o(cid:27) from the steel industry in Pennsylvania during the 1980s 6

(Sullivan and Wachter (2009)). Studies like those noted above are often conducted using a di(cid:27)erence-in-di(cid:27)erences framework, which does not account for potential general equilibrium e(cid:27)ects and thus complicates calculation of a trade shock’s e(cid:27)ect on the overall level of manufacturing employment (Muendler (2017)). Quantitative models, often drawing on empirical evidence from such studies, do o(cid:27)er such estimates as well as quanti(cid:28)cations of the impact of trade on social welfare. Caliendo and Parro (2015), for example, argue that increased trade with China explains approximately one-quarter of the decline in US manufacturing employment between 2000 to 2007, and that the growth of trade with China over this period increased US welfare, though, like Galle et al. (2017), they (cid:28)nd that gains vary across regions. Handley and Limªo (2017) (cid:28)nd that trade liberalization with China in the 2000s bene(cid:28)ts consumers via increased imported product variety. While changes in trade policy and increases in imports, particularly during the 2000s, have received considerable attention, other researchers interpret the long-run decline in the manufacturing employment share implicit in Figure 1 as driven by technology. Edwards and Lawrence (2013), for example, argue that the long, post-war decline in the share of US employment in manufacturing occurs (cid:16)irrespective of the changing developments in international trade (cid:29)ows, the size of the trade de(cid:28)cit, and other factors.(cid:17) A number of papers assess the role of particular technologies on manufacturing employment. Collard-Wexler and De Loecker (2015) describe the importance of the introduction of mini-mills in the US steel industry for subsequent gains in output and declines in employment. Acemoglu and Restrepo (2017) (cid:28)nd that US regions with an industrial mix that pre-disposes them to adopting more industrial robots have also experienced relatively larger employment declines, at a rate of approximately (cid:28)ve workers per robot. Similarly, Graetz and Michaels (2017) use cross-country and industry data to show that robot adoption relates to decreased work hours by middle- and especially low-skill workers. Another strand of research aims to decompose the respective roles of trade and technology on employment and wages. Goos et al. (2014) and Autor et al. (2015) argue that technological change has decreased the relative demand for routine tasks; the latter compares the results for computerization of routine tasks to increased Chinese import penetration in the United States and concludes that Chinese imports play a larger role in the decline of US manufacturing employment, especially after 2000. While this research uses careful measures to identify technology and trade, it remains susceptible to the possibility, highlighted in the anecdote presented in the introduction, as well as theoretical work in this area (e.g., Acemoglu (2002)), that a new technology’s invention or adoption may itself be in response to a trade shock. Bernard et al. (2006), Khandelwal (2010) and Bernard et al. (2011) show that US (cid:28)rms respond to import competition in part by upgrading their product mix. Bloom et al. (2016) (cid:28)nd evidence of technology upgrading within and across European (cid:28)rms that were more exposed to Chinese imports. In the US context, Autor et al. (2016) also (cid:28)nd that Chinese import penetration a(cid:27)ects manufactures’ innovative activities, though they document a nega- 4 tive relationship. Finally, interconnectedness is also found in the other direction. Fort 4In related research in labor economics, Clemens et al. (2017) show that imposing restrictions on 7

(2017) and Steinwender (2018) show that innovations in communications technologies facilitate trade. As a whole, this research highlights the di(cid:30)culties associated with clean identi(cid:28)cation of one force over another. 3 Employment and Value Added Reallocation Across Industries Examination of employment and output changes by industry provides useful context for the trends displayed in Figure 1, while also o(cid:27)ering evidence in support of both trade- and technology-based explanations for the overall decline in US manufacturing employment since the late 1970s. Figure 3 displays log changes in real value added, employment and import penetration for the twenty-one three-digit NAICS sectors that constitute manufacturing. Given the sharp drop in manufacturing employment after 2000 displayed in Figure 1, we provide separate decompositions for years before (left panel) and after (right panel) that year, ending the latter period before the Great Recession to avoid its impact. In each period, industries are sorted by their log change in real value added, from low to high. Figure 3 has three notable features with respect to identifying the in(cid:29)uence of trade and technology. The (cid:28)rst is the presence of Leather Products (316) and Apparel (315), whichexhibitdeclinesinbothemploymentandvalueaddedinbothtimeperiods. These sectors primarily encompass the production of labor-intensive goods such as clothing and footwear, commonly thought to be inconsistent with US comparative advantage. Apparel, in particular, has been subject to substantial tari(cid:27) and quota reductions in the United States during the period we study (Khandelwal et al. (2013)), and these liberalizations are re(cid:29)ected in the fact that it displays the largest increase in import 5 penetration across sectors between 1977 and 2000. A second suggestive feature of Figure 3 is the increase in the number of sectors exhibiting simultaneous declines in real value added and employment in the right panel. Indeed, as illustrated in Appendix Figure A.2, 52 percent of the 473 six-digit manufacturing industries that comprise manufacturing register such reductions between 2000 and 2007, versus 23 percent during the earlier time period. To the extent that this trend captures the exit of labor-intensive, low-labor-productivity (cid:28)rms within sectors whose products most overlap with Chinese manufacturers, this trend is consistent with the increase in Chinese import competition displayed in Figure 2 a(cid:27)ecting US employment, and the research into trade liberalization with China discussed above. On the other hand, as Figure 2 also illustrates, the 2000s is a period when (cid:28)rms’ use of computers and electronic networks increases. An intriguing possibility worthy of further attention, motivated by the anecdote in the introduction, is whether technology low-skill immigration induced adoption of more capital-intensive production techniques and shifts in product mix in the agricultural sector. 5Reallocation may operate through occupations as well as industries, presenting another challenge toidentifyingtheimpactsoftradeandtechnology. Thatis,thecharacteristicsthatmakeoccupations susceptible to o(cid:27)shoring, such as routineness, also render them susceptible to automation (Ebenstein et al. (2014); Oldenski (2014)). 8

Figure 3: Changes in Employment and Output by Industry 9

adoption during this period was hastened by trade liberalization with China. The third noteworthy feature of Figure 3 with respect to trade and technology is the presence of sectors such as Chemicals (325), Transportation Equipment (336) and Miscellaneous Products (339; second panels only), in which value-added rises even as employment falls. These divergent outcomes, and the large growth in labor productivity they imply, suggest labor-saving technological change. In automobiles, for example, the replacement of workers with robots is widespread. On the other hand, to the extent that import competition induces selection away from low-labor-productivity industries within sectors, trade might also be playing a role (Schott (2003, 2004)). Indeed, the industries within Miscellaneous Products with the largest loss and gain in employment between 1977 and 2000 are dolls and surgical instruments, respectively. A particularly interesting sector exhibiting rising output along with falling employment in recent years is Computers and Electronic Products (334). As pointed out in Houseman et al. (2011), and suggested by its presence at the bottom of both panels of Figure 3, this sector accounts for the vast majority of real value added growth in 6 manufacturing over our sample period. The two most in(cid:29)uential industries in terms of aggregate real value-added growth within this sector are Semiconductors (334413) and Electronic Computer Manufacturing (334111). The latter has experienced signi(cid:28)cant growth in Chinese import penetration and is particularly well-known for its o(cid:27)shoring and outsourcing. Physical production of hard disk drives, like many other consumer electronic devices, has moved almost completely o(cid:27)shore during our sample period, even as their design centers remain in the United States (Igami (2018)). The iPhone, in particular, is well known for being (cid:16)designed in California(cid:17) and assembled (cid:21) using physical inputs from many countries, including the United States (cid:21) in China (Folbre (2013)). The growing prevalence of such supply chains highlights a subtle but potentially important distinction between trade as import competition and trade as a technology. Although the bulk of US imports from China represent (cid:28)nished goods imported by US wholesalers and retailers (Bernard et al. (2010)), Figure 2 reveals that a growing share of manufacturing (cid:28)rms import goods directly. These direct imports may have di(cid:27)erent consequences than import penetration: empirical analysis of US manufacturing (cid:28)rms by Antr(cid:224)s et al. (2017) (cid:28)nds that while a (cid:28)rm’s presence in an industry subject to increasing levels of Chinese import penetration is associated with declining (cid:28)rm-level employment between 1997 and 2007, increases in the value of its direct imports from China are associated with either growing or no change in employment. In their quantitative model, the authors’ provide a rationale for this di(cid:27)erence, showing how greater access to foreign sourcing opportunities can allow importers to lower prices and raise output, even as non-importing (cid:28)rms shrink. Bernard et al. (2017) also (cid:28)nd that Danish (cid:28)rms exposed to increased import competition from China were more likely to o(cid:27)shore activities to Eastern Europe, which led to decreased domestic employment but not domestic output. Exploring the role of global value chains in the divergence between real output and employment is an important area for future research. 6Housemanetal.(2011)alsonotethatgrowthinmanufacturingrealvalueaddedmaybeoverstated due to mismeasurement of prices for imported inputs. 10

4 Reallocation of Manufacturing Employment Across and Within Firms In this section we dissect the overall shift in US manufacturing employment between 1977 and 2012 along (cid:28)rm and establishment margins of adjustment. We perform this decomposition using data from the Longitudinal Business Database (LBD) of the US Census Bureau, which links all private, non-farm employer establishments and (cid:28)rms overtimestartingin1977(JarminandMiranda(2002)). Eachestablishmentisassigned 7 a single industry code in each year based on its predominant activity. The data make a useful distinction between an (cid:16)establishment(cid:17) and a (cid:16)(cid:28)rm.(cid:17) An establishment denotes a single physical location where business transactions take place and for which payroll and employment records are kept, such as a manufacturing plant. In our analysis, as in o(cid:30)cial statistics, employees are grouped into industries based on the classi(cid:28)cation of the establishment in which they work. As a result, all employees in a manufacturing plant are classi(cid:28)ed as manufacturing employees, regardless of their occupation. A (cid:16)(cid:28)rm(cid:17) is an organizational structure that can include one or more establishments, and therefore can span multiple industries. To capture all manufacturing employment in the LBD, our decomposition includes all (cid:28)rms observed to have at least one manufacturing establishment at any point during the 1977 to 2012 sample period. The employment totals reported in this section are restricted solely to the manufacturing establishments at these (cid:28)rms; employment at their non-manufacturing establishments is analyzed later in the paper. We examine three mutually exclusive (cid:28)rm margins of adjustment: changes in employment within the continuing establishments of continuing (cid:28)rms (also referred to as the (cid:16)intensive(cid:17) margin of continuing (cid:28)rm-plants), changes due to the birth and death of establishments within continuing (cid:28)rms, and changes due to the birth and death of 8 entire (cid:28)rms. Figure 4 illustrates the results. The solid line displays overall US manufacturing employment, showing the same pattern since 1977 as in Figure 1. The dashed lines trace out the cumulative employment in year t along the margins of adjustment, in each case relative to the (cid:28)rms and plants present in base year 1977. So, for example, the (cid:28)nal value for the intensive margin indicates that (cid:28)rm-plants present in both 1977 and 2012 experience a decline in employment of approximately -0.8 million. Together, all three margins account for the -6.7 million overall decline in manufacturing 7We identify manufacturing plants based on an assignment of time-consistent NAICS codes developed by Fort and Klimek (2016) that ensure that the transition from SIC to NAICS does not result in spurious changes in the number of manufacturing workers based on changes in the set of activities considered (cid:16)manufacturing.(cid:17) While the resulting manufacturing employment totals from the LBD do not perfectly match the totals from the Bureau of Labor Statistics displayed in Figure 1, they are highly correlated over time. Our analysis drops records that are outside the scope of the County Business Patterns data, such as agriculture, and observations that are clearly erroneous, for example because of implausible payroll and employment numbers. 8We follow Haltiwanger et al. (2013) and de(cid:28)ne a (cid:28)rm death as occurring when all establishments of a (cid:28)rm exit from the LBD. Analogously, (cid:28)rm birth occurs when all a (cid:28)rm’s establishments are new to the LBD. While this approach avoids spurious (cid:28)rm birth and death due to merger and acquisition activity, future research into the extent to which these types of ownership changes are important factors in understanding manufacturing might be useful. 11

Figure 4: US Manufacturing Employment by Net Margins of Adjustment employment registered by the solid line, from 17.8 to 11.1 million. We (cid:28)nd that most of the change in US manufacturing employment between 1977 and 2012 (cid:21) 75 percent (cid:21) takes place within (cid:28)rms that already existed in 1977 1977 (consider the two lines "within continuing (cid:28)rm-plants" and "net plant birth/death within (cid:28)rms"). Most striking is the contribution of net plant birth/death within these (cid:28)rms, which by itself accounts for 63 percent of the overall change. Conversely, the set of (cid:28)rm-plants in continuous operation over the sample period is responsible for relatively little (cid:21) 12 percent (cid:21) of the overall decline, with most of that occurring during the early 2000s. The manner by which (cid:28)rms add or shed workers o(cid:27)ers clues about their structure and transition costs, as well as the nature of the shocks they face. Consider three possibilities. If automating existing plants is relatively cost-e(cid:27)ective, employment declines may be concentrated along the (cid:16)intensive(cid:17) margin (cid:21) that is, within establishments of ongoing (cid:28)rms. If technology upgrades are more e(cid:30)ciently accomplished by shuttering outmoded plants in favor of new facilities, employment declines may occur via the net 9 death of establishments within continuing (cid:28)rms. If entrepreneurs at entering (cid:28)rms have an edge in creating or implementing new technologies, as argued by Christensen 9For example, Brynjolfsson and Hitt (1998) describe a medical manufacturer’s experience transitioning to computer-integrated manufacturing. The (cid:28)rm’s initial attempt to do so at an existing plantfailedtogenerateproductivitygainsbecausecurrentworkersdidnotunderstandhowtoexploit the new processes. When the (cid:28)rm then opened a new plant with young employees, it realized such signi(cid:28)cant gains that it painted the plant windows black to prevent competitors from replicating its new techniques. 12

(1997), then resulting reductions in manufacturing employment may be driven by (cid:28)rm death, as outdated incumbents are pushed from the market. Responses to increased pressures of international trade can, of course, operate along the same margins. Trade liberalization with low-wage countries might render a US (cid:28)rm’s most labor-intensive products unpro(cid:28)table. To the extent that (cid:28)rms are able to reallocate production away from these goods within existing facilities, globalization may manifest as declines in employment along the intensive margin. But if plants are wedded to particular products, employment loss may be driven by net plant death within continuing (cid:28)rms. If a broad set of (cid:28)rms’ products is subject to increased import competition or if existing (cid:28)rms are unable to reallocate production within or across plants, trade competition may lead to the death of entire (cid:28)rms. The fact that net (cid:28)rm death accounts for just 25 percent of the overall decline in US manufacturing employment between 1977 and 2012 is surprising given the magnitude of the drop in employment over this period, as well as common expectations of how creative destruction associated with trade and technology shocks likely operate. Indeed, in the right panel of Appendix Figure A.3, we (cid:28)nd that net (cid:28)rm birth accounts for the bulk of employment growth among non-manufacturing (cid:28)rms(cid:21)(cid:28)rms that never have a manufacturing establishment(cid:21)over the same period. On the other hand, most of the decline in employment along the net (cid:28)rm death margin occurs in the 2000s, which, as discussed above, may plausibly be related to import competition from China. As illustrated in Appendix Figure A.4, we (cid:28)nd a similar break with respect to the number of US manufacturing establishments: according to the Census Bureau’s publicly available Business Dynamics Statistics (BDS), this series peaks in 1996. Overall, the small role of net (cid:28)rm death in the aggregate decline of US manufacturing employment suggests that incumbents may have an advantage relative to entrants. The relatively sharp drop in employment associated with net plant death within continuing (cid:28)rms in the early 2000s, along with the contribution of net (cid:28)rm death during that period, may help rationalize the large distributional losses associated with increased import competition from China found in the literature. That is, to the extent that (cid:28)rm and plant closures were geographically concentrated, displaced workers may have found it more di(cid:30)cult to (cid:28)nd new employment in their local labor market. On the other hand, the more-or-less constant decline of employment associated with net plant death within continuing (cid:28)rms prior to 2000 is consistent with (cid:28)rms continually replacing outmoded plants with new ones in response to a steady introduction of new technologies. To what extent do workers displaced by dying establishments (cid:28)nd employment at new plants? Simple descriptive regressions provide support for both trade and technology in plant turnover. For example, we (cid:28)nd a negative correlation between the probability of a plant’s death within a (cid:28)rm and the plant’s purchases of computers. This correlation disappears after 2000, presumably due to the ubiquity of that technology, but during the 2000s we (cid:28)nd another such correlation with respect to use of electronic networks to 10 control or coordinate shipments. In other words, there is heterogeneity within (cid:28)rms 10AsdiscussedfurtherinAppendixSectionA.2,thesecorrelationsarefoundbyregressingindicator variables for plant death over years t to t+5 on indicator variables for the noted activities in year t, 13

in terms of the establishments that adopt various technologies, and plants that do adopt these technologies have lower exit probabilities. With respect to trade, similar regressions indicate that before 2000, plant death within (cid:28)rms was correlated with increased import penetration in that plant’s industry. After 2000, when (cid:28)rm death becomes a more important margin in the aggregate decline, these correlations are no longer present at the plant level, but (cid:28)rms facing increased import competition from 11 China are more likely to exit. One potential explanation for this result is that the (cid:28)rms that could re-orient themselves away from import-competing industries did so early on, either by shuttering plants or switching industries. For (cid:28)rms specializing in import-competing products, however, increased import penetration led to death. The relatively small, -12 percent change in employment among continuing (cid:28)rmplants masks substantial gross (cid:29)ows associated with continuing (cid:28)rm-plants’ expansion and contraction. We illustrate the magnitude of these gross (cid:29)ows in Figure 5, which decomposes the three net margins displayed in Figure 4 into their constituent gross job creation and destruction parts. In each case, job creation margins are displayed in lines above zero, while their corresponding job destruction margins are displayed in similarly patterned lines below zero. Here, to compare gross margins over time, and in contrast to Figure 4, we break the 1977 to 2012 sample period into three intervals that begin in base years 1977, 1990, and 2000. As a result, the gross margins for any year t in Figure 5 are computed with respect to their nearest prior base year. So, for example, the (cid:28)nal values for the gross continuing (cid:28)rm-plant margins indicate that (cid:28)rm-plants whose employment grew between 2000 and 2012 account for positive 3.6 million of the change in US manufacturing employment between 2000 and 2012, while continuing (cid:28)rm-plants whose employment fell accounted for negative 5.0 million. The dominance of the intensive margin in gross employment changes represents another potentially fruitful area of study. To what extent is the adoption of new technologies, exposure to trade, or either importing or exporting associated with plant contraction? Plant expansion? Large levels of job creation and destruction at continuing (cid:28)rms also suggest a potentially important role for technology and trade in worker reallocation. Are some workers more likely than others to shu(cid:31)e among continuing plants? In Appendix A.2, we show that (cid:28)rms’ technology and trade activities are correlated with subsequent changes in their employment and output, which is consistent with a role for both trade and technology in the reallocation of activities across continuing establishments. Another noteworthy feature of Figure 5 is the decline of all three gross job creation margins over time. These decreases are indicative of a drop in US business dynamism that has been documented across all sectors by Decker et al. (2016). One potential explanationforthisdeclineisareductionin(cid:28)rms’responsivenesstoproductivityshocks due to rising adjustment frictions (Decker et al. (2018)), such as regulatory constraints, or the use of o(cid:27)shore rather than domestic capacity to make adjustments. Another is a reductionincompetition,perhapsasaresultofincreasingentrybarriersassociatedwith along with (cid:28)rm (cid:28)xed e(cid:27)ects. 11Unfortunately, given that trading is observed at the (cid:28)rm level, we are unable to examine whether plants that import are more or less likely to survive within (cid:28)rms over either period. 14

Figure 5: US Manufacturing Employment by Gross Margins of Adjustment adopting technology or adapting to globalization. De Loecker and Eeckhout (2017) document a steady rise in market power as measured by markups among US (cid:28)rms since the 1980s, with a sharp tick upwards in the early 2000s. A potentially intriguing area for further exploration is whether costs associated with trade or technology contribute to entry barriers. Using simple regressions of (cid:28)rm attributes on indicators for adoption and industry (cid:28)xed e(cid:27)ects, we (cid:28)nd across census years (cid:21) and display in Appendix Figure A.5 (cid:21) that (cid:28)rms purchasing computers and using electronic networks are signi(cid:28)cantly 12 larger and have higher labor productivity than non-adopters. Inspired by Acemoglu and Restrepo (2017), we (cid:28)nd similar premia for (cid:28)rms that import industrial robots (Harmonized System product code 84.7950.0000) starting in 1997. These adoption premia are analogous to the size and productivity premia found for importers and exporters in the international trade literature (Bernard et al. (2007)). As such, they may re(cid:29)ect the fact that adoption of technology, like expansion into foreign markets, requires the payment of high (cid:28)xed costs that only the largest and most productive (cid:28)rms (cid:28)nd it optimal to incur. Trade may also play a role in the decline of gross manufacturing job creation by pushing the US economy away from goods production and towards services. Pierce and Schott (2012) and Asquith et al. (2017) show that during the 2000s, industries with 12These regressions are described in greater detail in Online Appendix Section A.1. 15

relatively greater exposure to trade liberalization with China exhibit both suppressed job creation as well as exaggerated job destruction. To what extent might the US transition from goods to services occur within (cid:28)rms? Relatedly, the decline in gross manufacturing job creation along the (cid:28)rm birth and plant birth within continuing (cid:28)rm margins may indicate that smaller, more capital-intensive (cid:28)rms and plants are entering at the expense of larger, more labor-intensive establishments and (cid:28)rms. In fact, as shown in Appendix Figure A.4, we (cid:28)nd that the average number of workers per US manufacturing establishment fell 29 percent between 1977 and 2012, while the number of manufacturing establishments only begins to decline in the 1990s. Are these smaller entrants producing di(cid:27)erent goods more in line with US comparative advantage, or are they producing the same goods with a di(cid:27)erent technology? A (cid:28)nal question related to the gross margins displayed in Figure 5 is the extent to which the decline in business dynamism in other sectors of the US economy might be related to the actions of manufacturing (cid:28)rms, or vice versa. Such relationships may occur through various channels, such as local labor markets or input-output linkages between manufacturing and non-manufacturing industries. Below, we show that another important dimension of such contact is the fact that manufacturing (cid:28)rms possess a sizable presence in non-manufacturing industries. 5 Reallocation of Employment Across Regions While a signi(cid:28)cant portion of the literature on both trade and technology has exploited regional variation in the distribution of manufacturing activities to identify causal impacts, plant and (cid:28)rm relocation within the United States remains a relatively unexplored dimension of (cid:28)rm adjustment to trade and technology shocks. We (cid:28)nd substantial reallocation of manufacturing employment across US regions over time, as well as di(cid:27)erences in the extent to which regional declines in employment are driven by (cid:28)rm death versus continuing (cid:28)rms. Figure 6 plots US manufacturing employmentfrom1977to2012bythenineUSCensusregionsthatcomprisetheUnited States. Each bar represents manufacturing employment in a given year and region, and bars are shaded to correspond to the three intervals used in Figure 5: 1977 to 1989 (black); 1990 to 1999 (dark grey); and 2000 to 2012 (light grey). As indicated in the (cid:28)gure, manufacturing employment in New England, Mid-Atlantic, and East North Central declines more-or-less steadily over the sample period. In the rest of the country, by contrast, it is either relatively (cid:29)at or growing until 2000, after which manufacturing employment in all regions shrinks. Indeed, between 1977 and 2000, combined manufacturing employment in New England, Mid-Atlantic and East North Central falls by -2.3 million, while the increase for all other regions as a whole is 0.8 million. After 2000, the largest decline, in percentage terms, occurs in South Atlantic (-38 percent). Regions also display interesting variation in terms of the margins of (cid:28)rm adjustment. In Appendix Figure A.7, we show that employment loss due to net (cid:28)rm death is concentratedinNewEnglandandMid-Atlantic, whichtogetheraccountfor16percentage points of the overall 25 percentage point decline in US manufacturing employment 16

Figure 6: US Manufacturing Employment by Census Region attributable to that margin. East North Central, by contrast, stands out in terms of its disproportionate loss of employment within continuing (cid:28)rm-plants. Reallocation of manufacturing activity within the United States might shed useful light on reallocation internationally. Indeed, movement of US manufacturing employment from the north and east towards the west and south up to 2000 may have been a precursor to international o(cid:27)shoring. Bernard et al. (2013), for example, show that US labor markets exhibit substantial and persistent variation in relative skill endowments and wages over this period, and that labor markets with di(cid:27)erent relative wages tend to specialize in di(cid:27)erent groups of industries. Fort (2017) shows that US manufacturing establishments in high wage locations are more likely to fragment production, especially domestically. Anecdotal evidence suggests (cid:28)rms do in fact relocate in response tovariationinwagesacrosslocallabormarkets. RadioCorporationofAmerica(RCA), for example, continually moved production of its most labor-intensive products west and south in search of lower wages before moving it to Mexico in the 1990s (Cowie (2001)). Such activity is consistent with the Holmes (1998) (cid:28)nding that manufacturing employment is relatively low in more union-friendly states compared to neighboring right-to-work states, which are clustered in the South Atlantic, West Central, and Mountain regions whose manufacturing employment was stable or growing prior to 2000. Were such reallocations also a response to international competition? Were they facilitated by technologies that allow (cid:28)rms to serve customers from more remote, 17

lower-cost labor markets? Do incumbents have an advantage in making use of such technologies? A thornier question raised by Figure 6 is whether relocation within the United States, either within or across (cid:28)rms, coincides with labor-saving technology upgrades, as suggested by a the long-running decline in the average number of employees per establishment referenced above? If so, how can a causal impact of technology be identi(cid:28)ed? 6 Manufacturing Firms’ Non-Manufacturing Establishments Manufacturing (cid:28)rms can also have non-manufacturing establishments. In this section, we broaden our analysis to investigate how employment at manufacturing (cid:28)rms’ nonmanufacturing establishments has evolved, and in what non-manufacturing industries they participate. As noted earlier, we de(cid:28)ne a manufacturing (M) (cid:28)rm broadly to encompass any (cid:28)rm observed to have an M establishment during our 1977 and 2012 sample period. The non-manufacturing (NM) employment of M (cid:28)rms, therefore, is simply the sum of employment at any NM establishments owned by an M (cid:28)rm. While we focus on this comprehensive set of (cid:28)rms in order to capture all M employment, it is important to bear in mind that this de(cid:28)nition includes (cid:28)rms not traditionally thought of as manufacturers (cid:21) for example, big box retailers that may encompass relatively small food preparation facilities (cid:21) and that such (cid:28)rms might have an outsized impact on the trends in NM employment we display below. With this caveat in mind, Figure 7 displays M (cid:28)rms’ total employment across their M versus NM establishments. As indicated in the (cid:28)gure, NM employment rises moreor-less steadily until 2000, when it levels o(cid:27). As a result, M (cid:28)rms’ total employment rises until 2000 before declining afterwards due to the sharp drop in employment at 13 their M establishments. As illustrated in the left panel of Appendix Figure A.3 most of the growth in M (cid:28)rms’ NM employment occurs via net NM plant birth within continuing (cid:28)rms. The growing share of M (cid:28)rm employment at NM establishments might indicate that a growing number of workers at NM establishments is needed to support M production, that M (cid:28)rms’ scope is widening to include additional NM activities, or simply that employment growth at (cid:28)rms’ NM establishments re(cid:29)ects the broader shift of US 14 employment toward NM activities. Further insight into these explanations comes 13In work not reported here, we (cid:28)nd that the trends displayed in Figure 6 are sensitive to how M (cid:28)rmsarede(cid:28)ned. Forexample, requiring(cid:28)rmstohaveatleastsomethresholdlevelofemploymentin manufacturing in at least one year of the sample results in (cid:29)atter growth of NM employment over the sampleperiod. Inaddition,thegrowthofNMemploymentatM(cid:28)rms,evenwithourbroadde(cid:28)nition of manufacturing (cid:28)rms, is slower than the growth of NM employment at NM (cid:28)rms. This di(cid:27)erential is also worthy of further exploration. 14While recent research suggests that US manufacturers increasingly outsource ancillary services such as cleaning to domestic contractors (Dey et al. (2012); Berlingieri (2014); Katz and Krueger (2016)), suchactivitywould notbecaptured inFigure7 asit tracesNMemploymentwithinM (cid:28)rms. 18

Figure 7: Employment at Manufacturing Firms’ Manufacturing vs Non-Manufacturing Establishments from analysis of the particular activities occurring at M (cid:28)rms’ NM plants. Toward that end, we break NM industries into three groups based on their two-digit NAICS sectors: retail (NAICS 44 to 45), professional services (NAICS 51 to 56), and all other NM industries. Perhaps unsurprisingly, given the broad de(cid:28)nition of M (cid:28)rms noted above, we (cid:28)nd in Online Appendix Figure A.8 that about one-third of the overall growth in M (cid:28)rms’ NM employment between 1977 and 2012 is in retail, while another third falls into the (cid:16)other(cid:17) category. However, 32 percent of the increase in non-manufacturing employment at manufacturing (cid:28)rms is driven by professional services, which captures a wide range of often skill-intensive activities : information technology (NAICS 51); (cid:28)nance, insurance, real estate and leasing (NAICS 52-3); engineering and other technical services (NAICS 54); headquarters services (NAICS 55); and administrative support and waste management (NAICS 56). The growing use of workers in such industries may re(cid:29)ect the in(cid:29)uence of both trade and technology. For example, one action US manufacturers might take in response to growing import competition in goods is to move into (cid:16)neurofacturing(cid:17) (Leamer (2009)), either by diversifying away from goods production entirely or by making use of various communications and management technologies to focus on the engineering, design or marketing of goods rather than their physical production 15 (Bernard and Fort (2015, 2017)). 15Consistent with this explanation, Magyari et al. (2017) (cid:28)nds that, in certain cases, US manufac- 19

These(cid:28)ndingsraiseanumberofintriguingquestions. Doesincreasinguseofdesign, marketing and other management services facilitate the product di(cid:27)erentiation and upgrading US (cid:28)rms undertake to compete with producers from low-wage countries? DoesithelpexplaintherisingmarketpowerofUSproducersdocumentedinDeLoecker and Eeckhout (2017)? Do US manufacturing (cid:28)rms expand their service activities in the same geographic areas in which they used to produce physical goods? As illustrated in Appendix Figure A.6, though 46 percent of M (cid:28)rms’ NM employment growth takes place in the western half of the United States, the South Atlantic exhibits the fastest pace of employment growth, at 27 percent. Further analysis of the broader scope of US manufacturing (cid:28)rms’ activities across both geographic and regional dimensions seems promising. 7 Conclusion The decline in US manufacturing jobs and concerns over the competitiveness of US manufacturers in a global market place have sparked considerable commentary and research in recent years, including several articles in this journal by Charles et al. (2016), Baily and Bosworth (2014), Tassey (2014), and Houseman et al. (2011). A natural question arising in these discussions is whether trade or technology plays a larger role in the sector’s outcomes. As we have explained, we (cid:28)nd that question to be overly broad. It may also distract needed attention away from research into how to facilitate reallocation among displaced manufacturing workers. Given that few economists advocate for restricting either technology or trade, such research seems both timely and prudent. Instead, wehave soughtto gain new perspective onthe decline of US manufacturing employment by examining relatively unexplored dimensions of microdata tracking US manufacturing (cid:28)rms over time, and considering how patterns in those data might be explained by various mechanisms associated with trade, technology, and other forces. Here, we summarize a few of the empirical facts we report, and follow-up questions that are worth pursuing. We (cid:28)nd that 75 percent of the -6.6 million decline in manufacturing employment between 1977 and 2012 took place within continuing (cid:28)rms, largely through plant closures. Why is the primary adjustment within (cid:28)rms, and in the form of plant closures? What barriers to entry (cid:21) regulatory or otherwise (cid:21) might have dampened (cid:28)rm creation or suppressed (cid:28)rm destruction? How do entrants’ technology and production functions di(cid:27)er from those of incumbents and deaths? What are the implications of these plant closures and new production techniques for displaced workers? Manufacturing(cid:28)rms’activitiesoutsidemanufacturingmighto(cid:27)ersomecluesforthe persistence of incumbent manufacturing (cid:28)rms. Before 2000, the drop in manufacturing (cid:28)rms’manufacturingemploymentismorethano(cid:27)setbyincreasesinnon-manufacturing workers. After 2000, a sharp decline in those (cid:28)rms’ manufacturing employment and a (cid:29)attening of their non-manufacturing employment leads to a decrease in their total employment. Relatively high-skill professional workers (cid:21) like designers and engineers (cid:21) turing (cid:28)rms expand their NM employment in response to import competition from China. 20

account for approximately a third of the non-manufacturing workers added by manufacturing (cid:28)rms. Are incumbents (cid:28)rms better suited to engage in these activities? Does manufacturing (cid:28)rms’ greater focus on services mimic the growth in services that takes place across non-manufacturing (cid:28)rms, or does it point to an important role for the (cid:28)rm in building up capabilities that persist over time? Finally, trade and technology can interact with di(cid:27)erent parts of manufacturing in very di(cid:27)erent ways. Manufacturing (cid:28)rms that adopt speci(cid:28)c technologies, such as computers or industrial robots, are signi(cid:28)cantly di(cid:27)erent from those that do not: in particular, they are larger and more productive upon adoption. Importing is associated with di(cid:27)erent outcomes at the (cid:28)rm and industry levels: while exposure to greater import competition is associated with employment decline, (cid:28)rms increasing their use of imported goods conditional on such exposure can exhibit employment gains. Should direct use of imported goods be considered a technology? USmanufacturinghasmanydimensions: manufacturingandnon-manufacturingestablishments, overall trends of falling employment and rising value added, incumbents and non-incumbents, geographical movements within US regions, sunset and sunrise industries, di(cid:27)erences in (cid:28)rm-level choices regarding importing inputs and use of technology, and di(cid:27)erences across industries from import penetration and the spread of technology. Our understanding of how trade and technology a(cid:27)ect US manufacturing must seek to be multifaceted as well. References Acemoglu, D. (2002). Directed technical change. The Review of Economic Studies 69(4), 781(cid:21)809. Acemoglu, D., D. Autor, D. Dorn, G. H. Hanson, and B. Price (2016). Import competition and the great us employment sag of the 2000s. Journal of Labor Economics 34(S1), S141(cid:21)S198. Acemoglu, D. and P. Restrepo (2017, March). Robots and jobs: Evidence from us labor markets. Working Paper 23285, National Bureau of Economic Research. Antr(cid:224)s, P., T.C.Fort, andF.Tintelnot(2017). Themarginsofglobalsourcing: Theory and evidence from u.s. (cid:28)rms. American Economic Review 107(9), 2514(cid:21)64. Asquith, B. J., S. Goswami, D. Neumark, and A. Rodriguez-Lopez (2017). Us job (cid:29)ows and the china shock. Technical report, National Bureau of Economic Research. Autor, D., D. Dorn, and G. Hanson (2017, February). When work disappears: Manufacturing decline and the falling marriage-market value of men. Working Paper 23173, National Bureau of Economic Research. Autor, D., D. Dorn, G. H. Hanson, G. Pisano, and P. Shu (2016, December). Foreign competition and domestic innovation: Evidence from u.s. patents. Working Paper 22879, National Bureau of Economic Research. 21

Autor, D., D.Dorn, G.H.Hanson, andJ.Song(2014). Tradeadjustment: Worker-level evidence. The Quarterly Journal of Economics 129, 199(cid:21)1860. Autor, D. H., D. Dorn, and G. H. Hanson (2013). The china syndrome: Local labor market e(cid:27)ects of import competition in the united states. American Economic Review 103(6), 2121(cid:21)68. Autor, D. H., D. Dorn, and G. H. Hanson (2015). Untangling trade and technology: Evidence from local labour markets. The Economic Journal 125, 621(cid:21)646. Baily, M. N. and B. P. Bosworth (2014, February). Us manufacturing: Understanding its past and its potential future. Journal of Economic Perspectives 28(1), 3(cid:21)26. Barrot, J.-N., E. Loualiche, M. C. Plosser, and J. Sauvagnat (2017, August). Import competition and household debt. FRB of NY Sta(cid:27) Report No. 821. Becker, R., W. Gray, and J. Marvakov (2013). Nber-ces manufacturing industry database: Technical notes. Working Paper 5809, National Bureau of Economic Research. Berlingieri, G. (2014). Outsourcing and the rise in services. Working Paper 1199, CEPR. Bernard, A. B. and T. C. Fort (2015, May). Factoryless goods producing (cid:28)rms. American Economic Review Papers and Proceedings 105(5), 518(cid:21)523. Bernard, A. B. and T. C. Fort (2017). The Factory-Free Economy, Chapter Factoryless Goods Producers In the US, pp. Chapter 5. Oxford University Press. Bernard, A. B., T. C. Fort, V. Smeets, and F. Warzynski (2017). O(cid:27)shoring and reorganization. Technical report, Tuck School of Business. Bernard, A. B., J. B. Jensen, S. Redding, and P. Schott (2011). Multi-product (cid:28)rms and trade liberalization. Quarterly Journal of Economics 126(3), 1271(cid:21)1318. Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott (2007, August). Firms in international trade. Journal of Economic Perspectives 21(3), 105(cid:21)130. Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott (2010, May). Wholesalers and retailers in US trade. American Economic Review 100(2), 408(cid:21)413. Bernard, A.B., J.B.Jensen, andP.K.Schott(2006). Survivalofthebest(cid:28)t: Exposure to low-wage countries and the (uneven) growth of u.s. manufacturing plants. Journal of International Economics 68, 219(cid:21)237. Bernard, A. B., S. J. Redding, and P. K. Schott (2013, May). Testing for factor price equality with unobserved di(cid:27)erences in factor quality or productivity. American Economic Journal: Microeconomics 5(2), 135(cid:21)63. 22

Bloom, N., M. Draca, and J. V. Reenen (2016). Trade induced technical change: The impact of chinese imports on innovation, di(cid:27)usion, and productivity. Review of Economic Studies 83, 87(cid:21)117. Bown, C. (2016). Global Antidumping Database. The World Bank. Brynjolfsson, E. and L. M. Hitt (1998, August). Beyond the productivity paradox. Commun. ACM 41(8), 49(cid:21)55. Caliendo, L. and F. Parro (2015). Estimates of the trade and welfare e(cid:27)ects of nafta. The Review of Economic Studies 82(1), 1(cid:21)44. Charles, K. K., E. Hurst, and M. J. Notowidigdo (2016, May). The masking of the decline in manufacturing employment by the housing bubble. Journal of Economic Perspectives 30(2), 179(cid:21)200. Che, Y., X. Xu, and Y. Zhang (2017). Chinese import competition, crime, and government transfers in us. Journal of Comparative Economics. Christensen, C. M. (1997). The Innovator’s Dilemma. HarperBusiness. Clemens, M. A., E. G. Lewis, and H. M. Postel (2017, February). Immigration restrictions as active labor market policy: Evidence from the mexican bracero exclusion. Working Paper 23125, National Bureau of Economic Research. Collard-Wexler, A. and J. De Loecker (2015). Reallocation and technology: Evidence from the us steel industry. American Economic Review 105(1), 131(cid:21)71. Cowie, J. (2001). Capital moves: RCA’s seventy-year quest for cheap labor. The New Press. De Loecker, J. and J. Eeckhout (2017). The rise of market power and the macroeconomic implications. Technical report, National Bureau of Economic Research. Decker, R. A., J. Haltiwanger, R. S. Jarmin, and J. Miranda (2016). Where has all the skewness gone? the decline in high-growth (young) (cid:28)rms in the us. European Economic Review 86, 4(cid:21)23. Decker, R. A., J. C. Haltiwanger, R. S. Jarmin, and J. Miranda (2018). Changing business dynamism and productivity: Shocks vs. responsiveness. Technical report, National Bureau of Economic Research. Dey, M., S. H. Houseman, and A. E. Polivka (2012). Manufacturers’ outsourcing to sta(cid:30)ng services. ILR Review. Ebenstein,A.,A.Harrison,M.McMillan,andS.Phillips(2014). Estimatingtheimpact of trade and o(cid:27)shoring on american workers using the current population surveys. Review of Economics and Statistics 96(4), 581(cid:21)595. 23

Edwards, L. and R. Z. Lawrence (2013). Rising Tide: Is Growth in Emerging Economies Good for the United States? Peterson Institute for International Economics. Feenstra, R. C. (1996, March). U.s. imports, 1972-1994: Data and concordances. Working Paper 5515, National Bureau of Economic Research. Feler, L. and M. Z. Senses (2017, November). Trade shocks and the provision of local public goods. American Economic Journal: Economic Policy 9(4), 101(cid:21)43. Folbre, N. (2013, August). Not really made in china (or the united states). The New York Times. Fort, T. C. (2017). Technology and production fragmentation: Domestic versus foreign sourcing. Review of Economic Studies 84(2), 650(cid:21)687. Fort, T. C. and S. Klimek (2016). The e(cid:27)ects of industry classi(cid:28)cation changes on us employment compostion. Technical report, Tuck School of Business. Galle, S., A. Rodr(cid:237)guez-Clare, and M. Yi (2017). Slicing the pie: Quantifying the aggregate and distributional e(cid:27)ects of trade. Technical report, National Bureau of Economic Research. Goos, M., A. Manning, and A. Salomons (2014). Explaining job polarization: Routinebiased technological change and o(cid:27)shoring. American Economic Review 104(8), 2509(cid:21)2526. Graetz, G. and G. Michaels (2017). Is modern technology responsible for jobless recoveries? American Economic Review 107(5), 168(cid:21)73. Hakobyan, S. and J. McLaren (2016). Looking for local labor market e(cid:27)ects of nafta. The Review of Economics and Statistics 98(4), 728(cid:21)741. Haltiwanger, J., R. S. Jarmin, and J. Miranda (2013). Who creates jobs? small versus large versus young. The Review of Economics and Statistics 95, 347(cid:21)361. Handley, K. and N. Limªo (2017, September). Policy uncertainty, trade, and welfare: Theory and evidence for china and the united states. American Economic Review 107(9), 2731(cid:21)83. Holmes, T. J. (1998). The e(cid:27)ect of state policies on the location of manufacturing: Evidence from state borders. Journal of Political Economy 106(4), 667(cid:21)705. Houseman, S., C. Kurz, P. Lengermann, and B. Mandel (2011, June). O(cid:27)shoring bias in u.s. manufacturing. Journal of Economic Perspectives 25(2), 111(cid:21)32. Igami, M. (2018). Industry dynamics of o(cid:27)shoring: The case of hard disk drives. American Economic Journal: Microeconomics 10(1), 67(cid:21)101. 24

Jarmin, R. S. and J. Miranda (2002). The longitudinal business database. Center for Economic Studies Working Paper 02(17). Katz, L. F. and A. B. Krueger (2016). The rise of alternative work arrangements in the united states, 1995-2015. Working Paper 22667, NBER. Khandelwal, A. (2010). The long and short (of) quality ladders. Review of Economic Studies 77, 1450(cid:21)1476. Khandelwal, A. K., P. K. Schott, and S.-J. Wei (2013, October). Trade liberalization and embedded institutional reform: Evidence from chinese exporters. American Economic Review 103(6), 2169(cid:21)95. Langdon, D. and R. Lehrman (2012). The bene(cid:28)ts of manufacturing jobs. US Department of Commerce, ESA Issue Brief, 01(cid:21)12. Leamer, E. E. (2009). Heckscher-ohlin models for the post-industrial age. UCLA, mimeograph. Magyari, I. et al. (2017). Firm reorganization, chinese imports, and us manufacturing employment. Columbia University. Michaels, D. (2017, March). Foreign robots invade american factory (cid:29)oors. Wall Street Journal. Muendler, M.-A. (2017). Trade, technology, and prosperity: An account of evidence from a labor-market perspective. Technical report, WTO Sta(cid:27) Working Paper. Oldenski, L. (2014). O(cid:27)shoring and the polarization of the u.s. labor market. Industrial and Labor Relations Review 67. Oster, S. (1982). Intraindustry structure and the ease of strategic change. The Review of Economics and Statistics, 376(cid:21)383. Pierce, J. R. (2011). Plant-level responses to antidumping duties: Evidence from u.s. manufacturers.. Journal of International Economics 85(2), 222(cid:21)233. Pierce, J. R. and P. K. Schott (2012, December). The surprisingly swift decline of u.s. manufacturing employment. Working Paper 18655, National Bureau of Economic Research. Pierce, J. R. and P. K. Schott (2016a). The Surprisingly Swift Decline of U.S. Manufacturing Employment. American Economic Review 106(7), 1632(cid:21)1662. Pierce, J. R. and P. K. Schott (2016b). Trade liberalization and mortality: Evidence from us counties. Working Paper 22849, NBER. Schott, P. K. (2003, June). One size (cid:28)ts all? heckscher-ohlin specialization in global production. American Economic Review 93(3), 686(cid:21)708. 25

Schott, P. K. (2004, May). Across-product versus within-product specialization in international trade. The Quarterly Journal of Economics 119(2), 647(cid:21)678. Schott, P. K. (2008). The relative sophistication of chinese exporters. Economic Policy 53, 5(cid:21)49. Steinwender, C. (2018, March). Real e(cid:27)ects of information frictions: When the states and the kingdom became united. American Economic Review 108(3), 657(cid:21)96. Sullivan, D. and T. V. Wachter (2009). Job displacement and mortality: An analysis using administrative data. Quarterly Journal of Economics, 1265(cid:21)1306. Tassey, G. (2014). Competing in advanced manufacturing: The need for improved growth models and policies. Journal of Economic Perspectives 28(1), 27(cid:21)48. 26

Appendix for Online Publication This online appendix contains additional (cid:28)gures referenced in the main text as well as more detailed information about the regression results referenced in the main text. A Regression Detail A.1 Premia Regressions We examine the correlation between use of technology and both (cid:28)rm employment and labor productivity via a series of cross-sectional OLS regressions in each census year of the form ln(Attributet} = α+β1{Activityt}+ηt +(cid:15)t:. (A.1) f f j f The left-hand side variable is either the log employment or the log labor productivity (shipmentsdividedbyemployment)for(cid:28)rmf incensusyeart. The(cid:28)rstright-handside variable represents participation in one of the following technology or trade activities in census yeart: purchase of computers, use of electronic networks to control or coordinate shipments, direct importing of industrial robots (HS 84.7950.0000) or direct importing ofanygoodfromanycountry. ηt representsindustry(cid:28)xede(cid:27)ects. Weestimateseparate j regressionsforeachactivityandeachcensusyearfrom1977to2012. Dataforcomputer purchases is not available in 1997. Data for importing, importing robots, and use of electronic networks are not available before 1992, 1997 and 2002, respectively. Point estimates and ninety-(cid:28)ve percent con(cid:28)dence intervals for each activity and year are displayed in Appendix Figure A.5. As indicated in the (cid:28)gure, (cid:28)rms engaged in each of the examined activities are larger and more productive than those not engaging in the activities. These size and productivity premia generally shrink over time, though the decreases are considerably larger for the technology activities versus direct importing. A.2 Plant Death Regressions We examine the correlates of plant death within multi-establishment (cid:28)rms by estimating the following OLS regression, 1{Deatht:t+5} = α+β1{Activityt }+γln(Empt)+δ +ρt +(cid:15)t:t+5. (A.2) pf pf p f pf The left-hand side variable is an indicator for whether the plant exits between census years t and t + 5. After the constant, the second variable on the right-hand side represents indicators for whether the plant engages in a particular activity, such as purchasingcomputersorusinganelectronicnetworktocontrolorcoordinateshipments in census year t. The third variable on the right hand side is the natural log of plant employment, the fourth covariate represents (cid:28)rm (cid:28)xed e(cid:27)ects, and the (cid:28)fth covariate represents year (cid:28)xed e(cid:27)ects. We estimate this equation separately across census years 27

before and after 2000, i.e., 1977 to 1997 and 2002 to 2012. Data for computer purchases are available in all census years except 1997, when this information was not collected. Data for use of electronic networks are available starting in 2002. We also estimate a variant of equation A.2 in which we replace Activityt with either the change in import pf penetration or the change in import penetration from China in the plant’s industry between years t and t+5. Coe(cid:30)cientestimatesarereportedinthe(cid:28)rsttwocolumnsofTableA.1. Becausethe regressors are endogneous and no instrumental variables are employed, these coe(cid:30)cient estimates should be treated as correlations, with no claim of causality. Table A.1: Plant and Firm Death Regressions For comparison, we also report a series of analogous (cid:28)rm death regressions in the second two columns of Table A.1, 1{Deatht:t+5} = α+β1{Activityt}+ηt +ρt +(cid:15)t:t+5, (A.3) f f j f 28

where ηt captures industry (cid:28)xed e(cid:27)ects. In these regressions, we are also able to j investigate the association between (cid:28)rm death and being a direct importer of industrial robots (HS 84.7950.0000) in year t as well as being a direct importer or being a direct importer from China of any good in year t. As noted in the main text, we are unable to examine these relationships at the plant level given that trading is observed only at the (cid:28)rm level. A.3 Continuing-Firm Regressions To assess the potential role of trade and technology in US manufacturers’ employment changes within continuing (cid:28)rms, we examine how (cid:28)rm outcomes relate to various activities using (cid:28)rm-level OLS panel regressions of the form ∆log(Outcomet:t+5) = βActivityt +ηt +ρ +(cid:15)t:t+5. (A.4) f f j t f ∆log(Outcomet:t+5) represents the log di(cid:27)erence in (cid:28)rm-level manufacturing employf ment, total employment, real value added in manufacturing, or real value added in manufacturing per manufacturing worker between census years t and t+5. Activityt, f as above, represents one of several actions, including purchasing computers, using electronicnetworkstocontrolorcoordinateshipments, beingadirectimporterofindustrial robots (HS 84.7950.0000), being a direct importer of any good from any country, or being a direct importer from China. When considering the latter two activities, we also include contemporaneous t to t+5 changes in the analogous industry-level import penetration, that is, change in overall import penetration or the change in import penetration from China. These additions allow for the possibility, discussed in Section 3 of the main text, that import competition and direct foreign sourcing may have di(cid:27)erent associations with (cid:28)rm outcomes. We note that these regressions are purely descriptive and should not be interpreted as providing causal evidence. As an additional caveat, we note that regressions are unweighted. Results are presented in Table A.2, where the top and bottom panels display results for census years before and after 2000. The top panel reports the results of three regressions for each outcome variable, where the three regressions are separated into rows. The (cid:28)rst regression examines the relationship between computer purchases and the outcome variables while the second and third examine relationships with respect to being an importer or being an importer from China. Computer purchase data are not available in 1997, and importing data are not available until 1992. As a result, the number of observations for the (cid:28)rst regression is larger than for the second and third regressions. The bottom panel of Table A.2 considers years after 2000 and reports the results of (cid:28)ve regressions for each outcome variable. All regressions in this panel have the same number of observations. We note that observations are rounded to the nearest thousand per Census Bureau disclosure guidelines. Before 2000, computer purchasers exhibit declines in employment and real value added relative to non-purchasers, with the declines in the former being somewhat larger in absolute value. As a result, during this period, computer purchases are associated with increases in labor productivity. Results for being an importer or an 29

importer from China are similar. A second notable trend in this panel is that for all three activities, the coe(cid:30)cients in regressions considering total (cid:28)rm employment (column 2) are smaller than those for manufacturing employment (column 1), indicating that employment adjustment to the noted activities occurs disproportionately among manufacturing establishments. After 2000, we (cid:28)nd a di(cid:27)erent pattern of results for (cid:28)rms that purchase computers and are direct importers. These activities are now associated with rising employment and rising real value added. Moreover, we (cid:28)nd the same pattern of results for (cid:28)rms that use electronic networks to control or coordinate shipments. In contrast, (cid:28)rms that import industrial robots see relatively less manufacturing employment growth than (cid:28)rms that do not import these robots, though there is no signi(cid:28)cant relationship with their total employment and a positive and signi(cid:28)cant relationship with real value added and labor productivity. These results are consistent with the premise that technologymayreplaceworkersevenasitboostsoutput. Finally,importingfromChina isassociatedwithastatisticallysigni(cid:28)cantdecreaseinmanufacturingemploymentafter 2000, but no statistically signi(cid:28)cant relationship with total employment or real value added. Results for changes in either overall or Chinese import penetration at the industry levelindicatenegativecorrelationswithemploymentafter2000. TableA.2containstwo other suggestive results. First, being an importer in post-2000 years is correlated with relativelyhighergrowthinemploymentandrealvalueadded, whereasincreasedimport penetration in the (cid:28)rm’s initial and primary (based on employment) manufacturing industry is associated with statistically signi(cid:28)cant relative reductions in growth in both outcomes. Higher growth in Chinese import penetration is associated with relatively lower growth in manufacturing and total employment, while the relationship with real value added growth is negative but statistically insigni(cid:28)cant at conventional levels (p-value=0.12). Furthermore, the divergence in (cid:28)rm-level employment versus output correlations for robot importing and importing from China highlight the possibility that technology and trade may be factors in decreased manufacturing employment and increased output of US manufacturing (cid:28)rms. 30

Table A.2: Continuing-Firm Regressions 31

B Additional Figures Figure A.1: US Manufacturing Absorption Figure A.2: Employment versus Value Added Growth Across Six-Digit NAICS Sectors 32

Figure A.3: US Non-Manufacturing Employment by Net Margins of Adjustment Figure A.4: US Manufacturing Establishment Count 33

Figure A.5: Technology Adopters’ Size and Productivity Premia Figure A.6: US Manufacturing Firm Non-Manufacturing Employment, by Census Region 34

Figure A.7: US Manufacturing Firm Employment by Net Margin of Adjustment and Region 35

Figure A.8: US Manufacturing Firm Non-Manufacturing Employment, by Super NAICS Sectors 36