Who’s afraid of Virginia Wu? US employment footprints and self-sufficiency

Abstract

Globalization has brought about concerns of domestic job losses due to outsourcing to countries like China. The ‘employment footprint’ concept provides new insights into the implications of trade for employment. Using this approach for the period of 1995–2008, we analyze the relation of US jobs with international trade, particularly with China. Furthermore, we compare the US employment footprint with its labor endowment to assess if the country could be self-sufficient in terms of labor. We find that the US’s consumption increasingly depends on foreign workers. The country ‘consumes’ more labor than is nationally available; thus, self-sufficiency is not possible under realistic assumptions. Moreover, the US has benefited from jobs – especially in services – generated by the world economy. Referring to Albee’s famous play about living in illusions, we use ‘Virginia Wu’ as a Chinese version of ‘Virginia Woolf’ to argue that the perceived threat of China (Virginia Wu) is only an illusion.

KEYWORDS:

• Factor content of trade
• labor footprints
• input–output analysis

1. Introduction¹

The famous play ‘Who’s afraid of Virginia Woolf’ by Edward Albee (1962) is about a married couple that take refuge in false illusions. In this paper, we consider the implications of global trade during 1995–2008 on employment in the US. We explore whether global trade, particularly with China, has taken away domestic US jobs, or if that is just an illusion. We raise the questions: Has the US lost jobs due to global trade, particularly with China? Furthermore, can the US be self-sufficient in terms of labor? A major concern in advanced economies is that offshoring leads to job losses. The implicit reasoning is that unemployment would be less if offshoring would be replaced by local production. This begs the question whether this is possible in the first place; is the necessary labor available in the home country to replace the foreign workers? We answer this question by calculating what it takes to make the US self-sufficient in terms of labor. Our approach uses an empirical methodology that has become possible with the availability of global input–output tables. These tables allow us to calculate global employment footprints.

A country’s global employment footprint measures the global amount of labor that is embodied in the final products that this country consumes. The idea behind employment footprints is simple. Trade enables countries to use foreign labor, as embodied in imported goods and services. Not only the foreign labor that is embodied directly in the import of a – final or intermediate – product should be accounted for, but also the foreign labor that goes into the inputs that go into the production of this particular product. And the labor that goes into the inputs that go into the inputs, and so forth. This is the labor that is indirectly embodied.

The international fragmentation of production processes has increased in importance (Baldwin, 2006). Johnson and Noguera (2017) and Timmer et al. (2014), among others, provide empirical evidence on the fragmentation of production using a trade in value-added perspective and input–output techniques. The employment footprint concept builds upon these existing measures by focusing on trade in employment. The unbundling of production has two consequences. First, the contribution to the footprint of labor that is indirectly embodied has probably increased. Second, calculating employment footprints is difficult because international fragmentation has led to longer global supply chains. However, using global input–output tables allows us to include all indirectly embodied labor. We can thus trace global supply chains and calculate the employment footprint more precisely than is possible without these data.

We use the concept of ‘employment footprint’ as the number of workers, worldwide, required to produce the entire bundle of final products (goods and services) consumed in a certain country and year, in our case in the US. This method allows us to calculate how much global labor (of different skill-types and different origins) is currently at work for consumption in a specific country. We measure the number of domestic and foreign workers that go into US consumption. The foreign workers embodied in US consumption are the US imports of foreign workers. On the flip side, we consider the number of US workers that are exported, i.e. are embodied in foreign consumption of final products. Then we make different comparisons of employment footprints and labor endowments. We provide three measures.

First, we define the worker surplus as the labor endowments minus the footprint. The footprint tells how much labor is embodied in US consumption – including foreign labor – and the endowment expresses how much labor is available in the US. The difference (i.e. endowment minus footprint) gives the labor surplus and indicates the ‘net’ availability of labor (i.e. available labor after all requirements have been met and given the current trade). A labor deficit indicates that the US ‘consumes’ more labor than it actually endows. This is a ‘gains from trade.’

Second, we carry out a feasibility check. We examine whether the US has – in principle – sufficient labor to sustain its current consumption without trade. We assume that the US has the same technology and same level and type of consumption, but no trade. A labor deficit indicates that the current level of consumption cannot be sustained under self-sufficiency. Third, we calculate the self-sufficiency consumption ratio. This is the consumption level that could be sustained by US workers alone divided by the current consumption level. Note that in our calculations we assume that productivity in autarky is the same as with trade. This implies that our estimates for the autarky scenario are relatively optimistic as they incorporate possible dynamic productivity gains of trade.

This analysis is related to the concerns that outsourcing leads to job losses. As Rodrik (2017) points out, globalization has been associated with offshoring in advanced economies. The perceived threat to domestic jobs has increased, once production processes and value chains have become global (Hummels et al., 2001). In addition, several studies document potential negative effects of Chinese import exposure on the US labor market (Acemoglu et al., 2016; Autor et al., 2013; Caliendo et al., 2019; Ebenstein et al., 2012; Pierce & Schott, 2016). The employment footprint concept provides a way of looking at the labor effects of import competition.

The analysis based on the employment footprint of the US covers 14 consecutive years from 1995 to 2008. The paper is organized as follows. First, we provide the context and literature in Section 2, then we present the methodology in Section 3 and the data sources in Section 4. Subsequently, we discuss the results in Sections 5, and Section 6 concludes.

2. Literature

2.1. The employment footprint

Footprint indicators are increasingly popular in research, policy- and decision-making (Gomez-Paredes et al., 2015). They have been used for the analysis of ecology, energy, water usage, land, biodiversity, wages and inequality. The concept captures both direct and indirect repercussions along production and distribution chains (Cucek et al., 2012). Not only the direct impact of a particular final product is included, but also the indirect impact of supplying (foreign) intermediate products and supplying the intermediates for producing the intermediates. Footprints are computed by using input–output analysis (Arto et al., 2014; Gomez-Paredes et al., 2015; Wiedmann et al., 2006). The employment footprint builds on this concept and can be used to trace the global flows of labor that end up in the consumption of final products (i.e. final demand) of a particular country.

Employment footprints have been used before in a somewhat different context. Gomez-Paredes et al. (2015) apply employment footprints in relation to social sustainability issues of commodity chains, and Alsamawi et al. (2014) in the context of wage inequalities. The term ‘employment footprints’ is introduced by Alsamawi et al. (2014) to describe the total amount of workers (i.e. domestic workers supporting domestic consumption plus workers embodied in imported goods and services).

The employment footprint is constructed by linking employment accounts to trade flows using international input–output tables. This accounts for all workers, in whatever country, engaged in producing the final consumption bundle. All employment footprints together yield the global use of labor. Our method is identical to Alsamawi et al. (2014), but the application differs. Our main focus is on evaluating the labor consequences of self-sufficiency of the US. We do not focus on wage inequalities as Alsamawi et al. (2014). Another important difference is that Alsamawi et al. (2014) use the Eora database while our study uses the World Input-Output Database (WIOD).

2.2. Factor content of trade

Our paper fits into the factor content of trade literature which has been used to measure the gains from trade (Adao et al., 2017; Costinot & Rodriquez-Clare, 2018). The factor content of trade analyses, as well as our employment footprint analysis, are based on input–output (IO) models.² This literature goes back to the Hecksher-Ohlin (HO) model, extended by Vanek (1968), which states that countries export (import) their relatively abundant (scarce) factors. In general, the HO model does not perform well from an empirical point of view (see Feenstra, 2018 for a survey). Only when the basic assumptions of the model are relaxed, such as allowing for technological differences between countries, does the empirical performance improve (Davis & Weinstein, 2001). Factor content studies typically include all production factors (such as labor, capital, and land) that are embodied in the trade of goods and services. In contrast, our analysis focusses on one specific factor of production: labor.

One of the challenges in extensions to the HO model was to account for cross-country productivity differences, which is an issue relevant to our study. Leontief (1953) and other early work had insufficient data on the production techniques of foreign trading partners and resorted to US technology to estimate labor and capital embodied in imports from abroad. A widely cited study by Trefler (1993) introduced factor-augmenting productivity differences across countries into the HO model. He imputed international productivity differences to make the HO theorem perfectly fit the data on trade and endowments, then concluded they strongly correlated to cross-country variation in factor prices. However, there were two limitations in his study. First, Trefler (1993), like Leontief (1953) used the US IO-matrix to measure the factor content of trade globally for all countries. Reliance on technology coefficients of a highly developed and capital-abundant country to serve as a reference is problematic as it leads to a (downward) bias in the labor content of exports by countries with less efficient technologies. Second, Trefler (1993) did not account for trade in intermediate products, which became important with the global integration of production and the ongoing international fragmentation.

Subsequent studies have addressed the shortcomings, both through improved methodologies and better data (e.g. Marshall, 2012). Davis and Weinstein (2001) used the OECD's Input-Output Database to introduce the use of intermediates into the theoretical HO model and to take into account productivity differences. Reimer (2006) extended the approach of Davis and Weinstein (2001) by using a world IO-matrix to measure the factor content of trade when production technologies differ across countries and intermediate inputs are traded. A more recent contribution that employs international IO tables is Trefler and Zhu (2010). They used the Global Trade Analysis Project (GTAP) database in order to determine the correlation between relative factor endowments and the factor content of trade. However, Zhang (2015) criticized Trefler and Zhu’s approach as being economically misleading due to double-counting of re-exported intermediates in complex trade structures. Fisher and Marshall (2015) add to the criticism that the approach also assumes homogeneity and comparability of labor across countries.

The Fisher and Marshall (2015) critique is related to general aggregation limitations of all input–output-based analyses. Aggregate industries in one country produce baskets of goods and services that may differ from the baskets produced by the same industries in another country. A possible bias is that workers in different countries may specialize in different types of activities within the same industry. Our differentiation of workers based on skill-levels mitigates this problem to some extent. Educational attainment (reflected in skill-levels) can serve to proxy different types of activities performed by workers within an industry.

2.3. Our approach

We proceed in two steps. First, we use employment footprints to calculate the amount of labor – that we split into workers of different skill-levels and origins – embodied in final consumption of the US. We calculate the exports of domestic labor (i.e. domestic labor embodied in foreign final demand) and the imports of foreign labor (i.e. foreign labor embodied in domestic final demand). This enables us to determine the net imports of labor (reflecting the dependency of the US on foreign workers) between 1995 and 2008. Second, we calculate how much labor would – technically speaking – be required to sustain its actually observed consumption levels if the US were forced to abolish all trade (using various assumptions). We compare this US footprint under self-sufficiency with the US employment footprint in the actual case. We include both temporal and country-level comparisons.

3. Methodology

Our study differs from the standard factor content research in three respects. First, our focus is on employment footprints, not on testing the validity of the HO model. Second, in order to avoid the double-counting of labor inputs in trade, we draw upon the ‘trade in value-added’ concept (Johnson, 2014). We trace the domestic factors embodied in domestic production that are directly or indirectly contained in the final consumption of a partner country (even in the absence of direct bilateral trade between them). This demand-based perspective differs from related studies such as Groshen et al. (2005) and Stehrer and Stöllinger (2014), both of whom used IO tables to compare the number of jobs embodied in exports with the hypothetical number of jobs required to produce imports.³ This means that we are interested in the trade in factors, rather than in the factor content of trade. Third, we are not only interested in the trade in factors but in the full factor content of a country’s consumption. In this respect, we draw upon the footprints approach and want to know all the origins (including ‘home’) of the factors embodied in final consumption.

The starting point of the analysis is a world IO table (also known as a Global Multiregional Input–Output table, see Tukker & Dietzenbacher, 2013, for an overview), which looks as follows for the case of N countries.⁴ Assuming n industries in each country, ${\mathbf{\text{Z}}}^{RS}$ is the $n\times n$ matrix with intermediate deliveries and its element $z_{ij}^{RS}$ gives the delivery of goods and services (expressed in million US$) that industry i in country R sells to industry j in country S. The element $f_i^{RS}$ of the vector ${\mathbf{\text{f}}}^{RS}$ gives the delivery of goods and services from industry i in country R for household consumption and other domestic final demand purposes (including private investments, government consumption and investments, and changes in stocks) in country S. $\sum _j{z}_{ij}^{RS}+f_i^{RS}$ are the exports from industry i in country R to country S. For simplicity, we use the term consumption to refer to the total of all final demand categories. The element $x_i^R$ of the vector ${\mathbf{\text{x}}}^R$ gives the total output (or value of production) by industry i in country R. In country S, the element $v_j^S$ of the (row) vector $({\mathbf{\text{v}}}^S{)}^{\mathrm{\prime }}$ gives the value added (including wages and salaries, employers’ contributions, capital depreciation, indirect taxes, price-decreasing subsidies, and operating surplus or other income) generated in industry j in country S. Additionally, the element $l_j^S$ of the (row) vector $({\mathbf{\text{l}}}^S{)}^{\mathrm{\prime }}$ gives the input of labor in industry j in country S. Labor is measured in thousands of workers (not corrected for productivity differences). Note that there are no separate column vectors with exports nor separate row vectors with imports included in Table 1. This is because the N countries make up the entire world.

TABLE 1. The world input–output table.

	Intermediate use in:					Final use in:
	1	…	R	…	N	1	…	R	…	N	Total output
Product flows from
country 1	Z${}^{11}$		Z${}^{1R}$		Z${}^{1N}$	f${}^{11}$		f${}^{1R}$		f${}^{1N}$	x${}^1$
$⋮$	$⋮$		$⋮$		$⋮$	$⋮$		$⋮$		$⋮$	$⋮$
country R	Z${}^{R1}$		Z${}^{RR}$		Z${}^{RN}$	f${}^{R1}$		f${}^{RR}$		f${}^{RN}$	x${}^R$
$⋮$	$⋮$		$⋮$		$⋮$	$⋮$		$⋮$		$⋮$	$⋮$
country N	Z${}^{N1}$		Z${}^{NR}$		Z${}^{NN}$	f${}^{N1}$		f${}^{NR}$		f${}^{NN}$	x${}^N$
Value added	(v${}^1{)}^{\mathrm{\prime }}$		(v${}^R{)}^{\mathrm{\prime }}$		(v${}^N{)}^{\mathrm{\prime }}$
Total inputs	(x${}^1{)}^{\mathrm{\prime }}$		(x${}^R{)}^{\mathrm{\prime }}$		(x${}^N{)}^{\mathrm{\prime }}$
Labor inputs	(l${}^1{)}^{\mathrm{\prime }}$	$\dots$	(l${}^R{)}^{\mathrm{\prime }}$	$\dots$	(l${}^N{)}^{\mathrm{\prime }}$

The employment footprint gives the amount of work worldwide that is necessary for the final demands of a country (say R). Define $\mathbf{\text{A}}=\left[\begin{array}{ccc}\begin{array}{cc}{\mathbf{\text{A}}}^{11}& \cdots \\ ⋮& \ddots \end{array}& \begin{array}{c}{\mathbf{\text{A}}}^{1R}\\ ⋮\end{array}& \begin{array}{cc}\cdots & {\mathbf{\text{A}}}^{1N}\\ \text{.}\text{.}\text{.}& ⋮\end{array}\\ \begin{array}{cc}{\mathbf{\text{A}}}^{R1}& \cdots \end{array}& {\mathbf{\text{A}}}^{RR}& \begin{array}{cc}\cdots & {\mathbf{\text{A}}}^{RN}\end{array}\\ \begin{array}{cc}⋮& \text{.}\text{.}\text{.}\\ {\mathbf{\text{A}}}^{N1}& \cdots \end{array}& \begin{array}{c}⋮\\ {\mathbf{\text{A}}}^{NR}\end{array}& \begin{array}{cc}\ddots & ⋮\\ \cdots & {\mathbf{\text{A}}}^{NN}\end{array}\end{array}\right],\mathit{\omega }=\left(\begin{array}{c}\begin{array}{c}{\mathit{\omega }}^1\\ ⋮\end{array}\\ {\mathit{\omega }}^R\\ \begin{array}{c}⋮\\ {\mathit{\omega }}^N\end{array}\end{array}\right)$ where ${\mathbf{\text{A}}}^{RS}={\mathbf{\text{Z}}}^{RS}({\widehat{\mathbf{\text{x}}}}^S{)}^{-1}$ is the $n\times n$ matrix of input coefficients $a_{ij}^{RS}=z_{ij}^{RS}/x_j^S$, and ${\mathit{\omega }}^R={\mathbf{\text{l}}}^R({\widehat{\mathbf{\text{x}}}}^R{)}^{-1}$ is the vector of labor inputs per US$ of output, i.e. ${\omega }_j^R=l_j^R/x_j^R$. The vector with employment footprints is then given by (1) $\begin{array}{rl}& \left(\begin{array}{ccc}\begin{array}{cc}{\phi }^1& \cdots \end{array}& {\phi }^R& \begin{array}{cc}\cdots & {\phi }^N\end{array}\end{array}\right)\\ & =\left(\begin{array}{ccc}\begin{array}{cc}({\mathit{\omega }}^1{)}^{\mathrm{\prime }}& \cdots \end{array}& ({\mathit{\omega }}^R{)}^{\mathrm{\prime }}& \begin{array}{cc}\cdots & ({\mathit{\omega }}^N{)}^{\mathrm{\prime }}\end{array}\end{array}\right)\\ & \times \left[\begin{array}{ccc}\begin{array}{cc}{\mathbf{\text{M}}}^{11}& \cdots \\ ⋮& \ddots \end{array}& \begin{array}{c}{\mathbf{\text{M}}}^{1R}\\ ⋮\end{array}& \begin{array}{cc}\cdots & {\mathbf{\text{M}}}^{1N}\\ \text{.}\text{.}\text{.}& ⋮\end{array}\\ \begin{array}{cc}{\mathbf{\text{M}}}^{R1}& \cdots \end{array}& {\mathbf{\text{M}}}^{RR}& \begin{array}{cc}\cdots & {\mathbf{\text{M}}}^{RN}\end{array}\\ \begin{array}{cc}⋮& \text{.}\text{.}\text{.}\\ {\mathbf{\text{M}}}^{N1}& \cdots \end{array}& \begin{array}{c}⋮\\ {\mathbf{\text{M}}}^{NR}\end{array}& \begin{array}{cc}\ddots & ⋮\\ \cdots & {\mathbf{\text{M}}}^{NN}\end{array}\end{array}\right]\left[\begin{array}{ccc}\begin{array}{cc}{\mathbf{\text{f}}}^{11}& \cdots \\ ⋮& \ddots \end{array}& \begin{array}{c}{\mathbf{\text{f}}}^{1R}\\ ⋮\end{array}& \begin{array}{cc}\cdots & {\mathbf{\text{f}}}^{1N}\\ \text{.}\text{.}\text{.}& ⋮\end{array}\\ \begin{array}{cc}{\mathbf{\text{f}}}^{R1}& \cdots \end{array}& {\mathbf{\text{f}}}^{RR}& \begin{array}{cc}\cdots & {\mathbf{\text{f}}}^{RN}\end{array}\\ \begin{array}{cc}⋮& \text{.}\text{.}\text{.}\\ {\mathbf{\text{f}}}^{N1}& \cdots \end{array}& \begin{array}{c}⋮\\ {\mathbf{\text{f}}}^{NR}\end{array}& \begin{array}{cc}\ddots & ⋮\\ \cdots & {\mathbf{\text{f}}}^{NN}\end{array}\end{array}\right]\end{array}$(1) where M is the multiplier matrix $(\mathbf{\text{I}}-\mathbf{\text{A}}{)}^{-1}$, known as the Leontief inverse (or total requirements matrix). Element $m_{ij}^{RS}$ of the matrix ${\mathbf{\text{M}}}^{RS}$ indicates how much output from industry i in country R is directly and indirectly required per unit of final demand for the products produced by industry j in country S. ${\phi }^R$ gives the employment footprint for country R, which gives the amount of labor worldwide that is necessary to sustain the consumption pattern of country R. Note that (2) ${\phi }^R=\sum _{K=1}^N\sum _{J=1}^N({\mathit{\omega }}^K{)}^{\mathrm{\prime }}{\mathbf{\text{M}}}^{KJ}{\mathbf{\text{f}}}^{JR}$(2) This expression consists of: the domestic labor in country R that is embodied in the entire consumption bundle of country R (that is, $\sum _{J=1}^N({\mathit{\omega }}^R{)}^{\mathrm{\prime }}{\mathbf{\text{M}}}^{RJ}{\mathbf{\text{f}}}^{JR}$); and the foreign labor, again, embodied in the consumption of R (that is, $\sum _{K\ne R}\sum _{J=1}^N({\mathit{\omega }}^K{)}^{\mathrm{\prime }}{\mathbf{\text{M}}}^{KJ}{\mathbf{\text{f}}}^{JR}$). The latter term gives the imports of labor.

The question is whether the employment footprint ${\phi }^R$ in country R is larger (or smaller) than the actual labor force in this country. If larger, country R can be said to consume more labor than it actually has (given the current trade structure). The actual labor force is given by the number of workers employed in country R (i.e. $({\mathbf{\text{l}}}^S{)}^{\mathrm{\prime }}\mathbf{\text{e}}$, with $\mathbf{\text{e}}$ the summation vector consisting of ones) and unemployment in workers. Unemployed workers are workers not currently engaged in productive activities but actively searching for a job. The difference between the employment footprint and the number of employees (excluding unemployed workers) is equivalent to the imports of labor minus the exports of labor.

We also take worker productivity into account. We assume that differences in production technologies across countries are reflected by differences in factor costs. We calculate wage rates of different categories of workers by dividing the labor payments to these workers by the total number of workers of each category. Then we ‘translate’ the imports of foreign workers into so-called US-efficient workers. To illustrate this, suppose that we find that US consumption requires 360 high-skilled local workers in the Indian automobile industry. Suppose also that the wage rate of high-skilled workers in the US automobile industry is twice that of India. Then, the money that is paid for the 360 Indian high-skilled workers would have paid for only 180 US workers. We thus assume that the output produced by two high-skilled Indian workers in the automobile industry (using Indian technology) can be replaced by output produced by one US worker (using US technology). Dividing the number of high-skilled Indian workers in the automobile industry by two gives the number of US-efficient workers. This procedure is repeated for the foreign workers of all skill-types (high-, medium-, and low-skilled), industries and countries. This procedure is unbiased if labor payments perfectly reflect differences in labor productivity.

Our results depend on the assumptions related to productivity differences. We perform two robustness checks. First, in order to find out how sensitive these results are, we create lower and upper productivity bounds of –20% and +20%, respectively, after which we estimate effective US workers embodied in the US employment footprint. So, if a US worker is twice as productive as an Indian worker, we consider a lower bound where 1 US worker = 0.8×2 = 1.6 Indian workers, and an upper bound of 1 US worker = 1.2 × 2 = 2.4 Indian workers.

In our second robustness check, we take true labor productivity differentials into account. Here, labor productivity is defined as the industry's value added per worker (in that industry), which is used instead of wage rates to proxy the differences in labor productivities between US and foreign workers. Value added per worker is a broader measure that includes not just wages but also capital earnings. It is calculated by dividing value added at the industry-level by all workers employed in this industry (separately for each country). There is no disaggregation by worker skill-type because it is not possible to allocate capital earnings to workers of different skill-levels. Therefore, it is more appropriate to compare the US worker surplus based on the new value-added proxy of labor productivity with the corresponding surplus using an adapted wage-rate-based proxy. The adaptation is that the wage-rate-based proxy also does not differentiate between skill-levels (but only between industries).

For the feasibility check, we examine whether country R (i.e. the US) has – technically speaking – sufficient labor available to sustain its consumption without any trade. We calculate the footprint under the assumption that the country has the same technology and consumption as before, but no trade. That is, in the self-sufficiency case we assume that country R is forced to act under autarky with the means it has available. The employment footprint then becomes (3) ${\overline{\phi }}^R=({\mathit{\omega }}^R{)}^{\mathrm{\prime }}(\mathbf{\text{I}}-{\overline{\mathbf{\text{A}}}}^R{)}^{-1}{\overline{\mathbf{\text{f}}}}^R$(3) where ${\overline{\mathbf{\text{A}}}}^R=\sum _{J=1}^N{\mathbf{\text{A}}}^{JR}$ is the matrix with technical input coefficients. For instance, it gives how many dollars of steel are necessary for one dollar of car production, irrespective of the origin of the steel. In a situation of self-sufficiency, country R has to produce everything itself, including the inputs that go into the production (and the inputs that go into the inputs, etcetera). Similarly, ${\overline{\mathbf{\text{f}}}}^R=\sum _{J=1}^N{\mathbf{\text{f}}}^{JR}$ gives the consumption in country R under the assumption that all goods and services are now produced at home. Note that under self-sufficiency all the labor involved in producing for other countries in the current trade structure is released. It is now available for domestic production, including production to substitute for ‘lost’ imports.

We do not incorporate additional constraining factors in the analysis – such as the possible endogeneity of consumption to trade – because our goal is not to simulate autarky. This would not be possible with our input–output approach.

4. Data sources

Our primary data source is the 2013 version of the World Input-Output Database (WIOD). This database contains a time-series of world input-output tables (WIOTs) for the period 1995–2011.⁵ There are 40 countries and 35 industries included, covering more than 85% of world GDP, in addition to the ‘Rest of World’.⁶ The WIOTs were compiled by harmonizing national supply-use tables (or national input–output tables) and combining them with detailed international trade data. This provides a single and consistent source of global trade linkages involving intermediate and final trade flows between all industries and countries.⁷ The WIOD’s Socio-economic accounts (SEAs) give additional information on employment and wage shares of three skill categories of workers. Employment is defined as ‘all persons engaged’, which includes paid employees, the self-employed, part-time, and informal workers (the latter has been estimated). These disaggregated employment and factor payment data are available for 35 industries and 41 countries, which includes the ‘Rest of World’.⁸ All data have been harmonized to ensure international comparability and compatibility with the WIOTs. Differences in labor productivity across countries will be approximated by the factor cost data at the industry and skill-level.

To estimate the labor force (or domestic labor endowment), we used data on unemployed workers, which are then added to the number of employees engaged. For unemployment data, we use the International Labor Organization (ILO, 2015) database. In most cases, unemployment data were based on Labor Force Surveys (LFS). Whenever the ILO provided multiple data sources for unemployment, LFS data were preferred over official national sources because LFS use a consistent methodology across countries. The ILO also provides unemployment data as unemployment ratios, which we use as a robustness check. In this way, we were able to obtain data for nearly all countries and years in the WIOD database.

Although the data cover the years 1995–2011, our analysis focusses on the period until 2008 for two reasons. First, labor and capital compensation data are not available for non-EU countries in 2010 and 2011. This precludes productivity comparisons between EU and non-EU countries based on using factor costs as a proxy of worker productivity. Second, 2009 and 2010 were the years most influenced by the global financial crisis and thus may taint our results because of the general contraction of trade. This could imply that the worker composition of the consumption bundle became less diversified, with a higher share of domestic workers compared to previous years.

Employment data in numbers of workers also reflect labor participation rates, which are to some extent culturally determined and may differ across countries. Since the hours worked per worker differ across countries, we considered using hours worked instead of number of workers as our unit of measurement. This information is also provided by the SEAs, but data in hours are incomplete because only the EU, US, Japan, and other advanced countries (with the exception of China) have reliable data on this variable. The SEA creators derived data on hours worked for all other countries contained in the WIOD directly from employment data in numbers of workers (which does thus provide the same information). In addition, data on hours worked in the ‘Rest of World’ is not provided by the SEAs and would be difficult to estimate. Therefore, we use number of workers – initially without any cross-country productivity adjustment – as our baseline unit of measurement.

5. Results

We start by determining the composition of the US employment footprint. Then we evaluate the self-sufficiency of the US in terms of labor, followed by robustness and sensitivity checks.⁹

5.1. US employment footprint

We first use the employment footprint concept to compute the number of workers directly and indirectly embodied in the consumption bundle of the US from 1995 to 2008.

Table 2 displays the worker composition of the US employment footprint. The table shows the number of domestic and foreign workers embodied in US consumption (Table 2(A)), and their shares (Table 2(B)) in different years of the time period. In 1995, approximately 203 million workers were worldwide engaged, directly and indirectly, to produce the goods and services consumed in the US in that year. 124 million (61.1%) of them were domestic US workers and almost 79 million (38.9%) were foreign workers. Of all workers in the US employment footprint, only 9 million workers (4.5%) were from advanced nations outside of the US (EU-27 and other developed countries), while the remaining nearly 70 million foreign workers (34.3%) were from all other countries (including mostly developing and emerging countries).¹⁰ China alone provided more than 25 million workers or 12.6% of all the labor required to produce the US consumption bundle.

TABLE 2. US employment footprint, by region.

	1995	2000	2005	2006	2007	2008	Change 2008–1995
A. Workers embodied in US consumption (in thousands of workers)
US workers (1)	124,106	138,057	140,053	141,750	142,076	140,874	16,768
All foreign workers (2)	78,931	118,665	142,285	144,441	132,622	123,472	44,541
EU-27 workers	4066	6635	5662	5636	5161	5004	938
Other developed	5157	6827	6364	6267	5930	5428	271
China	25,562	29,574	45,172	46,337	43,641	37,607	12,046
Other emerging	15,638	27,913	28,841	29,905	24,951	22,283	6644
Rest of World	28,508	47,717	56,246	56,297	52,939	53,150	24,642
Footprint (3) = (1) + (2)	203,037	256,722	282,337	286,191	274,698	264,346	61,309
Notes: (1) shows the US workers embodied in US consumption (in thousands) and (2) shows the foreign workers embodied in US consumption (in thousands). The last column shows the change (in thousands of workers) between 1995 and 2008.
B. Workers embodied in US consumption (as percentage of all embodied workers)
US workers (1)	61.1	53.8	49.6	49.5	51.7	53.3	–7.8
All foreign workers (2)	38.9	46.2	50.4	50.5	48.3	46.7	7.8
EU-27 workers	2.0	2.6	2.0	2.0	1.9	1.9	–0.1
Other developed	2.5	2.7	2.3	2.2	2.2	2.1	–0.5
China	12.6	11.5	16.0	16.2	15.9	14.2	1.6
Other emerging	7.7	10.9	10.2	10.4	9.1	8.4	0.7
Rest of World	14.0	18.6	19.9	19.7	19.3	20.1	6.1
Total	100	100	100	100	100	100

Notes: (1) shows the US workers embodied in US consumption (as a percent share of all workers embodied in US consumption) and (2) shows the foreign workers embodied in US consumption (as percentage share of all workers embodied in US consumption). The last column shows the percentage point change in the shares between 1995 and 2008.

In 2008, the year just preceding the economic crisis, 264 million workers were embodied in US consumption (61 million workers more than 1995). The share of workers originating from developing countries (i.e. China, Other emerging, and ‘Rest of World’) rose. The total share of foreign workers peaked in 2006 with 50.5% and increased by 7.8 percentage points overall from 1995 to 2008, but the share of workers from emerging and developing countries increased by 8.4 percentage points. This implies that the share of workers from all advanced countries decreased by 8.4 percentage points (7.8 percentage points in the US itself, 0.6 percentage points in the EU-27 and Other developed countries). While the number of domestic US workers increased in absolute terms, from 124 to 141 million, the share of US workers in all embodied workers decreased to 53.3% in 2008 (and was as low as 49.5% in 2006). This is indicative of large and increasing US imports of labor.

There are two main findings coming out of Table 2. First, the workers involved in producing for US consumption diversified over the period because reliance on domestic workers declined (in relative terms). This is consistent with the increasing international specialization patterns in trade in that period. These findings are in line with Alsamawi et al. (2014) who find that consumption of developed countries is facilitated by large embodied labor flows from low-wage countries. They argue that this points to wage inequalities and ‘master-servant relationship’ patterns in many global supply chains.

The second observation is that the share of Chinese workers was large but only marginally increased in the period of observation from 12.6% in 1995 to 14.2% in 2008, peaking at 16.2% in 2006. Almost one-third of all foreign workers embodied in US consumption were Chinese, which indicates the relevance of China. However, the share of Chinese workers in all foreign workers in the US footprint decreased from 32% in 1995 (= 25,562/78,931) to 30% in 2008 (= 37,607/123,472). This trend was due to the large and growing share of ‘Rest of World’ workers. The share of ‘Rest of World’ workers in all foreign workers embodied in the US footprint (which were mainly workers from developing and emerging countries) increased from 36% (=28,508/78,931) to 43% (=53,150/123,472) in the time period.

Although the number (and the share) of foreign workers that is embodied in the US consumption bundle is impressive, it should be born in mind that the result is probably biased because productivity differences have not been taken into account. In general, the foreign workers that were imported by the US are less productive than domestic US workers, and thus have a smaller relative impact in terms of producing the US consumption bundle than it would appear based on Table 2. We will correct for productivity differences at a later stage.

The current debate on US job losses is focused on lost jobs in manufacturing. Table 3, therefore, splits the workers according to the industry group (manufacturing, services, other) they are working in. It gives the US manufacturing workers embodied in the US consumption bundle, the imports of manufacturing workers (i.e. foreign manufacturing workers embodied in US consumption), and the export of manufacturing workers (i.e. US manufacturing workers embodied in foreign consumption).

TABLE 3. Who works for whom? (in thousands of workers).

		1995	2000	2005	2006	2007	2008	Change 2008–1995
Manufacturing	US workers for US consumption	14,873	14,994	12,228	11,921	11,419	10,674	–4200
	Import of foreign workers	26,307	42,874	51,186	49,745	45,753	41,405	15,098
	Export of US workers	3744	3775	3205	3431	3638	3852	107
Services	US workers for US consumption	98,919	111,524	115,744	117,271	118,284	118,405	19,485
	Import of foreign workers	15,351	25,969	31,248	32,344	29,991	28,770	13,420
	Export of US workers	5364	5359	5161	5599	6247	6305	942
Other	US workers for US consumption	10,313	11,539	12,081	12,558	12,373	11,795	1482
	Import of foreign workers	37,273	49,821	59,851	62,351	56,878	53,297	16,024
	Export of US workers	766	525	484	540	606	712	–53
Total	US workers for US consumption	124,106	138,057	140,053	141,750	142,076	140,874	16,768
	Import of foreign workers	78,931	118,665	142,285	144,441	132,622	123,472	44,541
	Export of US workers	9873	9659	8850	9570	10,491	10,869	996

Notes: The first row next to each sector shows the number of US workers of that sector embodied in US consumption (in thousands); the second row next to each sector shows the number of foreign workers of that sector embodied in US consumption (in thousands); and the third row next to each sector shows the number of US workers of that sector embodied in foreign consumption (in thousands). The last column shows the change (in thousands of workers) between 1995 and 2008.

The number of US manufacturing jobs in US consumption initially increased until 1998 and then declined steadily over time, shrinking by 4 million between 1995 and 2008. This large reduction in US manufacturing jobs was not compensated by new US manufacturing jobs in foreign consumption. At the same time, 15 million manufacturing jobs were created overseas to directly or indirectly serve US consumption. The number of services jobs increased by almost 33 million in the time period. Nearly 20 million of them were newly created US jobs and 13 million were newly offshored services jobs.

Focusing on the number of US workers embodied in foreign consumption bundles, the results show that in 2008 nearly 11 million US workers were involved in this – more than 6 million of which in services. The changes in these numbers over time indicate that nearly 1 million new services jobs were induced by exporting related activities between 1995 and 2008. These 1 million new jobs could contribute to (partially) replace domestic jobs that may have been lost to globalization and import competition.

Thus, the perspective of lost jobs in manufacturing due to trade is incomplete because in a global value chain (GVC) context job losses are offset by job gains in other sectors. The data strongly suggest that this is propelled by a transition from manufacturing to services jobs. This pattern is consistent with a study by Timmer et al. (2013), which found that advanced nations were increasing their competitiveness by transitioning to (high-skilled) services activities. Note that the numbers do not yet reveal anything about the skill-type (quality) of jobs gained or lost by US and foreign workers. We discuss skill-distributions next to check the activities in which advanced nations specialize within production networks.

Table 4 splits all workers embodied in the US consumption bundle into two groups: domestic workers and foreign workers. For each of the workers, we can distinguish three skill-levels. They are defined by educational attainment categories using skill-shares provided in the WIOD Socio-economic Accounts (medium- and high-skilled workers typically correspond to workers having at least a secondary education, e.g. High School and vocational training). In 1995, 89% of US workers involved in US consumption were either medium- or high-skilled, compared to only 29% of the foreign workers. These numbers in 2008 were 91% and 37%, respectively. Among all workers embodied in US consumption, domestic US workers were highly educated relative to their foreign peers. Highly skilled workers typically have better salaries, and in Ricardian theory cross-country income differences are expected to reflect differences in labor productivity. This suggests that there may be significant productivity discrepancies between domestic and foreign workers. These discrepancies may, when properly accounted for, influence the capacity of US labor to be self-sufficient. We use this insight to take a deeper look at the impact of productivity in Section 5.2.

TABLE 4. US domestic and foreign workers in US consumption by skill-level (as percent of all domestic and foreign employment, respectively, embodied in US consumption).

	1995	2000	2005	2006	2007	2008	Change 2008–1995
Domestic workers
Low-skilled (1)	10.9	10.3	10.1	10.1	9.6	9.0	–1.9
Medium-skilled (2)	64.0	62.4	60.9	60.4	60.3	60.1	–3.9
High-skilled (3)	25.1	27.3	29.0	29.6	30.1	30.9	5.8
Total	100	100	100	100	100	100
Foreign workers
Low-skilled (1)	70.6	64.9	63.5	63.3	63.3	62.8	–7.8
Medium-skilled (2)	25.0	29.1	29.7	29.4	29.3	29.5	4.5
High-skilled (3)	4.5	6.0	6.8	7.3	7.4	7.7	3.3
Total	100	100	100	100	100	100

Notes: The top part of Table 4 indicates the domestic workers embodied in US consumption and the bottom part indicates the foreign workers embodied in US consumption with percent shares in each case summing to 100. The last column shows the percentage point change in the worker shares of each skill-level between 1995 and 2008.

5.2. Comparing the US employment footprint with US labor endowments

We use three different approaches to compare the labor that is necessary to sustain the actual US consumption bundle with US labor endowments (or labor force). In the first comparison, we consider only the raw (absolute) number of workers in the US employment footprint and do not account for differences in worker productivities between countries. Table 5 displays the employed US workers (row 1), the unemployment US workers (row 2), the US labor endowment (row 3) and the total number of workers embodied in US consumption, i.e. the US employment footprint (row 4). The difference in the endowment minus footprint gives the worker surplus and equals the difference between US exports of labor (US workers embodied in foreign consumption) plus all unemployed US workers in the labor force on the one hand and US imports of labor (foreign workers embodied in US consumption) on the other hand. For 2008, for example, we find that there is a worker deficit of –103,679 (Table 5, row 7). This is the same as exports of US labor (10,869, in Table 3) plus unemployment (8924) minus imports of labor (123,472, also in Table 3).

TABLE 5. Employment footprints and endowments (top part, in thousands of workers), the worker surplus (in thousands of workers and, in parentheses, as percentage of the labor force, middle part), and the self-sufficiency consumption rate (bottom part).

	1995	2000	2005	2006	2007	2008
(1) Employed workers	133,979	147,717	148,903	151,319	152,567	151,743
(2) Unemployed workers	7404	5692	7591	7001	7078	8924
(3) Labor endowment (employed + unemployed)	141,383	153,409	156,494	158,320	159,645	160,667
(4) US employment footprint (using the actual trade structure)	203,037	256,722	282,337	286,191	274,698	264,346
(5) US employment footprint (after efficiency adjustments)	136,379	152,236	155,944	157,909	158,268	157,528
(6) US employment footprint (self-sufficiency case)	135,753	152,547	155,188	157,591	158,065	157,151
Worker surplus (as a share of the labor endowment)
(7) Actual trade structure (no adjustments)	–61,653	–103,314	–125,843	–127,870	–115,053	–103,679
	(–43.6)	(–67.3)	(–80.4)	(–80.8)	(–72.1)	(–64.5)
(8) Actual trade structure (after efficiency adjustments)	5005	1172	550	412	1378	3139
	(3.5)	(0.8)	(0.4)	(0.3)	(0.9)	(2.0)
(9) As in (8) but with a 5% unemployment rate	–2064	–6498	–7275	–7504	–6605	–4894
	(–1.5)	(–4.2)	(–4.6)	(–4.7)	(–4.1)	(–3.0)
(9a) Actual trade structure, –20% bound	–5133	–10,043	–11,248	–11,544	–10,652	–9058
	(–3.6)	(–6.5)	(–7.2)	(–7.3)	(–6.7)	(–5.6)
(9b) Actual trade structure, +20% bound	–19	–4135	–4626	–4811	–3906	–2119
	(–0.0)	(–2.7)	(–3.0)	(–3.0)	(–2.4)	(–1.3)
(9c) Actual trade structure, no skills disaggregation.	–578	–4709	–4892	–5128	–4218	–2455
	(–0.4)	(–3.1)	(–3.1)	(–3.2)	(–2.6)	(–1.5)
(9d) as in 9c, but using value added per worker	–1144	–6368	–6113	–6838	–5848	–3966
	(–0.8)	(–4.2)	(–3.9)	(–4.3)	(–3.7)	(–2.5)
(10) Self-sufficiency case	5630	861	1307	729	1580	3516
	(4.0)	(0.6)	(0.8)	(0.5)	(1.0)	(2.2)
(11) As in (10) but with a 5.0% unemployment rate	–1439	–6809	–6518	–7187	–6403	–4517
	(–1.0)	(–4.4)	(–4.2)	(–4.5)	(–4.0)	(–2.8)
(12) Self-sufficiency consumption rate	98.7	96.8	96.0	96.0	96.5	96.6

Notes: (3) = (1) + (2); (7) = (3) – (4) and 100×[(3) – (4)]/(3); (8) = (3) – (5) and 100×[(3) – (5)]/(3); (10) = (3) – (6) and 100×[(3) – (6)]/(3); (12) = 100×[(1)/(6)]. 9a–9d are sensitivity checks related to the baseline results in 9 with 5% unemployment. In 9a foreign workers are assumed to be 20% more efficient relative to US workers than in the baseline case (using a wages proxy); in 9b foreign workers are assumed to be 20% less efficient relative to US workers than in the baseline case (using a wages proxy); in 9c wage rates are used to proxy labor productivity at the country/industry level only (not at the skill-level); and in 9d labor productivities are based on value added per worker (including capital compensation) at the country/industry level only (not at the skill-level).

In 1995, there was a discrepancy between endowment and footprint of 62 million workers. In other words, there is a labor deficit, which was 43.6% of the US labor force. This labor deficit share grows to 80.8% by 2006. This says that US consumers become more dependent on imports of labor over time. Clearly, the US consumed more labor than it had nationally available and could not be self-sufficient in terms of labor if workers have the same productivity across countries. Thus, there were positive benefits from trade in both cases (of 43.6% and 80.8%, respectively). However, given our insights from standard trade theory and the literature, the assumption that all workers have the same productivity across countries is an overly simplistic assumption because there are productivity differences.

In the second comparison of employment footprints and labor endowments we adjust for productivity differences between US and foreign workers following the wage rate-based proxy of productivity described in the methodology section.

For instance, for 1995 the imports of 79 million workers (Table 3) are equivalent to only 12 million US-efficient workers. This implies that US workers are found to be on average 6.4 times as productive as foreign workers. This relative wage discrepancy of foreign to US workers is identical to what Alsamawi et al. (2014) find based on the US employment footprint using the Eora database. This ratio increased over time to 7.4 in 2008, indicating an influx of less productive workers (especially from ‘Rest of World’). The average productivity difference between US and foreign workers hides an enormous variation across countries in the productivity differences of workers imported by the US. For example, based on our wage rate proxy of productivity, the average US worker was found to be about as productive as the average Australian worker, but 14 times more productive than an average ‘Rest of World’ worker. The import of US-efficient workers was only a little larger than the export of US workers, implying that trade was fairly balanced (when compared to the unadjusted numbers of workers).

US employment footprints after correction for productivity differences are given in row (5) of Table 5. For example, producing the US consumption bundle involved 158 million US-efficient workers globally in 2008. These were 141 million US workers and 17 million foreign US-efficient workers. The 141 million US workers embodied in the own consumption plus the 11 million US workers that are exported (and embodied in foreign consumption) make up the 152 million employed US workers. The US consumed in 2008 more workers than it had people employed. Made possible by trade, the US had net imports of 6 million US-efficient workers. Because there are 9 million US workers unemployed, the worker surplus is still 3 million US workers. As percentage of the labor force (Table 5, row 8), this is 2.0% in 2008. In terms of US-efficient workers, the worker surplus declined from 3.5% in 1995 to 0.3% in 2006 before rising again.

The results (Table 5, row 8) assume that all unemployed workers are willing and able to accept a job if so required. This is not realistic because of some frictional unemployment. In 2008 this implies that almost 6 million (of the 9 million, see row (2)) unemployed workers would have needed to take up a job. If we assume a natural unemployment rate of 5% as most economists do, only 1 million workers would be able to accept a job.¹¹ So, there would have been a true shortage of 5 million workers in 2008, implying that the US would not have been able to produce its consumption bundle. As a share of the labor endowment (Table 5, row 3), this gives the –3.0% reported in row (9).

In this more realistic scenario of a 5% unemployment rate, the US could not be self-sufficient in any year. The data in row (9) show that for all years between 1995 and 2008, the number of US workers required was always larger than the number of employed workers. While unemployed workers would theoretically be sufficient to bridge this gap if they all became employed, unemployment would have to fall to unrealistically low levels in many of these years for the US to remain self-sufficient. Even if we assume a 3.7% unemployment rate instead of 5%, which equals the lowest observed unemployment rate in the 1995–2008 time period (in 2000), the worker surplus would be close to 0% in the first few years and clearly negative in all other years.

In the third and last comparison of employment footprints and labor endowments we consider the feasibility check. We calculate the employment footprint, assuming that the US produces everything it consumes (i.e. with no trade) by using only its own technology and labor. This is the footprint given by ${\overline{\phi }}^R$ in eq. 3. Included are the production of the current imports and all foreign indirect inputs into the US production process. On the other hand, all US workers that were previously embodied in foreign consumption bundles are freed up now. The results for this self-sufficiency case are given in row (6) of Table 5. The US employment footprint in this case shows that about 157 million US workers would be required in 2008, which is about 3.5 million less than the US labor endowment of that year. The results imply that if the US used all of its workers in an optimal way, including unemployed workers, the US would have a slight 2.2% labor surplus (as a share of its labor force). In other words, unemployment would have fallen to 2.2% and the US could have sustained the same consumption levels it actually achieved that year. This worker surplus declined from 4.0% in 1995 to 0.5% in 2006 before increasing. This means that the US would only manage to be self-sufficient in 2006 if at least 90% of all unemployed workers would accept a job.¹² This would assume little to no frictional unemployment.

The last row in Table 5 gives the self-sufficiency consumption rate. It reflects benefits from trade and is defined as the employed workers in row (1) divided by the self-sufficiency footprint in row (6). The rationale is as follows. The 2008 consumption bundle requires in the actual situation with the current trade structure almost 152 million US workers. Some of these workers are exported and traded for foreign workers. Through this trade in embodied workers, the US is able to consume its 2008 bundle. In the case of self-sufficiency, however, the same bundle would require 157 million workers. The current 152 million employed workers would be able to generate a consumption bundle that is only 96.6% ($\approx$ 152/157). In most years, the loss of consumption is much larger and the lowest rate is found for 2005–2006 (96.0%).

Finally, note that our analysis may still overstate the self-sufficiency prospects of the US if additional factors are considered. For example, if the US did not import from China and Mexico, many consumer goods would have been much more expensive for US consumers. They would buy less, and US employment would be lower. In addition, in a world with internationally fragmented production processes, US export competitiveness is also determined by finding the cheapest, most reliable, and most flexible suppliers of intermediate inputs, which are often not located in the US itself. Therefore, factor prices (and hence input coefficients) will differ under self-sufficiency due to the switch from foreign to possibly less efficient domestic suppliers of intermediate inputs.

5.3. Robustness: productivity

Table 5 provides the results with respect to upper and lower productivity bounds in rows (9a) and (9b). For example, row (5) of Table 5 shows that 155,944 thousand US-efficient workers were embodied in the US employment footprint in 2005. At the –20% bound, 4164 thousand more US-efficient workers would be involved. As a share of the labor endowment in row (3), this is 2.6 percentage points more. Because more labor is needed, the labor shortage becomes larger. The labor deficit of –4.6% in row (9) of Table 5 is thus decreased by 2.6 percentage points, which yields –7.2% in row (9a). At the +20% bound, we find that 2776 thousand fewer US-efficient workers would be involved. This means that the labor shortage is relaxed and the –4.6% deficit reduces by 1.6 percentage points to –3.0% in row (9b). The results of the different calculations of US-efficient workers were similar also in other years in the period. Overall, the range in the respective worker deficits (given by the calculations with the –20% bound and the +20% bound) was always less than 4.5 percentage points and thus relatively small. What is more important, the large range of productivity differences does not affect our main finding that there was a labor shortage throughout the entire period and that this shortage grew until 2006 after which it declined in 2007 and 2008.

Row (9c) of Table 5 shows the US worker surplus using our adapted wage-rate-based proxy of labor productivity and row (9d) shows the revised US worker surplus using value added per worker. The worker surplus based on both the wage-rate proxy and the value-added proxy were negative throughout the period. Though the deficits were similarly negative in 1995 (–0.4% using wages vs. –0.8% using value added in 1995) this gap widened between 1995 and 2001 as the value-added proxy became more negative relative to the wage-rate proxy. For example, in 2001 the worker deficit was –2.3% using wages vs. –3.5% using value added, and this gap remained in later years. Our interpretation is that this reflects the growing US import of foreign workers from developing and emerging countries (see also Table 2(B)). Poorer countries typically have higher shares of capital relative to labor in total factor income (GDP) because workers have comparatively low wages. Estimating labor productivity differences ‘pure’ by using value added per worker (with capital compensation included) increases the measured productivities of workers from low-wage countries. Therefore, more US-efficient workers are embodied in the US employment footprint, and as imports from developing countries grew over time, the US worker deficit would be expected to increase relative to the baseline results.

The differences in both robustness checks to the baseline results are small, in the sense that they do not change our main findings. That is, despite the considerable changes in the assumptions regarding productivity, all results pointed at a worker shortage throughout the entire period. In all calculations, the shortage was found to increase until 2006, after which it declined in 2007 and 2008. The online supplementary materials file contains additional robustness checks related to the way unemployment data are obtained, the productivity of ‘Rest of World’ workers is measured, and assumptions regarding the substitutability of workers of different skill-types in autarky.

5.4. Role of China

In the discussion of trade-related job losses, the role of China stands out (Autor et al., 2013). Table 6 shows that there were only 2469 thousand Chinese workers in the US employment footprint in 2008 when measured in US-efficient workers. This equates to just 1.6% of all US-efficient workers embodied in the footprint. The raw numbers before efficiency adjustments are much higher. So, by taking productivity differences into account, the impact of China (in terms of the share of Chinese workers embodied in the US footprint) is small, but grows rapidly over time (from 0.6% in 1995 to 1.6% in 2008). The analysis thus confirms the growing impact of China, which is in line with existing studies investigating the effects of import competition from China (e.g. Autor et al., 2014; Caliendo et al., 2019). At the same time, the potential threat of job losses is limited given the small share of Chinese workers in US-efficiency terms.

TABLE 6. US imports of Chinese workers (in thousands of workers).

	1995	2000	2005	2006	2007	2008	Change 2008–1995
In thousands of Chinese workers
No efficiency adjustments	25,562	29,574	45,172	46,337	43,641	37,607	12,046
Including efficiency adjustments	812	1019	1939	2149	2341	2469	1657
As percent share of US employment footprint
No efficiency adjustments	12.6	11.5	16.0	16.2	15.9	14.2	1.6
Including efficiency adjustments	0.6	0.7	1.2	1.4	1.5	1.6	1.0

Note: The raw numbers before efficiency adjustments are the same as in Table 2(A,B).

Table 7 breaks down the US jobs that were created by consumption in foreign countries. US-China trade created 548 thousand new US jobs (via the increase in US exports) between 1995 and 2008. This is much more than for any other country and made up about half of the total increase in US jobs due to trade. About 319 thousand of these jobs were in services. In absolute numbers, the import of Chinese workers in 2008 was equivalent to 2.5 million US workers whereas the export of US workers to China was 0.8 million workers.

TABLE 7. Exports of US workers by destination country (in thousands of workers).

	1995	2000	2005	2006	2007	2008	Change 2008–1995
China	232	280	459	548	640	780	548
Mexico	513	945	666	701	736	737	224
India	65	75	146	187	219	232	167
Great Britain	459	554	500	591	645	602	143
Canada	1105	1200	1177	1203	1260	1243	138
Rest	7499	6605	5903	6340	6991	7275	–224
Total	9873	9659	8850	9570	10,491	10,869	996

Notes: The number of US workers embodied in foreign consumption of the country in each row (in thousands). The last column shows the absolute change between 1995 and 2008. Note that the table provides a disaggregation of the total exports of US workers, also shown in the last row in Table 3, by country.

Hence, while trade with China may have cost net US jobs, the numbers are not as impressive as conventional studies like Autor et al. (2013) suggest – because newly created jobs compensate for direct job losses caused by import competition. The public sentiment that China is causing massive job losses is therefore not correct because the discussion only focusses on job losses whilst newly created jobs are ignored.

6. Conclusions

Globalization can lead to job losses. One way of illustrating the effects of globalization on labor markets is to calculate so-called employment footprints. Global employment footprints are defined as the total number of workers (domestic and foreign) directly or indirectly involved in producing all final goods and services consumed in a country. These footprints help to shed new light on the employment implications of trade – including the dependence of countries on foreign labor and the threat of foreign workers for domestic jobs. We used the employment footprint concept to systematically evaluate the (hypothetical) self-sufficiency of domestic US labor by assuming US technology and consumption remains the same but that there is no trade.

In the US case study, we show that on the import side the share of foreign workers in the US employment footprint grew steadily over time. The trend was largely driven by workers induced in emerging and developing countries. Furthermore, most of the foreign workers were shown to be low-skilled. On the export side, we find that many new jobs were gained especially in the services industries. To illustrate the dependence on foreign labor, we show that the net imports of foreign workers by the US are larger than the available labor if we maintain the assumption of 5% frictional unemployment. This implies that the US does not have enough domestic workers to replace the imported foreign workers. This means that if the US would not trade, its domestic workers would not be able to produce the US consumption. With the same number of workers, the consumption bundle that would be possible under self-sufficiency was 1.3% smaller than the actual US consumption bundle in 1995, and 3.4% smaller in 2008. This indicates that the US clearly benefits from trade.

Our results complement those of Alsamawi et al. (2014), who focus on wage inequalities in relation to social responsibility. They also find, although in a different context, that high-income countries depend on low-income countries. Specifically, the US may not be able to sustain its actual consumption pattern without using (embodied) labor from foreign countries.

Dropping additional assumptions in the analysis would almost certainly further emphasize that US labor cannot be self-sufficient (in terms of maintaining consumption levels). For instance, price effects are important. Domestic inputs may not be perfect substitutes of foreign inputs even if differences in the production technologies between countries are accounted for.¹³ The finding that the US cannot maintain the same consumption levels under self-sufficiency (using only domestic workers) sends an important policy message: protectionism has clear drawbacks and some of the rhetoric in the public sphere from politicians and the popular press is misleading.

Despite widespread fears of a growing China rapidly taking away US jobs, we found that the effective share of Chinese workers embodied in the US employment footprint – after correcting for productivity differences – was small. We further suggest that future research should attempt to do this same analysis for more recent years (2008 onwards), once updated data on skills of workers become available.

Thus, while the evidence does suggest that some US jobs involved in producing for US consumption have been outsourced to China, we do not find that China had a particularly large negative impact on US jobs (and the same holds true for other developing countries). This is underlined by looking at the reverse perspective. US trade with China was responsible for the largest US job gains in terms of export-related activities between 1995 and 2008 (548 thousand new jobs), providing nearly 800 thousand US jobs overall in 2008. Therefore, it is equally important to consider how US labor has integrated in the world economy and the jobs that have been gained due to globalization.

Acknowledgements

We like to thank the two anonymous referees, Bart Los, Gaaitzen de Vries and Michiel Gerritse for their detailed comments and suggestions which greatly helped to improve this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

2 For an overview, see Foster and Stehrer (2010)

3 See also Stehrer (2012) who compares the demand-based measure of value-added flows (‘trade in value-added’) with the supply-based approach (‘value added in trade’) and applies them using the WIOD.

4 This follows standard IO methodology (see Miller & Blair, 2009). Bold-faced lower-case letters are used to indicate vectors, bold-faced capital letters indicate matrices, and italic lower-case letters indicate scalars (including elements of a vector or matrix). Subscripts indicate industries and superscripts indicate countries. Vectors are columns by definition, row vectors are obtained by transposition, denoted by a prime (e.g. ${\mathbf{\text{x}}}^{\mathrm{\prime }}$). Diagonal matrices are denoted by a circumflex (e.g. $\widehat{\mathbf{\text{x}}}$).

5 Although a new version of the WIOD was released in 2016 (with a time-period from 2000 to 2014, and which includes three more countries), we use the 2013 release because the Socio-economic accounts accompanying the 2016 release did not update employment data by skill-levels, which are crucial for our analysis. The OECD’s Inter-Country Input–Output (ICIO) tables also have an accompanying Trade in employment (TiM) database. However, the WIOD data was chosen rather than the OECD data for three reasons: (i) the OECD has not released the complete dataset publicly (employment figures published online are rounded to the nearest 100), (ii) the OECD provides no employment estimations for the ‘Rest of World’, which is also crucial for our analysis, and (iii) the OECD employment database includes no information on skill-levels.

6 A list of all countries is shown in Table A3 of the online supplementary materials. The 35 industries are based on the NACE revision 1, which corresponds to ISIC revision 3.

7 The WIOD is available at www.wiod.org. For more details on its construction, see Dietzenbacher et al. (2013) and Timmer et al. (2015).

8 The data in the satellite accounts are based on information available from the National Accounts and material collected from employment and labour force statistics. For more details on the construction of the satellite accounts, see Erumban et al. (2012). Industry-level employment for the ‘Rest of World’ aggregate (making up about 15% of world GDP but more than 30% in terms of world employment) is not included in the publicly released SEAs; however, confidential estimations, including a breakdown by skill-level, were gratefully provided by Gaaitzen de Vries. Gaaitzen de Vries was involved in the development of the WIOD satellite accounts. The importance of ‘Rest of World’ workers as a share of world employment also implies that they play a large role in calculating the relative productivity of foreign workers in the US employment footprint. Therefore, we provide a robustness check for our wage-rate proxy of ‘Rest of World’s productivity in the online supplementary materials based on the relative wages of workers in the largest emerging countries in the WIOD.

9 Self-sufficiency results for 39 other countries are provided in the online supplementary materials.

10 Other advanced countries include Australia, Canada, Japan, Korea, and Taiwan; Other emerging countries (i.e. other than China) include Brazil, India, Indonesia, Mexico, Russia, and Turkey; the ‘Rest of World’ aggregate combines all other countries in the world not included in any of the other categories, such that the employment footprint covers all countries.

11 That is, (160,667 thousand)*0.05 = 8 million persons remain unemployed under self-sufficiency, where 160,667 thousand is from row (3) in Table 5.

12 The worker surplus reduces to 729 thousand workers, which is 10.4% of the number of unemployed workers. Hence, 89.6% of the currently unemployed workers would have to accept a job offer.

13 For instance, foreign suppliers may be able to produce more cheaply due to economies of scale or country-specific idiosyncrasies, e.g. cultural factors, which are not accounted for by production technology. In addition, it has been shown that exporting firms are more productive and pay higher wages than non-exporters (Bernard et al., 2007). In the self-sufficiency case, all firms would be non-exporters. World IOTs are not currently detailed enough to distinguish between exporters and non-exporters, precluding any consideration of this in our analysis.