Policy Research Working Paper 10433 The Role of Global Value Chains for Worker Tasks and Wage Inequality Piotr Lewandowski Karol Madoń Deborah Winkler Macroeconomics, Trade and Investment Global Practice May 2023 Policy Research Working Paper 10433 Abstract This paper studies the relationship between participation in routine task intensity in non-offshorable occupations in global value chains, worker routine task intensity, and with- business services. Higher worker-level routine task inten- in-country wage inequality. It uses unique survey data from sity is strongly associated with lower wages, so global value 47 countries across the development spectrum to calculate chains participation indirectly widens the within-country worker-level, country-specific routine task intensity and wage inequality through this routine task intensity chan- combines them with sectoral measures of backward and for- nel. At the same time, global value chains participation ward global value chains participation. Higher global value directly contributes to reduced wage inequality, except for chains participation is associated with more routine-in- the richest countries. Overall, this analysis finds that global tensive work, specifically among offshorable occupations, value chains participation reduces wage inequality in most especially in countries at lower development levels. The low- and middle-income countries that receive offshored results by broad sectors contrast sharply: higher global value jobs but widens wage inequality in high-income countries chains participation is linked to a higher routine task inten- that offshore jobs. sity in offshorable occupations in the industry but a lower This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at piotr.lewandowski@ibs.org.pl (corresponding author), karol.madon@ibs.org.pl and dwinkler2@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Role of Global Value Chains for Worker Tasks and Wage Inequality• Piotr Lewandowski♦ Karol Madoń♣ Deborah Winkler ♥ Keywords: routine task intensity, global value chains, globalization, cross-country division of work, wage inequality JEL: J21, J24, J31, F66 • We thank Sébastien Dessus, Maryla Maliszewska, Marta Palczyńska, Bob Rijkers, Ben Shepherd, Jorge Tudela-Pye and the participants of the World Bank authors’ workshops “Leveraging trade for more and better job opportunities in developing countries” for their useful comments. The work was carried out under the overall supervision of Sébastien Dessus and Antonio Nucifora. This paper was financially supported by the World Bank. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank Group and its affiliated organisations, its Executive Directors or the governments they represent. Karol Madoń’s contribution was financed by the National Science Centre, project no. 2021/41/N/HS4/03640, agreement no. UMO- 2021/41/N/HS4/03640. All errors are ours.  Institute for Structural Research (IBS), IZA, and RWI. Warsaw, Poland. e-mail: piotr.lewandowski@ibs.org.pl, corresponding author.  Institute for Structural Research (IBS), and SGH Warsaw School of Economics. Warsaw, Poland. E-mail: karol.madon@ibs.org.pl  Global Trade and Regional Integration Unit, World Bank Group, Washington, D.C. USA. E-mail: dwinkler2@worldbank.org i 1. Introduction Traditional trade theory predicted that countries’ specialization in trade affects the international division of labor. Wealthier countries which tend to be relatively more endowed with skilled labor and technology have had a comparative advantage in the exports of skill- and technology-intensive goods and services. In contrast, developing nations have been relatively more abundant in low-wage labor and natural resources, thus specializing in labor- and resource-intensive goods exports. Both types of countries exported the goods and services that use their relatively abundant factors more intensively. Recently, however, countries specialize in the exports of tasks they have a comparative advantage in (Grossman & Rossi-Hansberg, 2008) rather than final goods and services. Technological change and trade liberalization have fostered the possibility of trading tasks, offering opportunities to developing countries to participate and upgrade in global value chains (GVCs) (Taglioni & Winkler, 2016). The “second unbundling” of corporate tasks has intensified this division of labor (Baldwin, 2014), as routine tasks are easier to offshore (Blinder & Krueger, 2013), especially in manufacturing (Rodrik, 2013). The decline in routine jobs in the United States, European Union, and some emerging countries since the late 1980s contributed to the polarization of job opportunities within countries (Autor & Dorn, 2013; Cortes et al., 2017; Goos et al., 2014; Jensen & Kletzer, 2010; Michaels et al., 2014; Reijnders & de Vries, 2018; Spitz‐Oener, 2006). A GVC consists of a series of value-adding tasks, from inception to selling a product or service for final consumption (World Bank, 2020). Richer countries perform more non-routine tasks that require creativity, data analytics, or guiding people. In contrast, poorer countries specialize in routine-intensive tasks that are often repetitive, well-structured, and require being exact and accurate rather than creative (Figure 1). What is the role of GVCs in explaining the division of worker tasks across countries? What is the GVCs’ contribution to within- country differences in job tasks, and as higher routine task intensity is strongly associated with lower earnings (Autor & Handel, 2013; de la Rica et al., 2020), to wage inequality? Do various forms of GVC participation differ in this regard? Figure 1. The average routine task intensity (RTI), by countries’ development level (GDP per capita), accounts for cross- country occupational task differences. Note: for each task content, the 0 is set at the United States average value, and 1 corresponds to one standard deviation of RTI in the United States. GDP per capita in PPP, current international $, country averages for 2011–2016. Source: Lewandowski et al. (2022). 1 Drawing on a unique survey dataset, this study examines the existence and nature of linkages between GVC participation and the routine task intensity (RTI) of workers across 47 countries at all developmental stages. Specifically, this paper systematically assesses how the nature of GVCs mediates this relationship, accounting for differences across sectors and types of occupations, particularly offshorable and non-offshorable occupations. Moreover, we evaluate how GVCs contribute to the task structures in the domestic labor markets, and to within-country wage inequality, as measured by the Gini coefficient. We distinguish between the direct – through wages – and indirect – through RTI – contribution of GVC participation to wage inequality. The relationship between GVC participation and RTI is depends on a country’s factor endowments which determine its type of task specialization in GVCs. In developing countries such as Indonesia, a higher backward GVC participation, i.e., the share of imported inputs used in export production, may be associated with a higher worker-level RTI. Such countries tend to have abundant low-wage labor and specialize in the production tasks of basic manufacturing GVCs, typically the final assembly stage. Thus, they rely strongly on imported inputs which they process for their semi-final or final exports. However, high backward GVC integration also characterizes countries specializing in more advanced manufacturing and services GVCs. Such countries are endowed with skilled labor and perform some routine tasks (e.g. customer service or accounting) and some non-routine tasks (e.g., IT support) (World Bank, 2020). Examples include Central Eastern European countries (Czechia, Hungary, Poland).1 The type of GVC participation in East Asian and Central Eastern European countries contrasts sharply with that of many Sub-Saharan African or Latin American countries specializing in commodities – agriculture and mining (Hanson, 2017). These countries show low backward GVC participation as they predominantly export upstream GVC tasks with low reliance on imported inputs and fewer opportunities to innovate and upgrade (Fernandez- Stark et al., 2011; Taglioni & Winkler, 2016; World Bank, 2020). They typically exhibit high forward GVC participation, namely a high share of domestic value added embodied in their direct partner countries’ exports (Borin & Mancini, 2015, 2019). As a result, higher forward GVC participation in commodity-exporting countries may be associated with a higher RTI, as upstream tasks in agricultural or small-scale mining GVCs are more likely to be routine-intensive.2 High forward GVC participation is also a feature of countries specialized in innovative GVC tasks (World Bank, 2020), but its expected relationship with RTI contrasts that of commodity exporters. In innovative countries, high value-added upstream tasks, such as research and design services, make up a larger portion of their domestic value added that is re-exported by their bilateral trading partners. These tasks tend to be non-routine. These country examples illustrate that the relationship between GVC participation and RTI may vary across sectors and countries with different development levels and models. It may also differ between backward and forward GVC participation. We make three key contributions. First, this study quantifies the relationship between GVC participation and worker task demand, which remains under-researched (Marcolin et al., 2016). Our PIAAC, STEP, and CULS survey data cover 47 countries at different development levels and types of integration into GVCs. We measure RTI at a worker level, applying the method proposed by Lewandowski et al. (2022). In the absence of direct 1 For instance, some East Asian countries that initially specialized in blue-collar jobs managed to increase their workforce's skill supply, upgraded in GVCs, and shifted towards more upstream and downstream activities (de Vries et al., 2019). Similarly, some Central Eastern European countries (Czechia, Hungary, Poland, Slovenia) have been upgrading from an assembly-line specialization towards more advanced activities (Kordalska & Olczyk, 2022; Timmer et al., 2019). 2 In agribusiness, for instance, routine tasks include seed sowing and harvesting. More downstream tasks, such as washing, chopping, packing, and applying bar codes on fruits and vegetables, are also routine. Assigning one specialized task to each worker, rather than having one worker perform a series of consecutive tasks, increases the RTI. 2 export measures at the task level,3 we link sectoral measures of GVC participation to RTI at the worker level in a given sector, drawing on the methodology of Borin & Mancini (2019) based on EORA data. We also control for technology use with a country-sector share of workers who use computers at work. The country-sector level globalization and technology measures are plausibly exogenous to the decisions of individual firms and workers. The closest study to ours is Lewandowski et al. (2022), but we use much more disaggregated measures of GVC participation (especially in manufacturing) and assess the relative role of forward and backward linkages. Reijnders and de Vries (2018) also provided evidence on the role of offshoring and technological change in GVCs in explaining the increase in non-routine occupational labor demand in a sample of 37 advanced and emerging countries.4 However, they assumed that occupations are identical globally and measured occupational task contents with American data (the Occupation Information Network – O*NET). We use worker-level RTI to account for cross-country task differences in comparable occupations. This is vital as theory suggests that offshoring leads to a global polarization of tasks within occupations (Grossman & Rossi- Hansberg, 2008) and occupational task demands indeed differ between countries (Caunedo et al., 2021; de la Rica et al., 2020; Lewandowski et al., 2022; Lo Bello et al., 2019).5 We find significant linkages between GVC participation and RTI. Importantly, these associations differ between backward and forward integration. Overall, backward GVC participation is not correlated with RTI, while higher forward GVC participation is associated with more routine-intensive work. Moreover, the strength of this relationship is negatively related to countries’ development – it is stronger in countries with relatively low GDP per capita and weaker in countries with high GDP per capita. Second, this study assesses how the nature of occupations and sectors mediates the relationship between GVC participation and worker-level RTI. Specifically, it investigates the role of occupations’ offshorability (Blinder & Krueger, 2013). The relationship between GVC participation and RTI may be particularly pronounced among workers performing offshorable tasks in low-skilled occupations. Workers in less developed countries have a comparative advantage in performing routine tasks due to their larger endowment with low-skill workers (Grossman & Rossi-Hansberg, 2008). Indeed, Lewandowski et al. (2022) found that the relationship between backward GVC participation and worker-level RTI is the strongest among workers in low-skilled occupations. We find a strong and significant relationship between GVC participation – both backward and forward – and RTI among workers in offshorable occupations, especially in less-developed countries. However, we find no such relationship among workers in non-offshorable occupations. In addition, acknowledging that sectoral specialization of countries matters for the global division of tasks (Hanson, 2017), this paper examines heterogeneity between sectors. Importantly, we find a contrasting relationship between GVC participation and RTI in the industrial and business services sectors. In industrial sectors, a higher GVC participation is associated with more routine-intensive work in offshorable occupations, confirming that a country’s GVC participation is driven by the manufacturing sector (Fernandes et al., 2022). In sharp contrast, a higher GVC participation in business services sectors is correlated with less routine-intensive work in non-offshorable occupations. 3 To understand how GVCs shape the division of tasks across countries, research would ideally relate measures of task exports to data on tasks’ routine intensity. GVC participation measures to date are only available at the sector or firm level for a given country. However, recent work has introduced new measures of income and job activities in exports where activity is defined as a sector-occupation pair (Kruse et al., 2023). 4 Reijnders and de Vries (2018) combine input-output data to decompose changes in occupational labor demand along the value chain, but their methodology does not allow differentiation between intensities of GVC participation. 5 Other strands of literature relating globalization to the demand for workers in routine jobs study the effects of global trade (Autor et al., 2015), the China trade shock on local labor markets (Aghelmaleki et al., 2022; Autor et al., 2013, 2016), as well as offshoring (Autor et al., 2016; Baumgarten et al., 2013; Ebenstein et al., 2014; Goos et al., 2014; Hanson, 2017). 3 Third, this study assesses the relationship between GVC participation and wage inequality (the Gini coefficient of hourly wages) within countries. Globalization may widen differences in RTI between workers in offshorable occupations and those in non-offshorable occupations and thus contribute to earnings inequality, as workers performing less routine-intensive tasks tend to earn more (Autor & Handel, 2013; de la Rica et al., 2020). Our study confirms that a higher RTI is associated with lower wages. Consequently, GVC participation can influence wage inequality through two channels: (i) indirectly through its relationship with RTI, (ii) and directly through diverse wage effects among different types of workers, especially in offshorable and non-offshorable occupations. Extensive literature studied the effects of offshoring on the relative demand for different occupations at the sectoral level, usually finding demand shifts with implications of inequality. It primarily differentiated between production and non-production workers and capturing relative demand for particular worker types with their share in the sector’s wage bill. It initially focused on goods offshoring in manufacturing – see the seminal studies on the United States by Feenstra and Hanson (1999, 1996), and the broader literature review in Crinò (2009) – generally finding an increase in the relative demand for non-production workers. Focusing on services offshoring, some studies found it increased the relative demand for skills in the United States and Western Europe (Crinò, 2010, 2012), or lowered the relative demand for non-production workers in German manufacturing (Winkler, 2013). Several studies focused on worker-level adjustments to trade and offshoring found a downward pressure on wages in low-skilled occupations and upward pressure on wages in high-skilled occupations in the United Kingdom and Germany (Geishecker & Görg, 2013; Koerner, 2022). Ebenstein et al. (2014) found that offshoring negatively affects individuals’ wages in the United States due to relocating workers from higher-wage manufacturing jobs to other sectors and occupations. Existing cross-country studies (Wolszczak-Derlacz & Parteka, 2018) find small negative effects of offshoring on the wages of low- and middle- skilled workers, but focus on high-income countries. In the meta-analysis of within-country studies, Cardoso et al. (2021) showed that offshoring benefits high-skilled workers and harms low-skilled workers, especially in the origin countries. However, Gonzalez et al. (2015) found that GVC participation has a relatively small impact on wage distributions and can reduce wage inequality among low-skilled segments of the labor force. Duarte et al. (2022) found that countries with medium levels of GVC participation tend to record higher income inequality than those with low or high levels of GVC participation. Our study’s novelty is quantifying labor market channels of globalization’s contribution to within-country wage inequality, in a cross-country setting that covers both developed and developing countries and accounts for occupations’ offshorability. The direct contribution – GVCs’ wage effects on different types of workers – reduces wage inequality within countries, while the indirect contribution – through linkages with RTI – increases it. The relative strengths of these contributions differ between countries at different development levels. We show that in countries that primarily receive offshored jobs, GVC participation reduces wage inequality despite widening the gap in RTI between offshorable and non-offshorable occupations. However, in rich countries that mostly offshore jobs, it widens wage inequality as GVC participation benefits mainly workers in non-offshorable occupations in services. This paper is structured as follows. Section 2 introduces the data, measurements, and descriptive analysis. Section 3 presents the model and regression results linking GVC participation to worker-level RTI, while section 4 focuses on the relationship between GVC participation and wage inequality. Section 5 concludes and outlines policy implications. 4 2. Data and descriptive evidence a. Data and measurement Our worker-level dataset covers 47 countries at different development levels (Table A4 in the Appendix). Most of the country coverage comes from the OECD’s Programme for the International Assessment of Adult Competencies – PIAAC (2019). During three rounds of the study (2011-2012, 2014-2015, and 2017-2018), data were collected in 37 countries. The sample sizes amount to a few thousand 16-65 years old individuals. We complement PIAAC with the Skills Towards Employment and Productivity – STEP (World Bank, 2017) survey data from nine low- and middle-income countries. The STEP data were collected in 2012-2014 among urban residents aged 15-64 and covered a few thousand respondents in each country. We also use the “skill use at work” module of the third wave of the China Urban Labour Survey (CULS, 2017), which directly implemented the STEP questionnaire and ensured comparability with other countries in our sample. The survey collected data from six large cities in China (Guangzhou, Shanghai, Fuzhou, Shenyang, Xian, and Wuhan) and covered about 15,000 individuals. Following Lewandowski et al. (2022), we create a worker-level task content measure of occupations across countries in the spirit of Acemoglu and Autor (2011). As the STEP surveys are urban surveys, for comparability we omit farmers and skilled agricultural workers (ISCO 6 from the sample in all countries. For methodological details, see Lewandowski et al. (2022). We calculate the worker-level routine task intensity according to the following formula: ( + ) (1) = ln( ) − 2 where, , , are routine cognitive, non-routine cognitive analytical, and non-routine cognitive personal task levels. Table A1 in Appendix A enlists survey items used to construct these task measures. Particular task measures and RTI are standardized using their mean and standard deviation in the United States. We use hourly wages in US dollars, adjusted for purchasing power parity, with a 99% winsorization. Wage data are available for 38 of the 47 countries in our sample (Table A4), we adjust the sample accordingly in the wage analysis.6 The country-sector level measures of GVC participation are based on the EORA database (Lenzen et al., 2012, 2013) and computed following the methodology of Borin & Mancini (2015, 2019). In particular, we use both backward and forward GVC participation measures. Both quantify value-added flows that cross at least two country borders. Backward GVC participation measures the share of imported inputs used in export production (% of total exports). Forward GVC participation captures the share of domestic value added embodied in a country’s bilateral partners’ exports (% of total exports).7 Finally, we follow Blinder & Krueger, 2013), dividing occupations into offshorable and non-offshorable. We assign occupations to groups starting at the 4-digit ISCO-08 level, depending on data availability. Most of the countries report occupations using 3- and 4-digit ISCO-08 codes. For clarity, Table A3 in Appendix A lists occupations with assigned offshorability groups (at the 2-digit ISCO level). 6 We use the full sample of 47 countries in the first part of the analysis to maximize variation across countries. 7 This measure avoids a double-counting problem prevalent in alternative measures of forward GVC participation. 5 b. Descriptive analysis In the first step, we visually explore the relationship between GVC participation and RTI at the country-sector level. There is no correlation between backward GVC participation (9%, insignificant, left panel of Figure 2) and the average RTI. Similarly, the scatterplot suggests only a moderate correlation with forward GVC participation which is negative (-17%, right panel of Figure 2). The definition of GVC participation does not specify the type of value-added crossing borders – ranging from low (e.g., raw materials) to high high-value-added tasks (World Bank, 2020). These weak relationships could thus mask heterogeneity across types of countries, sectors and occupations. Figure 2. The correlation between backward and forward GVC participation and the average routine task intensity (RTI), by country and sector. Backward GVC participation Forward GVC participation Note: for each task content, the 0 is set at the United States average value and 1 corresponds to one standard deviation of this particular task content value in the United States. GDP per capita in PPP, current international $, country averages for 2011–2016. Source: Authors’ calculations based PIAAC, STEP, CULS (tasks), World Bank (GDP), and EORA (GVC) data. In the second step, we relate GVC participation to average wages at the country-sector level, differentiating between the industrial and business services sectors. Overall, backward and forward GVCs participation measures positively correlate with average hourly wages at the country-sector level (Figure 3), suggesting positive productivity spillovers from firms participating in GVCs for workers. In the case of backward GVC participation, the correlation with wages in the industrial sector (37%) is stronger relative to the business services sector8 (23%, Figure 3, left panel). It is in line with the intuition that high backward GVC participation in the industrial sector (driven mainly by manufacturing sectors) is associated with assembly tasks of specialized sectors where hourly wages can be expected to be higher (think of, e.g., technicians in the automotive sector). There is, however, high dispersion, because high backward GVC participation can characterize low-wage countries specialized in limited manufacturing GVCs, but also richer countries specialized in more sophisticated GVCs In the case of forward GVC participation, the opposite finding holds. The correlation with average hourly wages in the business services sector (36%) is higher than in the industrial sector (18%, Figure 3, right panel). High forward GVC participation in business services is associated with high-value-added tasks such as product design or R&D, which earn higher hourly wages. The high dispersion again suggests that high forward GVC participation is associated with lower-wage commodity exporters and innovative countries. 8 Retail and wholesale trade, accommodation, food service, transportation, storage, information and communication, financial, real estate, professional and administrative service activities (G-N ISIC rev. 4 sections). 6 Figure 3. The correlation between backward and forward GVC participation and hourly wages, by country and broad sector. Backward GVC participation Forward GVC participation Note: Hourly wages are in PPP US $, top 1% of earners are excluded. Average wages are weighted with sectors’ output. Source: Authors’ calculations based PIAAC, STEP and EORA (GVC) data. 3. Global value chain participation and routine task intensity a. Econometric model The econometric model quantifies the relationship between GVC participation and the average RTI of workers and exploits variation between countries within sectors (especially within manufacturing). It broadly follows the specification of Lewandowski et al. (2022). In particular, we estimate pooled OLS regressions of the following form: = 0 + 1 + 2 + 3 + + (2) where is the RTI of individual in occupation in sector in country ; measures GVC participation in sector in country ; captures technology in sector in country ; are individual skills of worker in occupation in sector in country ; and are sector fixed effects. We use measures of backward and forward GVC participation in sector and country . The measures are standardized within the sample to allow for interpretation regarding their relative economic magnitudes. Importantly, these measures vary between narrowly defined sub-sectors within manufacturing. Additionally, we control for foreign direct investment (FDI) as a share of GDP to capture globalization more broadly. To capture technology, we use the share of workers in sector and country who use computers at work. These measures are based on the PIAAC and STEP survey questions about a worker’s personal computer use. We aggregate this worker-level information to the sector level to address potential endogeneity concerns, as the performance of particular tasks may require computers. We include a quadratic term, allowing for possible non-linear linkages between computer use and the RTI. We also include sector-level fixed effects (18 sectors of 1-digit International Standard Industrial Classification, ISIC rev. 4) and their interactions with a country’s GDP per capita (log, demeaned) to control for structural differences between countries. To control for individual characteristics and skill levels, we include variables for age (10-year age groups), gender, education level (primary, secondary, tertiary), and a test-based measure of literacy skills (four proficiency levels). The literacy test comprehensively quantifies individuals’ skills to understand, evaluate, use, 7 and engage with written texts in personal, work-related, societal, and educational contexts (PIAAC Literacy Expert Group, 2009). We estimate the regression for all workers, and two main subsamples: workers in offshorable and non- offshorable occupations. We apply the allocation proposed by Blinder and Krueger (2013), see Table A3 in Appendix A for details. In all worker-level regressions, standard errors are clustered at the country-sector level. b. Pooled sample regression results We start by regressing worker-level RTI against backward and forward GVC participation at the country-sector level and a set of controls (see econometric model, 2) in the pooled sample of 47 countries. Overall, we find no correlation between backward GVC participation and the average RTI, while a higher forward GVC participation is associated with a higher RTI of workers. The latter correlation decreases with rising development levels, as indicated by the negative interaction term between GVC participation and GDP per capita (Table 1, column 1). These findings confirm our intuition: workers in higher-income countries perform fewer routine-intensive tasks. Hence, a higher forward GVC participation in such country settings captures higher value-added tasks such as R&D rather than repetitive upstream tasks as would be the case in commodity-exporting countries. However, the relationship between GVC participation and the RTI of workers may differ between types of occupations. To shed light on this hypothesis, we divide the sample into offshorable and non-offshorable occupations, using the preferred classification proposed by Blinder and Krueger (2013). In line with our assumption, the correlation between these occupational groups differs. We find no correlation between GVC participation and RTI among workers performing non-offshorable occupations (Table 1, column 2). However, among workers in occupations classified as offshorable, higher backward and forward GVC participation is linked with a higher average RTI (Table 1, Columns 2 and 3). As backward and forward GVC participation measures are standardized within the sample, the larger (in absolute terms) coefficient of backward GVC participation suggests this variable’s relatively greater importance relative to forward GVC participation.9 The worker-level RTI is standardized with the United States mean and standard deviation, to provide a reference point and comparability with the widely used (Acemoglu & Autor, 2011) RTI measure based on the US O*NET data. The interpretation would be the following. An increase in backward (forward, resp.) GVC participation by one standard deviation is associated with a rise in worker-level RTI by 0.079 (0.053, resp.) the US standard deviations. The negative interaction terms with GDP per capita for both GVC measures imply that the relationship between GVC participation and RTI weakens with countries’ development levels. We provided the rationale for this finding in the case of forward GVC participation under the overall findings in the previous paragraph. Similarly, higher backward GVC participation in high-income countries captures more high-value added worker tasks – even in the assembly stage – while in lower-income countries, those tasks tend to be more repetitive (think of technician operating machines in the former versus assembly line workers in the latter). For example, in Ecuador (relatively low GDP per capita, about 1 log point below the sample average), one standard deviation higher backward GVC participation among workers performing offshorable occupations is associated with RTI higher by 0.133. In contrast, in Canada (relatively high GDP per capita, about 1 log point above the sample average), it is associated with RTI higher by only 0.025. Considering that backward GVC participation in both countries is 9 As a robustness check, we run models for backward and forward GVC participation measures separately, rather than combining them in one joint regression, and obtain similar results (Table 1A, Panel A in Appendix A). 8 at a similar level (12-14%), the positive association between backward GVC participation and the RTI is stronger in Ecuador, due to the interaction term with GDP per capita. Table 1. The relationship between GVC participation and RTI, total and by occupation type, standardized Dependent variable: worker level RTI (1) (2) (3) All Non- Offshorable workers offshorable Backward Global Value Chain participation (GVCB) share in exports (std.) 0.004 -0.016 0.079*** (0.019) (0.020) (0.024) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.017 0.027 -0.054* (0.029) (0.030) (0.032) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.019* 0.011 0.053*** (0.010) (0.011) (0.014) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.060*** -0.059*** -0.074*** (0.012) (0.013) (0.017) Ln(GDP per capita) –mean(Ln(GDP per capita)) 0.033 0.041 -0.014 (0.038) (0.038) (0.060) Observations 167,034 144,914 22,120 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for GVCB share and GVCF share are standardized. All regressions include controls for technology (computer use, computer use squared), FDI, skills, education, age, gender, sector FE, and sector FE interacted with GDP per capita. Source: Authors’ calculations based on PIAAC, STEP, CULS (tasks), World Bank (GDP), EORA data and Borin and Mancini (2015, 20 19) (GVC participation measures). c. Sector-specific regression results Next, we study differences between broad sectors by estimating regressions for specific subsamples. We distinguish three broad sectors – industry (B-F ISIC rev. 4 sections), business services (G-N ISIC rev. 4 sections), and other services (O-S ISIC rev. 4 sections). For details, see Table A2 in Appendix A. For the industrial sector, we find no correlation between backward GVC participation and RTI. In contrast, a higher forward GVC participation is linked to more routine tasks, although decreasing with development level (Table 2, Panel A, column 1). We also find no correlation between backward and forward GVC participation and the RTI among non-offshorable occupations (Table 2, Panel A, column 2). Both these patterns are consistent with the overall results for all sectors (Table 1). However, in the case of offshorable occupations, the higher the backward GVC participation is in a country and sector, the higher the workers’ RTI (Table 2, Panel A, Column 3). We also find a positive association between forward GVC participation and RTI among workers in offshorable occupations (Table 2, panel B, column 3), again confirming the pooled sample’s results (Table 1). However, a negative interaction term of GVC participation with GDP per capita in industrial sectors suggests that the development level counterbalances this relationship. In contrast to the overall results (Table 1), this interaction term is insignificant for backward GVC participation. Hence, workers in industrial sectors and countries specialized in smaller segments of GVC (e.g., assemblers of final products) tend to perform more routine-intensive tasks. For forward GVC participation, the interaction term with GDP per capita is negative and significant, in line with the overall results (Table 1). In countries with GDP per capita twice the average in our sample (comparable to the United States), the interaction term offsets the coefficient of forward GVC participation, so the overall association with RTI is 0.10 Our findings align with the literature: manufacturing of 10 We obtain similar results for manufacturing (ISIC rev. 4 section C) rather than industry – results are available upon request. 9 low-value-added, basic intermediates that require more routine-intensive work tends to be outsourced to less developed countries (factory economies), while the production of non-routine tasks remains in countries at higher development levels (Baldwin, 2013). Table 2. The relationship between GVC participation and RTI, total and by sector and occupation type, standardized Dependent variable: worker-level RTI (1) (2) (3) Panel A: Industry All Non- Offshorable workers offshorable Backward Global Value Chain participation (GVCB) share in exports (std.) 0.027 -0.007 0.092*** (0.024) (0.027) (0.029) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.004 0.018 -0.021 (0.032) (0.036) (0.040) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.030** 0.008 0.062*** (0.015) (0.018) (0.019) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.043*** -0.030 -0.079*** (0.017) (0.020) (0.023) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.096 0.006 -0.304*** (0.066) (0.081) (0.103) Observations 38,917 28,790 10,127 Panel B: Business services Backward Global Value Chain participation (GVCB) share in exports (std.) -0.055** -0.056** -0.002 (0.024) (0.027) (0.044) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.063* 0.083** -0.080* (0.037) (0.039) (0.047) Forward Global Value Chain participation (GVCF) share in exports (std.) -0.015 -0.020 0.019 (0.016) (0.017) (0.024) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.046*** -0.043** -0.064** (0.017) (0.017) (0.029) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.027 -0.017 -0.067 (0.049) (0.049) (0.078) Observations 71,979 63,003 8,976 Panel C: Other services Backward Global Value Chain participation (GVCB) share in exports (std.) 0.220*** 0.206*** 0.396*** (0.079) (0.079) (0.152) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.329*** -0.326*** -0.400 (0.114) (0.114) (0.257) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.029 0.023 0.116** (0.023) (0.023) (0.057) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.073** -0.077** 0.013 (0.032) (0.032) (0.064) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.210*** -0.126* -0.129 (0.076) (0.074) (0.180) Observations 50,843 48,133 2,710 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for GVCB share and GVCF share are standardized. All regressions include controls for technology (computer use, computer use squared), FDI, skills, education, age, gender, sector FE, and sector FE interacted with GDP per capita. Source: Authors’ calculations based on PIAAC, STEP, CULS (tasks), World Bank (GDP), EORA data and Borin and Mancini (2015, 2019) (GVC participation measures). In business services, we find a negative association between backward GVC participation and workers’ RTI (Table 2, Panel B, column 1). Due to the positive interaction term with GDP per capita, this effect is weaker in countries at higher development levels: it disappears in countries with GDP per capita twice the sample average, comparable to the United States, and becomes positive in countries with even higher GDP per capita (e.g., 10 Norway). In contrast to the overall sample (Table 1), these results are driven by non-offshorable occupations, whereas the coefficients among offshorable occupations are insignificant (Table 2, Panel B, columns 2 and 3). This could reflect that services are less tradeable than manufactured goods and involve more occupations that cannot be offshored (e.g., truck drivers, see also the list of occupations by offshorability in Table A3). In other services – much less integrated into GVCs than the other two broad sectors11 – higher backward GVC participation correlates with higher worker-level RTI (Table 2, Panel C, column 1), contrasting the overall sample results (Table 1). The net ‘effect’ depends on a country’s development level. Results for offshorable occupations are similar to the pooled sample: in countries with a GDP per capita lower than 170% of the average in our sample (comparable to Canada or Austria), backward GVC participation is associated with more routine intensive work, and in countries with GDP per capita above this level – with less routine intensive work (Table 2, Panel C, columns 1-2). Similarly, higher forward GVC participation is linked to higher RTI at the worker level (Table 2, Panel C, column 3). d. Robustness check: Occupation by skill intensity As a robustness check for differences between occupational groups, we re-estimate the regressions of Section c. We focus on occupational groups by their skill levels rather than between offshorable and non-offshorable occupations. We distinguish between by high-skilled (managers, professionals, technicians – ISCO 1-3), medium-skilled (clerical workers, sales and services workers – ISCO 4-5) and low-skilled (craft and related trades workers, plant and machine operators, elementary occupations – ISCO 7-9) occupations. This classification of occupations follows the standard typology of the International Labour Organisation, and was used by Lewandowski et al. (2022). These occupational groups perform tasks with different routine intensities. On average, workers in high-skilled occupations perform relatively non-routine tasks, workers in middle-skilled occupations moderately routine-intensive tasks, and workers in low-skilled occupations more routine-intensive tasks. Results for high-skilled occupations somewhat resemble those for non-offshorable occupations, while results for medium- and low-skilled occupations resemble those for offshorable occupations (see Table B2 in Appendix B). Importantly, we observe almost identical patterns in correlations between GVC participation and RTI for specific sectors. It confirms that distinguishing between industries is crucial for studying the relationship between GVC participation and labor market outcomes. Our results suggest that the relationship between GVC participation and RTI differs substantially between the industrial and business services sectors. In the industrial sector, higher GVC participation is associated with more routine intensive work in offshorable occupations that usually demand low to medium levels of skills. In business services, it is associated with less routine intensive work in non-offshorable occupations, which often require higher skills. 11 On average, backward GVC participation in other services is 17.4 pp lower than in industry, and forward GVC participation is 3.3 pp lower (differences estimated as broad sector fixed effects in regressions on GVC participation measures, controlling also for country fixed effects). 11 4. Global value chain participation and wage inequality within countries a. Econometric model and decomposition method As a higher RTI is negatively correlated with earnings, both at the occupation and worker level (Autor & Handel, 2013; de la Rica et al., 2020), GVC participation may widen wage inequality between workers in offshorable occupations and those in non-offshorable occupations. In this section, we study the contribution of GVC participation to wage inequality within countries. We distinguish between two channels: (1) the direct contribution of GVC participation to individual wages, and (2) the indirect contribution of GVC participation through its relationship with workers’ RTI. We calculate the Gini coefficient to quantify the relationship between GVC participation and inequality. The diagram in Figure 3 exemplifies our reasoning. Our analysis includes four steps. Figure 4. Diagram of wage inequality analysis. Source: Own elaboration. In the first step, we divide our sample into six subsamples by broad sector of employment (industry, business services and other services) and occupation (offshorable and non-offshorable), as introduced in section 3. For each subsample, we estimate the following Mincerian wage regression: = 0 + 1 + 2 + 3 + 3 + + (3) where, is the wage of individual in occupation in sector in country ; while the rest of the notation follows equation (2). Our key variable of interest in the wage regression is worker-level RTI. Based on the estimated coefficients for each right-hand side variable, we predict wages for workers in each of the six subpopulations. For each country, we then calculate the Gini coefficient, our baseline scenario. Appendix A gives a more detailed description, including formulas for the underlying methodology, which we outline below. In the second step, we assess the direct contribution of GVC participation to wage inequality. We assume a second scenario of no GVC participation and calculate predicted wages conditional on GVC participation values in equation (3) equal to zero (i.e., = 0) for each of the six subsamples. Then, we re-calculate the Gini 12 coefficient of predicted wages for each country in this second scenario. We define the direct contribution of GVC participation to within-country wage inequality as the difference between the Gini coefficient of wages calculated in the baseline scenario and the Gini coefficient of wages obtained in the scenario of no GVC participation. In the third step, we analyze the indirect contribution of GVC participation to wage inequality, through its relationship with workers’ RTI. Specifically, we use the estimated coefficients for the six sub-samples following equation (2) to predict worker-level RTI, now assuming GVC participation values equal to zero. In other words, we compute counterfactual workers’ RTI in the economy under the scenario of no GVC integration. We then use the estimated models for the six sub-samples following equation (3) to predict wages, conditional on these new counterfactual RTI values. To isolate the indirect contribution of GVC participation to wages through its relationship with workers’ RTI, we use the observed values of GVC participation in this wage model (rather than setting their values to zero), which is the counterfactual RTI scenario. We define the indirect contribution of GVC participation to wage inequality as a difference between the Gini coefficient of wages calculated in the baseline scenario and the Gini coefficient of wages obtained in the counterfactual scenario. In the fourth step, we calculate the total contribution of GVC participation to wage inequality. We set the GVC participation values in equation (3) to zero (as in the calculation of the direct contribution), and we use the counterfactual RTI conditional on no GVC participation (as in the calculation of the indirect contribution) to calculate wages using the original coefficients estimated in the six models. We identify the total contribution of GVC participation to wage inequality as the difference between the Gini coefficient of wages in the baseline scenario and the Gini coefficients of wages in this last scenario. Importantly, we estimate equations (2) and (3) for six sub-samples for different combinations of broad sectors (industry, business services and other services) and occupation types (offshorable and non-offshorable) to capture the likely different contributions of GVC participation in these worker subgroups. The relationship between GVC participation, RTI and wages may differ between the industrial and services sectors, as products created in the industrial sector are more tradeable and easier to fragment, especially in manufacturing. Also, offshorable occupations may be more vulnerable to wage adjustments than non-offshorable occupations. b. Sector-specific regression results In this subsection, we explore the contribution of GVC participation to within-country wage inequality, distinguishing between its direct and indirect contributions through workers’ RTI. Our approach likely provides an upper bound, as we use cross-sectional regression that describes the equilibrium allocation of tasks and wages across workers in different countries. GVC participation may be partly endogenous to comparative advantage in tasks and pre-existing wage-level differences. For this reason, we focus on the contribution of GVC participation to within-country wage inequality rather than cross-country differences in wage levels. Moreover, only a small share of cross-country differences in RTI can be attributed to globalization, as differences in technology use and skill supply play a much larger role (Lewandowski et al., 2022). We first estimate the relationship between RTI and individual-level wages in Mincerian wage regressions (equation 3), for each of the six subpopulations by broad sector and occupation type (Table 3). The regression results consistently show a significant and negative association between workers’ RTI and wages, in all types of occupations and sectors; that is, more routine intensive tasks tend to pay lower wages. The magnitude is the largest among offshorable occupations in business services, suggesting a particularly strong wage penalty for performing routine tasks in this sector (column 4). The wage penalty for more routine tasks is the second largest among offshorable occupations in industrial sectors (column 2). It is smaller in non-offshorable 13 occupations in the industrial (column 1) and business services (column 3) sectors, but not in other services (columns 5 and 6). Table 3. The relationship between RTI, GVC participation, and wages, by sector and occupation type, standardized Dependent variable: worker- Industry Business services Other services level wages (1) (2) (3) (4) (5) (6) Non- Offshorable Non- Offshorable Non- Offshorable offshorable offshorable offshorable Routine Task Intensity (RTI, std) -1.646*** -1.704*** -1.593*** -2.116*** -1.514*** -1.206*** (0.147) (0.206) (0.109) (0.238) (0.101) (0.247) Backward GVC participation 0.134 0.141 0.394** -0.589* 0.712 0.089 (GVCB) share in exports (std.) (0.165) (0.167) (0.186) (0.300) (0.558) (0.913) GVCB share (std.) * [Ln(GDP pc) 0.084 0.045 0.127 -0.196 0.519 1.182** –mean(Ln(GDP pc)] (0.059) (0.074) (0.080) (0.133) (0.391) (0.552) Forward GVC participation 0.404*** -0.146 0.761*** 0.123 -0.010 0.595 (GVCF) share in exports (std.) (0.147) (0.100) (0.140) (0.193) (0.371) (0.516) GVCF share (std.) * [Ln(GDP pc) 0.265*** -0.002 0.338*** -0.165 -0.166 -1.109*** –mean(Ln(GDP pc)] (0.058) (0.047) (0.077) (0.129) (0.161) (0.231) Ln(GDP per capita) – 0.268* 0.440** -0.017 0.446** 0.198 -0.061 mean(Ln(GDP per capita)) (0.157) (0.205) (0.106) (0.195) (0.668) (0.717) Observations 18,647 7,600 38,659 6,714 31,523 2,091 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for GVCB share and GVCF share are standardized. All regressions include controls for technology (computer use, computer use squared), skills, education, age, gender, sector FE, and country FE. The wage data for Canada, China, Hungary, Macedonia (FYROM), Peru, Serbia, Singapore, Sweden, and Türkiye are unavailable; therefore, these countries are excluded from the sample. Source: Authors’ calculations based on PIAAC, STEP, CULS (tasks), World Bank (GDP), EORA data , and Borin and Mancini (2015, 2019) (GVC participation measures). At the same time, the relationship between GVC participation and wages is significant only in some types of occupations and sectors. First, higher forward GVC participation is associated with higher wages among workers in non-offshorable occupations in the industrial and business services sectors. This could indicate high value-added tasks such as R&D upstream in GVCs which are more difficult to offshore and thus pay high wages. Second, more backward GVC participation is associated with higher wages only in business services – positively in non-offshorable occupations, and negatively in offshorable occupations. Due to the unavailability of wage data, we had to exclude Canada, China, Hungary, Macedonia (FYROM), Peru, Serbia, Singapore, Sweden and Türkiye from the previous analysis. Table 4 thus replicates the results of Table 2 for the reduced country sample when assessing the relationship between GVC participation and RTI. The estimated coefficients shown in Tables 2 and 4 differ slightly. Still, the findings are the same: more backward and forward GVC participation are associated with higher RTI among workers in offshorable occupations in the industrial sector, especially for countries at the average development level in our sample, but with a lower RTI among workers in non-offshorable occupations in business services (Table 4). 14 Table 4. The relationship between GVC participation and RTI, by sector and occupation type, standardized Dependent variable: worker- Industry Business services Other services level RTI (1) (2) (3) (4) (5) (6) Non- Offshorable Non- Offshorable Non- Offshorable offshorable offshorable offshorable Backward GVC participation -0.004 0.074*** -0.038* 0.041 0.204*** 0.353*** (GVCB) share in exports (std.) (0.025) (0.024) (0.021) (0.032) (0.049) (0.099) GVCB share (std.) * [Ln(GDP pc) 0.009 -0.014 0.078*** -0.030 -0.221*** -0.259 –mean(Ln(GDP pc)] (0.030) (0.031) (0.026) (0.031) (0.080) (0.199) Forward GVC participation -0.002 0.092*** -0.015 0.044* -0.057* 0.107 (GVCF) share in exports (std.) (0.019) (0.018) (0.018) (0.024) (0.032) (0.071) GVCF share (std.) * [Ln(GDP pc) -0.026 -0.080*** -0.039** -0.083*** 0.039 0.074 –mean(Ln(GDP pc)] (0.020) (0.021) (0.017) (0.028) (0.039) (0.079) Ln(GDP per capita) – -0.124** -0.139 0.017 -0.010 -0.037 -0.065 mean(Ln(GDP per capita)) (0.060) (0.125) (0.052) (0.075) (0.096) (0.138) Observations 21,023 8,364 44,351 7,578 34,540 2,326 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for GVCB share and GVCF share are standardized. All regressions include controls for technology (computer use, computer use squared), FDI, skills, education, age, gender, sector FE, and sector FE interacted with GDP per capita. The specification follows that in Table 2, but for consistency with Table 3, Canada, China, Hungary, Macedonia (FYROM), Peru, Serbia, Singapore, Sweden, and Türkiye (for which wage data are not available) are excluded. Source: Authors’ calculations based on PIAAC, STEP, CULS (tasks), World Bank (GDP), EORA data . and Borin and Mancini (2015, 2019) (GVC participation measures). c. Quantifying the direct and indirect contribution of GVC participation to wage inequality In this subsection, we use the estimated models to quantify the contribution of GVC participation to wage inequality in the reduced country sample of 38 countries, both directly and indirectly. We find that the direct contribution of GVC participation to wage inequality is negative in most countries (Figure 5a). In other words, higher GVC participation is linked to reduced wage inequality within countries. Some notable exceptions include the US and small countries intensively integrated into GVC, such as Ireland (high backward GVC participation) or Norway (high forward GVC participation). The results suggest a U-shaped relationship between GDP per capita and the direct contribution of GVC participation to wage inequality (Figure 5a). That is, the reduction in wage inequality is the strongest in upper-middle-income and bottom-high-income countries, but the smallest in low-income countries (which are weakly integrated into GVCs) and high-income countries. The Mincerian wage regressions suggest that the direct contribution reflects the positive role of forward GVCs for workers' wages in non-offshorable occupations in the industrial and business services sectors (Table 3). In sharp contrast, the indirect contribution of GVC participation to wages through its link with workers’ RTI widens wage inequality in most countries (Figure 5b). Contrasting relationships between GVC participation and RTI among different groups of workers drive this pattern. The Mincerian wage regressions suggest a negative relationship between the RTI of workers and individual wages in all sectors and occupation types (Table 3). So, a higher GVC integration is associated with wider within-country wage inequality through larger RTI gaps between workers in offshorable and non-offshorable occupations in different sectors. In most countries, the indirect contribution is smaller in absolute terms than the direct contribution. 15 Figure 4. The contribution of GVC participation to within-country wage inequality a) b) c) Source: Authors’ calculations based on PIAAC, STEP, CULS (tasks), EORA data and Borin and Mancini (2015, 2019) (GVC participation measures). 16 Finally, we combine the direct and indirect channels to assess the total (net) contribution of GVC participation to wage inequality within countries (Figure 5c).12 We find that GVC participation is linked to higher wage inequality in the top high-income countries, such as the US and Ireland, which in most cases is driven by the indirect contribution of GVCs through its links with worker RTI. At the same time, GVC participation is associated with reduced wage inequality in most low- and middle-income countries (in particular Kenya and Ghana, but also Mexico), as well as the bottom high-income countries (Central Eastern and Southern Europe, in particular, Czechia and Poland) where the direct reduction in wage inequality is stronger than the indirect contribution. The findings can be interpreted as follows. Our results suggest that in countries that mostly receive offshored jobs, GVC participation reduces wage inequality, despite widening the gap in the RTI of work between offshorable and non-offshorable occupations. However, in rich countries that mostly offshore jobs, GVC participation widens wage inequality as it benefits mainly workers in non-offshorable occupations in services.13 5. Conclusions and policy implications In this paper, we investigated the relationship between GVC participation and the RTI of workers and its contribution to within-country wage inequality. We used a unique dataset combining worker-level, country- specific RTI measures based on a pooled sample of survey data for 47 (38, resp.) countries at all development levels, applying the methodology of Lewandowski et al. (2022), with measures of backward and forward GVC participation at the country-sector level based on the method of Borin and Mancini (2019, 2015). We find that the relationship between GVC participation and the RTI of workers is complex and depends on the nature of GVCs, occupations and sectors. This study also finds that GVC participation contributes to wage inequality within countries directly and indirectly through its relationship with workers’ RTI. First, we found that the relationship between GVC participation and RTI differs between types of GVC participation: higher forward GVC participation correlates with more RTI of workers, while higher backward GVC participation does not. Importantly, these associations differ across occupational groups. In countries and sectors with higher GVC participation of either type, workers in offshorable occupations perform more routine- intensive work, with weaker associations for higher-income countries. At the same time, we find no such results among workers in non-offshorable occupations. Second, the industrial sector, which produces primarily tradable goods, follows this general pattern, whereas business services, which are tradable to a lesser extent, do not. Higher backward GVC participation is associated with less routine-intensive tasks in business services, especially among workers in non-offshorable 12 As the Gini coefficient is a non-linear measure, the sum of Gini coefficients with two separate shocks (direct and indirect effect) does not necessarily equal the Gini coefficient simulated with the same two shocks jointly (total effect). The residual, however, is relatively small compared to the total contribution (see Figure B1 in Appendix A). 13 Our approach may raise the question if the RTI and wage regressions can be estimated separately. We use seemingly unrelated regression (SUR) to test the validity of our results. We find that separate estimation is correct. First, we find no correlation between the residuals from RTI and wage models, suggesting that they are unrelated. Additionally, we confirm that error terms have fairly symmetric distributions required for the estimator to be unbiased in small samples. Second, the point estimates are consistent with those obtained from separate estimations. Minor differences occur due to slight differences in the estimation sample. Some individuals do not report their wages, resulting in slightly smaller sample sizes than RTI models (see Tables 3 and 4). The SUR approach requires samples of both models to be equal, so the RTI sample must be reduced to a wage sample. The SUR estimates are available upon request. 17 occupations, but with more routine-intensive tasks in other services. Focusing on the skill content of occupations as a robustness check shows that results for high-skilled occupations somewhat resemble those for non-offshorable occupations. In contrast, results for medium- and low-skilled occupations resemble those for offshorable occupations. Third, we studied the contribution of GVC participation to within-country wage inequality: direct and indirect through its relationship with the workers’ RTI. GVC participation is associated with larger wage inequality in most high-income countries, but with reduced wage inequality in most low- and middle-income countries. Its indirect contribution to wage inequality – widening the gap between the RTI of workers in offshorable occupations in the industrial sector and workers in non-offshorable occupations in business services sectors – is a crucial mechanism. Understanding the differences in the RTI of workers across the development spectrum and its relationship with fundamental factors – technology adoption, skill supply, and globalization – has important policy implications. The transition from routine to non-routine work has been a key dimension of structural change in labor markets, increasing worker productivity and earnings. Jobs with a higher non-routine content involve higher levels of technology, require higher skill levels, and offer higher earnings between and within occupations (Autor & Handel, 2013; de la Rica et al., 2020). Diverging effects of globalization on the RTI of different types of workers can thus contribute to wage inequality within countries. At the same time, cross-country differences in RTI, especially between high- versus low- and middle-income countries, are larger than implied by mere cross-country differences in skills supply, as they can be mainly attributed to differences in technology use (Lewandowski et al., 2022). Investments in education and skills in developing and emerging economies are frequently cited as necessary conditions to foster shared prosperity (World Bank, 2019). They are also often highlighted to counter the adverse labor market effects of increased technology adoption in developing countries. The mediating role of worker skills becomes even more urgent amid rapid advances in artificial intelligence, such as recent developments of Chat-GPT and GPT-4. While they are most likely required to achieve these goals, they are unlikely to be sufficient, given that differences in job task content are largely related to differences in technology use and participation in GVCs. In any case, policies to increase technology use and approaches to facilitate upgrading in GVCs should complement investments in skills, especially since technological change within GVCs tends to increase the relative demand for non-routine work (Reijnders & de Vries, 2018). Our study has limitations. First, it does not claim to have determined a causal effect. Since the survey data were collected once per country, only cross-sectional analysis is possible. The analysis therefore cannot capture wage changes over time or cases where GVC participation created new labor market segments that did not exist before. In the future, the second round of PIAAC data collection will allow running a quasi-panel study to study the relationship between changes in GVC participation, technology use, and the supply of skills, with the RTI of particular occupations in various countries. Second, the survey data do not distinguish between domestic and foreign-owned firms, so it is unclear if FDI correlates with RTI differences within sectors. Lewandowski et al. (2022) showed that FDI is not a significant factor behind RTI differences between sectors, but there may be a relationship within sectors. Third, adult skill surveys have greatly improved our understanding of skills supply and the quality of education worldwide. It is possible, though, that literacy or numeracy measures are insufficient to fully understand factors behind differences in the nature of work, task content of jobs, and productivity. Differences in managerial and interpersonal skills may also contribute to differences in organizing and performing work. These skills are unfortunately not measured in the same survey data that capture worker tasks. Finally, the estimated contribution of technology adoption to worker-level RTI may likely increase in the 18 future. Advances in artificial intelligence may more strongly affect business services tasks, the extent of offshoring, and thus the relationship between GVC participation and RTI. References Acemoglu, D., & Autor, D. (2011). Skills, tasks and technologies: Implications for employment and earnings. In D. Card & O. Ashenfelter (Eds.), Handbook of Labor Economics (Vol. 4, pp. 1043–1171). Elsevier. https://doi.org/10.1016/S0169-7218(11)02410-5 Aghelmaleki, H., Bachmann, R., & Stiebale, J. (2022). The China shock, employment protection, and European jobs. ILR Review, 75(5), 1269–1293. https://doi.org/10.1177/00197939211052283 Autor, D. H., & Dorn, D. (2013). The growth of low-skill service jobs and the polarization of the us labor market. American Economic Review, 103(5), 1553–1597. https://doi.org/10.1257/aer.103.5.1553 Autor, D. H., Dorn, D., & Hanson, G. H. (2013). The geography of trade and technology shocks in the United States. American Economic Review, 103(3), 220–225. https://doi.org/10.1257/aer.103.3.220 Autor, D. H., Dorn, D., & Hanson, G. H. (2015). Untangling trade and technology: Evidence from local labour markets. The Economic Journal, 125(584), 621–646. https://doi.org/10.1111/ecoj.12245 Autor, D. H., Dorn, D., & Hanson, G. H. (2016). The China shock: Learning from labor-market adjustment to large changes in trade. Annual Review of Economics, 8(1), 205–240. https://doi.org/10.1146/annurev- economics-080315-015041 Autor, D. H., & Handel, M. J. (2013). Putting tasks to the test: Human capital, job tasks, and wages. Journal of Labor Economics, 31(2), S59–S96. https://doi.org/10.1086/669332 Baldwin, R. (2013). Global supply chains: Why they emerged, why they matter, and where they are going. Global Value Chains in a Changing World, 13–59. https://doi.org/10.30875/3c1b338a-en Baldwin, R. (2014). Trade and industrialization after globalization’s second unbundling: How building and joining a supply chain are different and why it matters. In 5. Trade and Industrialization after Globalization’s Second Unbundling: How Building and Joining a Supply Chain Are Different and Why It Matters (pp. 165– 214). University of Chicago Press. https://doi.org/10.7208/9780226030890-007 Baumgarten, D., Geishecker, I., & Görg, H. (2013). Offshoring, tasks, and the skill-wage pattern. European Economic Review, 61, 132–152. https://doi.org/10.1016/j.euroecorev.2013.03.007 Blinder, A. S., & Krueger, A. B. (2013). Alternative measures of offshorability: A survey approach. Journal of Labor Economics, 31(2), S97–S128. https://doi.org/10.1086/669061 Borin, A., & Mancini, M. (2015). Follow the value added: Bilateral gross export accounting. Bank of Italy Temi Di Discussione (Working Paper) No, 1026. Borin, A., & Mancini, M. (2019). Measuring what matters in global value chains and value-added trade (Working Paper). World Bank. https://doi.org/10.1596/1813-9450-8804 Cardoso, M., Neves, P. C., Afonso, O., & Sochirca, E. (2021). The effects of offshoring on wages: A meta-analysis. Review of World Economics, 157(1), 149–179. https://doi.org/10.1007/s10290-020-00385-z Caunedo, J., Keller, E., & Shin, Y. (2021). Technology and the task content of jobs across the development spectrum (Working Paper No. 28681). National Bureau of Economic Research. https://doi.org/10.3386/w28681 19 Cortes, G. M., Jaimovich, N., & Siu, H. E. (2017). Disappearing routine jobs: Who, how, and why? Journal of Monetary Economics, 91, 69–87. https://doi.org/10.1016/j.jmoneco.2017.09.006 Crinò, R. (2009). Offshoring, multinationals and labour market: A review of the empirical literature. Journal of Economic Surveys, 23(2), 197–249. https://doi.org/10.1111/j.1467-6419.2008.00561.x Crinò, R. (2010). Service offshoring and white-collar employment. The Review of Economic Studies, 77(2), 595– 632. https://doi.org/10.1111/j.1467-937X.2009.00586.x Crinò, R. (2012). Service offshoring and the skill composition of labour demand. Oxford Bulletin of Economics and Statistics, 74(1), 20–57. https://doi.org/10.1111/j.1468-0084.2010.00634.x de la Rica, S., Gortazar, L., & Lewandowski, P. (2020). Job tasks and wages in developed countries: Evidence from PIAAC. Labour Economics, 65, 101845. https://doi.org/10.1016/j.labeco.2020.101845 de Vries, G., Chen, Q., Hasan, R., & Li, Z. (2019). Do asian countries upgrade in global value chains? A novel approach and empirical evidence. Asian Economic Journal, 33(1), 13–37. https://doi.org/10.1111/asej.12166 Duarte, R., Espinosa-Gracia, A., Jiménez, S., & Sánchez-Chóliz, J. (2022). New insights on the relationship between the involvement of countries in global value chains, and intra- and inter-country inequalities. Structural Change and Economic Dynamics, 63, 320–329. https://doi.org/10.1016/j.strueco.2022.11.001 Ebenstein, A., Harrison, A., McMillan, M., & Phillips, S. (2014). Estimating the impact of trade and offshoring on american workers using the current population surveys. The Review of Economics and Statistics, 96(4), 581–595. https://doi.org/10.1162/REST_a_00400 Feenstra, R. C., & Hanson, G. H. (1996). Globalization, outsourcing, and wage inequality. The American Economic Review, 86(2), 240–245. Feenstra, R. C., & Hanson, G. H. (1999). The impact of outsourcing and high-technology capital on wages: Estimates for the United States, 1979–1990. The Quarterly Journal of Economics, 114(3), 907–940. Fernandes, A. M., Kee, H. L., & Winkler, D. (2022). Determinants of global value chain participation: Cross- country evidence. The World Bank Economic Review, 36(2), 329–360. https://doi.org/10.1093/wber/lhab017 Fernandez-Stark, K., Bamber, P., & Gereffi, G. (2011). Workforce development in the fruit and vegetable global value chain. Duke Center on Globalization. Geishecker, I., & Görg, H. (2013). Services offshoring and wages: Evidence from micro data. Oxford Economic Papers, 65(1), 124–146. https://doi.org/10.1093/oep/gpr055 Gonzalez, J. L., Kowalski, P., & Achard, P. (2015). Trade, global value chains and wage-income inequality. OECD. https://doi.org/10.1787/5js009mzrqd4-en Goos, M., Manning, A., & Salomons, A. (2014). Explaining job polarization: Routine-biased technological change and offshoring. The American Economic Review, 104(8), 2509–2526. Grossman, G. M., & Rossi-Hansberg, E. (2008). Trading Tasks: A Simple Theory of Offshoring. American Economic Review, 98(5), 1978–1997. https://doi.org/10.1257/aer.98.5.1978 Hanson, G. H. (2017). What do we really know about offshoring? Industries and countries in global production sharing (SSRN Scholarly Paper No. 2980947). https://doi.org/10.2139/ssrn.2980947 20 Institute of Population and Labor Economics. (2017). The China Urban Labor Survey (CULS). Jensen, J. B., & Kletzer, L. G. (2010). Measuring tradable services and the task content of offshorable services jobs. In Labor in the new economy (pp. 309–335). University of Chicago Press. Koerner, K. (2022). The wage effects of offshoring to the East and West: Evidence from the German labor market. Review of World Economics. https://doi.org/10.1007/s10290-022-00471-4 Kordalska, A., & Olczyk, M. (2022). Upgrading low value-added activities in global value chains: A functional specialisation approach. Economic Systems Research, 0(0), 1–27. https://doi.org/10.1080/09535314.2022.2047011 Kruse, H., Timmer, M., de Vries, G., & Ye, X. (2023). Structural change in export activities: An exploration using occupations data. Groningen University, Mimeo. Lenzen, M., Kanemoto, K., Moran, D., & Geschke, A. (2012). Mapping the structure of the world economy. Environmental Science & Technology, 46(15), 8374–8381. https://doi.org/10.1021/es300171x Lenzen, M., Moran, D., Kanemoto, K., & Geschke, A. (2013). Building EORA: A global multi-region input–output database at high country and sector resolution. Economic Systems Research, 25(1), 20–49. https://doi.org/10.1080/09535314.2013.769938 Lewandowski, P., Park, A., Hardy, W., Du, Y., & Wu, S. (2022). Technology, skills, and globalization: Explaining international differences in routine and nonroutine work using survey data. The World Bank Economic Review, 36(3), 687–708. https://doi.org/10.1093/wber/lhac005 Lo Bello, S., Sanchez Puerta, M. L., & Winkler, H. (2019). From Ghana to America: The skill content of jobs and economic development (Working Paper No. 12259). IZA Discussion Papers. https://www.econstor.eu/handle/10419/196757 Marcolin, L., Miroudot, S., & Squicciarini, M. (2016). Routine jobs, employment and technological innovation in global value chains. OECD Science, Technology and Industry Working Papers, no. 2016/01. Michaels, G., Natraj, A., & Van Reenen, J. (2014). Has ICT polarized skill demand? Evidence from eleven countries over twenty-five years. The Review of Economics and Statistics, 96(1), 60–77. PIAAC Literacy Expert Group. (2009). PIAAC Literacy: A Conceptual Framework. OECD. https://doi.org/10.1787/220348414075 Reijnders, L. S. M., & de Vries, G. J. (2018). Technology, offshoring and the rise of non-routine jobs. Journal of Development Economics, 135, 412–432. https://doi.org/10.1016/j.jdeveco.2018.08.009 Rodrik, D. (2013). Unconditional convergence in manufacturing. The Quarterly Journal of Economics, 128(1), 165–204. Spitz‐Oener, A. (2006). Technical change, job tasks, and rising educational demands: Looking outside the wage structure. Journal of Labor Economics, 24(2), 235–270. https://doi.org/10.1086/499972 Survey of Adult Skills (PIAAC). (2019). https://www.oecd.org/skills/piaac/ Taglioni, D., & Winkler, D. (2016). Making global value chains work for development. World Bank. https://doi.org/10.1596/978-1-4648-0157-0 Timmer, M. P., Miroudot, S., & de Vries, G. J. (2019). Functional specialisation in trade. Journal of Economic Geography, 19(1), 1–30. https://doi.org/10.1093/jeg/lby056 21 Winkler, D. (2013). Services offshoring and the relative demand for white-collar workers in German manufacturing. In A. Bardhan, D. Jaffee, & C. Kroll (Eds.), The Oxford Handbook of Offshoring and Global Employment (p. 0). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199765904.013.0004 Wolszczak-Derlacz, J., & Parteka, A. (2018). The effects of offshoring to low-wage countries on domestic wages: A worldwide industrial analysis. Empirica, 45(1), 129–163. https://doi.org/10.1007/s10663-016-9352- 4 World Bank. (2017). STEP skills measurement surveys: Innovative tools for assessing skills (English). Social protection and labor discussion paper (no. 1421). http://documents.worldbank.org/curated/en/516741468178736065/STEP-skills-measurement- surveys-innovative-tools-for-assessing-skills World Bank. (2019). World Development Report 2019: The changing nature of work. https://www.worldbank.org/en/publication/wdr2019 World Bank. (2020). World Development Report 2020: Trading for development in the age of global value chains . https://www.worldbank.org/en/publication/wdr2020 22 Appendix A – Methodological details a. Measurements and classifications Table A1. The task items selected to calculate task content measures with the US PIAAC data Non-routine cognitive Task content Non-routine cognitive analytical Routine cognitive Manual interpersonal Solving problems Supervising others Changing order of tasks Physical Reading news Making speeches or - reversed (not able) tasks (at least once a month) giving presentations Filling out forms (at least Reading professional journals (any frequency) once a month) Task items (at least once a month) Making speeches or Programming giving presentations - (any frequency) reversed (never) Correlation with O*NET-based 0.77 0.72 0.55 0.74 measures Note: The cut-offs for the “yes” dummy in brackets. For the full wording of questions and definitions of cutoff see Lewandowski et al. (2022). O*NET-based measures are based on Acemoglu and Autor (2011). Source: Lewandowski et al. (2022). Table A2. Wide sectors aggregation, ISIC rev. 4/ NACE rev. 2 Section Tittle Wide sector B Mining and quarrying Industry C Manufacturing: Industry -Food and Beverages Industry -Textiles and Wearing Apparel Industry -Wood and Paper Industry -Petroleum, Chemical and Non-Metallic Mineral Products Industry -Metal Products Industry -Electrical and Machinery Industry -Transport Equipment Industry -Other Manufacturing Industry D Electricity, gas, steam and air conditioning supply Industry E Water supply, sewerage, waste management and remediation activities Industry F Construction Industry G Wholesale and retail trade; repair of motor vehicles and motorcycles Business services I Accommodation and food service activities Business services H Transportation and storage Business services J Information and communication Business services K Financial and insurance activities Business services L Real estate activities Business services M Professional, scientific and technical activities Business services N Administrative and support service activities Business services O Public administration and defense; compulsory social security Other services P Education Other services Q Human health and social work activities Other services R Arts, entertainment and recreation Other services S Other service activities Other services Source: Authors’ elaboration. 23 Table A3. Offshorability and task groups allocation by occupations, ISCO08 2-digit ISCO 08 Offshorability Task Title code group 11 not offshorable NRCP Chief Executives, Senior Officials and Legislators 12 not offshorable NRCP Administrative and Commercial Managers 13 not offshorable NRCP Production and Specialized Services Managers 14 not offshorable NRCP Hospitality, Retail and Other Services Managers 21 not offshorable NRCA Science and Engineering Professionals 22 not offshorable NRCA Health Professionals 23 not offshorable NRCP Teaching Professionals 24 not offshorable NRCA Business and Administration Professionals 25 offshorable NRCA Information and Communications Technology Professionals 26 not offshorable NRCA Legal, Social and Cultural Professionals 31 not offshorable NRCA Science and Engineering Associate Professionals 32 not offshorable NRCP Health Associate Professionals 33 not offshorable RC Business and Administration Associate Professionals 34 not offshorable RC Legal, Social, Cultural and Related Associate Professionals 35 not offshorable NRCA Information and Communications Technicians 41 offshorable RC General and Keyboard Clerks 42 not offshorable RC Customer Services Clerks 43 offshorable RC Numerical and Material Recording Clerks 44 not offshorable RC Other Clerical Support Workers 51 not offshorable NRM Personal Services Workers 52 not offshorable RC Sales Workers 53 not offshorable NRM Personal Care Workers 54 not offshorable NRM Protective Services Workers 61 not offshorable NRM Market-oriented Skilled Agricultural Workers 62 not offshorable NRM Market-oriented Skilled Forestry, Fishery and Hunting Workers 63 not offshorable NRM Subsistence Farmers, Fishers, Hunters and Gatherers 71 not offshorable NRM Building and Related Trades Workers (excluding Electricians) 72 not offshorable RM Metal, Machinery and Related Trades Workers 73 offshorable RM Handicraft and Printing Workers 74 not offshorable NRM Electrical and Electronic Trades Workers 75 not offshorable RM Food Processing, Woodworking, Garment and Other Craft and Related Trades Workers 81 offshorable RM Stationary Plant and Machine Operators 82 offshorable RM Assemblers 83 not offshorable NRM Drivers and Mobile Plant Operators 91 not offshorable NRM Cleaners and Helpers 92 not offshorable NRM Agricultural, Forestry and Fishery Laborers 93 not offshorable NRM Laborers in Mining, Construction, Manufacturing and Transport 94 not offshorable RM Food Preparation Assistants 95 not offshorable NRM Street and Related Sales and Services Workers 96 not offshorable NRM Refuse Workers and Other Elementary Workers Note: NRCA- Non-Routine Cognitive Analytical, NRCP- Non-Routine Cognitive Personal, RC- Routine Cognitive, RM- Routine Manual, NRM- Non-Routine Manual. Source: own elaboration based on (Acemoglu & Autor, 2011; Blinder & Krueger, 2013). 24 Table A4. List of countries used in the study Country name Country ISO3 Source Survey year RTI sample Wage sample Armenia ARM STEP 2013 yes yes Austria AUT PIAAC 2012 yes yes Belgium BEL PIAAC 2012 yes yes Bolivia BOL STEP 2012 yes yes Canada CAN PIAAC 2012 yes no Chile CHL PIAAC 2015 yes yes China CHN CULS 2016 yes no Colombia COL STEP 2012 yes yes Cyprus CYP PIAAC 2012 yes yes Czechia CZE PIAAC 2012 yes yes Denmark DNK PIAAC 2012 yes yes Ecuador ECU PIAAC 2017 yes yes Estonia EST PIAAC 2012 yes yes Finland FIN PIAAC 2012 yes yes France FRA PIAAC 2012 yes yes Georgia GEO STEP 2013 yes yes Germany DEU PIAAC 2012 yes yes Ghana GHA STEP 2013 yes yes Greece GRC PIAAC 2015 yes yes Hungary HUN PIAAC 2017 yes no Indonesia IDN PIAAC 2015 yes yes Ireland IRL PIAAC 2012 yes yes Israel ISR PIAAC 2015 yes yes Italy ITA PIAAC 2012 yes yes Japan JPN PIAAC 2012 yes yes Kazakhstan KAZ PIAAC 2017 yes yes Kenya KEN STEP 2013 yes yes Korea, Rep. KOR PIAAC 2012 yes yes Lao PDR LAO STEP 2012 yes yes Lithuania LTU PIAAC 2015 yes yes Macedonia, FYR MKD STEP 2013 yes no Mexico MEX PIAAC 2017 yes yes Netherlands NLD PIAAC 2012 yes yes New Zealand NZL PIAAC 2015 yes yes Norway NOR PIAAC 2012 yes yes Peru PER PIAAC 2017 yes no Poland POL PIAAC 2012 yes yes Russian Federation RUS PIAAC 2012 yes yes Serbia SRB STEP 2016 yes no Singapore SGP PIAAC 2015 yes no Slovak Republic SVK PIAAC 2012 yes yes Slovenia SVN PIAAC 2015 yes yes Spain ESP PIAAC 2012 yes yes Sweden SWE PIAAC 2012 yes no Türkiye TUR PIAAC 2015 yes no United Kingdom GBR PIAAC 2012 yes yes United States USA PIAAC 2012 yes yes Source: own elaboration. 25 b. Wage inequality analysis Baseline scenario In a first step, we divide the full sample into six groups by broad sector (industry, business, and other services) and type of occupation (offshorable and non-offshorable) and for each group estimate Mincerian wage regressions of the following form:14 = 0 + 1 + 2 + 3 + 4 ∗ + 5 ∗ (1) + 6 + 7 + + + where stands for hourly wages of individual , in occupation , in sector , and in country ; is backward and forward GVC participation in sector and in country ; is output in sector and in country ; measures technology in sector and in country ; are individual skills of worker , in occupation , in sector and in country ; and are, respectively, sector and country fixed effects. Based on the estimated coefficients from equation (1) and actual values for each right-hand side variables, we can predict wages ( ̂ ) for each individual in the six groups. Formally: ̂ = 0 + 1 + 2 + 3 + 4 ∗ + 5 ∗ (2) + 6 + 7 + + For each country, we then calculate the Gini coefficient ( ) of predicted wages: = ( ̂ ) (3) This is our baseline scenario. Scenario of no GVC participation In the second step, we assess the direct contribution of GVC participation to wage inequality ( ). This is based on the estimated models from equation (1), but based on predicted wages conditional on GVC participation values equal to zero ( ̂ ). Formally: ̂ = 0 + 1 + 2 ∗ 0 + 3 ∗ 0 + 4 ∗ 0 ∗ + 5 ∗ 0 ∗ + 6 (4) + 7 + + For each country, we then calculate the Gini coefficient ( ) under the assumption of no integration into GVCs: = ( ̂ ) (5) We describe the direct contribution of GVC participation to wage inequality ( ) as the difference between the Gini coefficients of wages calculated in the baseline scenario and in the scenario of no GVC participation: = − (6) 14 This model is equivalent to equation (3) in the main body of the paper. However, for simplicity reasons the expression + + ∗ + ∗ is noted as . 26 Counterfactual RTI scenario In a third step, we analyze how GVC participation indirectly contributes to wage inequality through its relationship with workers’ RTI ( ). Specifically, we estimate the model of workers’ RTI and then calculate 15 ̂ counterfactual worker-level RTI, assuming GVC participation values equal to zero ( ). Formally: = 0 + 1 + 2 + 3 ∗ + 4 ∗ + 5 (7) + 6 + + ̂ = 0 + 1 ∗ 0 + 2 ∗ 0 + 3 ∗ 0 ∗ + 4 ∗ 0 ∗ + 5 + 6 (8) + + ̂ We then use the estimated models from equation (1) to predict wages ̂ conditional on . To isolate the indirect contribution of GVC participation to wage inequality through RTI, we use the observed values of GVC participation in the wage model: ̂ ̂ = 0 + 1 + 2 + 3 + 4 ∗ + 5 (9) ∗ + 6 + 7 + + We describe the indirect contribution of GVCs participation to wage inequality ( ) as the difference between the Gini coefficients of wages calculated in the baseline scenario ( ) and the Gini coefficients of wages in the counterfactual RTI scenario ( ). = ( ̂ ) (10) = − (11) Total contribution of GVC participation In a fourth step, we calculate the total contribution of GVC participation to wage inequality ( ). We set the GVC participation values to zero (as in the calculation of the direct contribution), and we use the counterfactual ̂ RTI conditional on zero GVC participation ( , as in the calculation of the indirect contribution) to predict wages using the estimated coefficients in the models from equation (1). ̂ ̂ = 0 + 1 + 2 ∗ 0 + 3 ∗ 0 + 4 ∗ 0 ∗ + 5 ∗ 0 ∗ + 6 (12) + 7 + + We define the total contribution of GVC participation to wage inequality ( ) as the difference between the Gini coefficient of wages in the baseline scenario ( ) and the Gini coefficient of wages in this last scenario ( ). = (̂ ) (13) = − (14) 15 Equation (7) is equivalent to equation (2) in the main body of the paper. 27 Appendix B – Additional results Table B1. The Correlates of Routine Task Intensity (RTI) at the Worker Level, in the pooled sample, and by broad sectors, standardized (backward and forward GVC) Panel A: Pooled (1) (2) (3) (4) (5) (6) All workers Non-offshorable Offshorable All workers Non- offshorable Offshorable Backward Global Value Chain participation (GVCB) share in exports (std.) -0.000 -0.007 0.046** (0.010) (0.010) (0.019) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.011 -0.013 0.015 (0.007) (0.008) (0.014) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.019* 0.013 0.039*** (0.011) (0.012) (0.015) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.036 0.049 -0.038 (0.028) (0.031) (0.033) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.016 -0.006 -0.075 -0.006 0.006 -0.072 (0.038) (0.039) (0.060) (0.036) (0.036) (0.059) Observations 167,253 145,122 22,131 167,034 144,914 22,120 Panel B: Industry Backward Global Value Chain participation (GVCB) share in exports (std.) 0.019 -0.011 0.077** (0.024) (0.026) (0.030) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.015 0.026 0.005 (0.030) (0.032) (0.043) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.029* 0.011 0.048** (0.015) (0.018) (0.020) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.024 0.029 0.009 (0.031) (0.034) (0.046) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.112 -0.091 -0.364*** -0.089 -0.032 -0.334*** (0.070) (0.074) (0.110) (0.067) (0.075) (0.109) Observations 38,949 28,816 10,133 38,917 28,790 10,127 28 Panel C: Business services Backward Global Value Chain participation (GVCB) share in exports (std.) -0.086*** -0.093*** -0.014 (0.018) (0.020) (0.034) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.048** 0.058*** -0.054* (0.019) (0.020) (0.031) Forward Global Value Chain participation (GVCF) share in exports (std.) -0.021 -0.024 -0.001 (0.015) (0.016) (0.023) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.101*** 0.121*** -0.052 (0.031) (0.033) (0.039) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.065 -0.051 -0.136* -0.051 -0.038 -0.132* (0.044) (0.045) (0.075) (0.044) (0.045) (0.075) Observations 72,153 63,173 8,980 71,979 63,003 8,976 Panel D: Other services Backward Global Value Chain participation (GVCB) share in exports (std.) 0.265*** 0.249*** 0.436*** (0.075) (0.076) (0.149) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.371*** -0.363*** -0.478* (0.113) (0.114) (0.256) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.075*** 0.070*** 0.133*** (0.025) (0.026) (0.044) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.234** -0.235** -0.236 (0.104) (0.103) (0.224) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.199*** -0.199*** -0.292** -0.130** -0.134* -0.176 (0.070) (0.073) (0.135) (0.066) (0.069) (0.128) Observations 50,843 48,133 2,710 50,843 48,133 2,710 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for Computer Use, GVCB share and FDI/GDP are standardized. All regressions include controls for technology (computer use, computer use squared), FDI, skills, education, age, gender, sector FE, and sector FE interacted with GDP per capita. Source: Authors’ calculations based on Lewandowski et al. (2022) and PIAAC, STEP, CULS (tasks), and World Bank (GDP, taxonomy groups, government education spending), EORA data and Borin and Mancini (2015, 2019) (GVC participation measures). 29 Table B2. Pooled regression of backward and forward and by wide sectors and occupational groups, standardized (backward and forward GVC) Panel A: Pooled (1) (3) (2) (4) All workers Middle- High-skilled Low-skilled skilled occupations occupations occupations (ISCO 1-3) (ISCO 7-9) (ISCO 4-5) Backward Global Value Chain participation (GVCB) share in exports (std.) 0.004 -0.031* -0.059** 0.064*** (0.019) (0.017) (0.025) (0.022) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.017 -0.035 0.085** 0.023 (0.029) (0.026) (0.039) (0.030) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.019* 0.019* 0.009 0.046*** (0.010) (0.011) (0.017) (0.011) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.060*** -0.074*** -0.042** -0.045*** (0.012) (0.014) (0.016) (0.013) Ln(GDP per capita) –mean(Ln(GDP per capita)) 0.033 0.011 0.013 0.107** (0.038) (0.037) (0.046) (0.050) Observations 167,034 68,439 52,895 45,700 Panel B: Industry Backward Global Value Chain participation (GVCB) share in exports (std.) 0.027 -0.051** 0.038 0.056** (0.024) (0.020) (0.036) (0.024) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.004 0.066** -0.051 0.003 (0.032) (0.031) (0.044) (0.038) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.030** 0.017 -0.026 0.051*** (0.015) (0.018) (0.032) (0.016) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.043*** -0.006 -0.011 -0.060*** (0.017) (0.021) (0.034) (0.018) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.096 -0.109 -0.237** 0.016 (0.066) (0.074) (0.099) (0.060) Observations 38,917 11,245 4,208 23,464 Panel C: Business services Backward Global Value Chain participation (GVCB) share in exports (std.) -0.055** -0.037 -0.098** 0.030 (0.024) (0.028) (0.044) (0.032) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] 0.063* -0.070** 0.101* 0.124*** (0.037) (0.035) (0.053) (0.043) Forward Global Value Chain participation (GVCF) share in exports (std.) -0.015 0.010 -0.026 0.021 (0.016) (0.014) (0.020) (0.023) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.046*** -0.089*** -0.029 0.006 (0.017) (0.019) (0.018) (0.025) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.027 -0.008 -0.064 -0.054 (0.049) (0.050) (0.059) (0.067) Observations 71,979 24,754 32,362 14,863 30 Panel D: Other services Backward Global Value Chain participation (GVCB) share in exports (std.) 0.220*** 0.058 0.251** 0.640*** (0.079) (0.085) (0.100) (0.127) GVCB share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.329*** -0.242** -0.342*** -0.451*** (0.114) (0.123) (0.123) (0.145) Forward Global Value Chain participation (GVCF) share in exports (std.) 0.029 0.030 0.038 0.073 (0.023) (0.027) (0.031) (0.047) GVCF share (std.) * [Ln(GDP pc) –mean(Ln(GDP pc)] -0.073** -0.078** -0.082** 0.049 (0.032) (0.034) (0.042) (0.055) Ln(GDP per capita) –mean(Ln(GDP per capita)) -0.210*** -0.225*** -0.084 -0.024 (0.076) (0.079) (0.108) (0.113) Observations 50,843 31,609 15,051 4,183 Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors in parentheses. Standardized weights are used that give each country equal weight. The standard errors are clustered at a sector × country level. Measures for GVCB share and GVCF share are standardized. All regressions include controls for technology (computer use, computer use squared), FDI, skills, education, age, gender, sector FE, and sector FE interacted with GDP per capita. Source: Authors’ calculations based on Lewandowski et al. (2022) and PIAAC, STEP, CULS (tasks), and World Bank (GDP), EORA data and Borin and Mancini (2015, 2019) (GVC participation measures). Figure B1. The contribution of GVC participation to wage inequality, residual term. Source: Authors’ calculations based on Lewandowski et al. (2022) and PIAAC, STEP, CULS (tasks), EORA data and Borin and Mancini (2015, 2019) (GVC participation measures). 31