Embed Size (px)
Citation preview
Chains of Love?Global Production and the Firm-Level Diffusion of Labor Standards
Eddy MaleskyDept. of Political Science
Duke [email protected]
Layna Mosley
Dept. of Political ScienceUniversity of North Carolina at Chapel Hill
Word Count: 9,487September 28, 2017
Draft 4.19 Abstract: Under what conditions does the global economy serve as a means for the diffusion oflabor standards and practices? We anticipate variation among internationally engaged firms in theirpropensity to improve labor standards. Upgrading is most likely when a firm’s products exhibitsignificant cross-market differences in markups, making accessing high-standards overseas marketsparticularly profitable. Additionally, upgrading is more likely when lead firms attach a high salienceto labor standards. Therefore, while participation in global production induces “trading up”behaviors among firms overall, the effect strength varies across industries. We test our expectationsvia a survey experiment, which queries foreign firms operating in Vietnam about their willingness toinvest in labor-related upgrading. We find strong evidence for the effect of markups on upgradingchoices and suggestive evidence for the saliency mechanism.
Acknowledgement: For comments on previous versions of this paper, we thank Greg Distelhorst,Irfan Nooruddin, Ivan Png, Ryan Weldzius and participants in The Politics of Multinational Firmsconference (Princeton University, September 2016), the 2016 International Political EconomySociety meetings, and the 2017 meetings of the International Studies Association.
2
To what extent does participation in the global economy serve as a means for improving
labor standards and working conditions? While some worry that multinational production
encourages the competitive lowering of standards, others suggest that economic globalization not
only may facilitate greater efficiency and economic growth, but also that it creates incentives for
“trading up” (Vogel 1995). Firms operating in developing countries that want to export globally,
especially to richer countries, may need to improve their practices in order to comply with
destination market regulations and preferences. Governments, interested in positive trade balances,
support these improvements.
Scholars have applied the “trading up” logic to labor as well as environmental standards,
typically treating all internationally involved, developing-country firms as similar in their motivations.
Such accounts assume that the desire to service developed-country markets, coupled with
regulations and preferences in those markets, is sufficient to compel upgrading (e.g. Greenhill et al
2009, Prakash and Potoski 2006, Vogel 1995). By contrast, we expect variation among
internationally engaged firms. Upgrading is most likely when a firm’s products exhibit significant
cross-market differences in markups – making accessing high-standards overseas markets
particularly profitable. Additionally, upgrading is more likely when lead firms attach a high salience
to labor standards. Therefore, while participation in global production induces “trading up”
attitudes and behaviors overall, the strength of the effect varies across industries and firms.
Internationally active firms based in developing countries are central to trade-based diffusion
accounts. These firms engage in arm’s length transactional relationships with lead global firms.1
Lead firms may subcontract with multiple producers, in varied locations, and at multiple production
stages, for raw materials, intermediate inputs or finished goods (Locke 2013). Internationally
1 In this piece, we use “supply chains,” “value chains” and “commodity chains” interchangeably. See Gereffi 2014.
3
engaged developing country firms often are larger than their domestically focused counterparts
(Melitz and Ottaviano 2008). Many of these firms are foreign owned, generating foreign direct
investment (FDI) flows as well. In Vietnam, the setting for our study, wholly foreign-owned
enterprises account for over 71% of exports, and consequently the vast majority of employment in
exportable sectors (Nguyen and Ramstetter 2017). While multinational corporations (MNCs) and
their directly owned affiliates employed an estimated 82 million individuals worldwide in 2016 –
compared with 21 million in 1990 – a far greater number of individuals are employed in other firms
that subcontract production for multinationals (Shepherd 2013, OECD et al 2014, UNCTAD 2017).
If “trading up” alters labor conditions in low and middle-income countries, it is via these
types of developing country firms, which interact with – and are influenced by – actual and potential
supply chain partners abroad. Yet, international political economy scholars have paid little attention
to developing country firms in low and intermediate supply chain positions (Kim et al. 2017,
Osgood et al. 2016). Consistent with a “trading up” view, we expect that participation in global
supply chains generally creates incentives for developing country firms to upgrade labor-related
practices. Servicing foreign markets typically offers greater revenue opportunities and, in many
instances, requires the use of more skilled production techniques. By hiring and retaining the most
skilled local workers, firms can capture the material gains that accrue from servicing new markets.
We also expect that the incentives to “trade up” vary across firms. Firms in developing
countries will be most inclined to improve their standards when servicing those foreign markets that
offer relatively greater product markups. Higher markups serve to justify the investment in rights-
related improvements. Lead firms, shareholders, activists and consumers may further deepen firms’
incentives to upgrade. Labor-related issues are most salient to lead firms concerned about the
reputational risk associated with reports of child labor, hazardous working conditions, or forced
overtime (Gereffi 1994). In part, salience to firms reflects activists’ targeting of certain firms and
4
industries, often those producing branded or luxury products. Lead firms’ concerns also can stem
directly from shareholder or consumer pressures. And in capital intensive activities, better labor-
related practices may correlate highly with worker productivity and output quality (Distelhorst and
Locke 2017), offering sourcing agents and lead firms another reason to attend to worker rights.
We assess these expectations using firm-level data from foreign invested, manufacturing-
oriented firms operating in Vietnam. These 912 foreign-invested enterprises (FIEs) play a variety of
roles in their product markets, engaging in differing transactional relationships with lead, partner and
supplier firms. We ask firms about their overall willingness, measured as a percentage of operating
costs, to expend on upgrading, if such upgrading renders them eligible for contracts with overseas
lead firms. We find that multinational production indeed provides incentives for improving
workers’ treatment. Further, we establish that firms’ willingness to upgrade varies, both with the
location of the potential partner (Europe versus India) and with the type of product. These patterns
result not from the location of the destination market per se, but from the difference in markups
available in European versus in Indian markets, and from the salience to supply chain partners of
labor practices. We find strong evidence that participation in global supply chains is most likely to
motivate labor-related upgrading when higher markups are available in developed – versus
developing – country markets; we find more tentative evidence of upgrading when lead firms attach
high salience to labor rights.
I. Supply Chains and the Diffusion of Standards in the Developing World
Scholars and activists have long debated the consequences of global production for workers
in developing countries. Cost and time pressures can create incentives for labor rights violations,
especially for labor-intensive products with short life cycles (Locke 2013). Evidence for competition-
driven lowering of standards, however, is mixed (Adolph et al 2017, Mosley 2011). Indeed,
5
international economic ties may promote, rather than diminish, respect for labor rights. High
standards allow multinational affiliates to recruit the most skilled local workers. And multinationals,
interested in efficiency and standardization, often bring their (better) home country practices to
foreign affiliates (Garcia-Johnson 2000). Activists’ campaigns may create additional incentives for
multinational firms to act in “socially responsible” ways (Bartley 2007). Empirically, FDI is
associated with greater protection of labor rights, as well as with wage premiums (Mosley 2011,
Shepherd 2013).
Moreover, production for foreign markets may facilitate the diffusion of higher standards.
When some export markets have higher environmental standards, firms that want to service such
markets adopt the stricter requirements of destination markets (Vogel 1995). If these markets are
significantly large, firms will adjust all of their products (rather than only those destined for higher-
standards markets). Vogel’s “California effect” description referenced automobile manufacturers’
(both those based in the US and those based overseas) adoption of more stringent state of California
fuel economy requirements.
Scholars have since pointed to other “California effect” processes, exploring the
globalization-based diffusion of environmental outcomes, product standards and labor rights.
Prakash and Potoski (2006) suggest that developing country firms use the adoption of voluntary
environmental standards to signal their practices to overseas partners. These signals are particularly
valuable when domestic regulations are weak (Berliner and Prakash 2014), and when one’s trade
competitors also have adopted standards (Cao and Prakash 2011). Perkins and Neumayer (2012)
also report a robust association between automotive industry exports to highly regulated countries
and the stringency of developing country emission regulations.
Turning to worker rights, Greenhill et al. (2009) suggest that trade relationships can transfer
labor standards from destination markets to producer countries. The composition of a developing
6
country’s trade (in terms of labor rights in its export markets) is a significant predictor of its labor
standards. Relatedly, Adolph et al. (2017) investigate whether trade with a lower-standards country
(China) is associated with deteriorations in African countries’ labor rights.
What is notably absent from much of the “California effect” work is firm-level analysis. The
actions and incentives of firms are key to causal claims of diffusion. Businesses in developing
countries adopt more stringent standards in order to signal quality to potential buyers and supply
chain partners. Alternatively, multinational affiliates lobby host country governments to tighten
regulations, as a means of raising their local competitors’ costs (Garcia-Johnson 2000). And local
firms and governments worry that labor rights violations will reduce foreign demand, so they adopt
codes of conduct or improve legal protections (Distelhorst and Locke 2017, Locke 2013).
These firm- and industry-level mechanisms are typically assumed, however, rather than
tested. Greenhill et al (2009)’s country-level analysis is agnostic regarding the specific causal process
by which California effect improvements occur (also see Adolph et al 2017). Similarly, studies of
voluntary standards typically measure rates of adoption at the national or sectoral, rather than firm,
level (but see Berliner and Prakash 2014). But the incentives to upgrade should vary within
countries, as a function of industry and firm characteristics.
II. Developing Country Firms and the Diffusion of Labor Standards
Our baseline expectation is that internationally active firms in developing countries are
willing to expend significantly on labor-related upgrading. Downstream firms anticipate that
purchasing from foreign suppliers lowers their overall input costs and/or improves the quality of
their components. Upstream firms expect that supplying globally active firms creates greater demand
for their products and may allow them to capture higher markups. Supply chain participation also
facilitates, under some conditions, the transfer of more advanced production techniques (Distelhorst
7
et al 2016, Sutton 2013). All foreign-engaged firms therefore should be keen to gain access to
overseas supply chains and export markets.
The availability of markups in overseas markets is central to our theory. Product markups
represent the difference between an item’s marginal cost of production and its price. With perfectly
competitive markets and high price elasticity of demand, markups are small or non-existent. Firms
set prices equal to marginal cost. Empirical analyses in economics, however, find that markups
frequently exist (i.e. Gullstand et al 2014, Milburg and Winkler 2013). By allowing some firms to act
as price setters, anti-competitive regulations (including, inter alia, trade barriers) and market
concentration can create positive markups (Melitz and Ottaviano 2008).
Additionally, more productive firms exploit their competitive advantages to charge higher
markups (de Loecker and Warzynksi 2012). And consumers’ desire for differentiated products
(“love of variety”) reduces their elasticity of demand; firms that produce a new or different variety of
a good can earn higher markups (Broda and Weinstein 2006). International trade amplifies these
dynamics, as firms spread their fixed production costs more widely, increasing the returns from
offering a new variety. Indeed, de Loecker and Warzynski (2012) report that exporting firms charge
higher markups than their domestically oriented counterparts (also see Gullstrand et al 2014). Hence,
we expect that developing country firms will be more willing to invest in labor-related upgrading if
such improvements facilitate opportunities for higher markups abroad.
Developed country governments often link labor standards with trade liberalization, creating
an explicit – albeit sometimes ineffective – link between market access and worker rights (Lechner
2016). Firms, shareholders and consumers in developed markets also may devote attention to labor
rights violations in lead firms’ affiliates or subcontractors (Distelhorst et al 2016, Gereffi 2014,
8
Locke 2013),2 asking suppliers to participate in enterprise- or industry-level codes of conduct,
certification schemes and reporting requirements (Bartley 2007, Vogel 2006).
Despite doubts regarding the efficacy of corporate social responsibility programs, especially
where political institutions are weak (e.g. Berliner and Prakash 2015, Locke 2013), many developed-
country lead firms publicly express their preferences for higher labor standards. They devote
resources to influencing the behaviors of their supply-chain partners. We therefore expect
developing country firms to signal their willingness to upgrade labor standards as a means of
increasing their appeal as partners. Indeed, Distelhorst and Locke (2017) report that compliance
with labor and environmental standards leads to a four percent average annual increase in lead-firm
purchases from developing country manufacturing firms. Görg et al’s (2017) study of 2000 foreign
firms in 19 African states similarly reveals that corporate social responsibility considerations are
particularly salient for firms that export their output to developed (versus developing) nations.
H1: Internationally active firms in developing countries will be more willing to invest in labor-related improvements when offered the opportunity to export to developed, rather than developing, country markets. Product markups may vary not only between home and foreign markets, but also between
firms and industries within the same market (Milburg and Winkler 2013). In some industries, firms
are able set higher prices for the same good in some markets. In other industries, price
discrimination is – given the nature of consumers’ preferences – less feasible. Simanovska (2015)
reports evidence of higher markups for apparel in wealthier countries; she attributes this result partly
to consumers’ price sensitivity. Because higher-income consumers are less sensitive to price changes,
apparel firms apply larger markups in more affluent markets. Gullstrand et al (2014) offer evidence
2 Note that this mechanism relies, in part, on consumer demand for ethical consumption. See Hainmueller et al 2014.
9
for a similar dynamic in processed foods (although not in the wholesale sector). We also find (see
Section IV) evidence of variation across industries in the markup differentials available to
Vietnamese firms.
Given this variation, we expect that firms will differ in the intensity of their preferences over
destination markets. Firms that can capture greater markups via price discrimination will place a
high premium on access to developed (rather than developing) country markets. They will be more
willing to upgrade standards in order to service developed country markets, as the material benefit
from doing so is significant. In other industries, firms will want access to foreign supply chains, but
– given low markup differentials -- will be largely indifferent across foreign markets. The horizontal
dimension of Table 1 offers examples of industries with small and large markup differentials.
H2: Internationally active firms will be more willing to invest in labor-related improvements in industries characterized by high relative markups.
Even when cross-market differences in markups are similar, we expect a second industry-
level factor to matter -- the extent to which labor-related concerns are salient to lead firms. To what
extent do lead firms worry that they will incur reputational (and therefore material) costs if labor-
related problems are discovered in their supply chains? Salience often is the result of external
pressures. “Naming and shaming” is intended to strengthen multinational firms’ incentives to
address labor standards (Bartley and Child 2014). Human rights activists typically have called
attention worker rights violations in industries, such as apparel, with labor-intensive production and
the presence of branded products (Gereffi 2014).
Activists also have focused on firms that produce luxury (rather than necessity) goods, such
as handmade carpets or diamonds. Firms may respond by creating or joining certification and
labeling schemes, such as Rugmark; these initiatives also allow socially conscious consumers to link
10
their purchasing decisions with labor rights outcomes (Hainmueller et al 2014). Still other targets are
industries (toys, soccer balls) in which the use of child labor can be linked with the consumption of
finished goods by other children. Other activist campaigns have targeted commodity firms with
global brands, such as oil and gas companies. These campaigns, if successful, generate losses in
retail as well as investment markets.
Activist campaigns raising the visibility of labor issues for consumers are only one path by
which worker rights can become salient to lead firms. Another route, relevant to developing country
producers of intermediate goods, is lead firms’ supply chain management principles; these increasing
includes not only technical efficiency, but also ethical production, in directly owned affiliates as well
as in supply chain partners (Locke 2013). Such principles address shareholders’ ethical production
concerns; they may be more pronounced among publicly traded firms. Therefore, even if consumers
do not observe the conditions under which intermediate goods are produced, lead firms may
nonetheless emphasize high labor standards to suppliers. Multi-stakeholder initiatives such as ILO’s
Better Work Program – focused on apparel and other light final manufacturing – also bring
attention to labor conditions. A third, and related, pathway is the skill or knowledge-intensity of
production. When the production of a good relies on workers with advanced and specific skills,
firms have a greater interest in hiring and retaining the most qualified individuals. Better working
conditions enable firms to increase the retention of their most productive employees.
As the salience of labor rights increases, lead firms are more likely to condition their
sourcing decisions on respect for worker rights (Distelhorst and Locke 2017). For developing
country firms, higher salience means that labor-related upgrading is even more important as a
mechanism to gain access to markup differentials. We therefore expect, as the vertical dimension of
Table 1 indicates, that developing country enterprises will be even more inclined to invest in labor-
related improvements when supplier firms’ labor practices are highly salient to lead firms:
11
H3: Internationally active firms will be more willing to invest in labor-related improvements in in industries characterized by high salience of labor issues.
Table1:PredictedEffectsofMechanismsforLaborRightsImprovements
To test these hypotheses, we draw on a survey of FIEs in Vietnam. It is important to note
that our theoretical claims should apply to all internationally active firms in developing countries,
regardless of their ownership. Because foreign-owned firms are larger and more efficient than their
domestic counterparts, we expect that they represent the leading edge of diffusion. But, were we to
conduct a similar empirical study of internationally active, domestically owned firms, we would
expect similar patterns.
In studying developing country firms, we draw attention to entities that thus far received
limited attention from international political economists. The dominant theories relating global
production to labor outcomes are based largely on the behaviors of large lead firms from developed,
Low High
LowTreatmentEffect ModerateTreatmentEffect(i.e.Plastics/Rubber;Commodities) (i.e.FabricatedMetals;Chemicals)
ModerateTreatmentEffect LargeTreatmentEffect(i.e.Garments/Apparel) (i.e.Computers;Electronics)
DifferenceinMark‐Ups(Developed‐Developing)
SalienceofLaborPractices
Low
High
12
Western countries. Developing nations, however, now play important roles as sources of FDI; these
countries, including Brazil, China and India, accounted for 26 percent of global FDI outflows in
2016, down from 39 percent in 2014. Firms from these countries often invest in other low- and
middle-income locales, frequently in the same geographic region.
Moreover, much productive activity occurs through arms-length supply chain relationships,
rather than within the boundaries of multinational firms. For instance, Nike Inc.’s apparel,
equipment and footwear products are manufactured in 666 subcontractor factories worldwide,
located in 44 countries, and employing just over one million workers in total.3 Some of these
contracting firms are themselves large and multinational. Such arrangements also are common in
electronics; Taiwan-based Foxconn is the world’s largest contracting manufacturer.
The prevalence of supply chain production means that understanding “California effect”
upgrading requires a more deliberate consideration of non-lead firms. Economic sociologists have
long considered the structure and evolution of global value chains (e.g. Gereffi 1994, 2014). More
recently, Johns and Wellhausen (2016) posit that supply chain partnerships affect domestic firms’
willingness to expend political capital to protect their partners’ property rights and, therefore, host
governments’ willingness to breach contracts with foreign firms. Manger (2012) demonstrates that
lead firms based in the North, which desire cheaper and regular access to inputs produced in the
South, lobby their governments to conclude North-South preferential trade agreements. Similarly,
Osgood (2016) finds that supply chain participation is a key determinant of U.S. firms’ public
positions on trade liberalization. And Jensen et al. (2015) attribute the decline of U.S. firms’ anti-
dumping claims, even in the face of currency undervaluation, to supply chain linkages.
3 http://manufacturingmap.nikeinc.com/
13
We draw attention to the role of developing country firms in the diffusion of standards.4 We
expect that these firms – both current and potential supply chain participants – will be more inclined
to upgrade their labor-related practices when they transact with developed country-based firms. This
effect will be greatest when there is a significant difference in product markups between developed
and developing country markets, and when labor conditions are highly salient to lead firms.
III. Research Design and Estimation Strategy
To evaluate our expectations, we employ data from the 2015 Vietnam Provincial
Competitiveness Index survey of FIEs (PCI-FDI). Vietnam is an appropriate place to test our
theories; it is among the developing world’s most important FDI destinations. In both 2014 and
2015, FDI Intelligence, a division of the Financial Times, ranked Vietnam first among all emerging
economies in its Greenfield FDI Performance Index. Vietnam’s 2015 score of 6.45 means that it
attracted over six times more new investment capital than its share of global output predicts. The
next-highest ranked emerging economy was Hungary (4.32); China’s score was 0.41(Barklie 2016).
Importantly for our theory, the vast majority of Vietnamese FDI is in manufacturing, especially
garments, electronics, and food processing; inward FDI is typically targeted at foreign export
markets.
The fully anonymous PCI-FDI survey includes 1,584 foreign invested entities, drawn from
the 14 Vietnamese provinces with significant FDI activity.5 Vietnam’s General Statistical Office lists
12,571 eligible (tax-paying) foreign firms; the PCI samples from this set, using stratification to
ensure that firm age, legal type and industry are accurately represented. The survey had a response
rate of 25%, which climbs to 51% when incorrect addresses in the sampling frame are dropped.
4 Note that studies using supplier factory audits, as in Locke (2013), Locke et al (2013), rely on data from such firms, although lead firms (e.g. Hewlett-Packard, Nike) are the central agents. 5 Methodological details and background on the survey can be found at <http://www.pcivietnam.org>.
14
Responding firms therefore represent over eight percent of the entire population of foreign invested
projects in Vietnam since 1988. 6 Over 87 percent of PCI-FDI respondents indicate that they are
wholly foreign owned. Figure 1 provides details on the reported country of origin. South Korea,
Taiwan, Japan, and mainland China account for 68 percent of the businesses surveyed. If we include
investment from neighboring countries in Southeast Asia, the figure surpasses 80 percent.7
As Figure 2 shows, PCI-FDI respondents are concentrated in manufacturing (64 percent),
although no particular type of manufacturing dominates. In 2015, the three biggest industries after
general manufacturing were fabricated metal products (8.7%), rubber and plastics (6.4%), and
apparel/garments (6.4%). Motor vehicles, chemical products, machinery, and computers and
electronics follow, each with about 4% of the sample. In terms of employment, FIEs in Vietnam
tend to be larger than private domestic firms, by a factor of three. But, by international standards,
these firms are rather small: the average FIE has 220 employees, and 74 percent of FIEs have fewer
than 300 employees, for a median employment of 125. There is some large firm representation: 93
respondent firms employ more than 1,000 workers. FIEs also are relatively small in investment size,
with an average of $2.2 million in capital. Typically, FIEs in Vietnam are export-oriented (62% of
manufacturers engage in some form of export, to their home or to a third country). Some FIEs list
other foreign owned companies in Vietnam as the primary purchasers of their products. These
enterprises account for 71% of Vietnam’s total exports. The export propensity of FIEs in the
manufacturing sector is even higher, accounting for 80% of manufactures exports (Nguyen and
Ramstetter 2017).
6 The PCI research team further ensures that each year this survey is representative of the population of firms in Vietnam through random sampling with stratification. Seventy percent of respondents list themselves as the CEO, General Manager, or General Director of the operation; the rest include finance officers or others knowledgeable about operations. 7 These numbers correspond closely to the calculations made by the Ministry of Planning and Investment (MPI) and GSO. GSO, 2015. “The Enterprise Survey.” http://www.gso.gov.vn/Default.aspx?tabid=217
15
Figure1:NumberofFirmsfromEachCountryinSample
FIEs often focus on lower value-added activities such as final assembly. For instance,
motorcycle and autos are produced with kits: all of the high value inputs are imported from
elsewhere, and Vietnamese workers simply assemble the vehicle (Fujita 2011, Ngo 2015). Garment
production, Vietnam’s leading manufacturing industry, is similar. About 70 percent of Vietnam’s
textile and apparel production uses imported textiles and other inputs, predominantly from China
(ITA 2016a). As such, respondent firms are situated in the less rewarded segments of the value
chain, but they often aspire to move up the chain (Doner 2009).
While foreign firms in Vietnam sometimes are involved in global supply chains, most of
them are owned and managed independently of lead firms. Only 31% of sampled firms are part of a
11111111222222223334556688812131415
262730
7179
115219
351376
0 50 100 150 200 250 300 350 400
AustriaCosta Rica
CyprusIreland
MonacoRussiaSpain
SwedenCayman Island
IsraelItaly
MauritiusNew Zealand
PhilippinesSeychelles
Virgin IslandsHongKong - China
NorwayUnited Arab Emirates
CanadaBelgium
SwitzerlandIndia
SamoaBrunei
DenmarkIndonesia
NetherlandsThailandAustraliaGermany
FranceUnited Kingdom
MalaysiaUnited States
SingaporeChina
TaiwanSouth Korea
Japan
All Firms
11111111111111223333455555688
151919
4145
63127
192231
0 25 50 75 100 125 150 175 200 225 250
AustriaCanada
Cayman IslandCosta Rica
CyprusIsrael
MauritiusMonaco
New ZealandPhilippines
RussiaSeychelles
SpainVirgin Islands
HongKong - ChinaItaly
NorwaySamoa
SwitzerlandUnited Arab Emirates
IndiaBelgium
DenmarkGermany
IndonesiaThailand
BruneiAustralia
NetherlandsUnited Kingdom
FranceMalaysia
SingaporeUnited States
ChinaTaiwan
South KoreaJapan
Firms in Exporting Sectors
Second panel only includes firms in exporting sectors that responded to survey experiment.
16
Figure2:NumberofFirmsfromEachIndustryinSample
larger MNC. Among the 638 firms (nearly 70 percent of the sample) that are not part of a broader
multinational ownership structure, some (234 firms) export their main product mostly to their
country of ownership, while others (258 firms) export to third countries. Further, the vast majority
(88%) of FIEs in exporting sectors entered as 100% foreign-owned operations. Fewer than 6%
entered as joint ventures with domestic entities. Investment tends to be greenfield (creating new
entities), with only 7% entering by merging with or acquiring existing entities.
9
11
11
11
12
23
23
24
26
27
28
31
34
35
35
36
38
41
62
64
67
70
71
85
110
0 10 20 30 40 50 60 70 80 90 100 110 120
Real Estate
Agriculture/Aquaculture
M:Wood Products
Financial/Insurance
Electricity/Gas/AC
M:Food Processing
M:Basic Metals
M:Paper Products
M:Leather
M:Textiles
Other Services
M:Furniture
M:Electronic Equip.
M:Machinery
Construction
M:Computers/Electronics
M:Motor Vehicles
M:Chemicals
Information/Communication
M:Garments
Professional Services
M:Rubber/Plastics
M: Other
M:Fabricated Metals
Wholesale/Retail
All Firms
11
11
11
23
23
24
26
27
31
34
35
36
38
41
62
64
67
70
71
85
110
0 10 20 30 40 50 60 70 80 90 100 110 120
Agriculture/Aquaculture
M:Wood Products
Electricity/Gas/AC
M:Food Processing
M:Basic Metals
M:Paper Products
M:Leather
M:Textiles
M:Furniture
M:Electronic Equip.
M:Machinery
M:Computers/Electronics
M:Motor Vehicles
M:Chemicals
Information/Communication
M:Garments
Professional Services
M:Rubber/Plastics
M: Other
M:Fabricated Metals
Wholesale/Retail
Firms in Exporting Sectors
Second panel only includes firms in exporting sectors that responded to survey experiment; M: Denotes Manufacturing Sector.
17
The 2015 PCI-FDI asks a series of approximately twenty labor-related questions, which take
roughly fifteen minutes for respondents to complete. We test our theoretical expectation using an
experiment embedded in one of these items (see Figure 3). Of respondent firms, 478 FIEs were
assigned to the treatment group, while 434 were assigned to the control.8 We ask respondents to
imagine a scenario in which an international consultant has contacted the firm, as part of its efforts
to connect large multinationals with suppliers in emerging markets. The question states that, to be
shortlisted as a potential supplier for the multinational client, the Vietnamese firm would need to
adopt the multinational client firm’s Labor Code of Conduct for Suppliers. The code covers health
and safety regulations, limitations on overtime hours, and greater worker representation. As such, it
is typical of industry-wide, multinational firm, and supplier codes of conduct, which originated in the
late 1990s and are now widespread in both developed and developing countries (Locke 2013).
We describe the code as one that will increase operating costs, but also increase the
possibility of future orders. It is important to note that codes of conduct tend to increase variable
costs, requiring ongoing expenditures that vary with the level of output (i.e. limits on overtime,
greater worker capacity to bargain over wages, safety equipment for each worker). While some
elements of codes, such as building and fire safety, represent fixed costs, which do not vary with
each unit of output, codes largely imply increases in variable costs.9 As a result, the code’s promise
of access to overseas supply chains should prime respondents via the markup mechanism (higher
prices per unit), rather than via a size of market effect.
8 1,584 foreign firms responded to the PCI-FDI survey. We limit our analysis to 1,413 firms in sectors that have export potential, including agriculture, aquacultures, manufacturing and some services. Of these firms, 577 (36%) said the question was non-applicable, because they were targeting the Vietnamese market. Our empirical analysis therefore focuses on the 912 export-potential firms that responded to our survey experiment item (dropping 4 non-tradeable sectors: non- 1) construction; 2) other services; 3) finance and insurance; and 4) real estate). When we cluster standard errors (at the province, industry, or country of origin level), our sample sometimes drops to 886, because of missing data on industry. 9 We thank an anonymous reviewer for suggesting we clarify this point.
18
We use a contingent valuation approach to parsimoniously capture firms’ interest in labor
upgrading (Mitchell and Carson 1989, Cummings and Taylor 1999). The specific reforms necessary
to improve labor conditions may vary according to industry, production stage, manufacturing
technology and employment demographics. For firms engaged in cutting and sewing fabric, for
instance, fire safety is often a major concern. For businesses making plastic products, chemical
exposures are the most significant challenge. While one could ask multiple questions measuring
attitudes on various labor dimensions, some items would apply to only some firms. Aggregating
these items into an overall index of firms’ willingness to upgrade would be problematic. The
contingent valuation method allows us to measure the propensity for labor-related upgrading in a
way that is comparable across FIEs. We ask respondent firms to specify the maximum costs of
adjustments – ranging from 0 to 15%, as a percentage of current operating costs – they would be
willing to make to comply with the code. The adjustment cost options are consistent with prevailing
estimates of implementing internationally recognized labor codes of conduct.
Surveyed firms report being willing to spend, on average, 6 to 7 percent of operating costs
on labor-related improvements. This strikes us a significant amount, indicating a willingness to
expend markedly on global standards as a means of gaining access to new supply chain relationships.
The experimental part of this research comes in how the multinational firm is described. In one
version of the survey, it is a “large European company selling primarily to the European market”
(version A). In the other version, it is a “large Indian company selling primarily to the Indian
market” (version B).
A potential concern regarding our approach is social desirability bias: aware that labor
standards are often viewed positively, and given that the survey poses a hypothetical scenario, firms’
stated intentions may differ from how they would behave if such an opportunity were to present
19
itself.10 Because the PCI-FDI survey is sent by mail and respondents are guaranteed anonymity, we
are less concerned about bias that results from attempts to impress an interviewer. Further, neither
the Vietnam Chamber of Commerce and Industry (VCCI) or US-AID, the sponsors of the survey,
has specific connections to either hypothetical destination country. If firms were attempting to
impress, our point estimates would be biased upward. Nevertheless, what is most relevant for our
study is the difference between the treatment groups; if firms receiving the Europe treatment are
more inclined to attempt to impress, this is exactly the phenomena we seek to explain.
Figure 3: Survey Experiment Embedded in 2015 PCI Survey
10 Carrington et al 2014 and Hiscox et al 2014 discuss this problem as it relates to consumers’ propensity for ethical consumption.
G13: Imagine the following scenario: Your business has been contacted by an international consulting company, whose
primary job is to connect large multinational companies to suppliers in emerging markets. The consulting company would like to
shortlist your company, along with two other companies in your region, as potential suppliers of your product to a large
[European/Indian] company that sells primarily to the [European/Indian] market. To be eligible to be included on the shortlist, the
consulting company requires that your firm adopt the multinational’s Labor Code of Conduct for Suppliers. This Code of
Conduct includes greater representation for workers, limits on overtime work, and regulations to protect the health and safety of
workers. Adopting the Code of Conduct will allow you the possibility of future orders from this multinational and others like it,
but it also will increase your operating costs. Please tell us the maximum amount of adjustments - in terms of their financial costs
- that you would be willing to make in order be in compliance with the code of conduct and thereby eligible for the contract. To
make this easier, we have listed the costs as a share of your current operating costs:
Share of Operating Costs: (Please simply check the highest cost you would be willing to assume)
� <1% � 6% � 12%
� 1% � 7% � 13% � 2% � 8% � 14% � 3% � 9% � 15% � 4% � 10% � >15% � 5% � 11%
20
In evaluating the results of our intervention, the first empirical concern we confront is
balance. Although the survey experiment was randomized, 577 firms in exportable sectors
responded with “non-applicable.” A number of factors could contribute to this choice: first, despite
operating in an exportable sector, the respondent firm may view its primary sales target as domestic.
Second, in comparison to other questions on the survey, the prompt and contingent valuation
question were more time consuming and computationally intensive; or third, labor rights and
collective bargaining remain sensitive issues in Vietnam, rendering some firms averse to answering.
For our purposes, the key worry is whether these motivations correlate with the treatment, leading
to differential rates of item non-response that may bias estimated treatment effect sizes. We test for
this non-response bias and other sources of non-balance in covariates in Supplemental Information
A. We conclude that they pose no threat to inference.
Equation 1 estimates the result of our experimental treatment. Our dependent variable (y) is
the share of operating costs, from the contingent valuation survey item, that a firm is willing to
expend on compliance with the potential buyer’s Code of Conduct. We regress that number on our
treatment variable, which we code as 1 if the buyer was from India, and 0 if the buyer was from
Europe. The firms are indexed by i, and (p,s,c) denotes the province where the firm is located (p),
the industry/sector the firm operates in (s), and the country of origin of the investor (c). In
subsequent tests, we control for country (φ), industry/sector (π), and province (the primary sampling
unit, λ) fixed effects respectively. Thus,
( , , ) 0 1 ( , , ) ( , , )(1) + if export potential=1i p s c i p s c i p s c c s py India uE E M S O � � � �
The primary sampling unit for the PCI-FDI survey is the province, and firms are randomly
sampled from 14 provincial lists supplied by the national tax authority. Clearly, firms nested
together in the same province cannot be treated as independent draws from the underlying
distribution. Firms sharing a province are influenced by the same factor endowments, regulatory
21
environment, labor pool, infrastructure, and access to resources, violating the i.i.d. assumption. If
such firms also are from the same country, they likely interact regularly in formal business
associations or informal groupings (Wellhausen 2015). These associations also may represent
industries that put forward industry-specific complaints to provincial officials. As a result, each firm
from within the same province-country-sector triad provides less independent information than
firms from different groups. In such a setting, classical standard errors can greatly overestimate the
precision of the estimates. Thus the appropriate methodological response is to calculate cluster
robust standard errors (CRSE) at the country*industry*province level.11
IV. Empirical Results
Figure 4 displays the main results of our experiment. In the left panel, we plot the observed
kernel density distribution of firm responses to the contingent valuation question.12 Respondents
presented with the India treatment were far more likely to report a willingness to expend 5% of
operating costs, while those presented with the European variant were more likely to choose the
10% or 15% options.
The top panel of Table 2 reports the regression estimates for our experiment. Model 1
displays the results of the experimental treatment on the full sample, while Model 2 follows our
research design by limiting analysis to only firms in exportable industries. The coefficient in Model
11 Such a recommendation assumes that the number of clusters trends toward infinity; there are sufficient number observations within each cluster; and clusters are balanced in number of observations. Otherwise, test statistics will over-reject the null hypothesis and produce overly narrow confidence intervals. Under these circumstances, econometricians recommend recalculation of the standard errors using the “wild cluster bootstrap” procedure (Cameron et al. 2008). In Supplemental Information D, we implement this suggestion. Substantive results and significance tests remain very similar. We also present results using five alternative approaches to standard errors, for all estimations, in Supplemental Information E. Again, results remain very similar. 12 Substantive results remain the same, although hypothesis tests are underpowered, if we drop all firms that export to India or Europe, and if we drop all firms that are from India or Europe (see Supplemental Information F).
22
2 indicates that firms receiving the European treatment opted to pay about 0.57 percentage points
more in operating costs to comply with a hypothetical labor code of conduct. This treatment effect
is substantively meaningful, representing about 8.9% of the average answer of 6.42. Models 3, 4,
and 5 add country of origin, industry, and provincial level fixed effects, respectively. These
adjustments appear to increase the substantive effect of the experiment.
Figure 4: Unadjusted Results of Survey Experiment
The results are consistent with H1: firms offered an opportunity to sell to the European
market are more inclined to make costly adjustments. They also are consistent with recent private-
public efforts to govern labor rights and working conditions: for instance, the garment brands and
retailers that have signed the Bangladesh Accord on Building and Fire Safety – the stronger of
0.0
2.0
4.0
6.0
8.1
Kern
al D
ensi
ty
0 5 10 15Predicted Adjustment/Operating Costs (%)
India
Europe
n=886; India 421; Europe 465
Continuous Operating Costs
010
2030
40
Shar
e of
Obs
erva
tions
(%)
0-4 5-9 10-14 >15Predicted Adjustment/Operating Costs (%)
India
Europe
Binned Operating Costs
23
private sector initiatives there -- are concentrated in continental Europe and in the United
Kingdom.13
Table2:EffectofExportDestinationonLaborUpgrading(OLSResults)
Our results may be affected by data heaping: many respondents appear to have selected 0, 5,
and 10 percent, rather than intermediate values. Heaping could lead to imprecision in our estimates;
13 http://bangladeshaccord.org/signatories/
FullSample(1) (2) (3) (4) (5)
India ‐0.535* ‐0.567* ‐0.687** ‐0.690** ‐0.647*(0.297) (0.317) (0.326) (0.333) (0.335)
Constant 6.675*** 6.714*** 7.468*** 7.642*** 7.965***(0.202) (0.215) (1.927) (2.401) (2.360)
CountryFE No No Yes Yes YesIndustryFE No No No Yes YesSectorFE No No No No YesObservations 967 886 886 886 886R‐squared 0.003 0.004 0.059 0.080 0.097RMSE 4.564 4.598 4.569 4.572 4.564Clusters 607 556 556 556 556
FullSample(1) (2) (3) (4) (5)
India ‐0.144** ‐0.146** ‐0.157** ‐0.167*** ‐0.168***(0.057) (0.061) (0.063) (0.064) (0.064)
Constant 2.260*** 2.271*** 2.639*** 2.526*** 2.542***(0.040) (0.043) (0.381) (0.513) (0.517)
CountryFE No No Yes Yes YesIndustryFE No No No Yes YesProvinceFE No No No No YesObservations 843 769 769 769 769R‐squared 0.007 0.007 0.055 0.076 0.090RMSE 0.884 0.892 0.890 0.893 0.894Clusters 541 495 495 495 495
DV1:ShareofOperatingCostsFirmsWillSpendonLaborAdjustments(Continuous)
ThetablereportsOLSresultswithstandarderrors,clusteredatthe(countryXindustryXprovince)levelinparentheses(***p<0.01,**p<0.05,*p<0.1).ThefirstpanelteststhecontinuousdependentvariableandthesecondpanelteststhebinneddependentvariabledisplayedinFigure3.Model1usesthefullsampleofrespondents.Allsubsequentmodelstestresultsononlyfirmswithexportpotential.
ExportersOnly
ExportersOnlyDV2:ShareofOperatingCostsFirmsWillSpendonLaborAdjustments(Binned)
24
it also interferes with our assumption of a normal distribution for hypothesis testing (Heitjan and
Rubin 1991). To address this concern, we group the answers to the operating costs procedure into
four bins suggested by the peaks in the right panel of Figure 4 [(1) costs=0 percent; (2) 0<costs<=5;
(3) 5<costs<=10; (4) costs>10). We report the estimation using this re-scaled measure in the lower
panel of Table 2. In the unadjusted Model 2, the India treatment leads to a 0.15-point shift on the
four-point scale, which is statistically significant at the .05 level. Again, the effect size and precision
increase with the addition of country, industry, and provincial fixed effects.
Our results strongly suggest that FIEs in Vietnam are willing to make significant monetary
investments in labor-related upgrading, and that firms’ willingness to do so depends in part on the
location of potential lead firms. These findings are consistent with our first hypothesis. Goods sold
in the European market typically offer higher relative markups, giving firms a greater material
incentive to engage in behaviors that will grant them access to Europe-based supply chains.
Heterogeneous Treatment Effects by Industry
Our second and third hypotheses suggest that we also should observe differences among
developing country FIEs. We expect the largest treatment effects when a firm’s main product has
large markup differentials (between Europe and India, H2) and when a firm’s treatment of workers
is most salient for consumers, lead firms and supply chain partners (H3). Where markup differentials
are small and labor conditions are less salient, by contrast, our experimental treatment should have
little effect on firms’ contingent valuation responses.
Calculating markups requires detailed information on sales price and operating costs at the
product level (Gullstrand et al. 2014). Moreover, as we are interested in studying cross-country
differences in markups, data must be collected in a consistent manner across export destinations.
While others have calculated markups for India and for countries in Europe (De Loecker and
25
Warzynski 2012, De Loecker et al. 2016), few authors have attempted to calculate them using the
same survey instrument. We address this challenge by using proprietary data from the Chinese
Customs office, containing sales price data for over one million products cleared by the customs
agency in January 2010. These data are measured at the eight-digit Chinese country code level, and
they include information on the destination country.14 We then calculate the natural log of the
difference in markups of products destined for India and those for the European Union.15 We use
the PCI-FDI’s classification of products and services (the four-digit industrial classifications -- ISIC
Rev 4 -- of the United Nations Statistical Division)16 to match industries found in the PCI to markup
differentials calculated from the Chinese data.17 While China is not Vietnam, they do compete in
many of the same export product markets and they have comparable domestic economic structures.
Moreover, the thousands of sales for each product within the Chinese data allow us to more
precisely estimate markups.18 The Chinese data also reveal future opportunities for Vietnamese
firms: few surveyed Vietnamese FIEs currently export to India or Europe, but the Chinese data
shows the possible gains available to them from doing so.
In addition to markup differentials, we also expect that the salience of labor issues can
further motivate developing country firms to pursue labor-related upgrading. As we note above, the
salience of labor issues stems from targeting decisions by rights activists; the ability for consumers
14 Jarreau and Poncet (2012) have also used this data. See the China Customs website for details on this propiertary database < http://china-trade-research.hktdc.com/business-news/article/Facts-and-Figures/China-Customs-Statistics/ff/en/1/1X000000/1X09N9NM.htm> 15 To ease the calculation, we limited our analysis to the five most common European export destinations (Germany, France, UK, Italy, and Spain). 16 See <http://unstats.un.org/unsd/cr/registry/isic-4.asp> 17 As none existed, we wrote our own conversion table between the Chinese HS codes and the ISIC Rev 4 codes in the PCI dataset. 18 Using PCI data, we also calculated markups for Vietnamese firms. Given the small number of firms exporting to Europe and India, we instead calculated differentials between all developing and developed countries. Our markup measure based on Chinese data is highly correlated both with this PCI-based measure and with independent markup calculations for India (De Loeker et al. (2016)) and Europe (Chistopolou and Vermeulen 2012). See Supplemental Information H.
26
and shareholders to observe labor conditions throughout the supply chain; and firms’ desire to hire
and retain skilled labor. The multidimensional nature of salience renders it difficult to measure
across sampled Vietnamese FIEs. Below, we employ a comparison of a high-salience with a low-
salience Vietnamese industry, holding markup differentials constant. For the entire sample, we also
collect data on the share of intermediate goods used in production, by industry (Miroudot 2009,
50).19 Higher shares of intermediates could correlate – albeit imperfectly, as we discuss above -- with
less pressure for labor-related up grading. Figure 5 plots all firms in the PCI data (aggregated to the
two-digit industry level) by median mark-up differential and share of intermediate goods in
production. The figure is organized to align with the predictions in Table 1. According to our
theory, treatment effects should be strongest in the southeast quadrant (i.e. computers), where
markups and salience are highest; and weakest in the northwest quadrant (i.e. rubber/plastics),
where markups and salience are low.
We present results of the interaction between our treatment variable (India), and difference
in markups (Figure 6) as well as the share of intermediate goods (Figure 7), using the following
estimating equation:20
( , , ) 0 1 ( , , ) 2 3 ( , , )
2 3 ( , , ) ( , , )
(2) **
+X if export potential=1
i p s c i p s c s i p s c
s i p s c i p s c
y India Markup India MarkupIntermediate India Intermediate u
E E E E
E E
� � ' � '
� � ' �
It is important to note that while the survey experiment is randomized, the selection of firms into
specific industries is not. Thus, omitted confounders that might correlate with differences in
19 This data is only available at the two-digit level. 20 See Supplemental Information G1 and G2 for full regression results.
27
markups between these two industries could bias our analysis. We control for these with a matrix of
control variables (X) that are theoretically correlated with both markups and labor upgrading costs. 21
Figure5:DifferencesinMarkupsandSaliencebyIndustry
To guard against misinterpretation when the interaction effects not do vary at a constant rate
across the full distribution of the moderating variable (e.g. difference in markups) or there is not
sufficient common support (sufficient representation of both control and treatment group)
throughout the distribution of the moderator, we follow Hainmueller et al. (2017) in testing and
depicting interactions using both binning and kernel density procedures. The top rows in Figures 6
21 We present additional heterogeneous effects in Supplemental Information I.
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Food P
roces
sing
M:Leath
er
M:Textile
s
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Leath
er
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Textile
s
M:Leath
er
M:Leath
er
M:Textile
s
M:Garm
ents
M:Leath
er
M:Textile
s
M:Leath
er
M:Garm
ents
M:Textile
s
M:Textile
s
M:Leath
er
M:Leath
er
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Leath
er
M:Garm
ents
M:Leath
er
M:Leath
er
M:Leath
er
M:Textile
s
M:Garm
ents
M:Leath
er
M:Leath
er
M:Garm
ents
M:Leath
er
M:Textile
s
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Leath
er
M:Leath
er
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Textile
s
M:Textile
s
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Leath
er
M:Leath
er
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Leath
er
M:Leath
er
M:Garm
ents
M:Leath
er
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Textile
s
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Leath
er
M:Textile
s
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Textile
s
M:Textile
s
M:Textile
s
M:Garm
ents
M:Leath
er
M:Leath
er
M:Textile
s
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Garm
ents
M:Leath
er
M:Textile
s
M:Textile
s
M:Textile
s
M:Garm
ents
M:Leath
er
M:Textile
s
M:Textile
s
M:Textile
s
M:Garm
ents
M:Garm
ents
M:Textile
s
M:Textile
s
M:Textile
s
M:Textile
s
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Woo
d Prod
ucts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Woo
d Prod
ucts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Paper
Produc
ts
M:Woo
d Prod
ucts
M:Woo
d Prod
ucts
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Che
micals
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Rub
ber/P
lastic
s
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Basic
Metals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Fabric
ated M
etals
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Compu
ters/E
lectro
nics
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M:Electro
nic Equ
ip.
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Moto
r Veh
icles
M: Othe
r
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
Mining
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Mining
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
Agricu
lture/
Aquac
ulture
Agricu
lture/
Aquac
ulture
M:Furnitu
re
Agricu
lture/
Aquac
ulture
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
M:Furnitu
re
.2.4
.6.8
1
Shar
e of
Inte
rmed
iate
to T
otal
Tra
de F
low
s (%
)
-2 0 2 4 6
Difference in Markups between (Europe-India,ln)
Dashed lines depict mean values of the axes.
28
and 7 display the regression results on our continuous dependent variable; the bottom rows show
the effects on our categorical dependent variable.
All four panels of Figure 6 demonstrate that the experimental treatment does not have a
significant effect at low or medium markup differentials. Only when logged markup differences are
high (>2) do we observe a significant effect of the treatment. This analysis confirms our expectation
that firms are most willing to invest in labor upgrading when price inelasticity in destination markets
allows additional returns via price discrimination. These returns compensate for the costs of labor-
related improvements. Figure 7, by contrast, illustrates that the share of intermediate goods in
production has no effect on the treatment.
Figure6:HeterogeneousEffectofTreatmentbyMarkupsDifferential
L M H
-6
-4
-2
0
2
4
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
-2 0 2 4 6Moderator: Difference in Markups (ln)
Binning Estimator
-6
-4
-2
0
2
4
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
-2 0 2 4 6Moderator: Difference in Markups (ln)
Kernel Density EstimatorContinuous Dependent Variable
L M H
-1
-.5
0
.5
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
-2 0 2 4 6Moderator: Difference in Markups (ln)
Binning Estimator
-1
-.5
0
.5
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
-2 0 2 4 6Moderator: Difference in Markups (ln)
Kernel Density EstimatorBinned Dependent Variable
29
Figure7:HeterogeneousEffectofSaliencebyMarkupsDifferential
Quantitative Case Study of Salience
Given that our theoretical claims regarding salience are not captured adequately by the
intermediate goods share measure, we pursue a different empirical strategy for further testing of H3:
examining firms in two industries with similar markup differentials. The PCI-FDI includes a sizeable
and similar number of firms in apparel/garments and in plastic and rubber products. As Figure 5
L M H
-4
-3
-2
-1
0
1
2
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
.2 .4 .6 .8 1Moderator: Difference in Markups (ln)
Binning Estimator
-4
-3
-2
-1
0
1
2
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
.2 .4 .6 .8 1Moderator: Share of Intermediates
Kernel Density EstimatorContinuous Dependent Variable
L M H
-.6
-.4
-.2
0
.2
Mar
gina
l Effe
ct o
f 'In
dia'
on
Ope
ratin
g C
osts
.2 .4 .6 .8 1Moderator: Difference in Markups (ln)
Binning Estimator
-.6
-.4
-.2
0
.2M
argi
nal E
ffect
of '
Indi
a' o
n O
pera
ting
Cos
ts
.2 .4 .6 .8 1Moderator: Share of Intermediates
Kernel Density EstimatorCategorical Dependent Variable
30
illustrates, both industries are characterized by very small differences in Europe vs. India markups.
The globalization of production, as well as the elimination of trade barriers (including the 2004
phase-out of the Multifibre Arrangement for garments), has dramatically lowered markups in these
industries (Abraham et al. 2009). The thin markups are especially evident for producers at lower and
intermediate rungs in the supply chain, as the wide availability of potential suppliers allows lead firms
to reduce the prices paid for inputs (Milberg and Winkler 2013). Plastic bag production also is
fiercely competitive; U.S. firms routinely filing domestic trade complaints against exporting firms
throughout southeast Asia.22
Importantly, however, the salience of labor conditions varies markedly between these
industries. Apparel is a final product with a country of (at least final) origin on the label. During the
last two decades, as numerous activist groups have highlighted working conditions in the apparel
sector, and as industrial accidents have garnered widespread media attention, it has become easier
for consumers to locate information about production conditions (Bartley and Child 2014, Saideman
2007). Distelhorst and Locke (2017) find that much of the material reward for code of conduct
compliance is driven by apparel sector lead firms. By contrast, consumers have very little sense of
production conditions for plastic and rubber products. Some plastics outputs are used as
intermediate inputs; even when they are exported as final products – for surveyed Vietnamese firms,
the main product in this industry is “retail carrier bags” -- consumers know very little about their
production processes.
Comparing these industries therefore allows us to hold constant markups and isolate the
effect of salience on firms’ reported willingness to upgrade. Figure 8 presents the change in
operating costs reported by respondent firms in the two industries.23 As expected, for rubber and
22 See, for instance, ITA 2016b. 23 Full regression results, controlling for confounders, are produced in Supplemental Information J.
31
plastics, there is very little difference between firms receiving the India and European treatments. In
fact, firms in the European treatment were willing to pay marginally less in this low-visibility industry
(5.5% versus 6.6%).24 By contrast, in apparel, where markup differences are similarly low but
visibility is higher, firms in the European treatment group were willing to make changes worth 8.4%
of operating costs, compared with 6% in the India treatment group.
Figure8:ComparisonofTwoSectorswithNoMarkupDifferences
24 This may be explained, in part, by efforts to ban or heavily tax plastic carrier bags in Europe. Producers, aware of this, could assume that India will provide higher relative markups in the future.
8.4
6
01
23
45
67
89
Europe India
Wald F-test=4.27; p=.069Wearing Apparel (n=64)
5.5
6.6
01
23
45
67
89
Europe India
Wald F-test=.93; p=.32Rubber and Plastic (n=70)
Aver
age
Cha
nge
in O
pera
tion
Cos
ts (%
)
32
V. Conclusion
Multinational production can serve as an instrument for the improvement of labor
standards, if developing country firms that want to service foreign markets “trade up.” Most
accounts of “trading up,” however, pay little attention to firm- and industry-level variation in the
incentives to improve standards. We agree that the possibility of accessing global supply chains
motivates firms to improve their labor standards, but we point to important variation in firms’
propensity to do so. Firms most likely to upgrade are those that anticipate selling to developed
(versus developing country markets); those whose products have significantly higher markups in
some foreign destinations than in others; and those whose production conditions are more salient to
lead firms.
We test these expectations using a contingent valuation methodology, with an experimental
component, in a survey of foreign invested firms in Vietnam. We find compelling evidence that
FIEs are more likely to expend resources to improve labor costs if the overseas sales opportunity is
in developed (European) markets, rather than equally sizable markets in the developing world
(India). We find over half a percentage point difference, on average, in the share of operating costs
that such firm would devote to complying with a hypothetical international buyer’s labor-related
code of conduct. When we explore firm-level differences in responses to the experimental
treatment, we find strong support for the claim that the availability of higher markups influences
firms’ willingness to make improvements. We also find suggestive evidence that the salience of
labor conditions among lead firms enhances the propensity of developing country firms to upgrade.
Our results suggest that greater attention to supply chain participants, especially those that
are not lead firms – and that are not even necessarily large subcontractors – is warranted. In many
developing countries, such firms account for a significant proportion of employment. If “trading
33
up” – rather than “race to the bottom” -- operates, it is via these firms’ incentives and behaviors,
which are themselves influenced by consumer, shareholder and regulatory pressures in export
destinations.
In future research, we can draw further distinctions among internationally-active firms, as a
means of identifying additional mechanisms that may affect the willingness to upgrade, and as a way
of measuring the salience of labor rights issues across a broader spectrum of firms. For instance, we
can consider whether firms’ willingness to upgrade varies among the set of developed country
markets, or among large emerging market economies. Given the tendency of European consumers
as well as European lead firms to pay greater attention to ethical issues in supply chains, an
experimental design involving “the United States” rather than “Europe” as the location of the
supply chain partner might elicit a different response. Similarly, FIEs may respond differently to the
opportunity to export to China or Brazil, versus to India.
Future scholarship also can address how the dynamics identified in Vietnam – which allow
us to focus on the internal validity of our research findings – travel to other developing countries.
Vietnam is very active in the global economy, both as an FDI destination and as an exporter. It also
has experienced high rates of economic growth, which may allow firms more room to invest in
upgrading. And, while Vietnam has faced pressure – especially in the context of the TPP
negotiations – to reform its labor laws, the country’s political institutions do not provide workers
with the full complement of collective labor rights. It is therefore worth exploring how national
political and labor market institutions interact with firm-level incentives to upgrade.
34
References
Abraham, Filip, Jozef Konings and Stijn Vanormelingen. 2009. “The Effect of Globalization on Union
Bargaining and Price-Cost Margins of Firms.” Review of World Economics 145(1): 13-36.
Adolph, Christopher, Vanessa Quince and Aseem Prakash. 2017. “The Shanghai Effect: Do Exports to
China Affect Labor Practices in Africa?” World Development 89: 1-18.
Angrist, Joshua D., and Jörn- Steffen Pischke. 2009. Mostly Harmless Econometrics: An Empiricist’s
Companion. Princeton: Princeton University Press.
Barklie, Glenn. 2016. "Vietnam Leads Emerging Market Greenfield FDI Performance index." In fDi
Magazine. <http://www.fdiintelligence.com/Locations/Asia-Pacific/Vietnam/Vietnam-leads-
emerging-market-greenfield-FDI-performance-index?ct=true>
Bartley, Tim. 2007. “Institutional Emergence in an Era of Globalization: The Rise of Transnational
Private Regulation of Labor and Environmental Conditions.” American Journal of Sociology 113(2): 297-
351.
Bartley, Tim and Curtis Child. 2014. “Shaming the Corporation: The Sociological Production of Targets
and the Anti-Sweatshop Movement.” American Sociology Review 79: 653-679.
Berliner, Daniel and Aseem Prakash. 2014. “Public Authority and Private Rules: How Domestic
Regulatory Institutions Shape the Adoption of Global Private Regimes.” International Studies Quarterly
58: 793-803.
Berliner, Daniel and Aseem Prakash. 2015. “’Bluewashing’ the Firm? Voluntary Regulations, Program
Design and Member Compliance with the United Nations Global Compact.” Policy Studies Journal
43(1): 115-138.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How Much Should We Trust
Differences- in- Differences Estimates?” Quarterly Journal of Economics 119(1): 249–75.
35
Broda, Christian and David E. Weinstein. 2006. “Globalization and the Gains from Variety.” The
Quarterly Journal of Economics 121(2): 541-585.)
Cameron, A.C., Gelbach, J.B. and Miller, D.L., 2008. “Bootstrap-based improvements for inference with
clustered errors.” The Review of Economics and Statistics 90(3): 414-427.Cao, Xun and Aseem Prakash.
2011. “Growing Exports by Signaling Product Quality: Trade Competition and the Cross-National
Diffusion of ISO 9000 Quality Standards.” Journal of Policy Analysis and Management 30(1): 111-135.
Carrington, M.J., B.A. Neville, G.J. Whitwell. 2014. “Lost in Translation: Exploring the Ethical
Consumer Intention-Behavior Gap.” Journal of Business Research 67: 2759-2767.
Cummings, Ronald G., and Laura O. Taylor. 1999. "Unbiased value estimates for environmental goods:
a cheap talk design for the contingent valuation method." The American Economic Review 89 (3): 649-
665.
Christopoulou, R. and Vermeulen, P., 2012. “Markups in the Euro Area and The US over the Period
1981–2004: A Comparison of 50 sectors.” Empirical Economics, 42(1): 53-77.
De Loecker, Jan, and Frederic Warzynski. 2012. “Markups and Firm-Level Export Status: Dataset.”
American Economic Review. http://dx.doi.org/10.1257/aer.102.6.2437
De Loecker, Jan, Goldberg, P.K., Khandelwal, A.K. and Pavcnik, N., 2016. “Prices, Markups, and Trade
Reform.” Econometrica, 84(2): 445-510.
Distelhorst, Greg, Jens Hainmueller and Richard M. Locke. 2016. “Does Lean Improve Labor
Standards? Management and Compliance in the Nike Supply Chain.” Management Science 63(3): 707-
728.
Distelhorst, Greg and Richard M. Locke. 2017. “Does Compliance Pay? Firm-Level Trade and Social
Institutions.” Forthcoming, American Journal of Political Science.
36
Doner, Richard F., 2009. The politics of uneven development: Thailand's economic growth in comparative perspective.
Cambridge University Press.
Fujita, M. 2011. “Value chain dynamics and local suppliers’ capability building: an analysis of the
Vietnamese motorcycle industry.” In The Dynamics of Local Learning in Global Value Chains (pp. 68-99).
Palgrave-Macmillan UK.
Garcia-Johnson, Ronie. 2000. Exporting Environmentalism: U.S. Multinational Chemical Corporations in the
U.S. and Mexico. Cambridge: MIT Press.
Gereffi, Gary. 1994. “The Organization of Buyer-Driven Commodity Chains: How U.S. Retailers Shape
Overseas Production Networks.” In Gary Gereffi and Miguel Korzeniewicz, eds., Commodity Chains
and Global Capitalism (Westport, CT: Praeger), pp. 95-122.
Gereffi, Gary. 2014. “Global Value Chains in a Post-Washington Consensus World.” Review of
International Political Economy 21(1): 9-37.
Görg, Holger, Aoife Hanley, Stefan Hoffmann and Adnan Seric. 2017. “When Do Multinational
Companies Consider Corporate Social Responsibility? A Multi-Country Study in Sub-Saharan
Africa.” Business and Society Review 122(2): 191-220.
Greenhill, Brian, Layna Mosley, and Aseem Prakash. 2009. “Trade-based Diffusion of Labor Rights: A
Panel Study, 1986–2002.” American Political Science Review 103(4): 669-689.
Gullstrand, J., Olofsdotter, K. and S. Thede. 2014. “Markups and Export Pricing Strategies.” Review of
World Economics 150(2): 221-239.
Hainmueller, Jens, Michael J. Hiscox and Sandra Sequeira. 2014. “Consumer Demand for the Fair Trade
Label: Evidence from a Multi-Store Field Experiment.” Review of Economics and Statistics.
Hainmueller, J., J. Mummolo and Y. Xu 2016. How much should we trust estimates from multiplicative
interaction models? Simple tools to improve empirical practice.
37
Heitjan, D.F. and Rubin, D.B., 1991. “Ignorability and Coarse Data. The Annals of Statistics 19(4): 2244-
2253.
International Trade Administration (ITA). 2016a. “2016 Top Markets Report Technical Textiles
Country Case Study: Vietnam” http://trade.gov/topmarkets/pdf/Textiles_Vietnam.pdf
International Trade Administration. 2016b. “FACT SHEET: Commerce Initiates Antidumping Duty
Investigations on Imports of Polyethylene Retail Carrier Bags from the Socialist Republic of
Vietnam, Indonesia, and Taiwan,” http://enforcement.trade.gov/download/factsheets/factsheet-
prcb-ad-init-042109.pdf
Jarreau, J. and Poncet, S., 2012. “Export Sophistication and Economic Growth: Evidence from
China,”. Journal of development Economics 97(2): 281-292.
Jensen, J. Bradford, Dennis P. Quinn and Stephen Weymouth. 2015. “The Influence of Firm Global
Supply Chains and Foreign Currency Undervaluations on US Trade Disputes.” International
Organization 69(4): 913-947.
Johns, Leslie and Rachel L. Wellhausen. 2016. “Under One Roof: Supply Chains and the Protection of
Foreign Investment.” American Political Science Review 110(1): 31-51.
Kim, In Song. 2017. “Political Cleavages within Industry: Firm-Level Lobbying for Trade
Liberalization.” American Political Science Review 111(1): 1-20.
Kim, In Song, Helen V. Milner, Thomas Bernauer, Gabriele Spilker, Iain Osgood and Dustin Tingley.
2017. "Firms’ Preferences over Multidimensional Trade Policies: Global Production Chains,
Investment Protection and Dispute Settlement Mechanisms." Working Paper,
http://web.mit.edu/insong/www/pdf/conjoint.pdf.
Lechner, Lisa. 2016. “The Domestic Battle over the Design of Non-Trade Issues in Preferential Trade
Agreements.” Review of International Political Economy 23(5): 840-871.
38
Locke, Richard M. 2013. The Promise and Limits of Private Power: Promoting Labor Standards in a Global
Economy. New York: Cambridge University Press.
Locke, Richard M., Ben A. Rissing and Timea Pal. 2013. “Complements or Substitutes? Private Codes,
State Regulation and the Enforcement of Labour Standards in Global Supply Chains.” British Journal
of Industrial Relations 51(3): 519-552.
Manger, Mark S. 2012. “Vertical Trade Specialization and the Formation of North-South PTAs.” World
Politics 64(4): 622-658.
Melitz, Mark and Gianmarco I. P. Ottaviano. 2008. “Market Size, Trade, and Productivity.” The Review of
Economic Studies75(1): 295–316.
Milberg, William and Deborah Winkler. 2013. Outsourcing Economics: Global Value Chains in Capitalist
Development. Cambridge: Cambridge University Press.
Miroudot, S., Lanz, R. and Ragoussis, A., 2009. “Trade in Intermediate Goods and Services,” OECD
Trade Policy Working Paper No. 93, Paris.
Mitchell, Robert Cameron, and Richard T. Carson. 1989. Using Surveys to Value Public Goods: The Contingent
Valuation Method. Washington, DC: Resources for the Future Publishing.
Mosley, Layna. 2011. Labor Rights and Multinational Production. New York: Cambridge University Press.
Ngo, C.N., 2016. “Local Value Chain Development in Vietnam: Motorcycles, Technical Learning and
Rents Management.” Journal of Contemporary Asia 46(4): 1-26.
Nguyen, K. T. and E. Ramstetter. 2017. “Foreign Multinationals and Vietnamese Firm Exports.”
Preliminary Draft prepared for the Asia-Pacific Trade Seminars 13th Annual Meeting, Hanoi, 16-17
June 2017.
39
OECD, WTO and World Bank Group. 2014. Global Value Chains: Challenges, Opportunities and
Implications for Policy. Report prepared for submission to the G20 Trade Ministers Meeting,
Sydney, Australia, July 19, 2014.
Osgood, I., 2016. “Globalizing the Supply Chain: Firm and Industrial Support for US Trade.”
Agreements. Manuscript, University of Michigan.
https://ncgg.princeton.edu/IPES/2016/papers/S930_rm3.pdf
Osgood, I., D. Tingley, T. Bernauer, I.S. Kim, H.V. Milner and G. Spilker. 2016. “The Charmed Life of
Superstar Exporters: Survey Evidence on Firms and Trade Policy.” The Journal of Politics 79(1): 133-
152.
Perkins, R. and E. Neumayer. 2012. “Does the ‘California effect’ Operate across Borders? Trading-Up
and Investing-up in Automobile Emissions Standards.” Journal of European Public Policy 19(2): 217-
237.
Prakash, Aseem and Potoski, Matthew. 2006. “Racing to the Bottom? Trade, Environmental
Governance, and ISO 14001.” American Journal of Political Science 50(2): 350–364.
Shepherd, B. 2013. “Global Value Chains and Developing Country Employment: A Literature Review.”
OECD Trade Policy Papers, No. 156, OECD Publishing.
Simonovska, Ina. 2015. “Income Differences and Prices of Tradables: Insights from an Online
Retailer.” Review of Economic Studies 82(4): 1612-1656.
Sutton, John. 2013. Competing in Capabilities: The Globalization Process. Oxford: Oxford University Press.
United Nations Conference on Trade and Development (UNCTAD). 2017. World Investment Report:
Investment and the Digital Economy. Geneva: UNCTAD.
Vogel, David. 1995. Trading Up: Consumer and Environmental Regulation in a Global Economy. Cambridge:
Harvard University Press.
40
Vogel, David. 2006. The Market for Virtue: The Potential and Limits of Corporate Social Responsibility.
Washington: Brookings Institution.
Wellhausen, Rachel. 2015. The Shield of Nationality: When Governments Break Contracts with Foreign Firms.
New York: Cambridge University Press.
a
Chains of Love?
Global Production and the Firm-Level Diffusion of Labor Standards
SUPPLEMENTAL INFORMATION
Appendix A: Checking for Non-Response Bias and Balance of Confounders ......................................... b
Table A1: Item Non-Response and Balance of Confounders ................................................................................ b
Appendix B: Variation in Labor Rights by Industry from PCI Survey ........................................................ c
Appendix C: Variation in Labor Rights by Country of Origin from PCI Survey ......................................... d
Appendix D: Main Experimental Results with Wild Bootstrap Standard Error Correction ......................... e
Appendix E: Replication of Main Results with Full Sample and Alternative Approaches to Standard Errors ............................................................................................................................................................ f
Appendix F: Replication of Main Results Dropping All Firms from India or Europe & Doing Business in India or Europe........................................................................................................................................ g
Appendix G1: Full Interaction Results and Sensitivity Tests (Continuous Dependent Variable) ......... h
Appendix G2: Full Interaction Results and Sensitivity Tests (Binned Dependent Variable) ................. i
Appendix H: Correlation between Different Measures of Markups ......................................................... j
Appendix I: Additional Sub-Group Effects .................................................................................................. k
Table A2: Treatment Effects by Sub-Group ..............................................................................................................l
Appendix J: Mechanisms with Covariate Adjustment ................................................................................. m
Appendix K: Works Cited in Appendix ........................................................................................................ n
b
Appendix A: Checking for Non-Response Bias and Balance of Confounders Table A1 presents the results of a series of balance tests. The table indicates the mean and
standard error for both the European and Indian treatment groups, followed by the difference between the two. The first test is comforting, as it illustrates that the difference in non-response between the two groups was about 1.8 percentage points (36.4% for the India treatment and 34.6% for the Europe treatment) and not significantly different from zero.
Further tests look at differences in key firm attributes (e.g. entry mode, age) that might be associated with views regarding labor rights. Two further indicators measure pre-treatment differences in labor rights -- whether a union is currently allowed and the share of long-term workers with formal contracts. Finally, we look at the home country of the firm. Six firms (0.42% of our sample of exportable product-sector firms) are from India, while 96 firms (6.79%) are from Western Europe.
Four out of 27 potential confounders (roughly 15 percent) are statistically significant. This is exactly the share we would expect by chance when accepting a p-value of .1 or less. The small sample sizes and the lack of stratification by home country in the PCI survey could lead to non-balance. The randomization in the survey seems to have been well executed, however, and considering the potential dangers of controlling for non-balanced confounders in a survey experiment (Mutz and Pemantle 2015), we do not adjust our analyses to address non-balance.
Table A1: Item Non-Response and Balance of Confounders
Mean SE Mean SE India-Europe p-valueItem Non-Response=1 0.346 (0.018) 0.364 (0.018) 0.018 (0.491) 1,413CEO is male=1 0.921 (0.013) 0.926 (0.013) 0.005 (0.777) 836Years since registration (ln) 8.725 (0.234) 8.280 (0.245) -0.445 (0.189) 905Capital size ($1000s USD) 4,032 (447.9) 4,308 (485.3) 275.9 (0.676) 637100% Foreign Owned=1 0.860 (0.015) 0.903 (0.016) 0.043** (0.044) 912Multinational Corp.=1 0.308 (0.022) 0.312 (0.023) 0.004 (0.901) 869Entry through M&A=1 0.072 (0.013) 0.071 (0.013) -0.001 (0.955) 806Union in firm==1 0.662 (0.022) 0.637 (0.023) -0.025 (0.446) 868Workers under contract (%) 87.512 (1.204) 86.323 (1.250) -1.189 (0.493) 877Number of employees 208.763 (16.164) 231.967 (17.013) 23.204 (0.323) 900Profitable firm=1 0.547 (0.024) 0.547 (0.025) 0.000 (0.992) 817Loss Making firm=1 0.397 (0.024) 0.363 (0.025) -0.034 (0.325) 817Plan to expand business =1 0.481 (0.023) 0.498 (0.024) 0.017 (0.619) 912Export at all=1 0.619 (0.022) 0.638 (0.023) 0.019 (0.554) 912Customer is SOE=1 0.169 (0.016) 0.118 (0.017) -0.052** (0.026) 912Customer is government=1 0.046 (0.009) 0.025 (0.009) -0.021* (0.095) 912Customer is private firm=1 0.431 (0.023) 0.408 (0.024) -0.023 (0.480) 912Customer is foreign firm=1 0.515 (0.023) 0.502 (0.024) -0.012 (0.710) 912Export to home country=1 0.460 (0.023) 0.461 (0.024) 0.001 (0.986) 912Export to third country=1 0.372 (0.022) 0.376 (0.023) 0.003 (0.921) 912Vendor is SOE=1 0.136 (0.015) 0.104 (0.016) -0.032 (0.135) 912Vendor is private firm=1 0.699 (0.021) 0.680 (0.022) -0.019 (0.536) 912Vendor is household=1 0.228 (0.018) 0.177 (0.019) -0.051* (0.058) 912Inputs from in house=1 0.090 (0.012) 0.071 (0.013) -0.019 (0.307) 912Import from Home country=1 0.598 (0.022) 0.641 (0.023) 0.042 (0.190) 912Import from Third country=1 0.412 (0.023) 0.415 (0.024) 0.003 (0.936) 912Company from Europe=1 0.067 (0.012) 0.078 (0.012) 0.011 (0.508) 912Company from India=1 0.004 (0.003) 0.005 (0.003) 0.000 (0.923) 912Row 1 includes all firms in exporting sectors. Thereafter balance tests are restricted to firms in exportable sectors that responded to the survey experiment.
Confounders Europe India Difference Observations
c
Appendix B: Variation in Labor Rights by Industry from PCI Survey
Appendix B presents three confidence interval plots of PCI survey questions. Respondent firms answered prior to answering our experimental question. Responses are aggregated by two-digit industries, based on ISIC Rev4.
• Panel 1 (QG1.3): What proportion of your total workforce are long-term workers hired without formal contracts? ..................%
• Panel 2 (QG9): In the past THREE years, has your firm experienced a labor strike or work stoppage? Yes/No
• Panel 3 (QG12): In the past two years, how many inspection delegations (from MOLISA or DOLISA) have visited your business to inspect or monitor your company’s compliance with labor rights requirements?..................DELEGATIONS
• [Note: MOLISA is Vietnam’s Ministry of Labor, Invalids and Social Affairs. It is the central authority for labor inspection in Vietnam. Provincial governments operate Departments of Labor, Invalids and Social Affairs (DOLISA), which also carry out labor-related inspections.
Agriculture/AquacultureM:Food Processing
M:TextilesM:Garments
M:LeatherM:Wood ProductsM:Paper Products
M:ChemicalsM:Rubber/Plastics
M:Basic MetalsM:Fabricated Metals
M:Computers/ElectronicsM:Electronic Equip.
M:MachineryM:Motor Vehicles
M:FurnitureM: Other
Electricity/Gas/ACWholesale/Retail
Information/CommunicationProfessional Services
-5 0 5 10 15 20No Formal Contract %
90% confidence intervals
Workers w/o Contracts
AC10C13C14C15C16C17C20C22C24C25C26C27C28C29C31C32
DGJ
M
0 10 20 30 40Share of Firms %
Strikes in 3 Years
AC10C13C14C15C16C17C20C22C24C25C26C27C28C29C31C32
DGJ
M
0 .5 1 1.5 2 2.5Number of Labor Inspections
Inspections in 2 Years
d
Appendix C: Variation in Labor Rights by Country of Origin from PCI Survey
Appendix C presents three confidence interval plots of PCI survey questions that firms answered prior to answering our experimental question. Responses are aggregated by two-digit industries, based on ISIC Rev4.
• Panel 1 (QG1.3): What proportion of your total workforce are long-term workers hired without formal contracts?..................%
• Panel 2 (QG9): In the past THREE years, has your firm experienced a labor strike or work stoppage? Yes/No
• Panel 3 (QG12): In the past two years, how many inspection delegations (from MOLISA or DOLISA) have visited your business to inspect or monitor your company’s compliance with labor rights requirements?..................DELEGATIONS
Australia
China
France
Germany
Japan
Malaysia
Singapore
South Korea
Taiwan
Thailand
United Kingdom
United States
-5 0 5 10 15 20No Formal Contract %
90% confidence intervals
Workers w/o Contracts
Australia
China
France
Germany
Japan
Malaysia
Singapore
South Korea
Taiwan
Thailand
United Kingdom
United States
-5 0 5 10 15 20Share of Firms %
Strikes in 3 Years
Australia
China
France
Germany
Japan
Malaysia
Singapore
South Korea
Taiwan
Thailand
United Kingdom
United States
0 1 2 3 4Number of Labor Inspections
Inspections in 2 Years
e
Appendix D: Main Experimental Results with Wild Bootstrap Standard Error Correction
Clustering standard errors assumes that the number of clusters trends toward infinity; there are sufficient number observations within each cluster; and clusters are balanced in number of observations (Cameron et al. 2008, Kline and Santos 2012, Wooldridge 2013). Our analysis violates the second two assumptions, and therefore may be leading too over-rejection of the null hypothesis. Therefore, Appendix D addresses the statistical significance of the difference (0.5 percent) between the Europe and India groups. All models present the regression results from Table 2, Model 5 in the article text. We regress the operating cost response on the treatment variable (“India”), controlling for country, industry, and province fixed effects, and clustering standard errors at the (country*industry*province) level. We implement the Roodman (2016) Boottest procedure, which calculates wild cluster standard errors from 1,000 re-samplings and generates the confidence interval. This procedure follows Cameron and Miller’s (2015, p. 244) computationally intensive recommendation of repeated tests of the null hypothesis for each re-sampling. We plot the coefficients for firms receiving the India treatment on the x-axis. The dashed vertical line represents a treatment effect of zero.
These represent the estimated difference in operating cost expenditures reported by firms receiving the India treatment versus those receiving the European treatment. The y-axis plots the share of the distribution for each p-value. Two horizontal lines run across the graph. The solid line at p=.05 represents the 95% confidence interval (with gray diamonds). The dashed line at p=.1 represents the 90% CI.
We present four graphs. The first row presents a hypothesis test on firms in only exporting industries, and the second row demonstrates that the results are robust to including the full sample. The first graph in each row depicts results on the continuous measure of operating costs, and the second graph presents results from the binned measure of operating costs. Wald tests demonstrate that hypothesis tests are significant at the .1 level for the continuous measure and the .01 level for the binned measure.
-1.35 .0370.1
.2.3
.4.5
.6.7
.8.9
1
Rej
ectio
n p-
valu
e fro
m W
ild C
lust
er B
oots
trap
SEs
-2 -1.5 -1 -.5 0 .5 1 1.5 2India Treatment - Europe Treatment
Wald F-Test=3.74, P-Value=.065
Continuous Operating Costs
-.293 -.0430.1
.2.3
.4.5
.6.7
.8.9
1
Rej
ectio
n p-
valu
e fro
m W
ild C
lust
er B
oots
trap
SEs
-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5India Treatment - Europe Treatment
Wald F-Test=6.83, P-Value=.01
Binned Operating CostsFirms in Exporting Sectors
-1.25 .0330.1
.2.3
.4.5
.6.7
.8.9
1
Rej
ectio
n p-
valu
e fro
m W
ild C
lust
er B
oots
trap
SEs
-2 -1.5 -1 -.5 0 .5 1 1.5 2India Treatment - Europe Treatment
Wald F-Test=3.74, P-Value=.063
Continuous Operating Costs
-.282 -.0520.1
.2.3
.4.5
.6.7
.8.9
1
Rej
ectio
n p-
valu
e fro
m W
ild C
lust
er B
oots
trap
SEs
-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5India Treatment - Europe Treatment
Wald F-Test=7.63, P-Value=.005
Binned Operating CostsAll Firms
f
Appendix E: Replication of Main Results with Full Sample and Alternative Approaches to Standard Errors
Full Exporters Full Exporters Full Exporters Full Exporters Full Exporters(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
India -0.605** -0.647** -0.605** -0.647** -0.605** -0.647** -0.605* -0.647* -0.605** -0.647**(0.304) (0.319) (0.308) (0.324) (0.231) (0.251) (0.313) (0.335) (0.299) (0.323)
Constant 7.144*** 7.965*** 7.144*** 7.965*** 7.144*** 7.965*** 7.144*** 7.965*** 7.144*** 7.965***(2.138) (2.230) (2.157) (2.256) (1.466) (1.631) (2.277) (2.360) (1.776) (1.917)
Observations 967 886 967 886 967 886 967 886 967 886R-squared 0.091 0.097 0.091 0.097 0.091 0.097 0.091 0.097 0.091 0.097RMSE 4.537 4.564 4.537 4.564 4.537 4.564 4.537 4.564 4.537 4.564Clusters . . . . 14 14 607 556 172 167
Full Exporters Full Exporters Full Exporters Full Exporters Full Exporters(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
India -0.166*** -0.168** -0.166*** -0.168** -0.166*** -0.168*** -0.166*** -0.168*** -0.166*** -0.168**(0.064) (0.068) (0.064) (0.068) (0.054) (0.055) (0.060) (0.064) (0.061) (0.066)
Constant 2.350*** 2.542*** 2.350*** 2.542*** 2.350*** 2.542*** 2.350*** 2.542*** 2.350*** 2.542***(0.436) (0.459) (0.483) (0.493) (0.357) (0.405) (0.511) (0.517) (0.374) (0.398)
Observations 843 769 843 769 843 769 843 769 843 769R-squared 0.086 0.090 0.086 0.090 0.086 0.090 0.086 0.090 0.086 0.090RMSE 0.887 0.894 0.887 0.894 0.887 0.894 0.887 0.894 0.887 0.894Clusters . . . . 14 14 541 495 159 156
Variables
The table reports OLS results with standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). All models replicate Model 5 of Table 3, including country, industry, and province fixed effects. Odd numbered models display regression results on the full PCI sample. Even numbers restrict analysis to firms operating in exportable sectors. Five standard error approaches are employed: 1) Classical; 2) Robust; 3) Clustered at Primary Sampling Unit (Province); 4) Three-Way Clustering at Province, Country of Origin, and Sector; and 5) Province and Country of Origin.
Classic SE Robust SE Cluster Province Three-Way Cluster Cluster Prov./Country
Dependent Variable 1: Share of Operating Costs Firms Will Spend on Labor Adjustments (Continuous)
Dependent Variable 2: Share of Operating Costs Firms Will Spend on Labor Adjustments (Binned)
VariablesClassic SE Robust SE Cluster Province Three-Way Cluster Cluster Prov./Country
g
Appendix F: Replication of Main Results Dropping All Firms from India or Europe & Doing Business in India or Europe
Full Sample(1) (2) (3) (4) (5)
India -0.436 -0.436 -0.500 -0.520 -0.388(0.335) (0.335) (0.341) (0.349) (0.351)
Constant 6.598*** 6.598*** 7.375*** 7.720*** 8.239***(0.230) (0.230) (1.910) (2.409) (2.361)
Country FE No No Yes Yes YesIndustry FE No No No Yes YesSector FE No No No No YesObservations 776 776 776 776 776R-squared 0.002 0.002 0.049 0.073 0.099RMSE 4.582 4.582 4.551 4.553 4.531Clusters 474 474 474 474 474
Full Sample(1) (2) (3) (4) (5)
India -0.114* -0.114* -0.120* -0.133** -0.119*(0.064) (0.064) (0.065) (0.067) (0.068)
Constant 2.253*** 2.253*** 2.623*** 2.615*** 2.689***(0.046) (0.046) (0.375) (0.499) (0.497)
Country FE No No Yes Yes YesIndustry FE No No No Yes YesSector FE No No No No YesObservations 670 670 670 670 670R-squared 0.004 0.004 0.047 0.070 0.089RMSE 0.881 0.881 0.878 0.881 0.881Clusters 420 420 420 420 420The table reports OLS results with standard errors, clustered at the (countryXindustryXprovince) level in parentheses (*** p<0.01, ** p<0.05, * p<0.1). The first panel tests the continuous dependent variable and the second panel tests the binned dependent variable displayed in Figure 3. Model 1 uses the full sample of respondents. All subsequent models test results on only firms with export potential.
DV1: Share of Operating Costs Firms Will Spend on Labor Adjustments (Continuous)Exporters Only
DV2: Share of Operating Costs Firms Will Spend on Labor Adjustments (Binned)Exporters Only
h
Appendix G1: Full Interaction Results and Sensitivity Tests (Continuous Dependent Variable)
No Controls Controls No Controls Controls No Controls Controls No Controls Controls(1) (2) (3) (4) (5) (6) (7) (8)
India Treatment=1 -0.115 -0.138 -0.677 -1.182 0.054 -1.003 -0.728** -1.033(0.473) (0.546) (0.655) (0.741) (0.983) (1.063) (0.370) (0.765)
Difference in Markups (ln) 0.164 0.185 0.157 0.263 0.233* 0.234*(0.147) (0.161) (0.159) (0.174) (0.131) (0.130)
India*Difference in Markups -0.302 -0.468** -0.277 -0.585** -0.318* -0.318*(0.221) (0.226) (0.258) (0.266) (0.179) (0.178)
Share of Intermediate Goods 0.316 -0.465 0.123 -1.214 -0.300(0.797) (0.922) (1.353) (1.531) (0.930)
India*Share 0.157 0.662 -0.368 1.776 0.580(1.153) (1.260) (1.897) (1.993) (1.262)
Years since registration (ln) -0.122** -0.103** -0.123** -0.116*** -0.116***(0.053) (0.040) (0.054) (0.040) (0.041)
100% Foreign Owned=1 -1.304 -0.338 -1.376 0.012 0.011(1.735) (1.006) (1.717) (0.975) (0.977)
Multinational Corp.=1 -0.364 -0.558 -0.367 -0.584 -0.581(0.529) (0.384) (0.530) (0.388) (0.389)
Entry through M&A=1 -0.011 -0.294 0.012 0.020 0.018(0.805) (0.642) (0.798) (0.660) (0.662)
Number of employees -0.000 -0.000 -0.000 -0.000 -0.000(0.001) (0.001) (0.001) (0.001) (0.001)
Profitable firm=1 0.116 0.099 0.115 0.150 0.144(0.480) (0.359) (0.480) (0.371) (0.370)
Export at all=1 0.383 0.670* 0.356 0.842** 0.838**(0.493) (0.377) (0.504) (0.399) (0.403)
Constant 6.408*** 8.481*** 6.828*** 8.330*** 6.353*** 9.190*** 7.585*** 7.741***(0.729) (1.969) (0.551) (1.192) (1.034) (2.147) (1.072) (1.172)
Country FE Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes YesObservations 605 484 967 737 605 484 699 699R-Squared 0.082 0.131 0.075 0.107 0.082 0.133 0.116 0.116RMSE 4.625 4.548 4.524 4.464 4.633 4.554 4.460 4.466Clusters 413 348 607 489 413 348 458 458The table reports OLS results with standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Models 1 and 2 interact the treatment variable with the natural log of the difference in markups calculated from Chinese Customs Data. Models 3 and 4 interact the treatment with share of intermediate products. Models 5 and 6 combine both interactions. Model 7 and 8 replace the measure of markups from Chinese data with one created from Vietnamese PCI data.
DV Share of operating cost will spend on labor upgrades
Markup Difference (CH) Intermediate Share Combined Markup Difference (VN)
i
Appendix G2: Full Interaction Results and Sensitivity Tests (Binned Dependent Variable)
No Controls Controls No Controls Controls No Controls Controls No Controls Controls(1) (2) (3) (4) (5) (6) (7) (8)
India Treatment=1 -0.056 0.006 -0.095 -0.194 -0.107 -0.263 -0.152** -0.164(0.095) (0.110) (0.123) (0.143) (0.190) (0.217) (0.075) (0.150)
Difference in Markups (ln) 0.049* 0.063* 0.057* 0.085** 0.028 0.028(0.028) (0.032) (0.031) (0.035) (0.028) (0.028)
India*Difference in Markups -0.062 -0.102** -0.067 -0.141** -0.053 -0.053(0.041) (0.046) (0.049) (0.056) (0.038) (0.038)
Share of Intermediate Goods 0.142 -0.005 -0.131 -0.325 0.019(0.157) (0.184) (0.268) (0.313) (0.187)
India*Share -0.121 0.028 0.098 0.571 0.023(0.221) (0.244) (0.374) (0.416) (0.249)
Years since registration (ln) -0.020* -0.017* -0.020* -0.020** -0.020**(0.012) (0.009) (0.012) (0.009) (0.009)
100% Foreign Owned=1 -0.309 0.006 -0.336 0.053 0.051(0.377) (0.227) (0.370) (0.235) (0.236)
Multinational Corp.=1 -0.142 -0.172** -0.144 -0.178** -0.178**(0.117) (0.084) (0.118) (0.086) (0.086)
Entry through M&A=1 -0.092 -0.038 -0.085 0.023 0.022(0.176) (0.141) (0.175) (0.142) (0.143)
Number of employees -0.000 -0.000 -0.000 -0.000 -0.000(0.000) (0.000) (0.000) (0.000) (0.000)
Profitable firm=1 0.047 0.004 0.046 0.028 0.028(0.104) (0.078) (0.104) (0.080) (0.080)
Export at all=1 0.155 0.136* 0.147 0.157* 0.157*(0.107) (0.082) (0.109) (0.087) (0.088)
Constant 2.005*** 2.355*** 2.191*** 2.402*** 2.072*** 2.540*** 2.321*** 2.313***(0.138) (0.422) (0.105) (0.262) (0.201) (0.454) (0.251) (0.270)
Country FE Yes Yes Yes Yes Yes Yes YesProvince FE Yes Yes Yes Yes Yes Yes YesObservations 528 422 843 643 528 422 608 608R-Squared 0.087 0.135 0.065 0.107 0.087 0.139 0.121 0.121RMSE 0.905 0.894 0.885 0.871 0.906 0.894 0.867 0.868Clusters 369 311 541 437 369 311 408 408
DV Share of operating cost will spend on labor upgrades
Markup Difference (VN)
The table reports OLS results with standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Models 1 and 2 interact the treatment variable with the natural log of the difference in markups calculated from Chinese Customs Data. Models 3 and 4 interact the treatment with share of intermediate products. Models 5 and 6 combine both interactions. Model 7 and 8 replace the measure of markups from Chinese data with one created from Vietnamese PCI data.
Markup Difference (CH) Intermediate Share Combined
j
Appendix H: Correlation between Different Measures of Markups
Source 1 2 3 41 Difference between Markups (Europe-India,ln) China Customs Data (2010) 12 Median Markups in European Union Christopolu and Vermulen (2012) 0.3933* 13 Median Markups in India Stiebale and Venape (2016) -0.4109* 0.2158* 14 Difference between Markups (Developed-Developing, ln) Vietnam PCI (2016) 0.1022* 0.3222* 0.2451* 1
* Significant at .05 level.
Measure of Markups
k
Appendix I: Additional Sub-Group Effects We identify the firm’s sector, which is associated with both markup availability and visibility
of labor practices, as a key mechanism through which the survey experiment treatment operates. In addition to sector, what other firm characteristics might affect responses to the survey experiment? The table below explores a wide range of important subgroups. These include: (1) ownership type (100% foreign owned versus joint venture); (2) entry mode (greenfield, merger and acquisition, or part of an MNC); (3) home country; (4) primary pre-treatment export location; (5) employment size; (6) capital size; (7) previous performance and existing expansion plans; (8) compliance with the current Vietnamese Labor Code (providing contracts for all full-time employees); and (9) demographics of the firm’s current labor force.
Table A2 below offers insights into additional attributes of emerging market firms that might render them more inclined to invest in labor rights improvements as a means of improving their access to global supply chains. It is important to interpret these results with caution, however. First, none of these subgroup features were randomly assigned; they might be associated with unobserved confounders. Second, these analyses were not anticipated in our original research design, which primarily targeted sectoral differences and destination market country effects. Third, many of the groups were too small to use the clustered, bootstrap standard error approach, so we present all results with only robust standard errors, which might be overly efficient. Each row in the table represents a separate regression, where the analysis is restricted to firms in that category. We display subgroups in the first column of the regression table. The fourth column provides the constant, which reveals the share of operating costs that firms receiving the Europe treatment were willing to expend. The second column reports the India treatment effect, which can be interpreted as the difference in expenditure shares between firms receiving the India treatment and those receiving the Europe treatment. We identify several interesting patterns. First, there are wide differences across investor home country. Firms from Europe, Japan, and Korea are more willing to invest in upgrades to enter the European market than firms from other locations. For instance, there was nearly a two-percentage point gap in the share of operating costs that European-owned companies were willing to spend to sell their products in Europe (7.8%) as opposed to in India (5.9%). By contrast, the gap was one-tenth that size for firms from China and Taiwan. On the other hand, the pre-treatment destination of the company’s sales does not appear to matter much; given that few firms already were exporting to either Europe or India, however, we are less confident in this result.
Second, ownership and entry patterns appear important. Firms that are 100% foreign owned and not branches of MNCs have the greatest response to the treatment. These are likely the firms most capable of changing direction and switching downstream relationships. Additionally, firms that would be classified as small and medium-sized enterprises (SMEs) by both labor (between 10 and 200 employees) and capital size (between $100,000 and $1 million) appear to be most willing to pay for the opportunity to sell to the European market. Relatedly, the treatment effect is most pronounced for firms with non-college educated, domestic (vs. foreign) workers. Finally, pre-treatment levels of compliance matter: firms that report being 100 percent compliant with Vietnamese labor law – and perhaps therefore accustomed to investing in labor-related upgrading -- were more responsive to the possibility of entering the European market. While none of these subgroup effects detracts from our main results that FIEs are more willing to engage in costly labor upgrading when selling to foreign markets, the findings about country of origin and lack of attachment to larger MNCs suggest directions for future research.
l
Table A2: Treatment Effects by Sub-Group
m
Appendix J: Mechanisms with Covariate Adjustment
This graph replicates Figure 8, but controls for potential covariates that might be associated with both sector and treatment of labor (see Appendix H above), including the age of firm, ownership type (100% foreign owned or joint venture), employment size, whether the firm is part of a multinational organization, whether the firm currently exports, and a five-point measure of performance in the year before the survey. The substantive results of the model hold. In the sector characterized by a high salience of labor issues (garments), firms are significantly more willing (p<.1) to engage in labor-related upgrades when given the possibility of selling to the European market. In the lower-salience rubber/plastic sector, the effect is essentially zero. In fact, firms appear slightly more likely to upgrade when offered the opportunity to sell to India. The treatment effects for garments and rubber/plastic are again statistically significant from one another. To test this, we ran an interaction of India and a dummy for garments, limiting the analysis to only these two sectors. The coefficient on the interaction term revealed that, all else equal, the effect of the treatment was 5.67 percentage points more negative in garments. The Wald statistic for the difference was 6.53, which was significant at the .1 level (see additional output below). Results are robust to using the binned estimator.
Garments Rubber/Plastic Interaction Controls Garments Rubber/Plastic Interaction Controls(1) (2) (3) (4) (5) (6) (7) (8)
India Treatment=1 -2.435* 1.056 1.056 1.943 -0.440* 0.097 0.097 0.354(1.269) (1.088) (1.089) (1.274) (0.259) (0.231) (0.231) (0.279)
Garments Sector 2.935** 4.917*** 0.393 0.827**(1.307) (1.723) (0.271) (0.343)
India*Garments Sector -3.490** -5.666*** -0.537 -1.004**(1.671) (2.007) (0.347) (0.428)
Years since registration (ln) -0.059 -0.018(0.106) (0.022)
100% Foreign Owned=1 -6.287*** -1.439***(1.687) (0.391)
Multinational Corp.=1 -1.169 -0.317(1.102) (0.242)
Entry through M&A=1 0.155 -0.059(1.358) (0.302)
Number of employees -0.001 -0.000(0.001) (0.000)
Profitable firm=1 1.504 0.267(1.112) (0.245)
Export at all=1 0.166 -0.212(1.210) (0.267)
Constant 8.435*** 5.500*** 5.500*** 10.879*** 2.571*** 2.179*** 2.179*** 3.676***(1.101) (0.705) (0.706) (2.313) (0.223) (0.155) (0.155) (0.536)
Observations 64 70 134 103 59 57 116 89R-Squared 0.064 0.014 0.046 0.135 0.054 0.003 0.032 0.147RMSE 4.522 4.575 4.550 4.542 0.893 0.873 0.883 0.894
DV Share of operating cost will spend on labor upgrades
The table reports OLS results with robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). This table empirically tests the descriptive statistics presented in Figure 8.
Continuous DV Binned DV
n
Appendix K: Works Cited in Appendix
Cameron, A.C. and Miller, D.L., 2015. “A practitioner’s guide to cluster-robust inference.” Journal of Human Resources 50(2): 317-372.
Kline, P., and Santos, A. 2012. “A Score-based Approach to Wild Bootstrap Inference.” Journal of Econometric Methods 1(1): 23–41.
Mutz, D.C. and Pemantle, R. 2015. “Standards for Experimental Research: Encouraging a Better Understanding of Experimental Methods.” Journal of Experimental Political Science 2(2): 192-215.
Roodman, David. 2016. “Boottest” STATA .ado file. https://ideas.repec.org/c/boc/bocode/s458121.html
Wooldridge, Jeffrey M. 2003. “Cluster- Sample Methods in Applied Econometrics.” American Economic Review 93(2):133–38.
