Lights On: How Transparency Increases Compliance in Cambodian Global Value Chains

Abstract

This article evaluates the implementation of a transparency policy in garment factories in Cambodia through the Better Factories Cambodia program. Using a difference-in-difference approach that is often applied to control for endogeneity, the author finds that compliance improved following the implementation of the transparency policy. Compliance increased in a group of 21 critical compliance areas that represent fundamental worker rights relative to relevant comparison groups. Compliance among the least-compliant factories, however, did not increase relative to other factories, possibly reflecting limited access to resources.

Keywords

apparel industry global value chains compliance transparency Cambodia

Finding ways to improve working conditions in global value chains is a policy priority for governments, international organizations, unions, corporations, and nongovernmental organizations. Indeed, some have suggested that compliance with local labor laws and international labor standards is necessary to participate in global value chains. Voluntary monitoring (when firms or factories are responsible for monitoring their own compliance or compliance within their own supply chains) has been criticized for some time, and “independent” audits alone (that is, without additional measures) generate limited improvements in working conditions. The current debate centers around whether audits in combination with other measures have a higher chance of success and, if so, what other measures might be especially effective.

One possible other measure is transparency. In the private sector, financial transparency and disclosure are widely considered to be essential for corporate governance. With roots in Louis Brandeis’s famous 1913 Harper’s Weekly article that extolls the potential benefits of publicity, transparency theoretically increases the benefits of compliance (because stakeholders may accurately reward compliant firms) and increases the costs of noncompliance. Transparency has been promoted as a way to improve government and private-sector performance, but few empirical studies examine transparency’s effectiveness beyond corporate finance.

The Better Factories Cambodia (BFC) program offers a prominent exception. Motivated by the 1999 US-Cambodia Bilateral Textile Trade Agreement that formally linked market access to labor standards, the International Labor Organization (ILO)’s Better Factories Cambodia program monitors Cambodian garment exporters and assesses working conditions relative to ILO Core Labor Standards and Cambodian labor law. Widely lauded for its combination of a tripartite (government, labor, and business) approach, support for remediation, and significant improvements in working conditions, the BFC program was the model for ILO’s Better Work program that has spread to at least seven countries.

The BFC program began in 2001. Between 2001 and 2006, transparency (making factory-level audit information publicly available) was a key aspect of the BFC program. BFC ended transparency in 2006 with a move to a new monitoring system and in response to resistance from factory managers. BFC returned to public disclosure in October 2013 and announced in January 2014 that public disclosure would begin in March 2014. The disclosure policy focused on critical compliance issues, least-compliant factories, and union compliance. To focus on critical issues, BFC announced that they would post compliance of all factories in 21 critical compliance issues. To focus on least-compliant factories, BFC announced they would post compliance of the least compliant factories based on compliance with 52 legal requirements that include the 21 critical issues.

The goal of this article is to estimate the changes in compliance around the return to public disclosure. The first research question is, did compliance with 21 critical issues increase relative to compliance in other areas for all factories? The second is, did compliance in all compliance areas increase in the least-compliant factories relative to compliance in the same areas in the rest of the factories? These questions contribute to the debate surrounding the potential effectiveness of transparency outside of corporate finance and the debate over the most effective ways to improve working conditions in global supply chains.

Although these questions seem straightforward, the extensive program evaluation literature explains that finding a relevant comparison group and controlling for unobserved factors that may affect selection into treatment are necessary steps to answering them. The structure of the BFC transparency program, and data collected by the program, are ideally organized to meet both criteria. In particular, this article applies a difference-in-difference technique that is empirically appropriate for program evaluation.

This article contributes to several important academic and policy areas. In particular, the results illustrate that transparency may be effective in improving compliance in global supply chains and may productively complement monitoring and auditing programs. The next section explains how these results fit into these areas.

Literature Review: Supply Chains, Monitoring, BFC, and Transparency

Perhaps the most significant change in international trade over the past three decades has been the rise of global supply chains. The opportunity to export created opportunity for workers in developing countries (Robertson, Brown, Pierre, and Sanchez-Puerta 2009) but also brought international attention to poor conditions in exporting factories. Disasters, media reports, and the work of nongovernment organizations created urgency about working conditions for international firms and their consumers (Elliott and Freeman 2003; Jammulamadaka 2013; Sinkovics, Hoque, and Sinkovics 2016).

In response, lead firms of global supply chains took several approaches to assess and improve working conditions in their developing-country supplier factories. Lee (2016) surveyed the range of approaches, from corporate codes of conduct to international framework agreements. All involve some sort of monitoring. A consensus is emerging that suggests that self-monitoring and voluntary compliance are insufficient for improvements in working conditions because of insufficient incentives (Kuruvilla and Verma 2006; Boiral 2007; Locke 2013; Rossi, Luinstra, and Pickles 2014) or competing incentives (Barrientos 2013). The growing concern over private monitoring may have contributed to a proliferation of third-party monitoring programs that created a web of heterogeneous standards. As a result, factories complain about “monitoring fatigue” from having to meet a wide range of slightly different standards (Locke 2013). This fatigue can cause factories to focus on meeting the terms of specific codes on inspection day rather than working to raise working conditions generally.

The Better Factories Cambodia program was an innovative approach toward improving working conditions (Polaski 2006; Berik and van der Meulen Rodgers 2010). Rising Cambodian apparel exports that followed Cambodia’s transition to a market economy in the 1990s led to the 1999 US-Cambodia Bilateral Textile Trade Agreement. This agreement was unlike other agreements, either before or since, because it is the only one that has offered the opportunity to increase exports to the US market if a country improved working conditions in its factories. The United States, therefore, had to determine how working conditions would be evaluated. From several proposals, the United States decided that the International Labour Organization (ILO) would monitor Cambodian garment factories and determine whether Cambodia met the conditions described in the agreement (Kolben 2004). The ILO created the Better Factories Cambodia program in 2001 to fill this need.

In some ways, BFC is similar to other auditing programs. During unannounced visits, monitors assess working conditions in exporting garment factories. Monitoring teams contain at least two people who rarely assess the same factory twice. In Cambodia, monitors use an instrument containing questions designed to evaluate conditions and wage requirements relative to national law and international standards. Based on this comparison, BFC decides whether the factory is compliant on each question.

The BFC program is different from other programs in several ways. Unlike in most countries, the Cambodian government required BFC participation for permission to export. Therefore, the BFC program covers all exporting garment factories in Cambodia. The tripartite nature reduces monitoring fatigue to some degree because the buyers and other stakeholders were willing to accept ILO assessments in place of their own audits. Employing local assessors keeps assessment costs relatively low. The comprehensive BFC survey forms contain more than 250 questions. The ILO assessment team determines noncompliance on each question and assigns a 0 or 1 to each question to indicate compliance or noncompliance. The BFC remediation support helps factories address areas found to be noncompliant.

BFC monitors visit each factory approximately every eight months. Repeated visits generate a panel data set that makes it attractive for academic studies. Kotikula, Pournik, and Robertson (2015) described the evolution of the program and several of its social impacts. Studies generally have found a positive relationship between program participation and working conditions (Miller, Nuon, Aprill, and Certeza 2009; Adler and Woolcock 2010; Oka 2010a, 2010b; Shea, Nakayama, and Heymann 2010; Brown, Dehejia, and Robertson, 2014a, 2014b; Brown et al. 2016). Other studies have focused on other aspects of Better Work. Asuyama, Fukinoshi, and Robertson (2017) suggested that improvements in working conditions through BFC have improved factory performance. Bair (2017) studied Better Work Nicaragua’s hybrid governance structure. Posthuma and Rossi (2017) studied supranational supply chain governance using Better Work as an example.

The BFC program has evolved over time.¹ Factory visits between 2001 and 2005 identified significant violations with the goal of using follow-up visits to identify progress in problem areas. Early records are therefore less complete than later visits. Later visits began with the launch of an improved Information Management System (IMS) survey in December 2005. Ang, Brown, Dehejia, and Robertson (2012) found that BFC’s decision to end transparency in 2006 adversely affected compliance, and BFC returned to a policy of public disclosure in 2013.

The BFC case is important because prior research has generated mixed results about the effectiveness of transparency policies generally (e.g., Pozen 2018). Fung, Graham, and Weil (2007) found that government transparency policies that produce incomplete or irrelevant information can be ineffective or even counterproductive. To be successful, Weil, Fung, Graham, and Fagotto (2006) argued that transparency policies must produce information upon which decision-makers rely directly. In a way, Weil et al. (2006) predicted that transparency would matter in supply chains if the results produce information that is used in decision-making. Examples of global supply chain transparency, such as what we analyze in this article, however, are rare. Egels-Zandén and Hansson (2016) found that supply chain transparency is more helpful as a corporate tool than for consumers directly, which also seems consistent with Weil et al. (2006). This article extends this literature by evaluating the changes after BFC’s return to disclosure in October 2013.

The New BFC Transparency Program

The reasons for returning to transparency, as stated on the BFC transparency website, are to build on Cambodia’s reputation for decent working conditions and to “keep pace with competing industries where disclosure of ILO factory compliance data will soon be the norm,” support the Cambodian government’s enforcement efforts, induce changes in the chronically noncompliant factories, and “accelerate improvements in working conditions on critical issues.”² The new transparency program followed months of discussion with factory representatives (such as the Garment Manufacturer’s Association of Cambodia [GMAC]) and other stakeholders. The factories and subscribing buyers began to receive the results of transparency-focused monitoring reports in December 2013, and BFC posted results on the BFC website in March 2014.

The approximately 250 questions that make up the BFC compliance reports cover dozens of issues, including 21 critical areas, 31 noncritical areas, and a wide range of other areas. The ILO decided not to post the full compliance reports to help factories focus on three problem areas. Instead, for all factories, it posted compliance with the 21 critical areas (referred to as “critical issues”) and, for those factories with very low compliance, it posted the results for all 52 areas (referred to as “low compliance”). The online results also include information about union compliance, but this article focuses on the first two because union compliance includes a range of issues that are beyond the scope of this article. Salmivaara (2018) elaborated on Cambodian unionization.

To focus on fundamental worker rights, BFC identifies 21 critical issues that are the minimum requirement that BFC expects factories to maintain. Discussions between BFC, the Better Work Global staff, brands, unions, and other stakeholders identified each of the 21 critical issues and their legal foundation from either local labor law, international conventions, or signed memorandums of understanding (MOU). Some of the issues, such as child labor and forced labor, are “zero tolerance” issues for many, if not all, buyers. All factories with at least two visits have assessments posted online. Factories can request that BFC verify improvements before the results are posted online. Results posted quarterly include the factory name, assessment date, and number of critical issues in which the factory is compliant (e.g., 19 of the 21 issues).

The second component of the transparency program, “low compliance,” shifts the focus to the low-performing factories as measured by overall compliance in 52 measured areas that include an additional 31 issues besides the 21 fundamental worker rights described earlier. Table 1 lists the 52 issues. The first 21 are the critical issues discussed in the previous paragraph.³ All factories are assessed, and factories whose average compliance in the 52 areas falls two standard deviations below the mean are considered to be the low-compliant factories.⁴ The original low-compliant factories were identified prior to the program, but the potential for being included on the list of low-compliant factories may be ongoing. Although all factories are assessed, only the low-compliant factories, and their compliance scores, are listed on the website. Factories that are not listed in this group are those whose compliance is assessed to be above the minimum threshold described previously. All factories with at least three visits are included, and factories can request that BFC verify improvements before the results are posted online.

Table 1.

Transparency Issues

No.	Transparency issue description
1	Child labor is not found, or (where confirmed) is remediated
2	No forced labor
3	No discrimination against workers during hiring, employment, or termination based on their race, color, sex, religion, creed, ancestry, social origin, or political opinion
4	Women workers are not dismissed or forced to resign due to pregnancy
5	Women workers are not dismissed, and their employment status or seniority is not changed during maternity leave
6	No sexual harassment
7	Men and women doing work of equal value receive different pay
8	No discrimination against any worker due to the worker’s union membership or union activities (during recruitment, employment or termination)
9	Workers can freely form and join trade unions of their choice
10	The employer has not taken steps to bring the union(s) under its control
11	A worker’s job is not dependent on the worker not joining a union
12	Management does not interfere with workers or unions when they draw up their constitutions and rules, hold elections, or organize their activities, administration or finances
13	Workers are free not to join a union
14	The factory conducts regular emergency evacuation drills (every 6 months)
15	Emergency exits are unlocked during work hours
16	Sufficient number of emergency exit doors
17	Proper guards installed on all dangerous moving parts (not including needle guards)
18	Water for drinking is clean and sufficient
19	Minimum wages are paid correctly for all types of workers
20	Overtime wages are paid correctly for all types of workers
21	Management does not use fixed-term rotating contracts, or otherwise not include the entire period of continuous employment when determining workers’ entitlements to maternity leave, attendance bonus, seniority bonus, and/or annual leave
22	Management gets permission from the labor ministry before dismissing union leaders or candidates for union leadership
23	Emergency exit doors are accessible
24	Access paths are free of obstruction
25	Training to use fire extinguishers (one training/year)
26	Fire extinguishers are sufficient
27	Electrical wires and switches are well maintained
28	Safety warnings are posted on the main electrical switch boxes
29	Chemicals are stored properly
30	Chemicals are labeled in Khmer
31	Exhaust ventilation where chemicals are used
32	Proper personal protective equipment is provided
33	Management keeps one complete and accurate payroll
34	Sunday and holiday work payments are paid correctly
35	Severance payments (5%) are paid correctly
36	Required bonuses and allowances are paid correctly
37	Wage deductions are legal and authorized
38	Workers receive clear written payslips in Khmer
39	Overtime is limited to 2 hours per day and is voluntary
40	Workers are not punished if they refuse to work overtime
41	Attendance bonuses paid with authorized use of leave
42	Management gives workers annual leave (paid or unpaid) or annual leave compensation
43	Unused annual leave is paid out upon termination
44	Management gives workers paid sick leave
45	Women workers get at least 90 days off for maternity leave
46	Maternity leave payments are correct
47	Workers are allowed to take special leave (paid or unpaid)
48	Conciliation agreements are implemented (rights issues)
49	Arbitration awards are implemented (rights issues)
50	Management, including line supervisors, treats workers with respect
51	Monitors are allowed entry
52	All documents requested are provided

Notes: This table is adapted from the Better Factories Cambodia (BFC) Transparency website. The full table, which includes sample questions and legal references, can be accessed at http://betterfactories.org/transparency/uploads/CI_LC_Issues&Legal_Ref.pdf.

According to the BFC Transparency Database Report (Better Factories Cambodia 2018), the share of factories in full compliance with critical issues increased from 33 to 44% from the time before factories were eligible to be listed in the transparency database (dates vary by factory) to the time after they were eligible to be listed. The number of violations on 21 critical issues fell from 389 to 234 over the same period. Many categories demonstrated improvements, including emergency drills, open emergency exits, and having one complete and accurate payroll. Furthermore, the percentage of low-compliance factories fell from approximately 3% to about 2% between November 2017 and May 2018.

These simple summary statistics help frame the program and indicate increased compliance following the implementation of transparency. To go into more depth and answer the questions posed in the introduction, we first describe the compliance data.

Data

The empirical analysis draws upon the factory-level compliance reports. Technically, BFC does not claim to measure compliance. BFC measures noncompliance. For this study, we describe a lack of finding of noncompliance as compliance for the sake of prose and neither evaluate (second-guess) their noncompliance decision nor add additional information that would imply anything more than a lack of noncompliance. These reports cover all Cambodian garment exporters and include some factory characteristics (total employment, nation of ownership, and buyer relationship). Table 2 shows the factory counts by year and visit number and includes both new and revisited factories. The “visit number” headings refer to the number of BFC visits a factory has had (e.g., “1” represents the first visit, “2” represents two visits, and so on), and the numbers in the table are the number of factories that had been visited that many times. For example, in 2009, 25 factories had their first-ever BFC visit and 29 had their second BFC visit. The changing number of first-visit factories over time represents the entrance of new factories to Cambodia’s garment export sector. Factories are visited at least once per year and may be visited twice in a given year. The lower-left triangle is blank because, in 2001, no factory had been visited more than twice and, for example, no factory had been visited more than seven times in 2007. The 2006 new system implementation with full audits is evident in Table 2. The slight drop in 2017 reflects the fact that, at the time of this analysis, data were available through October 2017.

Table 2.

Factory Counts by Visit Number and Year (2001–October 2017)

Visit No.	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017
1	25	6		10	27	205	30	34	25	16	48	32	71	69	61	20	0
2	2	3	4	4	31	98	132	32	29	12	20	29	42	57	56	57	33
3		1		3	9	44	178	39	22	28	11	14	33	41	42	48	40
4				1	3	10	75	140	29	19	25	5	26	25	42	28	36
5					1	6	25	98	77	22	20	22	9	19	24	27	21
6						2	10	44	97	40	23	13	22	12	16	21	19
7							4	12	46	74	45	12	20	20	18	8	19
8								6	21	43	71	23	15	16	18	17	9
9								1	4	23	43	44	29	18	17	18	9
10								1		6	26	27	44	31	14	12	10
11									1		7	21	31	39	24	10	10
12										1	2	3	27	25	35	16	11
13												2	5	22	25	23	13
14													2	8	17	18	14
15														1	13	12	12
16														1		13	11
17															1	2	8
18															1		3
19																1	1
Total	27	10	4	18	71	365	454	407	351	284	341	247	376	404	424	351	279

Notes:“Total” represents the total number of factories assessed by BFC in a given year. The visit number refers to the number of times BFC has visited a given factory, and the numbers in the table represent the number of factories with that many visits in a given year. For example, in 2009, 25 factories were visited for the first time, and 29 had their second-ever BFC visit.

Changes in the assessment instrument over time, required to accommodate learning and shifting emphasis, make tracking individual questions difficult. Question codes were changed more frequently than individual questions; the same questions would be encoded differently over time. The resulting question set for this study is restricted to questions that directly imply compliance and could be consistently tracked through time. For example, questions such as “How many office staff are employed by the factory?” are not included as compliance questions. Questions coded on a continuous scale (not 0/1 compliance) describe plant characteristics when appropriate and were dropped in other cases. To maximize consistency over time, the individual questions from each instrument were matched manually. The match was reviewed by BFC staff. Questions that could not be consistently matched through time were not included in the data set.

Compliance issues outside the 52 compliance categories described in Table 1 are aggregated into a 53rd category labeled “all other questions.” Examples of questions in category 53 include, “Is there adequate ventilation?” and “Is the workplace clean?” Category 53 is used as a comparison group in the difference-in-difference analysis that follows.

For some exercises, presenting compliance for 53 categories is not practical. There are many ways to group individual questions. To simplify the presentation of the analysis, the 21 critical issue areas are grouped into seven critical issue groups (CIGs) and are listed in Table 3. The label of each group is used in the remainder of the article to identify each group in the results tables. Questions that do not fall into one of the CIGs are grouped into a category labeled as “Not 21 CI.”⁵

Table 3.

Critical Issue Areas into Seven Critical Issue Groups

No.	Critical issue label	Critical issue group
1	No unremediated child labor	Core
2	No forced labor	Core
3	No discrimination against workers	Core
4	No dismissal of pregnant workers	Core
5	No dismissal of workers during maternity leave	Core
6	No sexual harassment	Core
7	Equal pay for men and women	Core
8	No discrimination against workers based on union membership	Unions
9	Workers can join and form unions freely	Unions
10	No control of union by employer	Unions
11	Job is not dependent on union membership	Unions
12	No management interference with unions when they draw up their constitutions and rules, hold elections, or organize their activities, administration or finances	Unions
13	Workers are free not to join a union	Unions
14	Regular emergency evacuation drills (two drills/year)	Emergency
15	Emergency exit doors are unlocked during working hours	Emergency
16	Emergency exit doors are sufficient	Emergency
17	Dangerous machine parts have safety guards (not needle guards)	S_Guards
18	Water for drinking is clean and sufficient	Water
19	Correctly paid minimum wages	MinW
20	Correctly paid overtime wages
21	Bonuses, allowances, leaves count entire employment period	Bonus

Notes: MinW, minimum wages; S_Guards, safety guards.

Table 4 contains the number of total observations in each category, which is equal to the number of compliance categories times the number of factories times the number of time periods shown in Table 2 for each factory over the 2001–2017 period. The CIGs are clearly a minority of the total available questions, as shown by the much higher number of observations in the first category. Some categories are quite specific (such as safety guards, labeled “S_Guards” in Table 4) and therefore have only a few relevant questions. Others, such as the core labor standards, include a wide range of questions. As expected based on the “zero tolerance” requirements of many buyers, the core labor standards category has nearly perfect compliance and varies little over time and across factories. Others, such as unions and minimum wages (MinW), also exhibit over 96% average compliance.

Table 4.

Summary of Critical Area Groups

Group	Total observations	Mean compliance
Core	19,076	0.986
Unions	23,328	0.975
Emergency	12,227	0.836
S_Guards	4,416	0.837
Water	4,504	0.946
MinW	4,532	0.963
Bonuses	4,416	0.818
Not 21 CI	131,461	0.835
Total	202,936	0.870

Notes: Groups are defined in Table 3. Total number of observations includes all questions, factories, and visits included in the sample. Mean compliance is the simple (unweighted) arithmetic average across all factories, all visits, and questions within each category. The last category, “Not 21 CI,” aggregates all other compliance areas. MinW, minimum wages; S_Guards, safety guards.

Findings

To answer the two research questions posed in the introduction, this section contains two approaches. The first is a trend analysis. The trend analysis intuition is simple: Did the trend in compliance over time change when BFC implemented the new transparency program? The second is a difference-in-difference analysis. The intuition of the difference-in-difference analysis is straightforward: Did the compliance levels of the targeted group (either critical issues or low-compliance factories) change relative to compliance levels in relevant comparison groups?

Trend Analysis

Table 5 shows the number of factories by year from 2001 to 2017 for which we have compliance data and the mean compliance across all questions and factories without controlling for any potentially relevant variables. The mean compliance is the simple arithmetic average of the 0/1 compliance variable taken over all questions and all factories within each year. As 1 indicates compliance (and 0 noncompliance), higher numbers indicate higher average compliance. Note that the compliance averages rise from 77.1% in 2002 to 89.6% in 2010 but at a slower rate after 2006. From 2011 to 2017, overall average compliance rates fall from 88.2% to 85.7%.

Table 5.

Factory Count and Mean Compliance

Year	Number of factories	Mean compliance
2001	26	0.813
2002	9	0.771
2003	5	0.825
2004	19	0.833
2005	48	0.862
2006	270	0.800
2007	282	0.887
2008	290	0.885
2009	255	0.893
2010	234	0.896
2011	282	0.882
2012	235	0.860
2013	319	0.853
2014	375	0.863
2015	393	0.883
2016	333	0.883
2017	274	0.857

Notes: Author’s elaboration using data from BFC compliance report. Factory counts for 2017 are low because the data run through October. Compliance mean is calculated as the simple average of 0/1 compliance indicator across all compliance questions within each year. If a factory was visited twice in a given year, the compliance reports were averaged to get one annual average. The number of factories is different from the totals in Table 4 because some factories may have had multiple visits in a given year that would show up separately in Table 4 but not in Table 5.

At least two (not mutually exclusive) reasons explain why average compliance may change over time. First, firms that have higher- or lower-than-average compliance may enter or leave the sample. Figure 1 shows the level of compliance of factories as assessed in their first audit (“new” factories) and how that level and range of variation changes between 2001 and 2016 (note that the numbers along the x-axis are years, so that 1 represents 2001). The box-and-whisker plots in Figure 1 show the median, 25th, and 75th percentiles of compliance by year. Figure 1 shows that median compliance increased from 80 to 91% between 2001 and 2009. Median compliance fell to 86% in 2013, but recovered to 90% by 2016. These changes are small relative to the whole sample. Figure 1 also shows that the range of variation in compliance decreased substantially between 2001 and 2009 and increased modestly thereafter. The main point of Figure 1 is that compliance in new factories changed in ways similar to the overall mean shown in Table 5.

Figure 1.

First Visit Compliance Levels, 2001–2016

The second possibility is that firms change their compliance over time. Figure 2 shows the compliance by BFC visit. The largest change in compliance occurs between the first and second visit. BFC may pay more attention to firms with very low compliance than to firms with higher compliance. Figure 2 shows that median compliance rises from about 86% in the first visit to about 94% by the 15th visit. Note that the lowest compliance scores in the first visit tend to disappear as visits increase. The standard deviation falls from 16% in the second visit to approximately 7% in the 15th visit. Although the spread increases near the right side of the graph, in general the lower values are much closer to the median than they were during early visits. This result suggests that participating in the BFC program increases compliance, especially for the firms with the lowest initial compliance.

Figure 2.

Compliance by Visit Number, 2001–2016

Table 5 and Figures 1 and 2 show falling compliance in the latter half of the sample, whether this is measured in average compliance, new factory compliance, or factory visit. BFC announced a return to public disclosure in 2013. A natural question that arises is whether a noticeable change in the pattern of compliance occurred around the time of the policy change (October 2013) or the posting online in March 2014. For example, did some categories seem to go from falling compliance to rising compliance? As the focus on the transparency program was on low-compliance areas, did the low-compliance areas become more compliant?

One way to answer this question is to estimate the change in the compliance trend for different compliance categories. For example, compliance may have been falling until the implementation of the program (negative trend), and then may begin rising after the implantation of the program (positive trend). Such a change in trend is known in econometrics as a “trend break.” To formally evaluate the change in trend in different compliance areas, Table 6 contains the results from the trend break tests presented by Vogelsang and Perron (1998). Specifically, the tests identify the time period in which the trend changes (if at all). Table 6 also includes a description of the compliance trend over time. A full description of this approach is found in the Appendix.

Table 6.

Trend Breaks by Critical Issue Group

Category	Turning point	Trend pattern
Core	2012q3	High, constant after break
Unions	2017q1	High, constant after break
Emergency	2013q2	Falling until break, then rising sharply
S_Guards	2015q2	Falling until break, then constant
Water	2016q2	Falling until break, then constant
MinW	2009q1	High, constant after break
Bonuses	2009q4	Falling until break, then rising sharply

Notes: Turning points are estimated using the second additive outlier model of Vogelsang and Perron (1998), which is a model for identifying the unknown trend break in time series data. MinW, minimum wages; S_Guards, safety guards.

Table 6 shows roughly three kinds of patterns among the seven compliance categories described in Table 4. The first pattern, exhibited by the core, unions, and minimum wage compliance groups, is relatively consistent over time. The breaks identified by the trend break tests are relatively small and do not occur near the time of the policy change. These categories also exhibit the highest compliance rates in Table 4 and therefore may not have had much room for improvement.

The second pattern, shown by the safety guards and water groups, is consistently falling until the estimated break. At the break, the trend levels off and the compliance rates remain relatively stable. The third pattern, shown by the emergency and bonus categories, are falling trends and then, around the time of the policy change, show sharp increases in compliance. Note that emergency and bonus have the lowest average compliance in Table 4.

To illustrate these trends, Figure 3 shows examples of each of the three change types. In particular, Figure 3 shows the sharp increase that occurs in the emergency and bonus groups. Note that the estimated trend break in Table 6 for the bonus group is early and, in fact, seems to be identifying an earlier break. Figure 3, however, suggests that a second break occurs right around the time of the policy change, and after the policy change the compliance in the bonus category sharply increases. This sharp increase in the group with the lowest average compliance in Table 4 is consistent with the hypothesis that factories responded to the return to public transparency by increasing compliance in the groups that had the lowest previous compliance. These changes seem consistent with the theory and literature presented earlier.

Figure 3.

Patterns of Compliance Trends and Estimated Breaks

Difference-in-Differences with Ordinary Least Squares

One of the concerns about estimating the effect of a program is that unobserved factors (such as management quality) might affect the outcome that is hypothesized to be linked to an external policy change. That is, endogeneity might arise if firms become more compliant after the return to disclosure for reasons other than disclosure (for example, they may have management that was more likely to be compliant for other reasons). The intuition of the difference-in-difference approach is that the effect of a program, such as the implementation of transparency, would change the difference between the targeted group and the rest of the sample. As long as unobserved characteristics that might affect compliance of the factories remain constant, the change in that difference, also called the difference in the difference, is often considered to be the effect of the program in the program evaluation literature.

The program evaluation literature emphasizes the importance of accurately identifying the most relevant comparison group (Imbens and Wooldridge 2009). Simple “before and after” assessments can lead to inaccurate results when selection is important. Comparing “participants” with “non-participants” after the program is problematic because program “participants” may be different (e.g., endogenously selected), and it is possible, and perhaps even likely, that such differences (and not the program per se) explain differences in post-participation outcomes. Imbens and Wooldridge (2009) showed that randomized experiments, propensity score matching (PSM), and difference-in-difference models are three of the most common approaches used in evaluation. In the absence of a randomized experiment, Imbens and Wooldridge (2009) pointed out, and appealed to Smith and Todd (2005) for support, that the difference-in-difference approach provides a valid alternative to the PSM approach that addresses the same concern driving the PSM approaches.

The implementation of the difference-in-difference method is straightforward. Compliance is regressed on a dummy variable indicating membership in the treatment group before the transparency program (which is constant over time), a dummy variable equal to 1 is used for all periods after October 2013, and then the interaction between these two is measured. The estimation also includes a group of other control variables (year, size, region, and having a reputation-conscious buyer). The sample includes all factories over the 2001–2017 time frame, and the year effects absorb the effect of the earlier transparency period on compliance. The coefficient on the interaction term is the change in compliance in the treatment group relative to the change in the comparison group. The identification comes from the fact that the comparison groups did not have the same incentives to improve overall compliance.

The validity of the difference-in-difference approach requires finding a good comparison group. In the case of transparency in Cambodia, the program design offers two excellent and arguably exogenous comparison groups. The two comparison groups follow from the way that transparency was implemented. The first focuses on critical issues (for all factories). The comparison group is compliance in the non-critical issues. That is, we ask the question, did compliance increase in the critical issues relative to compliance in the rest of the compliance areas in the transparency period? As such, we compare “within” factories in the sense that we compare average compliance rates for different groups of questions within each factory and compare how the difference in average compliance across these question categories changes after the implementation of the transparency program.

The second focuses specifically on low-compliant factories. Here we ask the question, did the low-compliance factories increase compliance relative to other Cambodian factories? The comparison group includes the factories that are not identified as low-compliant factories. The low-compliant factories are identified prior to the transparency program and are followed over time so that both the treatment and the comparison groups are clearly identified. In particular, high-compliance factories (before the program) face little, if any, risk of being included in the program (i.e., they are very unlikely to be selected into the treatment category), which makes the comparison appropriate.

Under these conditions, Smith and Todd (2005) found that the difference-in-difference approach performs better than cross-sectional estimators (including the PSM approach described earlier).

Estimation Issues

The literature on global value chains generally, and BFC specifically, identifies four potential confounding factors that might affect the estimation results: changes in the global apparel market, country of ownership, relationship with international buyers, and factory size. Each of these, and how they are addressed in the estimation, are discussed in turn below.

Falling global apparel demand may reduce revenues, making it more difficult for factories to improve working conditions. A study by Robertson and colleagues found that falling demand after the global financial crisis (2007–2014) was associated with lower wages for apparel workers in Cambodia (Robertson, Lopez-Acevedo, and Savchenko 2019). Ruwanpura and Wrigley (2011) described how the financial crisis challenged Sri Lankan garment producers. By contrast, Beresford (2009) found that working conditions in Cambodia did not fall in response to an increasingly competitive environment when the MultiFibre Arrangement (MFA) ended at the end of 2004, but they did not evaluate falling demand more generally.

Figure 4 shows the change in the overall mean compliance rate and US apparel imports from Cambodia (measured as the quantity rather than value to avoid changes due to prices). US imports from Cambodia rose from 2002 until they fell during the 2007–2008 financial crisis. As with global trade, the import demand quickly recovered. Import demand rose until about 2012, when, with the exception of 2014, US imports from Cambodia fell. The simple (unconditional) correlation between the two is high (0.816). Compliance seems to fall before imports, however, raising questions about the direction of causality. To control for potential changes in demand over time, the empirical analysis includes individual year variables. These year variables also control for the earlier transparency period.

Figure 4.

Changes in Average Compliance and US Apparel Imports from Cambodia

Given that domestic factories may have less access to resources and support (Lipsey, Sjoholm, and Sun 2010), foreign ownership may be related to compliance decisions. Because the data include factories from the time they begin with BFC to the time that they close (or reach the end of the sample period), we can compare factories that close at any point during the sample period with those that reach the end of the sample period. As a rough proxy for parent-company support, Figure 5 compares factories that eventually close and those that survive, low-compliance status, and ownership region for all factories in all years. All factories in the sample fall into one of the four categories shown in Figure 5.

Figure 5.

Survival Status of Factories in Cambodia by Regional Ownership and Compliance Status, Various Years

Significant differences occur in patterns across the four panels. Cambodian-owned factories emerge most prominently in the lower right—the panel with factories that close and are in the BFC low-compliance category. By contrast, the panel in the upper right shows that Hong Kong–Taiwan–Macao factories most prominently feature as surviving with low compliance. To control for the possible effects of region of ownership, we include variables in the regressions that control for country of ownership.

Having a relationship with a reputation-sensitive buyer also affects compliance decisions (Oka 2010a, 2010b). Reputation-sensitive buyers may offer support or incentives to factories to improve working conditions. The BFC data include an indicator for factories that purchase BFC reports (prior to the new transparency program, factory-level reports were available by subscription). Thus, following Weil et al. (2006), who suggested that transparency may matter most for those who would use the data generated by transparency, these factories are coded as reputation sensitive. The factories included in this set are major international retailers and name brands, and, as such, are those that are often elsewhere characterized as “reputation sensitive.”

Figure 6 shows the average compliance rates over time for each of four categories: those with and without a reputation-sensitive buyer and those in and out of the low-compliance group. A clear difference is seen between factories that have a relationship with a reputation-sensitive buyer and those that do not. Compliance rates in the different groups follow similar paths over time. Factories with a reputation-sensitive buyer have consistently higher compliance than those without. Among the compliant factories, those with a reputation-sensitive buyer are more compliant than those without one; and among low-compliance factories, the same pattern exists—those with a reputation-sensitive buyer have higher average compliance rates over time than those without a reputation-sensitive buyer. To control for the possible effects of having a reputation-sensitive buyer, we include a binary variable representing that relationship.

Figure 6.

Compliance by Reputation-Sensitive Buyer and Compliance Status, 2001–2017

The last potential confounding variable is the size of the factory. Larger factories may have more resources than do smaller factories and, if improving compliance involves high fixed costs (such as installing an air conditioner), larger firms may be more able to afford these fixed costs. To control for the possible effects of factory size, we include a variable representing total factory employment.

Critical Issues

In this section we compare compliance in the 21 critical issues areas with compliance in all other areas for all factories before and after the new transparency program begins. If the program has a positive effect on compliance in the 21 critical areas, the compliance in these areas should increase relative to the other compliance areas after the program goes into effect.

To compare compliance in the 21 critical issues with compliance in other areas, each factory in each observed period has two observations. The first is the arithmetic average of the compliance indicators in the 21 critical issues. The second is the arithmetic average of the compliance indicators in all other compliance areas. The post-program variable is a binary variable equal to 1 for all dates after October 2013 (and 0 otherwise).

Table 7 presents the results of difference-in-difference regressions. The sample covers all factories in all years shown in Table 2 with two observations per factory. The first row shows the compliance rate in the 21 critical issues relative to the rest of the categories across all periods. These coefficients are statistically significant and consistently positive. Compliance in the 21 critical issues is higher, on average, than in other categories, even before the transparency program.

Table 7.

Difference-in-Difference Estimates of Compliance Changes in Critical Issues

	(1)	(2)	(3)
Variables	As compared to all other categories	As compared to NonCritCats	As compared to Cats 22–52
Critical issues	0.091***	0.033***	0.091***
	(0.003)	(0.002)	(0.003)
Post program	0.006	−0.009	0.006
	(0.009)	(0.008)	(0.009)
CI × Post	0.024***	0.042***	0.025***
	(0.005)	(0.004)	(0.005)
Constant	0.566***	0.591***	0.612***
	(0.017)	(0.015)	(0.019)
Observations	8,978	8,978	8,964
R-squared	0.263	0.232	0.263

Notes: Standard errors in parentheses. The “CI × Post” is the interaction between the critical issues variable and the post program (after October 2013) variable and represents the difference-in-difference estimate. Sample covers all factories in all years shown in Table 2 with two observations per factory (the mean compliance for critical issues and mean compliance for non-critical issues). Reputation-sensitive buyer, region, factory size, and year variables are included but not reported.

***

p < 0.01; **p < 0.05; *p < 0.1.

The second row shows the average overall compliance for time periods after October 2013 (when the transparency program goes into effect). These estimates are generally small and positive or, in the second column, small and negative. Table 5 shows that the decline started in 2011, so it seems unlikely that a shift in compliance toward the 21 critical issues from other categories caused by the transparency program explains the negative value in column (2) in the second row of Table 7.

The third row contains the difference-in-difference estimates: the interaction of the transparency program variable with the 21 critical issues variable. These coefficients are consistently positive and statistically significant. The main implication is that factories significantly improved compliance in the 21 critical issues relative to the rest of the compliance areas during the transparency period and is consistent with the hypothesis that the transparency program effectively increased compliance in the 21 critical issues.

Table 7 also contains the results from different comparison groups. Column (1) compares the 21 critical issues to all other categories. The second column compares the compliance of the 21 critical issues with 220 other compliance issues that are assessed by BFC but are not in Table 1 (those in group 53). The third column compares compliance in the 21 critical issues with the areas 22–52 listed in Table 1. The results are positive and statistically significant. The largest difference is between the 21 critical issues and compliance in the issues not listed in Table 1. The results are robust to the inclusion of the control variables described earlier (factory size, association with a reputation-conscious buyer, year effects, and region of ownership). The results are also robust to limiting the sample to the 2006–2017 period.

This section’s main message is that the transparency program is strongly associated with increasing compliance for the 21 critical issues. To the extent that the difference-in-difference approach effectively addresses concerns about endogeneity, these results are consistent with the hypothesis that the return to transparency contributed to increasing compliance in the critical issues.

Low-Compliance Factories

BFC classifies factories that fall two standard deviations below the mean compliance level as “low-compliance” factories. BFC’s decision about which firms meet this criterion offers a good opportunity for evaluation using the rest of the factories as a comparison group. In this context, the difference-in-difference approach asks the question, did low-compliance factories increase compliance more than factories not at risk of being posted online as part of the low-compliance group?

Table 8 contains the estimation results. The significant negative values in the first row show that the low-compliance factories were low-compliance factories prior to the transparency program. The small positive coefficient in the third row shows that overall compliance did not change much, on average, during the transparency period relative to the overall downward trend captured by the year variables (to save space, the year variables results are not shown but are available upon request).

Table 8.

Difference-in-Difference Estimates of Compliance Changes in Low-Compliance Factories

	(1)	(2)	(3)
Variables	As compared to all other factories	Compared to below average	Mid compared to above average
Low-compliance factories	−0.067***	−0.041***
	(0.008)	(0.009)
Below-average factories			−0.114***
			(0.003)
Post program	0.020*	0.024*	0.025***
	(0.011)	(0.014)	(0.009)
LCF x Post	−0.036**	−0.052***
	(0.016)	(0.017)
BAF x Post			−0.014***
			(0.005)
Constant	0.619***	0.606***	0.761***
	(0.022)	(0.028)	(0.018)

Observations	4,492	2,196	4,301
R-squared	0.162	0.171	0.446

Notes: Standard errors in parentheses. Below-average factories (BAF) represent factories below the average compliance level that are not in the low-compliance group in column (3). The “LCF × Post” is the interaction between the low-compliance factories (LCF) variable and the post program (after 2014) variable and represents the difference-in-difference estimate. The “BAF × Post” is the interaction between the below-average factories variable and the post program (after 2014) variable and represents the difference-in-difference estimate. The sample covers factories and years covered in Table 2 unless otherwise specified in column title. Reputation-sensitive buyer, region, factory size, and year variables are included but not reported.

***

p < 0.01; **p < 0.05; *p < 0.1.

The coefficients in the fourth row are, to the extent that the difference-in-difference approach addresses endogeneity, the estimated effects of compliance on the low-compliance factories (the treatment group) relative to the comparison group. The coefficients are negative and statistically significant: low-compliance factories experienced falling compliance during the transparency program relative to the other factories.

To explore the robustness of the difference-in-difference results, the three columns use different comparison groups. In column (2), we limit the sample to factories with below-average compliance. Along with the same caveat about endogeneity mentioned earlier, the estimated program effects are more negative. The results in column (3) compare the factories below the mean but are not in the low-compliance group with the rest of the factories. These factories also exhibit falling compliance following a return to disclosure but less than the least-compliant firms. One might expect that the risk of being listed might have induced an improvement in compliance, but Table 8 suggests that this is not the case.

The results in Table 8 are robust to the addition of the control variables. Additional controls for selection also generate similar results, which suggests that the results are not driven by changes in the composition of the sample over time. The results are also robust to limiting the sample to the 2006–2017 period.

Several characteristics of the low-compliance factories may explain their falling compliance during the transparency period. Figure 4 shows that Cambodian factories feature more prominently among factories that are more likely to close. Cambodian factories are especially prominent in the lower-right quadrant, which suggests that domestic factories struggle with both survival and compliance. Figure 5 shows that compliance falls most for low-compliant factories without a relationship with a reputation-sensitive buyer, suggesting that being associated with a reputation-sensitive buyer might increase compliance relatively more for the low-compliance factories than for factories with greater compliance.

As compliance rates seem to persist over time, but are falling most for the least-compliant factories, it is also interesting to compare the less-compliant firms by size. Figure 7 shows the size of the open and more compliant factories, the low-compliant factories that remain open, and the low-compliant factories that close. Figure 7 shows that the open low-compliant factories fall into two groups: large (approximately 1,100 workers) and small (approximately 400 workers). Note that low-compliant factories that have closed fall into the smaller group, that is, the closed low-compliant factories are very similar in size to the smaller but open low-compliant factories. By contrast, the open compliant factories tend to be much larger.

Figure 7.

Factory Sizes by Operational (Open or Closed) and Compliance Status

The policy implication of this pattern is that steps to improve compliance among the least-compliant factories might most effectively include access to resources and buyers that will help the factories grow. In other words, low compliance may be linked to resources and size (both of which are linked to foreign ownership), and a return to transparency may not provide the means for factories to surmount the barriers preventing compliance.

Simultaneous Estimates of Both Dimensions of Transparency

The estimation results in the previous two sections follow the structure of the BFC transparency program, but the overlap of the 21 critical issues in low-compliance factories may affect the results. Table 9 presents a schematic of the nine separate categories implied by the analysis in the previous two sections. The comparison of the change in compliance in the 21 critical issues with both compliance issues 22–52 and the remaining (approximately 200) compliance points (the group 53 described earlier) is represented in Table 9 as a comparison between column (1) and columns (2) and (3). The comparison of the change in compliance of the low-compliance factories with mid-compliant and high-compliant factories is represented in Table 9 as a comparison between the first row and the second and third rows. The overlap between the 21 critical issues and the low-compliant factories is very clear in Table 9 (the cell in the first row and first column).

Table 9.

OLS Results Difference-in-Difference for Individual Question-Factory Groups

		Question categories
		(1)21 Critical issues	(2)22–52 Transparency issues	(3)Other compliance points	Row sum(SE)
Factory categories	Low compliance	−0.013(0.017)	−0.048***(0.017)	−0.016(0.017)	−0.077***(0.030)
	Mid compliance	0.022***(0.005)	−0.011**(0.005)	−0.027**(0.005)	−0.016(0.011)
	High compliance	0.033***(0.005)	0.013**(0.005)	Reference
	Column sum (SE)	0.042**(0.019)	−0.046**(0.019)

Notes: This table contains the results from one estimation equation that includes the same controls and methodology as in Tables 7 and 8. Table includes only the interaction terms between each group indicator and the post-October 2013 variable. As such, these results represent the change in the difference between the reference category (indicated in the table above) and each of the other categories. Main effects of each category, the estimate of the post-October 2013 variable, reputation-sensitive buyer, region, factory size, and year variables are included but not reported. Standard errors (SE) in parentheses. OLS, ordinary least squares.

***

(**) indicates statistically significant at the 1% (5%) level.

To directly assess this overlap, we estimate the change in compliance for eight of the nine categories relative to the cell labeled “Reference” in Table 9. Specifically, we compare the change in compliance of the high-compliance factories (defined as factories with overall compliance scores above the median) for the compliance issues in previously described group 53 to the remaining factory-compliance issue combinations. This reference group is chosen because it is the farthest from the treatment groups, as seen in Table 9.

Table 9 contains the difference-in-difference estimation results with the same sample (all factories over the 2001–2017 time periods) and the same control variables. Rather than report the results for all included variables, Table 9 reports only the interaction between the post–October 2013 and individual category variables (the difference-in-difference estimates; results for all variables are available upon request). As such, the estimates in Table 9 show the change in compliance for each category after the return to transparency relative to the change in compliance for high-compliant factories in the non-transparency compliance issues.

Several important results emerge from Table 9. First, the overlap (of changes in compliance of the low-compliance factories in the 21 critical issues) does not drive the previous sections’ results. The improvement in compliance in the 21 critical issues appears in all factories except the low-compliance factories. The falling compliance of the low-compliance factories does not appear in the 21 critical issues. Instead, compliance falls in the remaining transparency issues. Note that compliance in all factories in transparency issues 22–52 fell as well, although it fell the most in the low-compliant factories. For all but the low-compliance factories, the increase in compliance in the critical issues was larger than the fall in compliance in the remaining transparency areas, leading to the net increase in compliance shown in the previous sections.

The results in Table 9 can be compared with the two previous sections by taking the sum of the individual category estimates across rows and columns. The row and column sums are also presented in Table 9. The results are qualitatively identical in the sense that the sums show that compliance in the 21 critical issues rose overall, and compliance for the low-compliance factories fell, relative to the compliance in the “other” compliance points in the most compliant factories.

Conclusion

Finding ways to improve working conditions in developing countries is a policy priority for many stakeholders, including unions, international buyers, governments, and factory managers. The idea of transparency has been advocated theoretically and has received some empirical support in the literature. This article evaluates changes in compliance following a targeted program of transparency in Cambodia’s garment supply chain factories. The program focused on three main areas: critical issues, low-compliance factories, and union compliance. Focusing on the first area, this article finds strong and positive results. The return to transparency is associated with a statistically significant (and economically significant) increase in compliance in a group of 21 critical issues that represent fundamental worker rights. In particular, compliance in these areas improved relative to the rest of the compliance issues. The most significant changes around the time of the implementation of the program were those areas that had the lowest prior compliance (emergency preparedness and bonuses). These results are important both for the program and for the literature more broadly because there are few examples of a targeted program, such as the BFC transparency program, that allow for identification of a clear comparison group. The main policy implication from these results is that targeted transparency programs seem to be an effective way to improve compliance for important compliance issues.

Conversely, the difference-in-difference results suggest that the program was not associated with rising compliance of the low-compliance factories listed online. In fact, compliance in these factories fell during the transparency period. The falling compliance results are robust to a variety of specifications, control variables, and selection issues. Low-compliance factories face a host of problems. Low compliance persists over time and many low-compliant factories are similar in size to the low-compliant factories that have closed. Larger low-compliant factories are associated with foreign ownership and may be more likely to survive. Association with a reputation-sensitive buyer is also associated with higher compliance, and this difference is especially stark within the group of least-compliant factories. The policy implications are that addressing low-compliance factories may require a wide range of policies that are associated with factory growth, including access to capital and markets, that may go beyond what transparency alone can deliver.

Looking forward, several questions emerge from this research, one of which relates to factory survival. Brown, Dehejia, and Robertson (2011) suggested that improvements in working conditions between the first and second BFC visit are associated with higher chances of survival. Understanding whether the low-compliance factories tried increasing compliance as a strategy for survival would help shape transparency programs for other countries into the future. Overall, however, the results here suggest that transparency seems to be an effective complement to audits as a way to improve working conditions among all but the least-compliant factories.

Footnotes

Appendix

Acknowledgements

I thank Better Factories Cambodia (BFC) staff, particularly Esther Germains, for help with the data and for useful comments. Comments from Sarah Ashwin, Matthew Amengual, Greg Distelhorst, participants at the MIT Special Issue Workshop, and participants at the 69th meeting of the Labor and Employment Relations Association (LERA) are especially appreciated. Research assistance from Jaime Sepulveda and Andrea Talero Bonilla is gratefully acknowledged. Data are managed by Better Work Global.

Stata programs are available upon request from the author at robertson@tamu.edu.

ORCID iD

Raymond Robertson

1

The BFC program evolved into a global Better Work program that now operates in Bangladesh, Haiti, Indonesia, Jordan, Nicaragua, and Vietnam. The Lesotho Better Work program ended in 2016. See . Better Work participation in Haiti was included as part of the HOPE II legislation passed by the US Congress.

2

See for this list and additional information.

3

Note that BFC also gathers data on a number of other areas beyond the 52 listed in . Later in the article these other areas are grouped into a single 53rd group.

4

The two-standard-deviation criterion comes from statistics, where, in a normal distribution, 95% of the values fall within two standard deviations from the mean. Values beyond two standard deviations are generally considered to be true outliers in the sense that it is very unlikely (a 2.5% chance) that factories would fall into this group by chance if they were actually complying at an “average” level.

5

The grouping collapses the individual issue areas into groups using the unweighted arithmetic average of compliance scores. Although using identical weights (that is, using no weights) imposes the assumption that all categories are valued equally, alternative weighting schemes would require additional justification that would go beyond the scope of the article.

References

Adler

Daniel

Woolcock

Michael

. 2010. Justice without the rule of law? The challenge of rights-based industrial relations in contemporary Cambodia. In Fenwick

Colin

Novitz

Tonia

(Eds.), Human Rights at Work: Perspectives on Law and Regulation, pp. 529–644. Oxford, UK: Hart Publishing.

Ang

Debra

Brown

Drusilla

Dehejia

Rajeev

Robertson

Raymond

. 2012. Public disclosure, reputation sensitivity, and labor law compliance: Evidence from Better Factories Cambodia. Review of Development Economics 16(4): 594–607.

Asuyama

Yoko

Fukinoshi

Takahiro

Robertson

Raymond

. 2017. Labor compliance and factory performance: Evidence from the Cambodian garment industry. Better Work Discussion Paper No. 23. February. Geneva and Washington, DC: International Labour Organisation and International Finance Corporation (IFC).

Bair

Jennifer

. 2017. Contextualising compliance: Hybrid governance in global value chains. New Political Economy 22(2): 169–85.

Barrientos

Stephanie

. 2013. Corporate purchasing practices in global production networks: A socially contested terrain. Geoforum 44: 44–51.

Beresford

Melanie

. 2009. The Cambodian clothing industry in the post-MFA environment: A review of developments. Journal of the Asia Pacific Economy 14(4): 366–88.

Berik

Günseli

Rodgers

Yana van der Meulen

. 2010. Options for enforcing labour standards: Lessons from Bangladesh and Cambodia. Journal of International Development 22: 56–85.

Better Factories Cambodia. 2018. Better Factories Cambodia transparency database report, 11th Cycle. September. Accessed at https://betterwork.org/dev/wp-content/uploads/2018/10/BFC-Transparency-Database-Report-Cycle-11th.pdf.

Boiral

Olivier

. 2007. Corporate greening through IS0 14001: A rational myth? Organization Science 18(1): 127–46.

10.

Brandeis

Louis

. 1913. What publicity can do. Harper’s Weekly. December 20.

11.

Brown

Drusilla

Dehejia

Rajeev

Rappaport

Ann

Davis

Mary

Robertson

Raymond

Babbitt

Laura

Voegeli

Elyse

, et al. 2016. The impact of Better Work. September. Accessed at betterwork.org/dev/wp-content/uploads/2016/09/Tufts-University-Final-IA.pdf.

12.

Brown

Drusilla

Dehejia

Rajeev

Robertson

Raymond

. 2011. Working conditions and factory survival: Evidence from Better Factories Cambodia. Better Work Discussion Paper No. 4. December. Geneva and Washington, DC: International Labour Organisation (ILO) and International Finance Corporation (IFC).

13.

Brown

Drusilla

Dehejia

Rajeev

Robertson

Raymond

. 2014a. Factory decisions to become noncompliant with labor standards: Evidence from Better Factories Cambodia. In Arianna Rossi, Amy Luinstra, and John Pickles (Eds.), Towards Better Work: Understanding Labour in Apparel Global Value Chains, pp. 232–50. Geneva and Basingstoke: International Labour Organisation and Palgrave Macmillan.

14.

Brown

Drusilla

Dehejia

Rajeev

Robertson

Raymond

. 2014b. Is there a business case for improving labor standards? Some evidence from Better Factories Cambodia. In Jennifer Bair, Doug Miller, and Marsha Dickson (Eds.), Workers’ Rights and Labor Compliance in Global Supply Chains, pp. 69–87. New York: Routledge.

15.

Egels-Zandén

Niklas

Hansson

Niklas

. 2016. Supply chain transparency as a consumer or corporate tool: The case of Nudie Jeans Co. Journal of Consumer Policy 39(4): 377–95.

16.

Elliott

Kimberly Ann

Freeman

Richard B

. 2003. Can Labor Standards Improve under Globalization?Washington, DC: Institute for International Economics.

17.

Fung

Archon

Graham

Mary

Weil

David

. 2007. Full Disclosure: The Perils and Promise of Transparency. New York: Cambridge University Press.

18.

Imbens

Guido

Wooldridge

Jeffrey M

. 2009. Recent developments in the econometrics of program evaluation. Journal of Economic Literature 47(1): 5–86.

19.

Jammulamadaka

Nimruji

. 2013. The responsibility of corporate social responsibility in SMEs. International Journal of Organizational Analysis 21(3): 385–95.

20.

Kolben

Kevin

. 2004. Trade, monitoring, and the ILO: Working to improve conditions in Cambodia’s garment factories. Yale Human Rights and Development Journal 7: 79–107.

21.

Kotikula

Aphichoke

Pournik

Milad

Robertson

Raymond

. 2015. Interwoven: How the Better Work Program Improves Job and Life Quality in the Apparel Sector. Washington, DC: World Bank.

22.

Kuruvilla

Sarosh

Verma

Anil

. 2006. International labor standards, soft regulation, and national government roles. Journal of Industrial Relations 48(1): 41–58. doi: https://doi.org/10.1177/0022185606059313.

23.

Lee

Joonkoo

. 2016. Global supply chain dynamics and labour governance: Implications for social upgrading. ILO Research Paper No. 14. May. Geneva: International Labour Organisation.

24.

Lipsey

Robert E.

Sjoholm

Fredrik

Sun

Jing

. 2010. Foreign ownership and employment growth in Indonesian manufacturing. NBER Working Paper No. 15936. Cambridge, MA: National Bureau of Economic Research.

25.

Locke

Richard

. 2013. The Promise and Limits of Private Power: Promoting Labor Standards in a Global Economy. New York: Cambridge University Press.

26.

Miller

Doug

Nuon

Vaasna

Aprill

Charlene

Certeza

Ramon

. 2009. Business as usual? Governing the supply chain in clothing post MFA phase out: The case of Cambodia. International Journal of Labor Research 1(1): 10–33.

27.

Oka

Chikako

. 2010a. Accounting for the gaps in labour standard compliance: The role of reputation-conscious buyers in the Cambodian garment industry. European Journal of Development Research 22(1): 59–78.

28.

Oka

Chikako

. 2010b. Channels of buyer influence and labor standard compliance: The case of Cambodia’s garment sector. Advances in Industrial and Labor Relations 17: 153–83.

29.

Polaski

Sandra

. 2006. Combining global and local forces: The case of labor rights in Cambodia. World Development 34(5): 919–32.

30.

Posthuma

Anne

Rossi

Arianna

. 2017. Coordinated governance in global value chains: Supranational dynamics and the role of the international labour organization. New Political Economy 22(2): 186–202.

31.

Pozen

David E.

2018. Transparency’s ideological drift. Yale Law Journal 128(1): 100–65.

32.

Raymond

Robertson

Brown

Drusilla

Pierre

Gaëlle

Sanchez-Puerta

Laura

(Eds.). 2009. Globalization, Wages, and the Quality of Jobs: Five Country Studies. Washington, DC: World Bank.

33.

Robertson

Raymond

Lopez-Acevedo

Gladys

Savchenko

Yevgeniya

. 2019. Globalisation and the gender earnings gap: Evidence from Sri Lanka and Cambodia. Journal of Development Studies. Accessed at https://doi.org/10.1080/00220388.2019.1573986.

34.

Arianna

Rossi

Luinstra

Amy

Pickles

John

(Eds.). 2014. Towards Better Work: Understanding Labour in Apparel Global Value Chains. Basingstoke, UK: Palgrave Macmillan.

35.

Ruwanpura

Kanchana

Wrigley

Neil

. 2011. The costs of compliance? Views of Sri Lankan apparel manufacturers in times of global economic crisis. Journal of Economic Geography 11(6): 1–19.

36.

Salmivaara

Anna

. 2018. New governance of labour rights: The perspective of Cambodian garment workers’ struggles. Globalizations 15(3): 329–46.

37.

Shea

Anna

Nakayama

Mariko

Heymann

Jody

. 2010. Improving labour standards in clothing factories: Lessons from stakeholder views and monitoring results in Cambodia. Global Social Policy 10(1): 85–110.

38.

Sinkovics

Noemi

Hoque

Samia F.

Sinkovics

Rudolph R

. 2016. Rana Plaza collapse aftermath: Are CSR compliance and auditing pressures effective? Accounting, Auditing & Accountability Journal 29(4): 617–49.

39.

Smith

Jeffrey

Todd

Petra

. 2005. Does matching overcome Lalonde’s critique of nonexperimental estimators? Journal of Econometrics 125(1-2): 305–53.

40.

Vogelsang

Timothy J.

Perron

Pierre

. 1998. Additional tests for a unit root allowing for a break in the trend function at an unknown time. International Economic Review 39(4): 1073–100.

41.

Weil

David

Fung

Archon

Graham

Mary

Fagotto

Elena

. 2006. The effectiveness of regulatory disclosure policies. Journal of Policy Analysis and Management 25(1): 155–81.