Paradigm Shift or Business as Usual? Workers’ Views on Multi-stakeholder Initiatives in Bangladesh

The scale of the tragedy at Rana Plaza in Bangladesh, in which more than 1,000 garment factory workers died when the building collapsed in April 2013, galvanized a range of stakeholders to take action to prevent future disasters and to acknowledge that business as usual was not an option. Prominent in these efforts were the Accord on Fire and Building Safety in Bangladesh (hereafter the Accord) and the Alliance for Bangladesh Workers’ Safety (hereafter the Alliance), two multi-stakeholder agreements that brought global buyers together in a coordinated effort to improve health and safety conditions in the ready-made garment industry. These agreements represented a move away from the buyer-driven, compliance-based model, which hitherto dominated corporate social responsibility initiatives, to a new This article is part of a project that was funded by the Volkswagen Foundation under its Europe and Global Challenges programme. The workers’ component was carried out as a collaboration between the London School of Economics and the BRAC Institute of Governance and Development, with supplementary funding from ESRC-DFID (ES/ L005484/1). We would like to thank Adiba Afros, Mahabub Rahman and Saiful Islam for their support in the research. In addition, we would like to remember our friend and colleague, Simeen Mahmud, who was a co-investigator on this project, but died on 19 March 2018.


INTRODUCTION The Objectives of the Article
The collapse of the Rana Plaza Tower on the outskirts of Dhaka, in Bangladesh, in April 2013 killed 1,134 and injured many more of the garment factory workers who were in the building at the time. The scale of the tragedy, and subsequent worldwide media coverage, turned it into a 'focusing event' (Birkland, 1998: 54), triggering policy action on health and safety by a range of institutional actors, and widespread acknowledgement that 'business as usual was not an option' (Labowitz and Baumann-Pauly, 2014: 11). Prominent among these responses were two initiatives -the Accord on Fire and Building Safety in Bangladesh (hereafter the Accord) and the Alliance for Bangladesh Worker Safety (hereafter the Alliance) -which brought global retailers and brands (hereafter lead firms) into collaboration with other concerned stakeholders to tackle problems in factories from which they sourced their garments. These initiatives, and the Accord in particular, have been hailed as 'historic' (All Party Parliamentary Group on Bangladesh, 2013: 36), 'game changing' (Ryan, 2013) and 'a new paradigm in the enforcement of global labour and human rights' (Anner et al., 2013: 2). They were seen to represent a move away from the buyer-driven, compliance-based model which hitherto dominated corporate social responsibility (CSR) initiatives in global supply chains, to a more collaborative multi-stakeholder approach.
There has been considerable interest within the CSR community in how these initiatives performed in the five years or so of their existence. There have been studies of managers in factories associated with the initiatives (Anner, 2018) as well as qualitative consultations with various Bangladeshbased stakeholders (Alamgir and Banerjee, 2019). There have also been surveys of workers since Rana Plaza (Ahmed and Nathan, 2014;Moazzem and Azim, 2018), but none have sought to explore the new initiatives from the perspectives of workers in the industry.
Our article addresses this gap in the literature. It is part of a larger project which assesses the impact of Rana Plaza from the standpoint of three primary sets of stakeholders: global lead firms, many sourcing from Bangladesh; Bangladeshi suppliers to global firms; and Bangladeshi garment workers producing for global markets. Our analysis draws on a survey of workers which was purposively designed to explore possible changes in their wages and working conditions as a result of the new agreements. We use our findings to speculate on whether the Accord did indeed represent a paradigm shift in the enforcement of global labour rights or was merely a continuation of 'business as usual' in a different guise. We begin with a brief account of how stakeholder initiatives, a key concept underpinning this analysis, have been theorized in the broader CSR literature before going on to discuss our empirical findings.

Stakeholder Theory and Global Value Chains: From Compliance to Collaboration
Early formulations of CSR envisaged entrepreneurs as potential agents of change who could transform the social order through the introduction of new progressive values (Falck and Heblich, 2007). This normative vision had to be abandoned once the focus shifted from the entrepreneur as individual to enterprises as legal entities in which managers were contractually obliged to maximize shareholder profits. Attempts to reconcile CSR objectives with shareholder interests led to more instrumental reformulations. The idea of 'strategic philanthropy' (Porter and Kramer, 2002: 56), for instance, argued that companies could enhance their reputations and thereby increase the value of their brands by engaging in socially responsible activities.
Stakeholder theory expanded the focus of this argument beyond the firm to encompass the range of external actors who were affected by, or could affect, a company's business activities, and could therefore be considered to have a stake in its activities (Freeman, 1984). It argued that corporations should maximize their immediate and longer-term success by factoring these stakeholders into its business plans, while distinguishing between 'key' stakeholders who had direct power to affect their profits, and 'minor' or 'emerging' ones whose significance varied according to the strength of their influence.
Arguments for CSR took on fresh life in the context of globalization as the pursuit of cost-cutting strategies by multinational corporations led to the relocation of the production of labour-intensive goods and services from the regulated, high-wage economies of the global North to low-wage, poorly regulated economies in lower-income countries. The resulting 'race to the welfare bottom' sparked off widespread 'anti-sweatshop' activism within Northern civil society aimed at exposing the violations of internationally agreed labour standards entailed in these strategies.
The threat to the reputations of multinational corporations and, by extension, to their profits posed by this 'emerging' category of stakeholders helped to shift corporate strategies from seeking to exploit the governance gap in supply chain capitalism to attempting to fill it through voluntary regulation. This took the form of corporate codes of conduct (CoCs), backed by periodic audits, which required supplying firms to comply with minimum labour standards as the condition for doing business with lead firms.
The adoption of this approach since the 1990s resulted in the rapid proliferation of codes, but subsequent evaluations reported mixed, often disappointing, results (Barrientos and Smith, 2007;Egels-Zandén, 2007;Lund-Thomsen et al., 2012;Oxfam, 2013). Impacts were generally limited to outcome standards, such as minimum wages and health, with little change in the less tangible 'process' rights to freedom of association and collective bargaining embodied in core International Labour Organization (ILO) principles. Impacts also tended to be restricted to first-tier factories which dealt directly with global buyers, bypassing lower-tier factories working on a subcontracted basis. In addition, the top-down approach of the compliance model led to considerable resentment on the part of suppliers, accompanied by efforts at evasion and deception.
However, even as disillusionment with corporate codes set in among some of its civil society advocates, support for CSR was growing among official development agencies where the turn to neoliberal ideologies of market-led growth carved out a prominent role for socially responsible business as a key agent of poverty reduction (Blowfield, 2005;Jenkins, 2005). Stakeholder theory was re-conceptualized yet again, this time to move beyond the instrumental concerns of individual corporations to a broader 'developmental' perspective which argued that, whatever the interests of particular stakeholders, they could reap mutual gains from collaborative efforts to improve working conditions in their value chains and, by extension, in workers' living standards in developing countries (Blowfield, 2005;Lund-Thomsen and Lindgreen, 2014).
The thinking behind this 'win-win' rationale was spelt out explicitly in relation to global garment value chains in a World Bank evaluation of the ILO's Better Works programme, 1 one example of the new collaborative approach (Kotikula et al., 2015). It pointed out that while lead firms might make sourcing decisions on the basis of profitability considerations, they had an incentive to improve working conditions in the factories from which they sourced in order to avoid reputational risk. Their suppliers might baulk at the costs of improving working conditions in their factories, but would gain if this helped to increase worker productivity or suppliers' competitive edge in gaining orders. Governments could benefit from improving working conditions in their export industries if it helped them attract foreign investment and boost exports. Finally, workers had an unambiguous stake in good working conditions because it enhanced their quality of life and could increase their productivity.
However, the shift from compliance to cooperation in the CSR literature did not silence all its critics. Scholars argued that, for all the new rhetoric about shared stakeholder interests, CSR initiatives had to operate in a wider political economy in which the raison d'etre of corporations continued to be defined, first and foremost, in terms of shareholder dividends (Margolis and Walsh, 2003). Some questioned whether cooperation was ever possible between unequally positioned stakeholders interacting with each other within an intensely competitive global economy. As Lund-Thomsen and Lindgreen (2014: 18) note: 'In this competitive context, there is very little scope for cooperation, beyond the limits set by international consumer markets'. These debates, and their different perspectives on stakeholder theory, provide the point of departure for our analysis of the collaborative initiatives put in place in Bangladesh after Rana Plaza.

Background to the Study: The Export Garment Industry in Bangladesh
The positive side of the story of the export ready-made garment (RMG) industry in Bangladesh is its spectacular growth. From a handful of factories in the late 1970s, the industry grew to 5,876 registered factories by 2013 while its share of the country's exports rose from 4 per cent in the 1980s to around 80 per cent. 2 It currently employs between 4 and 5 million workers, depending on estimates used (Labowitz and Baumann-Pauly, 2015). 3 While earlier studies reported that over 80 per cent of this labour force were women (War on Want, 2011), the rise of knitwear manufacturing has led to a shift in its gender composition so that men now make up around 54 per cent of the labour force (Farole and Cho, 2017). The rapid growth of the industry can be attributed to the apparently inexhaustible supply of cheap, largely female labour migrating from the countryside in search of employment. The industry also benefited from the efforts of successive governments to provide incentives to local capital as part of its drive to increase the country's export earnings. These efforts were reinforced by the early emergence of the Bangladesh Garment Manufacturers and Exporters Association (BGMEA) in the 1970s as a powerful lobby for employers' interests.
However, there is also a negative side to the story which relates to the working conditions which fuelled this growth. Bangladesh is signatory to most of the ILO's core labour conventions, but enforcement has been weak, deliberately so in the export RMG sector. The government passed a new Labour Law in 2006 4 which consolidated and updated previous legislation but with sections of the political elite themselves owners of garment factories, the new legislation was limited in scope. For instance, it retained restrictions on workers' rights, including requirements of 30 per cent representation of workers in a factory and of approval by the Labour Directorate before a union could be registered.
The weakness of the country's trade union movement partly reflects these factors. Many unions are poorly resourced and their efforts to organize within factories have been blocked by the government and met with violent repression by employers (Rahman and Langford, 2012). Left-leaning unions, 2. For trade information of the Bangladesh Garment Manufacturers and Exporters Association (BGMEA) see: www.bgmea.com.bd/home/pages/tradeinformation# (accessed 5 December 2018). 3. There are suggestions that there may be more than 7,000 factories in the country if unregistered, indirectly sourced factories are taken into account (Labowitz and Baumann-Pauly, 2015). This would increase the official estimates of workers by an additional million. 4. For more information on the 2006 Bangladesh Labour Law, see: www.ilo.org/dyn/natlex/ docs/ELECTRONIC/76402/110637/F-1265526237/BGD76402%20Eng.pdf in particular, have been systematically denied registration by the two main parties that dominated national politics since the 1990s (ibid.). As a result, Bangladesh has few factory-based unions and those that exist must rely on the support of national union federations. At the same time, most national federations are affiliated to the major political parties and have a long history of disruptive politics in which party interests generally dominate over those of the membership (ibid.). This confluence of factors mean that the industry has long been characterized by the absence of written contracts, by routine violations of health and safety regulations, low wages, long hours of overtime, high rates of labour turnover and low levels of trade union membership, estimated at around 5 to 10 per cent (Khan and Wichterich, 2015). Not surprisingly, Bangladesh has featured prominently in anti-sweatshop campaigns and seen a proliferation of CoCs by lead firms seeking to do business in the country. But these were mainly focused on working conditions within factories rather than infrastructural safety.
The collapse of Rana Plaza propelled health and safety concerns to the forefront of the policy agenda. The US immediately suspended Bangladesh from tariff-free access to its markets. The EU used the threat of suspension to launch the Bangladesh Sustainability Compact with the Bangladesh government, the US, Canada and the ILO in order to hold the government to time-bound commitments to improve fire and building safety and labour rights (ibid.). The Bangladesh government for its part adopted a National Tripartite Plan of Action for Fire and Structural Integrity and passed the Labour (Amendment) Act 2013 5 to make trade union registration easier, to improve health and safety regulations and to upgrade labour inspections.
The other prominent outcome was, of course, the adoption of the two multi-stakeholder initiatives. The Accord was an agreement between more than 200 international brands from 20 mainly European countries and two global union federations (UNIGlobal Union and IndustriAll Global Union) along with eight IndustriAll-affiliated Bangladeshi unions. It also included four international labour rights NGOs as witness signatories. The much smaller Alliance came into existence a few months later with 27 brands, overwhelmingly from North America, as signatories. Estimates (based on both registered and unregistered factories) suggest that the two agreements covered around 27 per cent of factories in the RMG sector and 45 per cent of its workers (Labowitz and Baumann-Pauly, 2015). 6 There has been some debate as to whether the two initiatives should be treated as essentially similar manifestations of new collaborative approaches 5. See: www.ilo.org/dyn/natlex/natlex4.detail?p_lang=en&p_isn=94286&p_classification= 01.02 6. The government took responsibility under its Tripartite Plan for oversight over factories that did not supply to members of Accord and Alliance. There is less information on this initiative but what exists suggests that it has not made much headway (Barrett et al., 2018).
to CSR or as fundamentally different. Barrett et al. (2018) support the former position. They argue that both initiatives emerged as 'innovation out of crisis' (ibid.: 11) and functioned in much the same way: their inspection teams examined whether conditions in firms supplying the signatory brands complied with the country's building code and drew up corrective plans where necessary to be implemented by managers within a prescribed time frame. Both agreements provided for the setting up of Health and Safety Committees in factories and for training in worker safety. Both were intended to last for five years with funding provided by annual contributions from signatory firms. 7 Donaghey and Reinecke (2018) offer a different interpretation. The Accord, they argue, embodied basic principles of industrial democracy. It was an agreement between global buyers and international union federations and their local affiliates. It included four international labour NGOs as witness signatories and the ILO as neutral chair. It was legally binding in that failure by signatory brands to abide by their commitments could be taken up in courts in their home countries. It also placed considerable emphasis on building workers' capacity for voice and representation at the core of its safety processes, working closely with local affiliates of its global union partners.
By contrast, Donaghey and Reinecke view the Alliance as a 'fairly traditional CSR approach' (2018: 15) but with collective self-regulation. It was led by anti-union US brands who were not willing to sign up to legally binding commitments. It did not include any unions among its signatories, although it did provide some local unions and a major Bangladeshi NGO with an advisory role. The authors describe its approach to factory-level change as one of 'problem-solving' rather than 'capacity building' (ibid.: 28).
The methodology used to select workers for our survey does not allow us to separate out the impacts associated with the Accord and Alliance on workers' voice and representation in order to test the hypothesis suggested by these debates. As we describe below, the majority of workers in our sample were from factories affiliated with the Accord either on its own or jointly with the Alliance. However, in light of the claims made for the potential for industrial democracy embodied in the Accord, and the significant representation of Accord-affiliated factories in our sample, we are able to explore a different hypothesis: that affiliation with the Accord was associated with changes that went beyond the health and safety concerns which had provided the initial impetus for the agreement to include impacts relating to workers' voice and representation.
7. At the end of its five-year mandate, Alliance withdrew from Bangladesh. Accord went to court to seek an extension to complete the process of remediation. In May 2019, it came to an agreement to continue its work for another 281 working days after which a national council would take over.

Methodology and Description of Sample
Our research set out to explore the perspectives of workers on possible changes associated with the Accord and Alliance since their inception. In 2017, we carried out a survey of 1,500 workers from 240 factories in five sites in and around Dhaka, where much of the export industry is located. Ideally, we would have liked to have sampled workers from the factories covered by the survey of managers carried out as part of the same project. However, we were reluctant to allow managers from these factories to help us select our sample because of obvious dangers of selection bias and efforts to influence workers' responses.
We opted instead to select our sample from specific locations within or next to Dhaka district -Savar, Ashulia, Gazipur, Narayanganj and Dhaka city -based on the fact that they had the greatest concentration of the factories included in the management survey. We then focused on one neighbourhood in each location where the workers in these factories were most likely to reside. In each of these five neighbourhoods, 1,000 garment worker households were identified and 200 women and 100 men randomly selected from each neighbourhood sub-sample. This gave us a total sample of 1,500 workers (1,000 women and 500 men) evenly distributed across the five locations. 8 The survey was carried out in workers' homes with their consent, by prior appointment and with assurance of full confidentiality. The first step in our analysis was to classify the workers in our sample according to whether their factories were affiliated with either of the initiatives, but we faced the problem that many had either not heard of these agreements or did not know their factory's affiliation status. We consulted the Accord, Alliance and government websites to fill in this information. Table 1 provides basic information on the factories in our sample. 9 It uses BGMEA criteria to distinguish between small (<600 workers), medium (600-2,499) and large (2,500+) factories. Around 40 per cent of factories, but just 17 per cent of workers, fell into the small category, 39 per cent of factories and 37 per cent of workers fell in the medium category and 21 per cent of factories but as many as 46 per cent of workers fell in the large category. From the entire sample of workers, 29 per cent were affiliated to the Accord, 52.5 per cent to both the Accord and Alliance (A&A) and just 1.5 per cent were affiliated to the Alliance on its own. It is therefore important to note that when we refer to 'A&A-affiliate' factories in this study, we are referring to factories that are either affiliated with the Accord on its own or jointly with the Alliance. The remaining 17 per cent of workers from factories that were not affiliated to either agreement was treated as a form of 'control' group. The distribution of A&A affiliation by factory size 8. In the end, less than 10 per cent of the workers in our sample came from factories covered by the managers' sample. 9. A more comprehensive report of our findings can be found in Kabeer et al. (2019). varied in predictable ways. Only 32 per cent of workers in small factories were in A&A-affiliated factories compared to 86 per cent and 96 per cent in medium and large factories respectively. Workers in affiliated factories, particularly the larger ones, were more accurately informed about affiliation status of their factories than those in non-affiliated factories who were more likely to say that they 'did not know'. Table 2 provides a brief description of the workers in our sample. It compares workers in small factories with those from medium and large factories combined. 10 Among workers in our sample, 66 per cent were female (a reflection of the survey design), with little variation by factory size and A&A affiliation. The mean age of workers was 26.5 years, their mean years of education was 5.9; 77 per cent were married and the vast majority (87 per cent) had migrated to their current location specifically in response to opportunities in the RMG sector. They had spent a mean of 5.6 years in the industry and 3.6 years in their current factory.
For the analysis, we disaggregated the main question motivating this article into two sub-questions. First, to what extent did wages and working conditions differ between A&A-affiliated and non-affiliated factories? Studies have generally shown that wages and working conditions are better in the larger, often first-tier factories which deal directly with lead firms and are hence under greater international scrutiny than those at the smaller, usually subcontracted end. We wanted to know whether A&A affiliation made any difference to wages and working conditions, regardless of factory size. And secondly, could any differences that we found be attributed to the efforts of the two initiatives?
Our analysis explored these questions using objective indicators of wages and working conditions along with subjective indicators of workers' views and perceptions of these issues, including their views on whether and what kinds of changes had occurred in the aftermath of Rana Plaza. Table 3 provides descriptive statistics on the differences in these objective and subjective indicators by A&A affiliation and factory size, distinguishing between small factories and medium/large ones 11 and will be discussed in the course of the analysis.
In the next stage of the study, we used regression analysis to separate out the possible influence of A&A affiliation and factory size on selected indicators of wages and working conditions as well as on changes reported in these since Rana Plaza. The results of the regression analysis are reported in 10. Since the main purpose of our investigation is to assess the difference made to wages and working conditions by A&A affiliation independent of factory size, keeping large and medium categories of factories separate would have left us with excessively small sample sizes of non-A&A workers (76 and 26 respectively) with low precision of estimates. 11. It should be noted here that there was considerable similarity between medium and large factories in these indicators, justifying our decision to treat them as one category (Kabeer et al., 2019).   Source: see Table 1.
Tables 4a, 4b and 5. We used Ordinary Least Squares as our preferred regression model for both continuous and binary variables. The main reasons for selecting linear probability models for binary outcomes are for simplicity of interpretation and comparability of the specifications throughout the paper.
Since the focus of our study was to identify correlates of wages, working conditions and voice in the factories rather than making predictions, Ordinary Least Squares is likely to produce the same conclusions as non-linear models (such as logit or probit). 12 We decided to conduct a number of focus group discussions (FGDs) after a preliminary analysis of the survey data in order gain a better understanding of our quantitative findings. Seven FGDs were carried out between August and September 2018 in the areas from which our sample was drawn, four with male workers and three with female. Each group consisted of between six and eight participants, together adding up to a total of 54 participants. They included helpers, quality controllers, finishers, ironmen, a supervisor and a large number of machine operators. Seven of them were from Accordaffiliated factories, 11 from Alliance-affiliated factories, four from factories affiliated to both, nine from factories described as 'compliant', 14 from 'non-compliant' factories and nine did not know the status of their factories.

Wages and Working Hours
We begin our analysis with the issue of wages, identified in our survey as one of the workers' most important priorities. Wages are also an important priority for employers but in the reverse direction. Given that the competitive advantage of the RMG sector in Bangladesh lies in the low costs of its labour, employers have sought to use their considerable influence with the government to keep wages at very low levels. The monthly minimum wage for unskilled workers in the sector was first set in 1994 at 930 Bangladeshi Takas (US$ 23.25). It remained frozen at that level for over a decade until massive protests by workers in May 2006 forced a rise to 1,662 Takas (US$ 24.80). It was subsequently raised to 3,000 Takas (US$ 43.40) in 2010, to 5,300 Takas (US$ 65) in 2014 and then, in December 2018, to 8,000 Takas (US$ 95). Each raise was preceded by a period of intense agitation by workers, a point we will return to later.
12. This was borne out by a comparison with results using logistic regression analysis for binary outcome indicators. Regression analysis was also done for these outcome indicators by adding various control variables for worker characteristics. The direction and significance of coefficients for A&A affiliation and factory size were largely similar. Note: Standard error in parenthesis. *, **, *** significant at 10%, 5% and 1% level respectively. Job security and benefits index were constructed with 2 indicators each on a scale of 0 to 2. Source: see Table 1. Note: Standard error in parenthesis. *, **, *** significant at 10%, 5% and 1% level respectively. Dependent variable in Column 1 is whether the factory has health and safety committee. WPC effectiveness index is constructed by adding five indictors (viz. whether workers can openly express their views to the WPC; whether the WPC looked out for their complaints; whether they think the management gave importance to the WPCs; whether they believe that the committee was effective; and whether it gave satisfactory solutions) to range between 0 and 5. Source: see Table 1. Note: Standard error in parenthesis. *, **, *** significant at 10%, 5% and 1% level respectively. Source: see Table 1.
Monthly earnings are made up of basic salary, overtime (OT) earnings and any bonuses offered by a factory. Workers in our survey reported a mean basic salary of 7,299 Takas for the previous month, increasing to 8,669 with overtime payments and to 8,993 Takas with attendance bonus (the main form paid) (see Table 3). Average basic salaries in our sample were thus higher than the prevailing minimum wage of 5,300 Takas, even in the small non-A&A affiliated factories. However, only 45 per cent of the workers in our survey believed that their wages were fair. This may reflect the fact that wages generally fell short of various estimates of the 'living wage' which attempt to take account of the effects of inflation on the cost of living. These vary from the Asia Floor Wage region-wide estimate of 36,385 Takas (US$ 454) to the Fair Labour Association estimate for Bangladesh of 7,797 Takas (US$ 97).
The length of the working day has received considerable attention in the literature on export garment manufacturing because, as Locke (2013) points out, long hours of overtime appear to have acquired the status of a norm in this sector. The 2006 Bangladesh Labour Law stipulated that the daily work shift could not exceed 8 hours while the upper limit to overtime was 2 hours. Workers reported an average of 8 hours of work a day, but the average amount of overtime when work pressure was high was 3.3 hours, exceeding the legal limit. In addition, 83 per cent reported that overtime was compulsory. Asked about their working hours, including overtime, 65 per cent of the workers declared they were either satisfied or indifferent while 35 per cent declared themselves dissatisfied. Since the table does not show much difference in overtime hours worked by factory size or A&A affiliation, variations in satisfaction with working hours clearly reflected other factors.
The indicators used to capture different aspects of wages for the regression analysis were basic salary, basic salary with overtime payment and bonuses, and whether workers considered their salaries to be fair. The indicators for demands on workers' time were the number of hours worked, including overtime, whether overtime was compulsory and whether workers were satisfied with their hours of work. The results of the regression analysis support the finding that there was very little difference in basic salaries and working hours by either affiliation status or factory size (see Table 4a). Significant differences in earnings by affiliation status only emerged once overtime payments and bonuses were added to basic salaries, with differences ranging from around 1,000 Takas among workers in larger affiliated factories to around 300 Takas in smaller ones.
The results also suggest that while compulsory overtime was widespread, its likelihood was lower in A&A-affiliated factories, particularly the larger ones. As we noted earlier, incentives such as higher overtime payments and bonuses appeared more common in these factories. Workers in A&Aaffiliated factories were more likely to believe that they earned a fair wage than those in non-A&A factories and more likely to report satisfaction with their working hours. Factory size did not appear to make much difference to these indicators, but the coefficient for the interaction term suggested that workers in smaller factories affiliated with A&A were not very different from smaller non-A&A factories in terms of compulsory overtime or belief in fairness of salaries.
Our FGDs provided further insights into workers' perspectives on wages and working hours. On the question of wages, complaints from participants echoed the wider literature about the effects of inflation. They particularly highlighted the issue of rents, their single largest monthly expense, as a major source of tension. Not only did housing rents rise at a higher rate than the price of other essentials but increments in wages were frequently wiped out by the accompanying hike in rents. As one male worker with Accord complained: 'If our salary is raised by two Takas, our rent is increased by four Takas'. 13 The discussion about working hours helped to disentangle different sources of satisfaction and dissatisfaction. Some workers welcomed the opportunity to do overtime ('within reason') because it increased their earnings, allowing them to improve their living standard, support their families or save for the future. Lower-paid workers, such as helpers and less skilled machinists, had no choice but to do overtime to make ends meet. Several women with young children said they would have preferred to forego additional overtime earnings in order to get home earlier.
Paradoxically, some workers identified reductions in working hours as a major source of dissatisfaction. Reductions occurred as a result of stricter enforcement of legal limits on working hours since Rana Plaza, but they had been accompanied in many factories by an increase in the hourly production targets that workers were expected to meet. According to one female worker with Accord, 'We now have to complete 10 to 12 hours of work within 8 hours . . . whereas previously we may have produced 120 pieces per hour, now we have to produce 150 to 200 pieces per hour'. 14 Pressure to work harder also increased with rising wages, with managers demanding higher productivity to compensate for higher wages. Complaints about the unfairness of wages thus reflected not only their inadequacy relative to rising costs of living, but also their failure to reflect increasing work intensity.

Job Security and Social Benefits
The 2006 Bangladesh Labour Law made it compulsory for employers to issue written appointment letters and identity cards to all workers and upgrade workers to permanent status after they completed the agreed probation period of a year. The law also entitled workers to one day's holiday in the week; 13. Focus group discussion, Savar, 3 August 2018. 14. Focus group discussion, Narayanganj, 21 September 2018. 10 days of paid holiday in the year, 11 days of festival leave, 16 weeks of paid maternity leave and up to 14 days of paid sick leave. These provisions have enormous potential to contribute to workers' job security, occupational mobility and social welfare. Appointment letters serve as a de facto contract, specifying terms and conditions and enabling workers to prove their status as employees with the full rights of employees. Identification cards provide proof of employment, should workers seek to enforce their rights in the labour courts. They also record the expertise and experience of workers and hence allow them to bargain for appropriate wage levels when they move to new factories. Permanent status entitles workers to full legal rights, including written notice of dismissal or wages in lieu and entitlements to holidays and paid leave allow workers to rest, recuperate and spend time with their families.
As detailed in Table 3, over 95 per cent of workers had permanent status, 95 per cent had identity cards, 62 per cent had written letters of appointment, 91 per cent had paid leave and 90 per cent had maternity leave. According to our FGDs, these rules had existed previously, but many workers had either not known of their existence or had not known how to take advantage of them. Reports about the actual observance of these rights varied between the better factories where holidays were observed, child care support and medical facilities provided and requests for leave respected, and those where procedures to claim rights were deliberately made complicated, where leave was often denied and maternity benefits withheld. Table 4a reports on the results of regression analysis on selected indicators of job security and paid leave. These included: an index of job security based on whether the worker received written letters of appointment and possessed an identity card (its value ranged from 0 to 2); and a social benefits index based on whether they reported paid leave and maternity leave (its value ranged from 0 to 2). In addition, in order to capture whether these formal provisions gave workers a greater sense of job security, we asked them if they believed that they could be fired at any time. Around 37 per cent of the overall sample said that they believed this to be the case.
The regression results suggest that A&A affiliation was positively associated with both the indexes and that it reduced the likelihood that workers believed that they could be fired at any time. Smaller factories reported lower values for both indexes, but factory size did not appear to affect perceived job security. The results for the interaction terms suggest the positive association between A&A affiliation and the job security and benefits indexes varied by firm size, with weaker 'effects' in smaller factories.

Health and Safety
Problems of health and safety in factory conditions have a long history within the RMG industry. The UK Parliamentary Group's report on the Rana Plaza disaster drew attention to the succession of audits over the years that routinely reported on these problems: 'In lesser infringements, fire extinguishers may be out of date or exit routes may be blocked. More serious misdemeanours may see gates between floors locked, significant overcrowding on the factory floor, windows barred, and electrical equipment exposed' (All Party Parliamentary Group on Bangladesh, 2013: 23). But despite a series of deadly accidents in the RMG sector prior to Rana Plaza, both buyers and suppliers turned a blind eye to hazardous factory conditions.
The situation changed dramatically after Rana Plaza. The EU Sustainability Compact prioritized government action on this front. The 2013 Labour Act 15 included better regulation of stairway and gangways, provision of safety equipment, regular fire drills and strengthened factory inspection. It also required the setting up of Health and Safety Committees (HSCs) in all factories with 50 or more workers in order to coordinate efforts around health and safety within the factory. Worker representatives on the committees were either to be drawn from trade union members in the factory or, in the absence of a trade union, appointed by members of the Worker Participation Committees (WPCs; see below). In addition, of course, health and safety featured centrally in the Accord and Alliance agreements.
Our survey suggests that these provisions had been implemented in most factories: 90 per cent of workers in our survey reported an HSC in their factory and 85 per cent had received health and safety training. We also asked workers a number of questions to elicit their subjective perceptions of safety. The overwhelming majority of workers (95 per cent) said that they felt safe working in their present factory while 90 per cent said they could refuse to go into a building if they thought it was unsafe.
The FGDs strongly supported this overall picture of improvement. Almost all the participants had undergone health and safety training, often multiple times, and spoke knowledgeably about the changes they observed: regular inspection by engineers, the closing down of unsafe factories and the renovation of others, the setting up of the HSCs, the installation of fire extinguishers on every floor, the widening of exit staircases, regular fire drills, first aid boxes, the need for masks while working and the importance of keeping walkways and aisles clear of obstructions. Several workers confirmed that they felt sufficiently empowered to refuse to work in an unsafe environment: 'If we see a crack on the wall or are concerned with our safety, now we can certainly speak up' said one male worker (with unknown factory status) in an interview. 16 'Everyone has this right. We have been watching various videos, learning about our rights and safety through different organizations like Accord and Alliance' commented another male worker from an Accord-affiliated factory. 17 The results of the regression analysis in Table 4b suggest these improvements in health and safety conditions went further in A&A-affiliated factories: regardless of factory size, their workers were significantly more likely to report HSCs in their factories, to have undergone training in health and safety and to believe that their factories were safe. However, there was no significant difference in the likelihood of workers saying they would not enter an unsafe building, suggesting that this particular change was widespread across factories.

Awareness, Representation and Voice
Although policy responses to Rana Plaza gave central place to health and safety provisions, there was also renewed attention to questions of workers' awareness, voice and representation, reflecting widespread consensus on the view put forward by Human Rights Watch (2015: 4): 'If workers at Rana Plaza had more of a voice, it is entirely possible that the circumstances that led to the thousands of deaths and injuries could have been prevented'. This was also one of the reasons why, as Donaghey and Reinecke (2018) point out, the Accord agreement stressed the importance of building workers' voice as part of its safety processes.
The 2013 Labour Act relaxed restrictions on trade union registration and strengthened other mechanisms for worker representation. The ILO estimated that 527 new trade unions had registered in the garment sector since 2013 so that there were 611 trade unions and 48 union federations by 2018. 18 Worth noting is the observation by a local newspaper that the newly registered unions were 'overwhelmingly affiliated' with Solidarity Center, a labour NGO allied to US's largest trade union federation (Siddiqi, 2017: 68). Animosity towards left-wing unions, on the other hand, remained strong on the part of factory owners and their efforts to register continued to be blocked by the BGMEA, often with the backing of the government (Hossain and Hossain, forthcoming;Siddiqi, 2017).
The 2013 Labour Act also sought to strengthen Workers' Participation Committees. The 2006 legislation stipulated that WPCs made up of representatives of managers and workers had to be set up in any factory with 50 or more workers in order to provide a platform for social dialogue, particularly important in factories where there were no trade unions. While half of the committees should have been nominated by union leaders or elected by workers, in reality, their membership was mainly nominated by 17 Table 3). In terms of knowledge, it is indicative of the relative significance attached to various forms of governance within factories that 71 per cent of the workers knew about codes of conduct but only 40 per cent knew about the country's labour laws. Most learnt about the codes from posters and information provided by management, practices that clearly did not extend to the country's labour laws. It is also indicative of current priorities that while a vast majority of workers (85 per cent) had undergone training in health and safety, only 6 per cent of them had received training on rights, negotiations and collective bargaining.
To explore the issue of voice and representation, the survey included questions about the role of trade unions and WPCs. Only 4 per cent of the workers said that there was a trade union in their factory, 35 per cent said that there was no trade union in their factory and, tellingly, 61 per cent said they did not know whether there was one or not. None of the workers was associated with a trade union outside the factory. By contrast, 78 per cent of workers said that there was a WPC in their factory and, significantly, 80 per cent of these said that members of the WPC were elected. Since elections to WPCs were only made mandatory in 2014 after Rana Plaza, we can assume this is a relatively recent change.
Asked about the role of the WPCs, 83 per cent of workers agreed that they could openly express their views to the WPC; 77 per cent said that the WPC looked out for their complaints; 64 per cent believed that management gave the WPCs importance; 66 per cent believed that the committees were effective, while similar percentages believed that they gave satisfactory solutions (Kabeer et al., 2019). We used these responses to develop a WPC effectiveness index with a scale of 0 to 5. Workers in A&A factories systematically reported higher values than those in non-affiliated factories, regardless of factory size: their overall index was 3.8 compared to 0.8 in non-affiliated factories.
The results of the regression analysis suggest that affiliation with A&A was positively associated with knowledge of codes of conduct and labour laws, with the likelihood of a WPC in the factory, with the likelihood of elected WPCs and with the WPC effectiveness index (see Table 4b). These indicators did not vary by factory size but the interaction term between factory size and A&A affiliation suggested that among A&A-affiliated factories, smaller factories were less likely to have a WPC than larger ones and were less likely to have effective WPCs.
The FGDs touched on the different channels through which workers could express their grievances: supervisors, line managers, 'compliance' madams, WPCs, complaint boxes and Alliance telephone help lines. In line with the survey results, most FGD participants considered the WPCs more effective than other mechanisms but they varied considerably in their assessments of WPCs' achievements. The more positive among them valued the fact that they could take complaints to WPC members who were often able to resolve them through discussions with the appropriate level of management. However, it appeared from the examples they gave that their complaints generally related to a limited, and often mundane, range of issues: fresh soap for the bathroom; fixing bathroom taps; replacing the water filter; supporting requests for leave; arbitrating disputes between workers and remonstrating with abusive supervisors.
The less positive argued that these were the only issues that WPCs were in a position to deal with. They pointed out that WPC members were workers like themselves; they were not given time off to carry out their duties and faced the same pressures to complete their production targets. Nor could they risk their jobs by raising controversial issues. Some FGD participants went further to express the view that once a worker was elected to the WPC, they became a tool of management. In the view of a male worker from an Accord-affiliated factory, the WPCs were more effective if their members had been backed by Accord or Alliance: 'Just like the government, it is not enough for the worker representatives to be elected, they need to be given control and power to be more effective'. 19 On the question of trade unions, the FGDs confirmed that there was negligible trade union presence within the factories. Many added that anyone found out to be a union member was likely to be sacked. The discussion therefore focused on trade union federations and their role in promoting workers' rights. Here knowledge and opinions varied. First, there were several workers, mainly women, who were not very clear about the role of these federations. Second, there were those, most often from A&A-affiliated factories, who said that they did not feel the need to approach federations as their problems were generally solved within the factory.
The third focus group had stronger and largely positive views about trade unions. The question of unfair dismissal was singled out as an issue which trade unions were better placed to handle than WPCs. Once workers were dismissed, they had no access to WPC members since they were barred from crossing the factory gates. Trade unions, on the other hand, had offices outside the factories and provided a safe space for workers' complaints.
However, it also became clear that FGD participants were not always talking about the same kinds of federations. Some were talking about local affiliates of the global federations who signed Accord and whose power derived from this association: 'These unions have the telephone number of every single buyer and employer. If there is a problem and the federation inform the buyers, that factory will not get any orders' said one male worker (unknown factory status). 20 Members of these unions clearly worked closely 19. Focus group discussion, male worker from an Accord-affiliated factory, Gazipur, 14 September 2018. 20. Focus group discussion, Savar, 7 September 2018. with Accord, they were able to participate in the inspection processes, to take up workers' grievances directly with managers, and to threaten legal action, often invoking the authority of the Accord to press their case (see also Zajak, 2017). This strand of unionism was thus associated with social dialogue, negotiation and legal action.
Other FGD participants were referring to the local union federations, often unregistered, which generally supported workers in the 'normal' (in other words, non-compliant) factories. These were described as older federations, whose leaders had been around from 'much earlier'. The workers in these factories were less likely to have WPCs, or effective WPCs, they had little scope for solving their problems within the factories and were more likely to take their complaints to outside 'worker leaders'. The unions in question did not always have the resources to fight legal cases. Their power lay in their ability to mobilize street-level protests. As a male worker from a non-affiliated factory told us: 'Worker leaders organized and protested to increase the wage to 5,300 Takas. I hear they are planning to do this again soon. It would be good for us'. 21 A female worker from a non-affiliated factory, who had participated in such a protest in the past, recalled: 'The workers were organized by outsiders who are worker leaders. They threw bricks at the windows of our factories and demanded that all of us join them to show solidarity. They said, "why shouldn't you join us, we are all workers"'. 22

Dignity at Work
Our focus so far has been on aspects of wages and working conditions that are either mandated by the country's labour laws and/or covered by corporate codes of conduct. However, some aspects of working conditions, specifically relationships and behaviour within factories, are difficult to regulate and hence do not appear in the codification of labour standards at national, corporate or international levels. Yet, as studies have suggested, these aspects have long been a problem within the RMG industry (Mahmud and Kabeer, 2006;Siddiqi, 2003). This was borne out by our survey, in which 75 per cent of workers stated that they had experienced mistreatment in their current factory, with somewhat higher percentages reporting it in non-A&A factories (see Table 3). Verbal abuse featured most frequently: 89 per cent of the workers observed it in relation to others, 81 per cent maintained that it occurred frequently and 87 per cent reported that they had experienced it themselves. Other forms of 21. Focus group discussion, Narayanganj, 21 September 2018. In fact, thousands of workers did go out on strike in January 2019 to protest for higher wages. 22. Focus group discussion, Mirpur, 10 August 2018. mistreatment observed by workers included humiliating forms of behaviour (68 per cent), use of threats (46 per cent), regulated toilet breaks (24 per cent) and physical abuse (24 per cent). Supervisors were singled out by 80 per cent of the workers as most verbally abusive while 9 per cent singled them out as most physically abusive. The main reasons given for mistreatment were making mistakes (77 per cent), failure to meet production targets (61 per cent), working too slowly (33 per cent) and absence from work (22 per cent).
Among the FGDs, only a minority of participants, mainly from the larger, A&A-affiliated factories, believed that behaviour on the factory floor had improved over time. They explained this in terms of new mechanisms for registering their complaints, greater awareness on the part of workers and greater willingness to speak up, as well as to the pressures exerted on their behalf by external actors, including buyers and unions. The rest testified to widespread abuse, linking it with employers' overriding preoccupation with meeting production targets. Some believed that levels of abuse had increased since Rana Plaza as a result of the intensification of workloads noted earlier: 'No one can be absent or late for work, no one can pause in their work, no one can leave their work station at any time . . . these regulations did not exist before, they have become stricter now. If we cannot meet the production target, then there is no end to their scolding'. 23 While most FGD participants focused their anger on supervisors as the immediate perpetrators of abuse, some offered more systemic explanations of abusive behaviour. As a female worker from an A&A factory pointed out, supervisors had their own production targets: 'There is no point in always complaining about the supervisor . . . . The in-charge is above the supervisor, he will beat up the supervisor if he is unable to meet his targets'. 24 A male worker from an A&A factory told us how he came to realize that abusive behaviour was built into the logic of factory life. He had once fallen behind on his work and been summoned along with his newly recruited quality manager to the general manager's office. While he fully expected the tirade of abuse he received, he was shocked by the level of abuse directed at his new manager: This is how the verbal abuse system works . . . above this man is the in-charge, above him is the production manager, then the assistant general manager and then the general manager. The abuse starts at the top and gets passed down to each lower level . . . And I know if my job was at risk, I would also put pressure on those below me. 25 23. Focus group discussion, male worker from a non-compliant factory, Narayanganj, 21 September 2018. 24. Focus group discussion, female worker from an A&A factory, Narayanganj, 21 September 2018. 25. Focus group discussion, male worker from an A&A factory, Gazipur, 14 September 2018.

Workers' Perspectives
Our analysis so far has found that, independent of factory size, A&A affiliation was significantly and positively associated with many, though not all, of our measures of wages and working conditions. However, the crosssectional nature of our data means that we cannot attribute these differences to efforts associated with the two agreements since it is equally possible -and indeed very plausible -that their global signatories were already working with factories with better working conditions before Rana Plaza. We therefore turn to a series of questions in our survey which sought workers' own views about change to help us with the issue of causality. These asked workers if they had heard of Rana Plaza. In fact, all the workers in the survey had heard of Rana Plaza; indeed 10 per cent knew of someone who had died or been injured. They were then asked whether and what changes they had observed taking place since Rana Plaza. Finally, they were asked why they thought these changes had taken place.
The questions about perceived changes in working conditions were asked in relation to the issues discussed earlier. Responses could vary between 'improvements', 'deterioration' or 'no change' (see Table 3). Improvements were most frequently mentioned in relation to healthier working environment (87 per cent); building safety (85 per cent); and improved safety training (83 per cent). The frequency with which improvements were reported for other issues was lower and more varied: the opportunity to complain (68 per cent); timely payment of salary (63 per cent); salary and benefits (59 per cent); behaviour of supervisors (57 per cent); behaviour of managers (56 per cent); job security (47 per cent); opportunities for training and promotion (26 per cent); overtime hours (18.7 per cent); and sexual harassment (13 per cent). A deterioration in conditions was most likely to be reported in relation to overtime (40 per cent). For all other conditions, around 10 per cent or less reported deterioration while higher percentages did not report any change.
In order to explore whether we can attribute these reported changes at least partly to the influence of A&A, we carried out regression analysis to estimate the strength of the association between the changes in question and A&A affiliation while controlling for other possible direct or mediating influences. Given its importance in predicting variations in factory conditions, and the greater likelihood of A&A affiliation in larger factories, we included factory size as one of these influences. While visits by inspectors were associated with A&A remediation processes, the indicator we chose was whether workers had spoken with inspectors as evidence of pro-active engagement. We also hypothesized that more effective WPCs (as measured by a WPC effectiveness index of 4 or 5) were more likely to be associated with improvements in working conditions. Finally, we expected that workers who joined their current factory before Rana Plaza were more likely to be aware of changes that took place subsequently. The results of our regression analysis (see Table 5) can be summarized as follows: Workers in factories with A&A affiliation were more likely than those in non-affiliated factories to report improvements in building safety, healthy environment, safety training, opportunities for complaint, opportunities for training, timeliness of payments and sexual harassment. They were significantly less likely to report improvements in behaviour of supervisors (a reflection perhaps of the issues raised in the FGDs), while differences in terms of salary and other benefits, overtime hours, job security and opportunities for promotion were not significant.
Factory size on its own did not appear to be significantly associated with indicators of improvement except that smaller factories were less likely to report improvements in building safety than larger ones and (unexpectedly) more likely to report improvements in management behaviour. The coefficients for the interaction between factory size and A&A affiliation suggest that the positive impact of A&A affiliation on improvements in safety training, timely payments and opportunities for training were weaker in smaller factories. They also suggest that among small factories, A&Aaffiliated factories were less likely to report improvements in the behaviour of management, salary and benefits and opportunities for promotion than non-affiliated ones. Direct interaction with inspectors by workers in the past year was positive, associated with improvements in health and safety conditions, management behaviour, timely payments, job security and sexual harassment.
The results for the WPC effectiveness index were particularly striking. Workers who reported more effective WPCs reported positive changes in all our indicators of change. In other words, the effectiveness attributed to these WPCs appeared to be borne out by their ability to bring about important changes within their factories. Lastly, workers who had joined their factory before Rana Plaza were more likely to report improvements in salary and benefits, health and safety conditions, opportunities for complaint, opportunities for training and promotion and timeliness of payments. Asked what they believed to be the most important reasons for the changes reported, 88 per cent of the workers believed it was pressure from buyers, while an additional 31 per cent explicitly identified the Accord or Alliance (see Table 6). Given that many workers had not heard of the Accord or the Alliance, it is possible that responses identifying 'buyer pressure' included pressure exerted by the Accord and Alliance. Buyers were thus perceived as the primary guardians of workers' welfare and rights. Other drivers of change mentioned by workers included: government policy (43 per cent); owners' commitments (25 per cent); BGMEA (9 per cent); the media (7 per cent); and pressure from trade unions (0.4 per cent). These results suggest a number of important points. They support the hypothesis that efforts associated with A&A affiliation were either directly responsible for improvements in various aspects of working conditions or else they created a hospitable environment for improvements, such as those mandated by the amended labour laws. This hypothesis is further supported by the significance of the association between the likelihood of improvements and pro-active inspection processes which are a major feature of the A&A initiatives. It is also indirectly supported by the finding that improvements in working conditions were more likely to be reported by workers who had joined their current factory before Rana Plaza and were thus in a position to observe subsequent changes.
The role of the WPCs is interesting. While workers from A&A-affiliated factories were generally more likely to report effective WPCs, it was those who reported the WPC effectiveness index of 4 or 5 who were most likely to report improvements in all our indicators of change, an effect independent of A&A affiliation. These workers came from just 40 of the 240 factories in our sample. So, while the WPCs may indeed represent a nascent form of workplace democracy of the kind that Donaghey and Reinecke (2018) associate with the Accord, the greater effectiveness of some WPCs relative to others within Accord-affiliated factories suggest other factors mattered as well, for example, variations in quality of factory management.
A final important point to make relates to aspects of working conditions where improvements appeared slow or non-existent, the 'sticking points' in the processes of change associated with the Accord and Alliance. The first relates to basic salaries. While more than half the workers in our survey believed that there had been an improvement in salaries and benefits since Rana Plaza, this improvement did not appear to be associated with A&A affiliation (Table 5) nor was there any significant difference in basic salaries by factory affiliation, once factory size had been controlled for (Table 4a).
The second relates to working hours. Overtime was compulsory in most factories, affiliation status made no difference to the length of the working day and a high percentage of workers were dissatisfied with their hours of work (Table 3). While many more workers in the survey reported a deterioration in their working hours since Rana Plaza, regardless of factory affiliation status (Table 3), our FGDs suggested that this might reflect the increase in work intensity that came with stricter controls over overtime hours rather than an increase in working hours. A third area of continuity relates to the mistreatment of workers. Around half of the workers reported an improvement in such behaviour since Rana Plaza, but again this did not appear to be associated with A&A affiliation (Table 5). However, it is worth noting that workers in A&A-affiliated factories were somewhat less likely to report such mistreatment (Table 4b). The final area relates to mechanisms for workers to voice their concerns. We noted evidence of change in terms of the spread of elected WPCs, but we also noted the continued absence of trade unions, regardless of factory affiliation.

Employers' Perspectives
We set out in this article to explore whether new collaborative approaches to CSR exemplified by the A&A in Bangladesh had overcome some of the limitations of previous compliance-based approaches. Our analysis of the garment workers' survey would lead us to offer a qualified yes to this question: important aspects of wages and working conditions, particularly conditions relating to structural safety, were considerably better in A&Aaffiliated factories; workers believed that there had been improvements in many of these conditions since Rana Plaza; and most of them attributed these improvements directly to A&A or indirectly to buyer pressure. However, it became clear from the findings of interviews carried out with managers of A&A-affiliated factories as part of the larger project (Rahman and Rahman, this issue) that the perspectives of the workers in our survey were circumscribed by their position at the bottom of the value chain. They had limited knowledge of power dynamics higher up the global value chain. The managers, on the other hand, were the stakeholders who interacted most directly with the buyers and were directly responsible for implementing the two initiatives. Their initial response to A&A had been favourable: they believed it offered a coordinated framework for improving health and safety conditions within the industry and the opportunity to improve the country's tarnished reputation.
This early enthusiasm appeared to have given way over time to greater ambivalence. This partly reflected technical problems associated with A&A procedures -duplication of inspection efforts, inconsistency in findings, miscommunications, unrealistic timetables and so on. But their greatest bitterness was reserved for the fact that while buyers paid for the costs of inspection, they failed to share the greater financial burden associated with the remediation and relocation measures recommended by inspectors (see also Barrett et al., 2018). This went against the expectations created by Article 22 of the Accord: 'participating brands and retailers will negotiate commercial terms with their suppliers which ensure that it is financially feasible for the factories to maintain safe workplaces and comply with upgrade and remediation requirements instituted by the Safety Inspector'. 26 On the contrary, the purchasing practices of lead buyers were antithetical to the financial feasibility of local suppliers. Not only did they fail to increase the procurement prices paid to their suppliers, which would have been one way to offset some of the remediation/relocation costs incurred by the latter, but the prices they paid continued their steady long-term decline. According to Anner (2018), between 2011 and 2016, prices declined by 10.7 per cent for items exported to the US and by 9.04 per cent for items exported to Europe. As he points out (ibid.: 7), these declines had little to do with changes in the price of raw materials or fluctuations in exchange rates.
Anner (ibid.: 15) also notes a steady reduction in delivery times. In 2011, the major global brands gave factories in Bangladesh an average of 94 days to complete an order; by 2016, this had dropped to 86 days, a reduction of 8.14 per cent. This translated into the intensified pressure to meet production targets which workers attributed to management. With wages going up while purchasing prices went down, it is not surprising that profit margins declined among Bangladesh suppliers -by 13.3 per cent between 2011 and 2016 -leaving them with a mean profit margin of 7.69 per cent (ibid.: 7). Rahman and Rahman (this issue) report that 70 per cent of the managers they interviewed reported profit margins of between 0 and 5 per cent.
It would appear therefore that while the shift from compliance to cooperation has been accompanied by a number of important improvements, it has left long-standing asymmetries in power relationships within global value chains largely intact. 27 The persistence of these asymmetries of power goes some way towards explaining the 'stickiness' in employer practices revealed by our survey: the resistance to paying a basic living wage; reliance on compulsory overtime and increasing intensification of workloads; mistreatment of workers who fail to meet production targets; and active hostility towards trade unions who might seek to challenge these practices.

Revisiting Stakeholder Theory: The Empirical Approach
While the persistence of exploitative practices on the factory floor may be worse in Bangladesh than elsewhere, they are not unique to it. As Locke (2013: 126) points out, poor working conditions, excessive working hours, precarious employment practices and low wages appear endemic to production for global supply chains. This means we cannot understand the persistence and pervasiveness of these practices by focusing on individual factories or countries. We need to move from a narrow 'spotlight' perspective which confines our gaze to the locus of production to a 'floodlight' approach which illuminates the broader political economy of supply chain capitalism within which these production processes are located.
This broader perspective would draw attention to the growing concentration within the retail sectors of the US and Europe which has endowed a small number of global brands and retailers with disproportionate bargaining power vis-à-vis the large and growing numbers of suppliers dispersed in low-wage economies across the developing world. It would also reveal how the rise of the new 'fast fashion' business model has fuelled the demand by consumers in affluent countries for a greater variety of clothing every year at ever-cheaper prices, proving particularly inimical to the promotion of decent working conditions in supplier factories (Oxfam, 2004;Taplin, 2014).
Empirical stakeholder analysis, as opposed to normative or instrumental stakeholder theory (Egels-Zandén and Sandberg, 2010), suggests the reverse of the win-win model cited earlier. The major brands and buyers have been able to use the unequal distribution of bargaining power within global chains to compel their suppliers to meet the competitive pressures generated by the fast fashion model, producing increasing varieties of products within shorter lead times and at decreasing prices. While they may insist on supplier compliance with their codes of conduct to avoid reputational risk, they do not necessarily reward compliant factories by improving the terms they offer. 28 Faced with these practices, suppliers must reduce their profit margins, pay their workers lower wages, demand longer hours of work, hire contract or migrant labour and take shortcuts on safety standards. Most of these responses are evident in Bangladesh.
Other stakeholders also have restricted 'room for manoeuvre'. Governments may have the institutional capacity and political will to enforce their own laws but are often inhibited from doing so by their fear that it will increase the costs of doing business, drive away global buyers and lead to a 28. See: https://apparelinsider.com/survey-raises-questions-purchasing-practices/ (accessed 7 December 2017). massive loss of jobs. 29 Workers who bear the brunt of these practices have little choice but to put up with them since they have neither the organizational voice to challenge their employers nor the option of seeking more decent work opportunities.
Consumers have been only intermittently visible in stakeholder analysis. Nor is it clear what kind of stake they have in improving the working conditions of those who produce their clothes. Some authors believe that consumers have demonstrated their willingness to pay higher prices for clothes made in decent working conditions (Ross, 1997). Others are less sanguine. As Taplin (2014: 73) observes, 'Western consumers have become accustomed to cheap fashion and for the most part appear unwilling to pay more for items that are untainted by exploitative practices'. In place of a benign version of stakeholder analysis, therefore, Taplin offers a more pessimistic reading, arguing that the current practices in global value chains have created a situation 'where the "villains" are many and the innocent caught up in the manifold uncertainties that such a model produces' (ibid.). The question then is what, if anything, can be done to protect the 'innocent' and give them a greater voice in shaping the CSR agenda.

Is 'Shared Stakeholder Responsibility' Possible?
In his review of various initiatives undertaken to improve labour standards within global value chains, Locke (2013: 2) concludes that improvements cannot be left to any single set of stakeholders or any specific set of approaches. They require new institutional arrangements that would reallocate costs and rewards among all stakeholders engaged in these value chains, transcending traditional boundaries between producers and consumers, buyers and suppliers, NGOs and corporations and advanced and developing countries. Given the power dynamics that characterize supply chain capitalism, it is difficult to envisage what these arrangements might be and how they might materialize, but his conclusion does shift the search for solutions away from a 'spotlight' perspective to a more holistic 'floodlit' approach. By way of conclusion, we assess recent recommendations for advancing the transition to a fairer set of institutional arrangements, drawing on Sobhan (2014) to highlight the specific nature of the challenges that characterize the Bangladesh context.
The Stern Center has outlined what it terms a 'shared responsibility model' to address remaining health and safety problems in the RMG industry in Bangladesh after the Accord and Alliance (Barrett et al., 2018). It recommends an international task force, led by Bangladeshi stakeholders, who will coordinate efforts to raise the funds necessary to complete this task and oversee the implementation process. Along with the Bangladeshi government and RMG employers, they envisage financial contributions from international actors, including Western buyers as well as governments in countries that import garments from Bangladesh -the EU and US together account for 80 per cent of the country's exports.
While there are self-evident reasons why the Bangladesh government and RMG employers should contribute to improving conditions in the sector, the case made by the Stern proposal for involving Western buyers and governments is based on appeals to a sense of fairness and 'special obligations'. Yet the commitments frequently expressed by these actors to economic justice for workers in global value chains surely demands a much stronger and more institutionalized model of shared responsibility, one based on the fairer distribution of rights as well as responsibilities across the value chain rather than on special pleading on behalf of its weakest stakeholders?
Most global buyers, as we have seen, have not shown much inclination in this direction despite the normative, instrumental and developmental arguments in its favour, but their governments are in a position to make a difference. These governments currently siphon off a substantial share of the revenue generated by value chain production, in the form of 15 per cent tariffs imposed on imports of garments into the US 30 and 19 per cent value added tax levied on all goods and services sold in the EU (Bain, 2018;Norfield, 2011). The concerns these governments have expressed for the rights of the global workforce that makes clothing for their citizens would be more credible if they were backed by commensurate obligations on their part. These could be met by routinely redistributing some of the revenue they collect from imports from Bangladesh and other lower-income exporting countries back to the countries in question both to ensure that their industries have the resources to provide decent working conditions and to promote mechanisms within civil society to hold the industry accountable.
The question of national regulation is clearly problematic in Bangladesh. The failures of government oversight that contributed to the conditions prevailing in the RMG sector are part of a broader crisis of governance failure within the country. While there is no doubt that the government needs to take much greater responsibility for the rights of its workers, as it must for the rights of all its citizens, the present state of governance in Bangladesh suggests that this is unlikely to happen in the foreseeable future.
What the government may be willing to strive for, given its stake in protecting its highest export-earning industry, is to build the RMG sector into what Sobhan (2014: n.p.) terms 'an island of good governance . . . in an ocean of mal-governance'. Such an approach will lead inevitably to the further bifurcation of an already bifurcated industry (Barrett et al., 2018). It will give rise to a small island of large, regularly-inspected 'good' factories, with internationally acceptable labour standards, exporting to Western countries, surrounded by an ocean of 'bad' and 'ugly' factories with poor working conditions and minimum oversight (similar in other words to the informal working conditions that prevail in the rest of the country) which export to countries like Turkey and Russia which have few concerns about workers' rights. Garment employers clearly have a stake in the survival of their industry. Whether a BGMEA whose past leadership has been dominated by those most resistant to regulation will be able to take a more enlightened view in the future remains to be seen. But garment employers are not a homogeneous group. Many have improved conditions in their factories independently of international pressure. They can either push the BGMEA in the direction of achievable labour standards as the industry's contribution to a shared responsibility model or set up their own associations in order to provide an example of enlightened leadership for the rest of the industry.
Other recommendations within the CSR literature relate to initiatives by trade unions, labour rights NGOs, the ILO and others to build skills and organizational capacity within garment factories to improve working conditions and comply with national regulations. As Locke suggests, each of these may have limited impact on their own but they can mutually reinforce each other to amplify impact. Of particular importance from the Bangladesh perspective are those initiatives that seek to build the voices of workers in the industry -the stakeholders with most to gain from improvements in labour standards and workers' rights. Labour movement scholars and activists have focused primarily on trade unions as the key institutional actors to bring about change. While the troubled history of trade unionism in Bangladesh has meant that progress has been exceedingly slow, it has not been absent.
The recent scholarship on this issue has been motivated by the promise of industrial democracy held out by the Accord. Zajak (2017Zajak ( : 1015, for instance, has argued, that the presence of the Accord provides a 'shadow of protection' to efforts by local unions affiliated to its signatories to address workers' grievances while the interactions between these unions and international organizations and alliances have increased their strategizing skills. These skills and experiences are unlikely to disappear when the agreements come to an end. But nor did the struggle for workers' rights come into existence with the inception of these agreements. As longer-standing scholarship on the industry points out, localized protest movements have been occurring on a sporadic basis for over two decades but they have been mainly led by unregistered factory-based unions and leftwing federations outside the spotlight of global value chain research (Khan, 2001;Mahmud and Kabeer, 2006;Rahman and Langford, 2012;Siddiqi, 2017). Such actions testify to the growth of a 'practical' as opposed to a 'discursive' class consciousness among workers (Rahman and Langford, 2012), one that was not in evidence in research from the 1980s and early 1990s but has clearly evolved over time. It is this practical consciousness that was at play in 2006 when two small independent leftist unions were able to mobilize militant mass action for higher minimum wages on a scale never previously seen in the history of the industry. It is also this practical consciousness that has fuelled subsequent 'wildcat', often violent, agitations that have led to periodic increases in the minimum wage. As a number of authors point out, this form of 'collective bargaining by riot' 31 or 'collective labour militancy recast as senseless vandalism' (Ashraf and Prentice, 2019: 104) can be seen as an understandable response to the persistent repression of more legitimate forms of collective bargaining. It has proved remarkably effective at least in achieving wage rises.
This older scholarship has also pointed out that it was not just trade unions on their own that contributed to the growth of worker consciousness. Different sections of civil society, including human rights activists, feminist groups, legal organizations and progressive policy think-tanks, who are active in promoting democratic process and the rule of law within the country, have also been active in supporting the struggle for labour standards and workers' rights in the garment sector. They do not view these struggles as distinct from each other (Ashraf and Prentice, 2019;Hossain and Hossain, forthcoming;Mahmud and Kabeer, 2006;Rahman and Langford, 2012;Siddiqi, 2017). They recognize that success in promoting workers' rights in the most significant industry in the country may help catalyse change in other sectors of the economy, a version of the 'lighthouse effect'. 32 At the same time, they are also aware that gains in relation to workers' rights in the garment sector are unlikely to be sustained in the absence of citizens' rights in the wider society. For these organizations, the process of building voice and representation among workers in garment sector has to be carried out as part and parcel of efforts to build a broader culture of rights within the country -otherwise CSR initiatives, however well-intentioned, will continue to remain mired in the short-term instrumentalism of its more powerful stakeholders.