Does Corporate Social Responsibility Make a Difference on Labor Conditions? a Case Study in Southern China

Does Corporate Social Responsibility make a difference on labor conditions? A Case Study in Southern China Yiu Por Chen Assistant Professor Public Services Graduate Program DePaul University & IZA e-mail: [email  protected] edu Phone: 312/362-8441 Fax: 312/362-5506 1 Abstract The corporate social responsibility code of conduct (CSRC) has been the model of corporate governance (CG) used by multinational corporations (MNCs) for their suppliers.Using the transaction cost perspective, this study argues the CSRC system may reduce some transaction costs for MNCs in planning and implementing corporate social responsibility (CSR). However, the CSRC may suffer from opportunism at the supplier level, which may undermine CSRC outcomes. By utilizing a unidimensional item response model with a randomized survey of factory workers of 12 toy manufacturers in Southern China, this study provides a unified measure to evaluate the effects of CSRCs on labor conditions from workers’ voices.Furth ermore, this study offers fresh empirical evidence to show the ways opportunism may undermine CSRC effects on labor relations. JEL code: M14, M23, J28, J50, J81, J83, J88 Keywords: Corporate Social Responsibility, Multinational Corporations, Outsourcing, Safety, LaborManagement Relations, Working Conditions, Workers’ Rights, Labor Market Policy 2 â€Å"There are lots of firms and lots of changes in labor practices among them, so there is no lack of data.To judge whether formal labor market rules produce worse employment outcomes, as claimed by orthodox analysts, one could contrast employment between firms with more or less rigid internal rules. † ? Richard Freeman (2005: 19-20), suggesting that a microanalysis of workers and firms will be a path forward to understanding the effect of labor institutions. 1. Introduction Corporate social responsibility (CSR) has been rapidly institutionalized and now covers the vast majority of industrial sectors around the globe.Donalds on (1996) highlights that 90% of all Fortune 500 companies have established codes of conduct, which are by far the most common mode of CSR used by corporations today. Furthermore, 51% of German firms claim to have codes of conduct, compared to 41% in the UK, and 30% in France (Schneider and Barsoux, 1997). Brytting (1997) also found that 52% of the larger Swedish companies had codes of conduct. In fact, CSR is largely the response of multinational corporations (MNCs) to the accusation of being the major actors in â€Å"new exploitations† of less developed countries (LDCs) (Chan, 2001). For the outsourcing MNCs, one of the important aspects of CSR is to ensure that workplaces and labor conditions in poor nations are not â€Å"too† terrible (even if it is worse than in the workplaces of developed countries). Theoretically speaking, the original approach to CSR suggested in the literature is a stakeholder approach. The stakeholder approach advocates that firms should not only maximize profit but also behave as good citizens of the community and take account of the needs of other stakeholders that may be affected by the firms’ production.This approach to CSR involves all related stakeholders in the development and implementation of the CSR (Freeman, 1984). In reality, the corporate governance 1 CSR in general is closely linked with principles of â€Å"sustainable development†, that is, enterprises should be obliged to make decisions based not only on the financial/economic factors but also on the social and environmental consequences of their activities. In terms of labor condition issues related to the outsourcing process, there are numerous reports and news about the exploitation of labor.The interested reader can visit some labor NGOs’ websites, for example, www. AMRC. org, China Labor Watch, ILO, etc. 3 (CG) approach (or firm-centered approach), which uses corporate social responsibility codes of conduct (CSRC) to regulate th e suppliers’ operations in LDCs, is more prevalent. 2 From an institutional theoretical perspective, this article argues that the high transaction cost incurred prevents the stakeholder approach from bringing all stakeholders together and developing strategies to deal with the externality produced by the outsourcing of production.Instead, the current CSR practice has reduced the original stakeholder approach from the CG approach to CSR, the CSRC. In practice, while the CSRC may get around the transaction cost of getting all stakeholders together and reduce potential conflicts that may arise during the CSRC planning and implementation process, this CSRC may be vulnerable to opportunism (or agency problem) at the supplier levels (Jiang, 2009). As the codes are externally imposed on the suppliers, they can exploit the information asymmetry to disregard some of the codes.Because factory interviews and surveys for suppliers’ management could hardly reveal the true faces of labor conditions, an effective measure to reveal workers’ voices at suppliers’ level is called for. However, systematic survey data are needed to verify the validity of CSRC to labor conditions. In addition, there is no unified method to understand the effect of CSRC on labor conditions. 3 This article provides a theoretical argument and a unified empirical approach to analyze labor conditions under the CSRC and the opportunism effect at the supplier side.Utilizing a unidimensional item response theory (UIRT) model, this study looks at the CSRC effect on 16 labor conditions in three dimensions: labor bargaining and association rights, violation of law, and safety and living environment. These labor conditions data are generated from a randomized survey of 12 toy factories in Southern China, including 10 CSRC suppliers for McDonald’s, Mattel, Disney, and Wal-Mart, and 2 non-CSRC factories as a comparison group. 2In particular, some MNCs and business associations would use Corporate Social Responsibility Codes Certifications (CSRCC) as measures to regulate suppliers’ conduct, such as Nike and the like. 3 For example, the Office of Child Labor, Forced Labor, and Human Trafficking at the Bureau of International Labor Affairs, U. S. Department of Labor mentioned in the China’s country report that â€Å"Because of China's repressive political system, it was not possible to obtain any information directly from China.There are no Chinese non-governmental organizations (NGOs) active in this area, and foreign NGOs do not have access. Accordingly, it was impossible to corroborate or reject allegations by various labor and human rights organizations that children are working in export industries which produce for the United States market† (Office of Child Labor, Forced Labor, and Human Trafficking, Bureau of International Labor Affairs). 4 The UIRT model effectively compares the relative seriousness of a labor problem with the lab or union problem, which is the most serious labor problem found in this study.It shows that, first, the CSRC may have a positive effect on labor conditions. CSRC suppliers have less serious labor condition problems than factories with no CSRC. Second, because different CSRCs may have different coverage and monitoring intensity, different CSRCs may result in different labor condition issues. Non-CSRC factories are found to have six serious labor problems; McDonald’s, Mattel, Disney, and Wal-Mart have four, two, one, and zero serious problems, respectively.Third, suppliers working with more than one CSRC may reduce the possibility of opportunism at the supplier level. While suppliers with 1 CSRC would have 2 problems as serious as the labor union problem, suppliers with 2 CSRCs would have no such problem. However, the CG approach to CSRC may suffer from opportunism. This study documented some typical suppliers’ measures to circumvent CSRC auditing: fire drill, trained qu estion, giving a holiday, cleaning the factory, and safety drill.In addition, the subsequent analysis shows that different suppliers’ measures to circumvent the CSRC may represent a diverse level of opportunism. The subsequent sections are organized as follows: the second section explores the effect of the stakeholder approach and CG approach to CSRC and their effects on labor conditions; the third section describes the survey, basic observation from the data, and the methodology; the fourth section provides the estimation results; the last section discusses the results and draws some conclusions. 2.Stakeholder versus corporate governance approaches to CSR and labor conditions: A transaction cost perspective With the globalization of production in recent years, MNCs from developed countries migrated their manufacturing (high labor cost) operation to LDCs for the sake of cost minimization or to be closer to the consumer market. Due to the regulations of the host government, th eir production is usually outsourced to local suppliers. The intense competition among suppliers may induce lower bidding price for outsourcing deals from MNCs.This low bidding price might, in turn, result in the downward pressure of labor cost, safety equipments (Jiang, 2009; Pun, 2005a ; b). As a result, the globalization of production may even make working conditions weaker than it was before the market opened (Lee, 5 1995; Chan, 2003). Such worker conditions have been reported in mass media, generating concern from consumers, NGOs, policymakers, and academics. For example, in China, labor relations have been changing since the market reform began (Zhu and Warner 2004 a ; b). Workers usually suffered from â€Å"new exploitations† (Chan, 2001). In response to the recent surge of complaints from NGOs, media, and consumer movements, outsourcing MNCs’ corporate social responsibility (CSR) has been designed to meet the consumer demand for a â€Å"clearer† product with fewer exploitations and/or a better working environment and safety for the LDC worker. 5 In fact, the original stakeholder theory was mapped out by Freeman (1984). The theory attempts to ascertain which groups are stakeholders in a corporation and thus deserve management attention. A company’s stakeholders are therefore all those who are influenced by, or can influence, a company’s decisions and actions.This theory intends to address the â€Å"Principle of Who or What Really Counts. † A firm, as a member of the stakeholder, cannot survive without taking into account the needs of the other stakeholders (Freeman, 1984). Figure 1: Original stakeholder model (Freeman, 1984) 4 Richard B. Freeman (1994) conducted one of the earliest surveys on labor standards under globalization of production. For further CSR code description, please go to http://www. csreurope. org/ 5 Bjorner (2004) shows that consumers would prefer to pay more for a â€Å"clean† product in the market.See also Harrison and Scorse’s (2003) study on how globalization impacts compliance with labor standards. 6 The CSR problem may be viewed from the institutional perspective: when a firm’s operations create externality, the stakeholders, those who have been affected by the externality, could bargain with the firm and achieve welfare improvement as long as the property rights of each party are well defined and the information cost for bargaining is low (Coase, 1960). Under these conditions, an agreement that guarantees the interests of stakeholders could be made.In the case where the externality is produced by MNCs’ outsourcing, the CSR is therefore an institution (or the rule of game) that may help to remedy past and potential conflicts of interest among stakeholders and produce a more sustainable future development. However, the stakeholder approach of CSR may not be sustainable due to the complexity of the agreement, the high cost of property righ ts’ identification, and the high information cost of bargaining. First, there is doubt about the willingness of MNCs to design a valid CSR practice that would help alleviate workers’ conditions at the supplier level (AMRC, 2004:10).Second, it is sometimes difficult to identify all the stakeholders because they are often locally determined. The transaction cost may be significant as stakeholders who can work together are needed to conceptualize and implement CSR. 6 Third, even when one can draw a rough picture about potential stakeholders, the question is how these very different – usually polarized – stakeholder views can successfully collaborate from the drafting the CSR to its implementation and monitoring (Dubinsky, 2002).The stakeholder approach to CSR is thus considered to have a higher transaction cost than the CG approach because it has to define all stakeholders and bring them together to develop the CSR design. In general, the CG approach to CSR is an expression used to describe what some see as a company’s obligation to be sensitive to the needs of all stakeholders in its business operations (Hill and Jones, 1992). 7 This is a top-down approach that aligns with the MNCs’ board (Husted, 2003).However, in the case of outsourcing decision, according to Williamson (2008), the choice of form of governance is subject to transaction costs due to asset specificity, uncertainty, frequency and their transaction costs, among others. These transaction costs would then determine the form of outsourcing (market or vertical integration). 8 The CG approach to CSR thus coincides with the outsourcing form and uses external social auditors instead of relying on stakeholders to monitor the behaviors of suppliers. 6 Murray (2002) suggests that even when companies have elegant statements of principle, the complexity is the accountability.Local-level involvement is needed and involves a complex regime of stakeholders (Murray, 2002: 41). 7 Most of the CSR has similar objectives. See, for example, Leipziger (2001) for SA8000, one of the guides to the new CSR code. 7 Figure 2: Corporate governance approach: the firm is at the center. Source: Fassin (2008) Figure 5. In practice, the CG approach to CSR usually employs the CSRC as a means to control the behavior of suppliers. 9 From the MNCs’ point of view, the CSRC may have less uncertainty and may be a more cost-effective market solution. First, MNCs can use their existing internal system to monitor the suppliers.Second, MNCs can employ â€Å"external specialists† such as auditing firms instead of the stakeholders to â€Å"plan† and â€Å"monitor† the implementation of CSRC at the supplier level. 10 One of the potential pitfalls of the CG approach to CSR is that even when the MNCs aim to implement the CSRC for these suppliers/suppliers, monitoring issues prevail and are subject to opportunism (Brown, 2002). 11 In some cases, the CG app roach to CSRC has been criticized as a â€Å"lame duck† by labor NGOs, even if the MNCs are indeed willing to safeguard better working conditions for 8Williamson (2008) has extended the transaction cost economics approach to outsourcing decision of a firm. The contractual schema suggested that key attributes of transactions such as asset specificity, uncertainty, and frequency and their transaction costs will determine the form of outsourcing (market or vertical integration) (Williamson, 2008: 8). The schema assumed the outsourcing form would be determined and stabilized as long as the property rights are well defined and reliably enforced by courts. 9 See Kotler and Lee (2005) for their suggestions on using CSR from a corporate governance point of view and from a marketing perspective. 0 See, for example, the social auditing of Pricewaterhouse. Also see the establishment of SA 8000 by Social Accountability International (SAI) as a newly formed labor standard that responds to the need of MNC’s code of conduct. Their website: http://www. sa-intl. org/index. cfm? fuseaction=Page. viewPage=473 11 Dara O’Rourke (2001, 2004), an urban planning professor at U C Berkeley, has shown that there has been a problem in implementing the CSR code to the workshop level and the code’s impacts on workshops’ working condition and safety may be not as expected. 8 uppliers’ workers. This is because of the existing monitoring problems at the implementation level . 12 There may be two basic problems at the supplier level. First, the suppliers may not want to comply with the procedure that the MNCs require (AMRC, 2004, 65; Jiang, 2009). There are many measures from suppliers to MNCs’ CSRCs. Second, the â€Å"social auditing† monitoring procedures may be biased and possibly diluted by suppliers’ measures (O’Rourke, 2002). 13 Auditors sometimes leave out some important factors during their investigations (See O†™Rourke, 2000, 2001, 2002). 4 This leads to agency problems associated with CG approach to the CSRC (Goodpaster, 1991; Winn, 2001). In the CG approach, there is high monitoring cost and information asymmetry among stakeholders, especially between local suppliers and the MNCs. 15 In this regard, the CSR movement also encountered challenges about its validity at the theoretical and implementation levels. 16 To validate the theoretical discussion, the testable hypotheses are as follows: H1: If the CSRC is â€Å"good† for workplace conditions, then we should see suppliers with CSRC that have better labor conditions than â€Å"no CSRC† factories.H2: In addition, if a different CSRC content has a different effect, we should expect a systematic difference among CSRC firms. H3: Moreover, if the CSRC suffers from opportunism, different measures to get around the CSRC at the subcontract level may affect CSRC effects. 12 In fact, a number of â€Å"social auditors† have b een created in response to the new â€Å"demand† from MNC. However, the incentive for these auditors to reveal the true situation at the workshop level is in doubt (O'Rourke, 2000). 13 The Designs of Codes vary greatly across companies and industries.For example, the World Worldwide Responsible Apparel Production (WRAP) certification requested that the independent monitors be paid by the company under investigation. That causes serious doubt of the independency of auditor. Also, there is no requirement to interview workers during the factory investigation (Jenkins, 2002: 26). 14 O’Rourke (2002) investigates the social auditing process of Pricewaterhouse Coopers in a garment factory in Shanghai. Pricewaterhouse Coopers used a standard factory monitor process for factory management, but not for workers.He also shows that there may be management bias and flaw reporting on the auditing process. The management bias includes failures to collect information from workers and f ailures to access restraints on freedom of association. 15 Dubinsky (2002) documented the garment worker of suppliers in L. A. for GUESS. While the CSRC in GUESS was intended to enhance workers’ condition, the workers interviewed reveal no material improvement in their working conditions and were threatened and penalized when they speak out. 6 For more information of the consultants for CSR code, please visit: http://www. thecsrgroup. com/ 9 H4: Finally, if CSRC can increase transaction cost of opportunism at the supplier side, more CSRC would reduce the effect of opportunism. To assess the validity of these hypotheses, this study employs data generated from a random survey of workers from 12 suppliers of branded toy MNCs in Shenzhen and evaluates the CSRC effect using a unidimensional item response theory model. The following section describes the survey and the methodology of this study. . Survey and Methodology This section will discuss the survey and the methodology used this study. Section 3. 1 provides details on the survey design and the sample frame. Section 3. 2 compares the CSR code of conduct of branded companies with the mandate of the International Council of Toy Industries (ICTI) and International Labor Organization (ILO) conventions. Section 3. 3 provides a ranking of CSRC tasks in terms of level of difficulty. Section 3. 4 demonstrates the methodology used in this study. 3. The survey The survey was conducted at the Shenzhen and Dongguan areas in Southern China, both highly concentrated with labor-intensive industries, in particular, toy factories. 17 The survey, which took place from June 2006 to March 2007, focused on the toy industry’s CSRC. To systematically examine the impact of CSRC on labor conditions, this survey first used the industrial phone book and identified a sampling frame before the randomized selection of 12 toy factories in Shenzhen and Dongguan areas at Southern China.The selection of the sample is as follows: Their size should be comparable (usually around 1000 workers in each factory). This study identified 10 code factories within which there were eight suppliers with single CSR (Mattel, McDonald’s, Disney, Wal-Mart) and two suppliers with two CSRCs (Disney & Wal-Mart, Mattel, and Wal-Mart). The survey also included two suppliers with similar technologies that produce only for the domestic market as a comparison group. Twenty 17 Please see Appendix 1 for the background of the setting, toy industry in China, and the reasons for choosing the sites in this study. 0 workers at the entry level were interviewed in each factory, and all the workers were interviewed anonymously. 18 A total of 240 workers were surveyed in this study. 19 3. 2 A Comparison of Corporate Social Responsibility Codes of Conduct Table 1 shows a comparison between the statements of the study firms’ CSRC, the International Council of Toy Industries (ICTI) and the International Labor Organization’s c onventions according to publicly available information. The first column shows ILO conventions that are very detailed and cover the labor conditions mentioned.However, the ILO conventions operate at the state level. That is, unless a state signed the conventions, ILO convention restrictions would not be imposed at the local level. The ICTI codes at the association level and other firms’ CSRC followed the ILO conventions, but with less detail. 20 Table 1 is divided into the following sections: violation of law, bargaining rights and freedom of association, workshop safety and living environment, and monitoring procedures. In terms of violation of law, all the codes indicated that there should no child labor under 14 years of age and no forced labor.However, with the exception of the ICTI code, no individual CSRC mentions maternity leave. As for the section on labor bargaining rights and freedom of association, Disney shows its respect for rights for association and collective bargaining, while the ICTI and Wal-Mart show respect for rights of association only. With regard to safety and living environment, Mattel’s CSRC shows the most detailed description, followed by the ICTI; Disney and McDonald’s CSRCs show similar content and are less detailed than those of Mattel and ICTI. Wal-Mart’s CSRC shows the least detail in this aspect.For the monitoring mechanism, the ICTI CSRC mentioned auditing at the initial stage and in production, as well as a follow-up auditing during the production cycle. The ICTI also mentioned that the factory audit could review the employment record, books, and interview worker privately and the audit could be unannounced to reduce the opportunism. While no individual CSRC gives such detailed monitoring mechanism description, Disney makes it clear that they can review records and books, and suppliers cannot subcontract their work further. McDonald’s mentions that the supplier’s management 18Understa nding that suppliers may have pressure to workers not to reveal the true factory information to strangers, I have instructed the surveyors to talk to perspective workers and develop trust between each other. Surveyors would explain the survey purpose and ask the agreement of perspective respondents. The survey is conducted outside of the factory areas, usually at local restaurant during workers’ lunch time, to eliminate influence from suppliers. 19 The survey questionnaire can be provided upon request. 20 More detailed descriptions of the CSRC for ICTI can be found in its website: http://www. toyicti. rg/info/codeofbusinesspractices. html. 11 should put an effort to monitor the CSRC standard, and show that the auditor can enter workers’ dormitory and interview the workers. Among other CSRCs, only Wal-Mart shows that it has increased efforts to monitor the suppliers in both announced and unannounced factory audits. Mattel, however, does not show any monitoring informati on to public. In a nutshell, individual firms’ CSRC could be very different and their monitoring efforts may differ from earlier observations. The next section offers basic observations about the ranking of the difficulty CSRC tasks from a survey of 12 factories. 12Table 1: A comparision of Toy Industries' Corporate Social Responsibility Code of Codes with ILO convent International Labor Organization (ILO) convent International Council of Toy Industries (ICTI) Violation of Law child labor (;15 yrs) (C59) minimum age of 14 * under age labor (;16 yrs) No Maternity leave (C3) maternity benefits as provided by law Illegal Over Time work (C1) (C6) ( C41) (171) by law Forced Labor (C29, C105) Labor Bargaining Rights and Freedom of Association Bargaining Rights (C98) Bargaining Channel Employee Association (C87) Compliant to Client Labor Union (C87) Labor department Use Collective Bargain (C154) Strike Use Safety nd Living Environment Fire Safety (C155) Workshop Safety (C155) Food p roblem Live environment problem other problem no discrimination (C111) compensation Coercion and Harassment notification to employee monitor and inspection: Unannouned Factory Audit no forced or prison labor is employed Disney 15 yrs (but 14 yrs if law allowed) Mattel specified by country's law McDonald 15 yrs (but 14 yrs if law allowed) yes, only if no schooling 60hr/week, 1 day/week off no forced labor Wal-Mart not allow >14 yrs old 48/week +12 hr OT, 1/week day off except special business circumstances no forced labor overtime must be voluntary and documented, rest days with local law payroll deductions must comply with law and mattel requirment 72hr/6days, 14/24hrs no forced labor freel employee representation by local law upplier need to respect right for association lawful union activities supplier need to respect right for Collective Bargain lawful collective bargain in lawful and peasceful manner, without penalty, interference supplier need to respect right for association e mergency exits, trained emergency evacuation; machinery safeguards >= local laws and employees trained facilities or appropriate provisions for meals and other breaks housing, rooms, sanitary facilities meet basic needs; ventilated, fire safety by local laws toilet facilities meet local hygiene requirements , and are properly maintained standards set by law fire safety adequate lighting and ventilation potable drinkng water, sanitary facilites, health and safety housing envriomental friendly race, color, gender, religion, disability with the local law or local standard, choose the higher one rogram and system for fires, spills and natural disasters, emergency exit unlocked have programs to address health and safety, first aid, medical treatment canteens must be safe, sanitary, meet basic needs dormitories must be safe, sanitary, meet basic needs minimize environmental impact fire exit esential safety equipment, first aid kit, emergency care potable drinkng water, adequate, clean res trooms, appropriately-lit work station race, color, gender, religion, disability with the local law no unhealth and hazardous enviroment nontoxic material, envrionmental friendly race, color, gender, religion, disability with the local law legally mandated rates eed to have local lanuage post the CSR terms to employee 1 to 2 factory manager shall be monitor the compliance of standard yes yes, review employment record and books, yes, but no number specified, 26% of 16000 in 2006 (6% increase) can go to housing accommodations can go to housing accommodations others Audit inculdes: Initial , Follow-up, In-Production. no further subcontractor can be used. can private interviews with employees >16000 audit for 8873 suppliers in 2006 Source: International Labor Organization (ILO) conventions, International Council of Toy Industries (ICTI), The Walt Disney Company Code of Conduct for Manufactures, Mattel, Inc. Global Manufacturing Principles, McDonald's Code of Coduct for Suppliers, Wal-Ma rt Stores, Inc.Standard for Suppliers *but notwithstanding the foregoing, that C138 Minimum Age Convention (1973) and C182 Worst Forms of Child Labor Convention (1999) of the International Labor Organization apply Code posted or available for all employees in local language. need to have local lanuage post the CSR terms to employee 13 3. 3 Ranking the difficulty of CSRC tasks Before going into the methodological analysis, let us examine the level of the labor problem from the survey data. This survey considers each labor problem as a task for the supplier to accomplish. The seriousness of each labor problem can be considered as the level of difficulty to accomplish the task. Table 2 shows the percentage of all respondents saying â€Å"the task is a problem† = â€Å"1† and â€Å"the task is not a problem† = â€Å"0. 21 The larger the mean in the variable, the greater the number of respondents who answered â€Å"1,† and the more problem is perceived in tha t task. The variations in these variable means are large. All the respondents cited â€Å"1† in response to â€Å"Union,† which had a mean of â€Å"1†. This is considered to be the â€Å"most difficult problem. † The two other â€Å"most difficult problems† are â€Å"complaint to client† and â€Å"labor association,† with 99% and 98%, respectively, with â€Å"1† on average. As only few respondents raise â€Å"other problems†, its mean is the lowest at 2% on average, which indicates that suppliers consider this the â€Å"easiest problem†. Table 2: Corporate Social Responsibility Tasks' Diffculty level Item Obs Mean Std. Dev.Min Labor Bargaining and Association Rights Labor Union 240 1. 00 0. 00 1 0. 99 Complaint to Client 240 0. 11 0 Employee Association 240 0. 98 0. 14 0 Bargaining Rights 238 0. 45 0. 50 0 Bargaining Channel 240 0. 73 0. 45 0 Labor department Use 240 0. 48 0. 50 0 Collective Bargain 240 0. 96 0. 1 9 0 Strike Use 240 0. 62 0. 49 0 Violation of Law child labor (>14 yrs) 238 0. 11 0. 31 0 under age labor (>16 yrs) 238 0. 34 0. 47 0 No Maternity leave 238 0. 82 0. 39 0 Illegal Over Time work 238 0. 85 0. 36 0 Safety and Living Environment Fire Safety 235 0. 03 0. 18 0 Workshop Safety 235 0. 26 0. 44 0 Food problem 235 0. 83 0. 38 0 Live environment problem 235 0. 29 0. 46 0 0. 02 other problem 235 0. 3 0 Note: the respond â€Å"1† means â€Å"the task is a problem,† while â€Å"0† means â€Å"the task is not a problem† in respondent's factory. â€Å"labor union† has 100% respond â€Å"1† and that mean it is the â€Å"hardest problem† for suppliers. â€Å"other problem† has 2% respond with â€Å"0† and is considered to be the most â€Å"easiest problem† for suppliers. Max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 21 Please see Appendix 2 for a detailed breakdown of the labor condition distribution. 14 3. 4 Methodology T able 2 presents the distribution of the CSRC task and the problem perceived. However, both the respondents’ knowledge of the problem and the suppliers’ characteristics may bias the analysis of CSRC effect on labor conditions.The respondents’ knowledge and the suppliers’ effort to implement CSRC are, however, unobservable. To correct these unobservable biases and obtain a consistent estimate of the likelihood of suppliers’ problem, this methodology section proposes to use the unidimensional item response theory (UIRT) (or the Rasch) model. In general, the item response theory (IRT) is a body of theories describing the application of mathematical models to data from questionnaires and tests as a basis for measuring abilities, attitudes, or other variables. 22 The IRT model is based on the idea that the probability of getting an item correct is a function of a latent trait or ability. The UIRT model is a member of IRT family which applies to dichotomo us data. 3 The UIRT model is usually used in test analysis, which can analyze the relative difficulty level of an examination question by removing the individual (the examinee’s) unobservable influence on the answer of the question. It is used for statistical analysis and development of assessments, often for high-stake tests such as the Graduate Record Examination (GRE). 24 Using the GRE as example, a person with higher intelligence would more likely be able to correctly respond to a question in an intelligence test. This study introduces UIRT model to evaluate the impacts of CSRC on labor conditions. The UIRT model in this study estimates the probability of a worker getting a positive response in a list of the ifferent tasks by taking into account each task’s difficulty level and the different abilities, knowledge, 22 IRT models apply mathematical functions that specify the probability of a discrete outcome, such as a correct response to an item, in terms of person a nd item parameters. Person parameters may, for example, represent the ability of a student or the strength of a person's attitude. Item parameters include difficulty (location), discrimination (slope or correlation), and pseudoguessing (lower asymptote). Items may be questions that have incorrect or correct responses, as well as statements that allow respondents to indicate the level of agreement. 3 In general, the UIRT model estimates these outcomes using two types of predictors – a person’s ability and the test item’s difficulty level. A person’s ability and the test item’s difficulty level are given as â€Å"X† in the right side of the equation. â€Å"Y† is a person’s response to a test item (or a survey item) and is given on the left side of the equation. 24 Among other methods, IRT provides a basis for evaluating how well assessments and individual questions on assessment work. In education, psychometricians apply IRT to a chieve tasks as developing and refining exams, maintaining banks of items for exams, and equating the difficulties of successive versions of exams (for example, to allow comparisons between results over time). 15 nd willingness level of each respondent and individual supplier. The UIRT model is particularly useful in this study because this model can effectively take into account an individual’s knowledge of the problem and each supplier’s willingness to carry out the task. With the estimated outcomes on hand, one can compare the relative difficulty level of any task by using the easiest task (or the hardest task) that suppliers could fulfill. The last section shows that the â€Å"other problem† is the easiest problem among others; this study will therefore use the â€Å"other problem† as the basis for the relative difficulty in ranking problems in the statistical analysis. 25In our context, the UIRT model is defined as follows: Let us consider a sample o f individuals who answer the survey as if taking a test, and assume the probability, Pij , that worker i’s answer â€Å"yes† (or â€Å"1†) responds to task j and depends only on a parameter, ai , representing the respondent’s ability (which can represent the joint outcome of a worker level understanding of the firm and other factors such as suppliers’ characteristics) and another parameter, b j , representing the difficulty level of the task attempted. Assume further that for some monotone transformation, M. M ( P ) = ai ? b j ij (1) for all i and j. The odds of getting a task right decreases with task’s difficulty (and thus the minus sign before b j ). That is, an additive representation is postulated for the suitably transformed probabilities, Pij . When M is the inverse logistic transformation,M ( Pij ) = ln( Pij ) = ai ? b j 1 ? Pij (2) (ln is the natural logarithm), we have UIRT model. 25 We can also use â€Å"union† as a base to obtain a more precise estimation of the probability of getting a task â€Å"right† as the responses are all zeros. As this article focuses on the relative difficulty level of tasks, it is reasonable to use the easiest task as the base. 16 The UIRT model has certain very desirable statistical properties for the estimation of these parameters. With the UIRT model, this grouping method can be approximated to estimate parameters. 26 We can then compare the estimated M ( Pij ) to determine the relative difficulty level of each task. robability of getting a particular task with â€Å"1† can also be calculated The as: exp( M ( Pij )) /(1 + exp( M ( Pij ))) . 27 With these desirable properties of the UIRT model, this study can allow different groupings of CSRC and evaluate their effects on labor conditions by comparing the parameters. 4. Results 4. 1 Basic Result Table 3 shows the benchmark model that pooled all the factories together to investigate the overall picture of lab or conditions using the UIRT model and the â€Å"other problem† as the basis of comparisons. As shown in Table 2, response â€Å"0† means the worker thinks that there is â€Å"no problem† with a particular task, while â€Å"1† means the task â€Å"has problem†.Therefore, using the estimated coefficients, the â€Å"union† coefficients serve as the upper bound and the â€Å"other problem† as the lowest bound; the interpretation of the coefficient is therefore an assessment of the relative significance of the problem against â€Å"union† and â€Å"other problem. † That is, when a task has similar significance as the â€Å"union,† it means that the task is â€Å"very difficult† for suppliers, while a task that is not significantly different from the â€Å"other problem†, it means that the task is also an â€Å"easy problem† for suppliers. Table 3 shows (from left), the fixed effect model, the fixe d effect with probability weighted, and the random effect model, respectively. All the models have passed their test of specification. Comparing 26Specifically, it can be shown (Lord & Novick, 1968, p. 429) that a respondent’s raw score (number of tasks correct) is a minimal sufficient statistic for his/her ability. This leads to a practical implementation of the model in that statistical estimates of abilities and item parameters can be obtained by proceeding as if everyone with the same raw score has exactly the same ability. The parameter estimates are commonly computed using conditional maximum likelihood CMLE and yield consistent estimates of item parameters (Andersen, 1973). 27 Obviously, the Pij is unobservable, as are ability and item parameters. Estimates of the Pij (and, hence, the ai and b j could be obtained if it were possible to give a respondent the same question about a task on repeated occasions, and his/her responses were independent over trials. Clearly, th is is not possible. Alternatively, the probabilities could be estimated if a worker with similar ability could be identified. Assuming a group of people with the same ability, as individuals respond to items independently, the observed proportion of individuals within the group who respond positively to task j is an estimate of the probability that any given person from that group passes the task. 17 the fixed effect model and the model with probability weighted, the ranking of variables’ coefficients is in the same order. There is no qualitative difference between the two models.In fact, the coefficient value and the ranking are identical between the fixed effect model in the first column and the random effect model in the third column, suggesting that heterogeneity is not a problem and the random effect model is not necessary. 28 Therefore, this study will be based on the fixed effect model for extended models in the subsequent sections. â€Å"Union,† â€Å"complain t to client,† and â€Å"employee association† are the top three difficult tasks for suppliers. The estimated probability of union, complaint to client, and employee association in the fixed effect model is 1, 0. 9998, and 0. 9996, respectively, suggesting a 100%, 99. 98%, and 99. 96% chance of getting these tasks with a problem. These estimated probabilities are very consistent with the statistics in Table 2. This reflects that CSRC may not be very helpful for labor union and freedom of association activities.All the tasks in the model are positively and significantly different from the â€Å"other problem,† suggesting that these problems warrant our attention, except the fire safety problem. The reason why the fire safety problem is not frequently cited as a problem may be attributed to the fact that after a fatal factory fire in Shenzhen area in the early 1990s, a workshop-level fire safety law has been introduced. Since then, labor NGOs and the local governmen t have paid attention to fire safety in workplaces. 29 After a brief overview about the problem with the pooled data, the next section tests hypotheses about the effectiveness of the different CSRCs and the problem of opportunism at the supplier level. 28When reading across the columns, one should avoid directly comparing the coefficients among different models as the assumptions are different. However, we can focus on the relative importance of variables in each model. 29 For a detailed discussion about the labor movement and the history about the event, read AMRC (2004: 41-82). 18 Table 3: Unidimensional Item Response Model for Labor Conditions Fixed Effect Item Coef. Labor Bargaining and Association Rights Union 22. 71 (1. 82) Complaint to Client 8. 37 (0. 9) Employee Association 7. 85 (0. 76) Bargaining Rights 3. 87 (0. 61) Bargaining Channel 5. 03 (0. 59) Labor department Use 3. 99 (0. 58) Collective Bargain 7. 25 (0. 73) Strike Use 4. 54 (0. ) Violation of Law Child labor (>14 yrs) 1. 96 (0. 6) Under age labor (>16 yrs) 3. 38 (0. 61) Maternity leave 5. 51 (0. 62) Illegal Over Time work 5. 74 (0. 63) Safety and Living Environment Fire Safety 0. 71 (0. 7) Workshop Safety 3. 01 (0. 61) Food problem 5. 58 (0. 65) Live environment problem 3. 18 (0. 61) constent Number of obs F( 16, 239) Prob > F Pseudo R2 Linktest: _hat 4045. 00 10551. 86 0. 00 0. 49 0. 00 1. 00 P>|z| 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 Fixed Effect with p-weighted Coef. P>|z| 28. 30 (0. 59) 10. 70 (0. 92) 10. 52 (0. 78) 4. 72 (0. 61) 6. 14 (0. 61) 5. 09 (0. 6) 9. 39 (0. 75) 5. 33 (0. 61) 2. 56 (0. 66) 4. 03 (0. 61) 6. 2 (0. 62) 7. 99 (0. 67) 1. 58 (0. 78) 4. 47 (0. 61) 7. 28 (0. 65) 4. 74 (0. 62) 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 Random Effect Coef. 29. 19 (9. 73) 8. 44 (0. 9) 7. 92 (0. 76) 3. 88 (0. 61) 5. 05 (0. 6) 3. 99 (0. 58) 7. 31 (0. 72) 4. 56 (0. 6) 1. 96 (0. 6) 3. 38 (0. 61) 5. 55 (0. 62) 5. 79 (0. 63) 0. 71 (0. 7) 3. 01 (0. 61) 5. 62 (0. 65) 3. 18 (0. 61) 4. 06 (0. 58) 4045. 00 F( 16, 239) Prob > F /lnsig2u P>|z| 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 31 0. 00 0. 00 0. 00 0. 04 0. 00 0. 00 0. 00 0. 31 0. 00 0. 00 0. 00 0. 00 4045. 00 88251. 79 0. 0 0. 55 1. 00 (0. 13) 0. 00 (0. 01) 0. 00 1. 00 1. 00 (0. 13) _hatsq 0. 00 (0. 01) Other Problem is the basis for the comparison Jacknife standard error in parenthesis 60. 68 0. 00 -4. 55 (0. 29) sigma_u 0. 10 (0. 02) Prob >= chibar2 0. 40 Likelihood-ratio test 0. 08 19 4. 2 Which CSR code is better? As shown in Table 1, different branded companies may have differences in the CSRC statement. This section aims to evaluate the effect of an individual brand company’s CSRC on labor conditions and to understand the relationships between company’s CSRC statement and CSRC outcomes. Table 4 shows the UIRT model for labor conditions by different CSRCs.As there are suppliers with more than one CSRC, this study will pool suppliers with same CSRC to estimate the difficulty level of relative tasks. For example, as the survey has one supplier with Disney only but two suppliers with Disney & Wal-Mart, this study combines these three supplier data into the UIRT model. As discussed in the methodology section, the UIRT model can then essentially extract the common information from these Disney suppliers and investigate the aggregated â€Å"Disney CSRC effect. † Table 4 ranks the CSRC according to the number of tasks similar to the difficulty level of â€Å"union,† descending from left (the easiest task) to right (the most difficult task).In Table 4, columns 1 to 5 show suppliers with a larger number of problems and the lowest number of problems similar to the difficulty level of the problem â€Å"union. † They are non-CSRC suppliers (with 6 problems), McDonald’s (5 problems), Mattel (2 problems), Disney (2 problems), and Wal-Mart (with 0 problem). The first obser vation is that the most serious problems are concentrated on â€Å"labor bargaining and association rights†. When focusing on individual CSRC in this section, the â€Å"non-CSRC† suppliers in cited in the first column show that five tasks have the same difficulty level as union. In contrast, there are no such tasks for the Wal-Mart suppliers, suggesting that they may commit less serious labor bargaining and association rights problems.While we focus on the tasks with the lowest bound of difficulty, Disney suppliers may have more (6 tasks) tasks that are insignificantly different from the â€Å"other problem. † This means that Disney’s suppliers may be less likely to have labor bargaining and association problems. Why so? Can the mission statements of CSRC give some useful information? As Table 1 shows, Disney’s mission statement provides a more detailed description than other CSRCs with respect to the labor bargaining and association problems. Howe ver, while there is no explicit statement about labor bargaining and association problems in the McDonald’s CSRC, its suppliers are more likely to commit violations over other CSRCs, as shown in the second column of Table 4. On 20 he whole, it seems that the CSRC could have some positive impacts on labor bargaining and association problems, and the effects would be dependent on the details of the CSRC statements. 30 Let us now discuss the violation of laws. While the non-CSRC suppliers may have serious problems regarding maternity leave, McDonald’s suppliers may have serious problems with illegal overtime work when compared with the labor union problem. This is consistent with McDonald’s CSRC statement – the most concise among other CSRCs – on illegal overtime work. When focusing on the insignificant problem, an interesting pattern emerged: the tasks that are not as serious as â€Å"other problems† are inversely distributed along the columns , except for Wal-Mart suppliers.Disney suppliers continually outperform other CSRCs with no tasks considered significant, while Wal-Mart suppliers show significance in all the tasks. While Table 1 does not show any significant difference between Wal-Mart’s CSRC compared to the other CSRC in this issue, the results may be affected by the fact that these suppliers do not produce goods only for Wal-Mart. The next subsection analyzes the number of CSRCs and its the impact on labor conditions. In terms of safety and living environment issues, all suppliers under this study do not indicate significant problems on fire safety. However, the non-CSRC suppliers are consistently worse than other suppliers in terms of workshop safety and food problem.Indeed, food seems to be a common problem among suppliers, except for Disney suppliers. Consistent with results in labor bargaining and association rights, Disney suppliers outperform other CSRC suppliers because no task is significantly dif ferent from the â€Å"other problem. † In all the three sections, Disney suppliers continually outperform other CSRCs. This result seems to be consistent with Disney’s CSRC statement, shown in Table 1. Some suppliers work for more than one CSRC, which may mean more transaction costs in avoiding the CSRC. It is interesting to explore how the number of CSRCs affect firms. 30 Certainly, we can just observe the CSRC statement, but efforts to enforce the statement terms are unobservable.As discussed in the methodology section, the UIRT model is a tool to remove unobservable influences and obtain consistent estimates of the difficulty level of tasks. 21 Table 4: Unidimensional Item Response Model for Labor Conditions by different CSR code No CSR code McDonald's Mattel Item Coef. P;|t| Coef. P;|t| Coef. P;|t| Labor Bargaining and Association Rights Union 42. 38 0. 00 41. 63 0. 00 19. 98 0. 00 (4. 33) (4. 15) (3. 01) Complaint to Client 42. 38 0. 00 41. 63 0. 00 5. 53 0. 00 (4 . 33) (4. 15) (0. 94) 42. 38 41. 63 19. 98 Employee Association 0. 00 0. 00 0. 00 (4. 33) (4. 15) (3. 01) Bargaining Rights 42. 38 0. 00 -0. 03 0. 26 0. 28 0. 77 (4. 33) (0. 03) (0. 3) Bargaining Channel 42. 38 0. 00 21. 30 0. 00 19. 98 0. 00 (4. 33) (2. 27) (3. 01) Labor department Use 0. 00 0. 87 20. 67 0. 00 2. 69 0. 00 (0. 00) (2. 01) (0. 68) Collective Bargain 42. 38 0. 00 41. 63 0. 00 4. 80 0. 00 (4. 33) (4. 15) (0. 89) Strike Use 20. 77 0. 00 41. 63 0. 00 1. 48 0. 05 (2. 30) (4. 15) (0. 74) Violation of Law Child labor (;14 yrs) Under age labor (;16 yrs) Maternity leave Illegal Over Time work 0. 00 (0. 00) 19. 92 (1. 77) 42. 38 (4. 33) 0. 87 0. 00 0. 00 0. 00 -0. 03 (0. 03) -0. 03 (0. 03) 21. 92 (2. 44) 41. 63 (4. 15) 0. 00 (0. 00) 0. 00 (0. 00) 21. 84 (2. 43) 0. 26 0. 26 0. 00 0. 00 0. 28 (0. 93) 0. 51 (0. 9) 3. 83 (0. 77) 5. 23 (0. 99) 0. 30 (0. 93) 2. 29 (0. 78) 4. 01 (0. 88) 0. 77 0. 57 0. 00 0. 00 Disney Coef. 23. 12 (13. 27) 23. 12 (13. 27) 6. 32 (13. 94) 4. 63 (14. 02) 3. 24 (14. 00) 4. 87 (13. 94) 8. 00 (19. 73) 23. 12 (13. 27) 2. 94 (13. 92) 4. 48 (14. 02) 6. 49 (14. 24) 5. 35 (14. 05) 1. 13 (14. 02) 3. 27 (14. 01) 7. 97 (28. 15) P;|t| 0. 09 0. 09 0. 65 0. 74 0. 82 0. 73 0. 69 0. 09 0. 83 0. 75 0. 65 0. 71 21. 01 (1. 94) Safety and Living Environment Fire Safety Workshop Safety Food problem 0. 00 (0. 00) 21. 01 (1. 94) 20. 89 (1. 91) 0. 87 0. 00 0. 00 0. 34 0. 31 0. 00 0. 75 0. 01 0. 00 0. 94 0. 82 22 0. 78 4. 3 Is more CSRC better?This section focuses on the number of CSRCs: if more CSRCs increase transaction costs for suppliers, which in turn results in violations of the CSRC, then there should be less violation with regard to labor conditions as the number of CSRCs increases. Table 5 shows the UIRT model for labor conditions according to the number of CSR code. The nonCSRC supplier column is the same as in Table 4, while the â€Å"1 CSRC† column has a pool of suppliers with a single CSRC and â€Å"2 CSRC† suppliers include Matt el ; Wal-Mart and Disney ; Wal-Mart. The pattern of violations in labor conditions in Table 8 is consistent with the hypothesis that more CSRCs is better to control opportunism at the suppliers’ level; the suppliers of 2 CSRCs constantly outperform the 1 CSRC suppliers and the non-CSRC suppliers on the control of labor conditions.For example, in terms of labor bargaining and association rights, the number of tasks with a similar significant coefficient as union is 5 in â€Å"No CSRC† suppliers, 2 in â€Å"1 CSRC† supplier, and none in â€Å"2 CSRC† suppliers. All the estimation results in â€Å"violation of law† and â€Å"safety and living environment† show that the tasks relative to coefficients are reduced as the number of CSRC increases. These results are consistent with the perspective that an increase in the number of CSRCs may increase the transaction costs of opportunism at the suppliers’ level. However, suppliers may also have measures to reduce governance effects from CSRC. The effects of different measures of opportunism are investigated in the next section. 23 24Table 5: Unidimensional Item Response Model for Labor Conditions by number of CSR code No CSR code 1 CSR code 2 CSR codes Item Coef. P;|t| Coef. P;|t| Coef. Labor Bargaining and Association Rights Union 42. 38 0. 00 39. 19 0. 00 19. 40 (4. 33) (0. 31) (3. 12) Complaint to Client 42. 38 0. 00 39. 19 0. 00 6. 11 (4. 33) (0. 31) (0. 89) Employee Association 42. 38 0. 00 39. 19 0. 00 5. 57 (4. 33) (0. 31) (0. 76) Bargaining Rights 42. 38 0. 00 19. 10 0. 00 2. 02 (4. 33) (0. 26) (0. 67) Bargaining Channel 42. 38 0. 00 20. 14 0. 00 3. 92 (4. 33) (0. 20) (0. 64) Labor department Use 0. 00 0. 87 19. 73 0. 00 3. 51 (0. 00) (0. 25) (0. 58) Collective Bargain 42. 38 0. 00 23. 61 0. 00 5. 21 (4. 33) (0. 98) (0. 6) Strike Use 20. 77 0. 00 20. 24 0. 00 3. 51 (2. 30) (0. 26) (0. 64) Violation of Law Child labor (;14 yrs) Under age labor (;16 yrs) Maternity l eave Illegal Over Time work Safety and Living Environment Fire Safety Workshop Safety Food problem Live environment problem 0. 00 (0. 00) 21. 01 (1. 94) 20. 89 (1. 91) 0. 87 0. 00 0. 00 0. 87 15. 47 (0. 87) 14. 77 (14. 65) 22. 30 (0. 42) 18. 70 (0. 23) 2030. 00 0. 64 0. 00 0. 32 0. 00 0. 00 0. 43 (0. 76) 2. 98 (0. 68) 4. 48 (0. 79) 2. 72 (0. 67) 1335. 00 0. 40 0. 00 (0. 00) 19. 92 (1. 77) 42. 38 (4. 33) 21. 01 (1. 94) 0. 87 0. 00 0. 00 0. 00 17. 05 (0. 41) 19. 27 (0. 24) 20. 85 (0. 24) 24. 31 (15. 01) 0. 0 0. 00 0. 00 0. 11 1. 61 (0. 64) 1. 76 (0. 67) 4. 07 (0. 67) 4. 30 (0. 70) P;|t| 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 00 0. 01 0. 01 0. 00 0. 00 0. 57 0. 00 0. 00 0. 00 0. 00 (0. 00) Number of obs 680. 00 Pseudo R2 0. 84 Other Problem is the basis for the comparison Jacknife standard error in parenthesis 4. 4 Suppliers’ Measures to circumvent CSRC auditing 25 When considering other transaction costs of governing with the CSRC, suppliers may apply different strategies to get around the CSRC auditing. Because most CSRC auditing would give advance notice to suppliers, suppliers would have room for measures to circumvent the auditing.This study has documented some of the common supplier measures to circumvent CSRC auditing: fire drill, trained question, giving a holiday, cleaning the factory, and safety drill. Certainly, suppliers may use more than one measure to get around the CSRC auditing; it is likely that suppliers with more labor problems would use more measures to cover up their problems. This section first looks at the number of measures that may be associated with labor problems. We then investigate how the different measures affect labor conditions. Table 6 shows the UIRT model for the number of measures to circumvent CSRC auditing. The first column of Table 6 shows the result of â€Å"no measure. A response of â€Å"no measure† to circumvent CSRC auditing is associated with no significant measures that are different from â€Å"o ther problem,† suggesting that a supplier without any measure for CSRC auditing is performing relatively well in terms of labor conditions. The second column, i. e. , â€Å"one measure† to circumvent CSRC auditing, shows a very different pattern. Although no task is as difficult as the union problem, all tasks show a significant coefficient, except the fire safety problem. The third column, â€Å"two or more measures,† indicates an even stronger coefficient in all the tasks than the column with one measure. Furthermore, the task â€Å"complaint to client† is as strong as â€Å"union† in this column.When comparing the three columns in Table 6, it appears that the more the measures suppliers use, the higher the likelihood they commit more labor problems. This is consistent with the transaction cost hypothesis that opportunism may increase the transaction cost of governing with the CSRC. It is not necessary that all measures have an equal effect on lab or problems. The next section shows the effect of individual measure of CSRC auditing on labor problems. 26 Table 6: Unidimensional Item Response Model for Labor Conditions by Number of Supplier's Measure No Measure One Measure Two or More Measures Item Labor Bargaining and Association Rights Union 19. 88 0. 22 20. 47 0. 0 34. 57 0. 00 (15. 58) (3. 66) (4. 30) Complaint to Client 5. 35 0. 79 7. 34 0. 00 34. 57 0. 00 (19. 88) (1. 20) (4. 30) Employee Association 5. 35 0. 79 7. 34 0. 00 20. 80 0. 00 (20. 03) (1. 21) (2. 85) Bargaining Rights 1. 64 0. 92 2. 40 0. 00 17. 16 0. 00 (15. 09) (0. 77) (2. 19) Bargaining Channel 3. 47 0. 82 4. 24 0. 00 17. 84 0. 00 (14. 91) (0. 74) (2. 13) Labor department Use 3. 20 0. 83 3. 85 0. 00 17. 42 0. 00 (14. 89) (0. 71) (2. 17) Collective Bargain 19. 88 0. 22 6. 63 0. 00 19. 85 0. 00 (15. 58) (1. 03) (2. 30) Strike Use 2. 94 0. 85 3. 78 0. 00 18. 47 0. 00 (15. 15) (0. 73) (2. 18) Violation of Law Child labor (>14 yrs) -14. 1 0. 34 1. 26 0. 09 15. 87 0. 00 (14. 61) (0. 74) (2. 18) Under age labor (>16 yrs) 2. 24 0. 88 2. 59 0. 00 16. 91 0. 00 (15. 11) (0. 77) (2. 12) Maternity leave 2. 77 0. 86 4. 60 0. 00 18. 89 0. 00 (15. 14) (0. 76) (2. 20) Illegal Over Time work 5. 23 0. 80 6. 39 0. 00 19. 09 0. 00 (19. 94) (1. 06) (2. 26) Safety and Living Environment Fire Safety 0. 00 1. 00 0. 73 0. 40 12. 91 0. 32 (22. 56) (0. 86) (12. 82) Workshop Safety 0. 00 1. 00 2. 03 0. 01 16. 24 0. 00 (22. 56) (0. 78) (2. 09) Food problem 4. 13 0. 79 5. 10 0. 00 21. 50 0. 16 (15. 39) (0. 86) (15. 24) Live environment problem 2. 19 0. 89 2. 67 0. 00 16. 88 0. 00 (15. 10) (0. 77) (2. 2) Number of obs 267. 00 1777. 00 1321. 00 Pseudo R2 0. 52 0. 51 0. 52 Other Problem is the basis for the comparison Jacknife standard error in parenthesis Supplier's measures to circumvent CSRC auditing include fire drill, trained question, giving a holiday, Cleaning the Factory, and Safety Drill. 27 4. 4 Suppliers’ Measures to circumvent CSRC auditing Table 7 s hows the UIRT model for individual measures to circumvent CSRC auditing. The measures in this study include fire drill, safety drill, giving a holiday, trained question, and cleaning the factory. As the safety drill estimation is not converged, Table 10 shows only the four other supplier measures.Table 7 ranks the measures according to the number of the most serious tasks (compared to union) from the left, namely, fire drill (6 tasks), giving a holiday (4 tasks), trained question (0 task), and cleaning the factory (0 task). Why do fire drill suppliers have the most tasks similar to the union coefficient? As discussed in the last subsection, fire safety is one of the most serious areas that the Chinese local authorities are concerned with, and it is the most frequent task to be checked during a factory visit. If a supplier has not even met the basic standard of the fire safety, it is reasonable to infer that other tasks are likely to be a problem as well. This is similar in the case of holidays for workers – meaning that the supplier tries to prohibit workers from revealing this fact.As giving holidays for a busy factory is costly, suppliers are likely to create more than one problem in the task. â€Å"Trained question† is one of the most common ways that suppliers try to get around CSRC auditing during the worker interview. Suppliers try to train workers with â€Å"model answers† that fit the CSRC auditing interview questions. The â€Å"trained question† effect may be very limited as only trained workers model their answers to the interview question, and this cannot cover other the problems in the factory. It is clear that the â€Å"trained question† may have fewer problems than â€Å"fire drill† and â€Å"giving a holiday. † â€Å"Cleaning the factory† was found the least serious measure.Only suppliers who are confident with their labor conditions would use â€Å"cleaning the factory† to polish th eir workshop. 28 Table 7:Unidimensional Item Response Model for Labor Conditions by Supplier's Measure Fire Drill Give Holiday Trained question Cleaning Factory Item Coef. P>|t| Coef. P>|t| Coef. P>|t| Coef. P>|t| Labor Bargaining and Association Rights Union 40. 28 0. 00 39. 74 0. 00 35. 79 0. 00 21. 80 0. 00 (1. 05) (3. 83) (5. 39) (2. 87) Complaint to Client 40. 28 0. 00 39. 74 0. 00 21. 55 0. 00 8. 79 0. 49 (1. 05) (3. 83) (2. 80) (12. 70) Employee Association 40. 28 0. 00 22. 22 0. 24 21. 55 0. 00 7. 68 0. 00 (1. 05) (18. 58) (2. 98) (1. 00) Bargaining Rights 18. 38 0. 29 0. 00 0. 75 17. 88 0. 00 3. 5 0. 00 (14. 33) (0. 00) (2. 78) (0. 75) Bargaining Channel 18. 38 0. 00 39. 74 0. 00 18. 36 0. 00 4. 47 0. 00 (0. 45) (3. 83) (2. 62) (0. 73) Labor department Use 21. 11 0. 00 16. 38 0. 32 18. 07 0. 00 4. 38 0. 00 (0. 61) (16. 06) (2. 69) (0. 71) Collective Bargain 40. 28 0. 00 39. 74 0. 00 20. 60 0. 00 6. 54 0. 00 (1. 05) (3. 83) (2. 81) (0. 85) Strike Use 40. 28 0. 00 17. 62 0. 0 0 19. 30 0. 00 4. 85 0. 00 (1. 05) (1. 31) (2. 88) (0. 74) Violation of Law Child labor (>14 yrs) 0. 00 0. 45 18. 35 0. 00 16. 41 0. 00 2. 01 0. 01 (0. 00) (1. 70) (2. 61) (0. 73) Under age labor (>16 yrs) 18. 38 0. 29 18. 00 0. 00 17. 51 0. 00 3. 33 0. 00 (14. 3) (2. 04) (2. 60) (0. 75) Maternity leave 21. 11 0. 25 22. 22 0. 27 19. 09 0. 00 5. 22 0. 00 (14. 97) (19. 80) (2. 72) (0. 74) Illegal Over Time work 40. 28 0. 00 22. 22 0. 27 19. 85 0. 00 6. 14 0. 00 (1. 05) (19. 80) (2. 72) (0. 85) Safety and Living Environment Fire Safety 0. 00 0. 39 0. 00 0. 75 14. 16 0. 00 0. 72 0. 40 (0. 00) (0. 00) (2. 54) (0. 85) Workshop Safety 0. 00 0. 38 0. 00 0. 75 16. 77 0. 00 2. 84 0. 00 (0. 00) (0. 00) (2. 70) (0. 75) Food problem 40. 28 0. 00 39. 74 0. 00 20. 59 0. 00 6. 11 0. 00 (1. 05) (3. 83) (2. 87) (0. 89) Live environment problem 18. 38 0. 00 21. 51 0. 00 17. 53 0. 00 2. 88 0. 00 (0. 45) (2. 15) (2. 72) (0. 5) Number of obs 85. 00 408. 00 1440. 00 2452. 00 Pseudo R2 0. 83 0. 88 0. 49 0. 50 Other Problem is the basis for the comparison Jacknife standard error in parenthesis except the fire estimation, which indicates standard error in parenthesis. â€Å"Safety Drill† is not converged in estimation. 29 Robustness Check For convenience, the estimation of the UIRT model in Table 7 uses the â€Å"other problem† as a base. However, as the â€Å"other problem† is not exactly along zero in the data, there may be a problem in estimating the â€Å"true probability† of the task. More importantly, there may be problem of the UIRT assumption about the zero covariance among items in the estimation.This robustness check uses the â€Å"union† as a base in Table 8, which reruns the UIRT model in Table 5. If the zero covariance assumption is violated, using the upper bound of the problem, the union, as the base of estimation will give very different results. However, if the zero covariance assumption is not violated, as this estimation uses  "union† as a base, the coefficient will be interpreted exactly opposite to that given in Table 5; those problems with larger and significant coefficients given in Table 5 should now be close to zero and insignificant in Table 8 and vice versa. Most of the coefficients here are negatively signed, suggesting that they are â€Å"weaker† than the union problem.Non-CSRC suppliers are seen to have six problems (complaint to client, employee association, bargaining rights, bargaining channel, collective bargain, and maternity leave), â€Å"1 CSRC† has two problems (complaint to client and emplo