Business associations and allegations of human rights wrongdoings
Andrea Biswas Tortajada - Statistics for International Relations Research II
- 1. Introduction
- 2. Research proposition and hypothesis
- 3. Data and measurements
- 4. Methods
- 5. Results
- 6. Conclusion
- 7. Annex 1
- 7. Bibliography
Since 2011, the international community has sought to achieve corporate respect for international standards aimed at securing dignity and equality for all through the multi-level, multi-stakeholder, transnational, and relational governance system advocated by the United Nations Guiding Principles for Business and Human Rights (UNGPs) (Assadi 2011; Backer 2014; Burger 2003; Prenkert 2014). This innovative governance tool has called for several actors to take action at multiple levels to strengthen corporate human rights due diligence (Backer 2014; Burger 2003; Taylor 2011). Yet, ten years after their unanimous endorsement by the Human Rights Council and despite corporate pledges to redefine its role with society and the environment, UNGP adoption remains despairingly low, and allegations of severe negative human rights impacts by transnational companies continue to rise. This has been captured by different benchmarks ranking corporate commitments and actions to respect human rights (Human Rights Resource Centre 2014; Salcito, Wielga, and Singer 2015).
However, at a time when allegations of severe negative human rights impacts by transnational companies continue to rise, corporate membership to industry associations seems to signal an improvement in practices (Eccles (2020); Hogan, Rhodes, and Lawlor (2020);Watch (2020)). This literature suggests that, when it comes to social sustainability, companies are more responsive to private peer-to-peer governance structures than they are to governments threats of heightened regulation, changes in the institutional investment environment, and pressure from civil society groups (Lin, Mao, and Wang 2017; Marquis and Tilcsik 2016; Rosenberg 2011).
2. Research proposition and hypothesis
Building on this premise, this project proposes that the governance spaces created by private transnational industry associations help to advance corporate human rights commitments, thereby preventing external sources from making allegations of serious human rights abuses against a given company. This will be tested by looking at whether company membership in industry-specific associations and the United Nations Global Compact - a global sustainability platform focused on corporate social responsibility corporate governance and mobilizing business to pursue social objectives - have an impact on whether allegations of severe human rights impacts are raised against a company.
This project tests the following three hypotheses:
Ho: UNGC membership negatively affects the chances that allegations of negative human rights impacts will be raised against a company,
Ho: Membership to industry-specific business associations negatively affects the chances that allegations of negative human rights impacts will be raised against a company.
Ho: Membership to industry-specific business associations has a higher influence than UNGCP-membership in negatively affecting the chances that allegations of negative human rights impacts will be raised against a company.
3. Data and measurements
This project uses the 2019 dataset produced by the Corporate Human Rights Benchmark (CHRB), which is available for download on their website. This group assesses the performance on human rights of the world’s 199 largest publicly traded companies in the agricultural, apparel, extractives, ICT manufacturing and automotive manufacturing industries. This focus on the businesses with the largest market capitalisation leads to a significant bias that is partially addressed through the inclusion of at least six companies per continent, spread across different regions, and ten companies per industry to ensure a minimum of geographical and industry balance in the pool of assessed firms. As a result of this sampling bias, nevertheless, it is unlikely that a clear differentiation between laggard and leader regions and industries can be drawn (CHRB 2021).
Whilst there are not quantitative base units to measure corporate human rights performance the benchmark used here is widely employed in the business and human rights field. Moreover, the CHRB has collaborated with some 400 stakeholders from the business, government, civil society, academia, finance, and legal sectors to revise and update its methodology and ensure that it provides an accurate proxy for companies’ actions to prevent and address negative human rights impacts. Nevertheless, the scope of this analysis is limited to the end of the production value chain and does not look at the impact that company practices have along the distribution, retailing, end-use, re-use and disposal of the products and services they offer.
The full CHRB dataset includes over 80 indicators divided into six areas, namely: corporate human rights related policy commitments and governance, corporate systems and processes to implement such policy commitments (embedment), grievance mechanisms and remedies, performance regarding human rights practices, responses to allegations of negative impacts and transparency. Each of these categories has been assigned a different weight to produce a company’s total CHRB score. This assignment will focus on a reduced number of independent variables as shown below and described in the next section.
3.1. Defining the independent and dependent variables
The dependent variable Allegations is a categorical binary variable where: - 0 indicates that no allegations of negative impacts have been made against a company in 2019 - 1 indicates that at least one allegation has been made against a company has been made in 2019
Out of the 199 companies included in the study, 89 companies had at least one serious allegation connected to them.
Other independent variables include:
 “HQ”: Country of company domicile is a categorical nominal variable
 “Region”: categorical nominal variable capturing where the parent company is domiciled out of four regions, namely North America, Europe and Central Asia, East Asia and Pacific, and the rest (a category that comprises Latin America and the Caribbean, South Asia, Sub-Saharan Africa, and the Middle East and North Africa).
 “Sector”: categorical nominal variable capturing industry or economic sector from which the company derives at least 70% of its revenue. Information and Communication Technology (ICT) Manufacturing, Apparel, Agricultural (including food and beverage), Extractives (including oil, gas and mining), Finance, and Automotive Manufacturing.
 “Total”: numerical continuous variable aggregating scores awarded for six themes described below. Scores are given in halves and integer units from 0 to 26 points. The higher the total, the stronger a company’s commitment to respecting human rights.
 “Commitments (Governance & Policy Commitments, Score/8)”: numerical continuous variable with a maximum possible score of 8 points. Assesses a company’s public acknowledgement of its responsibility to respect human rights and the way this informs corporate policies and explicit commitments approved by the highest levels of corporate management.
 “HRDD (Embedding Respect & Human Rights Due Diligence, Score/12)”: numerical continuous variable with a maximum score of 12 points. Assesses the extent of a company’s systems and processes established to implement the company’s policy commitments in practice. It comprises  “Embedding Respect (Score/2)” and  “Human Rights Due Diligence (Score/10),” both of which are numerical continuous variables.
 “Remedy (Remedies & Grievance Mechanisms, Score/6)”: numerical continuous variable with a maximum score of 6 points. Focuses on the extent to which a company provides remedy in addressing actual adverse impacts on human rights.
 “UN”: dichotomous variable that indicates that a company is member to the UN Global Compact (1) or not (0).
 “Association”: dichotomous variable that indicates that a company is member to industry-specific associations (1) or not (0). Whenever one or more industry groupings were relevant, a company needed to be member to at least one of them to score (1).
3.2 Corporate Human Rights Peformance
Total scores reveal that in average, companies’ scores are rather low, with a mean score of 9.374 out of 26 possible points. Looking at each category in more detail, we notice that the lowest scores were obtained in the Human Rights Due Diligence part of the CHRB. Out of a maximum of 12 points, the mean score was 3.131 points. Companies scored the highest in terms of commitments to respecting human rights, averaging 3.638 out of 8 possible points. Moreover, there were companies that obtained nil points in the total score as well as in individual components of the index. The five companies that have taken no steps whatsoever to make a public commitment to respecting human rights, and thus obtained a total score of zero in the benchmark, were all based in China: Heilan Group (apparel), Kweichow Moutai (agriculture), Shenzhou International (apparel), Youngor Group (apparel) and Zhejiang Semir Garment (apparel).
Table 1. Corporate Human Rights performance, summary statistics.
|Variable||N||Mean||Std. Dev.||Min||Pctl. 25||Pctl. 75||Max|
CHRB Scores by region and sector
Figures 2.a and 2.b below show a regional and sectoral breakdown of firm representation and average scores.
North America outnumbers other regions with 81 included firms whilst there is only one company from the MENA region (Middle East and North Africa): Saudi Aramco, a firm in the extractives industry, with a score of 1 out of 26 possible points. The highest scoring companies, with 25 points each, were Eni and Unilever, both based in Europe. Companies in that region had a mean score of 13.21. North American companies scored, on average, 8.6 points, 1.5 points above companies in East Asia Pacific and 4 points higher than firms in Latin America and the Caribbean.
The CHRB focuses on four high-risk sectors: agricultural products, apparel, extractives and Information Communication Technologies (ICT) manufacturing. As can be seen in Figure 2.b, ICT manufacturing is lowering the global average score down as it is the poorest performing sector. Extractives and agriculture, the two most scrutinised global industries, have the highest scores.
Allegations of negative human rights, by region and sector
Figures 3.a and 3.b visualise the relationship between allegations of negative human rights impacts and membership to the United Nations Global Compact and sector-specific industrial groupings disaggregated by region and sector.
The effect that being a member to the United Nations Global Compact and industry-specific groupings has on whether a company faces allegations of negative human rights impacts will be tested using a binomial logistic regression model. This model is used given its suitability to assess the relationship between a binary response variable (yes/no membership) and other explanatory variables, granted that it satisfies the following main assumptions (statisticssolutionsAssumptionIndependence2015?):
Appropriate outcome structure. The dependent variable needs to be binary and in this case it refers to whether companies face allegations of negative human rights impacts.
Observations need to be independent of each other. This assumption is met as the CHRB dataset does not rely on repeated measurements or matched data.
Low levels of multicollinearity among the independent variables. This is tested in section 5.2 using the variance inflation factor.
Independent variables and log odds are linearly related. This was done by looking at the p-values of the selected model. As these were small enough, the hypothesis of these coefficients being zero was rejected, suggesting that the linearity with log odds held for this dataset.
A large sample size. The original dataset consisted of 16 variables for 199 observations with no data omissions.
Moreover, logistic regressions do not need to comply with the same assumptions than linear models to. For instance, no linear relationship between the dependent and independent variables is required and neither do the error terms (or residuals) ought to be normally distributed. Lastly, homoscedasticity is not required.
The following section shows the results of different models tested using different variables and controls.
The table below shows three different models. Model 1 and 2 uses total score in the Corporate Human Rights Benchmark and corporate membership to sector-specific associations and the UNGC as main independent variables. Model 3 looks only at the effect of association to sustainability groups as variables whist Model 4 adds total benchmark scores. Other models were tested using individual components of the CHRB index, especially the variable capturing corporate activities to remedy negative human rights impacts (“Remedy”). In all of these cases, the results were similar to those presented on Table 2 and thus the total score was used as a preferable variable given that it captures corporate commitments and concrete actions to ensure companies respect human rights.
|Model 1||Model 2||Model 3||Model 4|
|Predictors||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI|
|(Intercept)||0.247 \*\*\*||0.137 – 0.430||0.201 \*\*\*||0.107 – 0.361||0.350 \*\*\*||0.205 – 0.580||0.205 \*\*\*||0.109 – 0.372|
|Total||1.145 \*\*\*||1.083 – 1.216||1.113 \*\*\*||1.057 – 1.177||1.120 \*\*\*||1.055 – 1.192|
|UN1 \[Yes\]||0.824||0.409 – 1.621||1.603||0.890 – 2.897||0.871||0.430 – 1.728|
|Association1 \[Yes\]||2.005 \*||1.050 – 3.833||3.055 \*\*\*||1.702 – 5.570||1.986 \*||1.038 – 3.804|
In these three models, total CHRB scores and membership to industry-specific groupings are statistically significant at the 99.9% but the positive signs of the parameters indicate companies have higher odds of facing allegations of negative human rights impacts. This counterintuitive finding will be tested in subsequent models that take Model 4 as a base onto which controls for region of company domicile and sector were added.
|Model 5||Model 6||Model 7||Model 8|
|Predictors||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI|
|(Intercept)||0.127 \*\*\*||0.048 – 0.307||0.195 \*\*\*||0.091 – 0.397||0.118 \*\*\*||0.042 – 0.303||0.245 \*\*||0.101 – 0.559|
|Total||1.124 \*\*\*||1.058 – 1.200||1.131 \*\*\*||1.061 – 1.209||1.137 \*\*\*||1.066 – 1.218|
|UN1 \[Yes\]||1.035||0.497 – 2.130||0.886||0.421 – 1.831||1.078||0.496 – 2.324||1.756||0.880 – 3.553|
|Association1 \[Yes\]||1.825||0.918 – 3.628||1.896||0.948 – 3.798||1.723||0.820 – 3.621||3.032 \*\*||1.575 – 5.958|
|Sector \[AP\]||2.147||0.838 – 5.661||2.280||0.878 – 6.106||1.954||0.785 – 4.968|
|Sector \[EX\]||1.244||0.514 – 3.042||1.294||0.516 – 3.278||1.128||0.473 – 2.694|
|Sector \[IT\]||1.663||0.623 – 4.514||1.619||0.596 – 4.470||1.188||0.459 – 3.085|
|Sector \[MS\]||3.712||0.738 – 21.840||4.003||0.780 – 24.151||3.179||0.665 – 18.159|
|Region \[East Asia and|
|1.231||0.537 – 2.843||1.238||0.526 – 2.935||1.093||0.481 – 2.487|
|Region \[Europe and|
|0.786||0.326 – 1.856||0.719||0.291 – 1.739||1.135||0.500 – 2.584|
|Region \[Other\]||0.917||0.213 – 3.443||0.904||0.199 – 3.613||0.774||0.174 – 3.010|
In Model 5, 6 and 7, the only consistently statistically significant parameter was the total CHRB score, which also had a positive sign suggesting that the odds of a company facing allegations of negative human rights impacts is higher for better-scoring firms. Further prove that this variable is relevant, once it was removed, like in Model 8, all other independent variables cease to be of statistical significance.
Models 9 to 12 tested interactions between the different independent variables but due to their lack of statistical significance, they are not further considered for analysis. This can be seen in Annex 1.
5.1 Model selection
In order to select a Model amongst the 8 possible choices, it is useful to use the Akaike Information Criterion (AIC), run a Likelihood Ratio (LR) test as well as look at the Receiver operating characteristic (ROC) curve and Area under the curve (AUC).
AIC values favour Model 2 (246.872) and Model 4 (248.719). We can also run a LR test for these two models only to assess the goodness of fit of Model 2 (nested model) and 4 (complex model). This tests the Ho that there is no significant difference in models against the Ha that the full model with additional term is better. Given the p-value of 0.6958, it is not possible to reject the null hypothesis that there is no significant difference in these two models, making Model 2 a good selection.
## Likelihood ratio test ## ## Model 1: Allegations1 ~ Total + Association1 ## Model 2: Allegations1 ~ Total + UN1 + Association1 ## #Df LogLik Df Chisq Pr(>Chisq) ## 1 3 -120.44 ## 2 4 -120.36 1 0.1529 0.6958
The ROCs and AUCs for Models 2 and 4 are show in Figure X below. Both models have a fair performance, which explains why the ROC curves for Model 2 and Model 4 overlap and their AUC values are the same (0.73).
## Warning in verify_d(data$d): D not labeled 0/1, assuming No = 0 and Yes = 1! ## Warning in verify_d(data$d): D not labeled 0/1, assuming No = 0 and Yes = 1!
Given the results of these different tests, it is possible to select the more parsimonious Model 2.
5.2 Model diagnostics
This section takes a closer look at some of the binary logistic regression assumptions. Regarding multicollinearity among the independent variables, this can be tested using the variance inflation factor (VIF). As a general threshold, a VIF value higher than 5 indicates a problematic level of collinearity. Model 2 however, does not present a problem in this regard and neither are there any outlying observations given low Cook’s distances.
## Error in vif(TAS): could not find function "vif"
Table 4 shows the results for the selected Model. The two independent variables (Total CHRB score) and membership to industry-specific associations) are both positive and statistically significant, suggesting that higher scoring companies have 1.113 higher odds ratios of facing an allegation of negative human rights impacts and companies that have joined industry-specific groupings have 2.005 higher odds ratios to also face allegations of human rights harms.
|(Intercept)||0.201 \*\*\*||0.107 – 0.361|
|Total||1.113 \*\*\*||1.057 – 1.177|
|Association1 \[Yes\]||2.005 \*||1.050 – 3.833|
With this information, it is possible to calculate the marginal effects for membership association on predicted values of raised allegations of human rights harms.
We can also calculate average marginal effects (AMEs) for Model 2, which suggests that a company that has joined an industry-specific association has, on average, 15.30% higher probabilities of facing allegations for causing human rights harms. In a similar way, each percentage point increase in total CHRB scores raises the probability of a company facing allegations of negative human rights impacts.
## factor AME SE z p lower upper ## Association1Yes 0.1530 0.0739 2.0699 0.0385 0.0081 0.2979 ## Total 0.0224 0.0048 4.6239 0.0000 0.0129 0.0318
The results obtained do not support the hypotheses made above since the signs of the predictors are positive rather than negative, which was the initial expectation. This is probably due to either a series of statistical omissions or incomplete theoretical considerations. These elements are discussed in more detail in the concluding section.
Contrary to what was proposed in this paper, membership to the United Nations Global Compact and industry-specific groupings are both positively associated with companies facing allegations of negative human rights impacts in 2019. This counter-intuitive result provides an opportunity to reassess this research’s theory and estimation procedures (Kennedy 2005). Theory wise, it could be that the membership requirements to join both the UNGC and industry associations are low, which means that companies with poor corporate human rights performance are able to join. This also suggests the presence of an issue with reverse causality: companies facing allegations of negative human rights are more likely to join these associations in an effort to improve their reputation as well as gain access to resources and knowledge on how to identify and manage human rights impacts. As such, further research should look at membership screening procedures to assess whether entry requirements regarding human rights commitments need to be met.
Besides insufficient theory, the analysis could yield different results if panel data were used and the model could control for entry requirements and year of membership and of the allegations raised against a given company. This would be useful to assess if there is a temporal component to the relationship between allegations and association membership. Additional research would benefit from obtaining panel data for this set of companies and test the model lagging the variables caputuring membership to the UNGC and industrial groupings.
7. Annex 1
Additional Models testing interaction terms.
|Model 9||Model 10||Model 11||Model 12|
|Predictors||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI||Odds Ratios||CI|
|(Intercept)||0.147 \*\*\*||0.045 – 0.429||0.169 \*\*||0.055 – 0.475||0.076 \*\*\*||0.021 – 0.234||0.106 \*\*\*||0.029 – 0.342|
|Total||1.108 \*||1.008 – 1.225||1.073||0.975 – 1.186||1.198 \*\*\*||1.089 – 1.335||1.130 \*||1.011 – 1.277|
|UN1 \[Yes\]||1.104||0.505 – 2.395||0.471||0.121 – 1.730||1.073||0.491 – 2.327||0.360||0.085 – 1.410|
|Association1 \[Yes\]||1.692||0.800 – 3.575||1.788||0.847 – 3.782||3.772 \*||1.053 – 13.884||4.964 \*||1.338 – 19.197|
|Region \[East Asia and|
|0.809||0.175 – 3.521||1.229||0.522 – 2.911||1.333||0.559 – 3.223||1.350||0.565 – 3.270|
|Region \[Europe and|
|0.537||0.088 – 2.950||0.684||0.271 – 1.675||0.706||0.287 – 1.701||0.646||0.254 – 1.588|
|Region \[Other\]||0.388||0.015 – 4.852||0.933||0.206 – 3.715||0.928||0.202 – 3.770||0.947||0.206 – 3.866|
|Sector \[AP\]||2.354||0.902 – 6.338||2.330||0.895 – 6.275||2.449||0.929 – 6.713||2.592||0.973 – 7.231|
|Sector \[EX\]||1.291||0.509 – 3.307||1.318||0.516 – 3.411||1.409||0.556 – 3.623||1.507||0.578 – 4.025|
|Sector \[IT\]||1.656||0.608 – 4.589||1.647||0.599 – 4.628||1.602||0.589 – 4.440||1.638||0.591 – 4.654|
|Sector \[MS\]||4.075||0.778 – 25.372||4.091||0.768 – 25.489||4.283||0.837 – 26.093||4.475||0.831 – 28.985|
|Total \* Region \[East Asia|
|1.054||0.908 – 1.236|
|Total \* Region \[Europe|
and Central Asia\]
|1.031||0.904 – 1.178|
|Total \* Region \[Other\]||1.122||0.821 – 1.655|
|Total \* UN1 \[Yes\]||1.100||0.974 – 1.244||1.132||0.997 – 1.290|
|Total \* Association1|
|0.918||0.814 – 1.028||0.893||0.785 – 1.006|
Assadi, Muzaffar. 2011. “Discourse on Human Rights: Narratives Beyond “Political”.” The Indian Journal of Political Science 72 (2): 377–86. http://www.jstor.org/stable/42761422.
Backer, Larry Catá. 2014. “Governance Polycentrism – Hierarchy and Order Without Government in Business and Human Rights Regulation.” SSRN Scholarly Paper ID 2373734. Rochester, NY: Social Science Research Network. https://doi.org/10.2139/ssrn.2373734.
Burger, Michael. 2003. “Bi-Polar and Polycentric Approaches to Human Rights and the Environment.” Columbia Journal of Environmental Law 28 (2): 371–94. https://heinonline.org/HOL/P?h=hein.journals/cjel28&i=377.
CHRB. 2021. “CHRB Methodology.” Corporate Human Rights Benchmark. 2021. https://www.corporatebenchmark.org/chrb-methodology.
Eccles, Robert G. 2020. “Human Rights Really Aren’t All That Important: Just Ask 200 Leading Companies.” Forbes. March 9, 2020. https://www.forbes.com/sites/bobeccles/2020/03/09/human-rights-really-arent-all-that-important-just-ask-200-leading-companies/.
Hogan, Benn F, ML Rhodes, and Mary Lawlor. 2020. “Irish Business and Human Rights,” November, 42.
Human Rights Resource Centre, Business &. 2014. “Steep Rise in Allegations of Human Rights Abuse as Boom in Investment Brings Hope of Prosperity. Business and Human Rights in Eastern Africa: A Regional Briefing Paper.” Business & Human Rights Resource Centre. https://media.business-humanrights.org/media/documents/files/media/documents/eastern-afr-briefing-bus-human-rights-apr-2014.pdf.
Kennedy, Peter E. 2005. “Oh No! I Got the Wrong Sign! What Should I Do?” The Journal of Economic Education 36 (1): 77–92. https://doi.org/10.3200/JECE.36.1.77-92.
Lin, Yupeng, Ying Mao, and Zheng Wang. 2017. “Institutional Ownership, Peer Pressure, and Voluntary Disclosures.” The Accounting Review 93 (4): 283–308. https://doi.org/10.2308/accr-51945.
Marquis, Christopher, and András Tilcsik. 2016. “Institutional Equivalence: How Industry and Community Peers Influence Corporate Philanthropy.” Organization Science 27 (5): 1325–41. https://doi.org/10.1287/orsc.2016.1083.
Prenkert, Jamie Darin. 2014. “Conflict Minerals and Polycentric Governance of Business and Human Rights.” In Law, Business and Human Rights, by Robert Bird, Daniel Cahoy, and Jamie Prenkert, 203–29. Edward Elgar Publishing. https://doi.org/10.4337/9781782546627.00015.
Rosenberg, Tina. 2011. Join the Club: How Peer Pressure Can Transform the World. W. W. Norton & Company.
Salcito, Kendyl, Chris Wielga, and Burton H. Singer. 2015. “Corporate Human Rights Commitments and the Psychology of Business Acceptance of Human Rights Duties: A Multi-Industry Analysis.” The International Journal of Human Rights 19 (6): 673–96. https://doi.org/10.1080/13642987.2015.1029284.
Taylor, Mark B. 2011. “The Ruggie Framework: Polycentric Regulation and the Implications for Corporate Social Responsibility.” *Etikk i Praksis
- Nordic Journal of Applied Ethics* 5 (1): 9–30. https://doi.org/10.5324/eip.v5i1.1731.
Watch, Human Rights. 2020. “World Report 2020: Rights Trends in Holding Companies to Account: Momentum Builds for Corporate Human Rights Duties.” Human Rights Watch. January 2020. https://www.hrw.org/world-report/2020/country-chapters/global-2.