The Effect of Sustainability Standard Uncertainty on Certification Decisions of Firms in Emerging Economies

Abstract

Voluntary sustainability standards that establish global rules for firms’ environmental and/or social conduct and allow for verification of firm compliance via third-party certification hold the promise to govern firms’ sustainability conduct in a globalizing world economy. However, the recent proliferation of competing and overlapping global sustainability standards that have been developed by various stakeholders with different agendas, creates uncertainties for firms that likely reduce their propensity to adopt any standard. Without widespread adoption these standards cannot effectively govern firm conduct and in contrast create barriers for firms’ access to export markets. We suggest that the uncertainties associated with competing standards and the effect of these uncertainties on standard certification decisions are especially large for firms in emerging economies because these firms lack access to information about current and future standards and the resources to obtain certifications to multiple standards. We theoretically propose and empirically identify three distinct sources of sustainability standard uncertainty: (i) diversity of customer requirements, (ii) dynamism of customer requirements, and (iii) the unpredictability of the future evolution of standard and propose that each of these sources reduces firms’ propensity to obtain certification to any standard. Our empirical results based on certifications to food safety standards by a sample of 97 Mexican food exporting firms confirm that three distinct sources of sustainability standard uncertainty exist and that all of them negatively impact certification. We discuss ethical implications and offer recommendations for both suppliers as well as standard setting organizations.

Appendices

Appendices

Appendix 1: Sample Survey Questions

Questions capturing uncertainty (all items used a 7-point agree/disagree format)
Diversity of customer requirements	α = 0.62
1. Customers’ requirements are very diverse, i.e., different customers impose different requirements.
2. Customers from different regions of the world differ in the requirements they impose on us.
Dynamism of customer requirements	α = 0.60
1. Customer requirements regarding our firm’s conduct are constantly changing.
2. I expect that our customers will impose new requirements on our firm in the near future.
3. New customer requirements are constantly being developed and introduced.
Future evolution of standards	α = 0.69
1. Our current investments in specific standards will still be recognized in 5–10 years.
2. It’s not clear whether investments in specific standards will still be recognized in 5–10 years.
3. It’s not clear how standards are going to evolve in the next 5 years (e.g., become more stringent, less stringent, etc.)
4. It’s not clear what new supplier requirements will emerge in the next 5 years.
Questions asking about access to information on standards
Where do you get information about standards and requirements? (check all that apply)
a. International customers
b. Domestic customers
c. Internet or magazines
Does your company actively search for information about environmental requirements? (5-point never/frequently scale)
Question asking about supply chain role
Please indicate your company activity (check all that apply)
a. Producer
b. Processor
c. Distributor/agent/exporter

Appendix 2: Model Performance Diagnostics

Table 3 shows McFadden’s ‘pseudo’ R² statistic (McFadden 1974), which compares the likelihood statistic of the model as specified to the likelihood of the model with only a constant (i.e., if the only proportion of successes in the dataset were used to predict the probability of success for each observation). Unlike the usual R² in OLS, McFadden’s pseudo R² does not describe the proportion of variance explained by the model. Specifically, it is computed as

$${\text{McFadden}}\;{\text{pseudo}}\,\;R^{2} = 1 - \frac{{\ln \left( {\hat{L}_{\text{U}} } \right)}}{{\ln \left( {\hat{L}_{\text{R}} } \right)}},$$

where $\hat{L}_{U}$ and $\hat{L}_{R}$ are the likelihood of the model as specified and with only a constant, respectively (Hoetker 2007). Thus, if a model has a McFadden pseudo R² of 0.58, the likelihood of the model is 58 % higher than the constant-only alternative. The real meaning of this information is, as Hoetker argues, unobvious; as with many model criteria, it is most useful as a means to compare models rather than as standalone information.

Along with pseudo R², Table 3 also displays the area under the receiving operator characteristic (ROC) curve, which is widely used in medical diagnostic test design, and is a very useful metric for binary-outcome models (Zweig and Campbell 1993). The ROC curve measures the ability of a model to accurately distinguish true positive outcomes ($\hat{y} = 1 |y = 1$, where $\hat{y}$ is the predicted value of the dependent variable) from false positive outcomes $(\hat{y} = 1 | y \ne 1)$. The curve is drawn by plotting the true positive rate (called the ‘sensitivity’) against the false-positive rate (1—the ‘specificity,’ or the true negative rate) for various ‘success cut-off points’ between 0 and 100.²

The area under the ROC curve quantifies the ability of the model to distinguish between positive and negative outcomes. The naïve 50/50 guess has an area of 0.5, whereas a perfect model has an area of 1. The value of the area is interpreted as the probability that a randomly selected true-positive observation will have a higher predicted $\hat{y}$ than a randomly selected true-negative observation. So, for in the case of the model in Table 3, with ROC area of 0.883, a randomly selected firm that has in fact certified to a food safety standard will have a higher predicted value $(\hat{y})$ than 88.3 % of the firms have not certified to a food safety standard.