WORKING PAPER (November 15, 2010)

CHOOSING AMONG COMPETING ENVIRONMENTAL AND LABOR STANDARDS: AN EXPLORATORY ANALYSIS OF PRODUCER ADOPTION1

Andrea M. Prado Stern School of Business 44 W 4th Street, Suite 7-157 New York, NY 10012 Phone: 212-998-0224 E-mail: [email protected]

WORKING PAPER (November 15, 2010)

1

I am indebted to Adam Brandenburger, Andrew King, Melissa Schilling, Robert Salomon, Michael Katz, William Greene, Melissa Gonzalez, and Manuel Arriaga for important input. Deepak Hegde, J.P. Eggers and audiences at the 2010 Academy of Management Conference, the 2010 Business Policy and Strategy (BPS) Dissertation Consortium, the Informing Green Markets Conference at Michigan University (June, 2010), and the 2010 Consortium for Cooperation and Competition (CCC) gave valuable comments. Financial support from the Stern School of Business, the Fulbright program, and the Latin American and Caribbean Environmental Economics Program (LACEEP) is gratefully acknowledged.

CHOOSING AMONG COMPETING ENVIRONMENTAL AND LABOR STANDARDS: AN EXPLORATORY ANALYSIS OF PRODUCER ADOPTION

ABSTRACT There has been a proliferation of environmental and labor certifications, such as Fair Trade, Rainforest Alliance, and SA 8000. This phenomenon is observed in many industries, including coffee, tourism, apparel, aquaculture, and electronics. Although these programs can have important social and environmental impacts (especially in developing countries, where institutions are weak and governments often do not have enough resources to enforce regulation), the effects of having multiple certification programs competing in an industry are not yet understood. Certification programs have been studied as a mechanism for producers to communicate unobservable organizational attributes to exchange parties. Such signals convey information about a producer’s environmental and labor practices, allowing buyers/consumers to factor them into their purchasing decisions. Previous research on self-regulation has focused on whether firms adopt a single certification. This paper explores what happens in industries in which multiple environmental and labor certifications are available to producers. I extend our understanding of self-regulation by analyzing how managers choose among various certification programs with different characteristics and deployed via different strategies. The empirical context of this paper is the flower industry in Colombia and Ecuador, two of the most important flower growers in the world. I use a combination of fieldwork and quantitative analysis of a unique dataset on flower exporters in these two countries. This exploratory analysis suggests that signaling theory might not suffice to explain producers’ certification choice. Other characteristics, such as the certifying organizations’ promotion strategies and firms’ social connections, seem to influence producers’ decisions. Results suggest that industries with competing certification programs might face a “race to the bottom,” where certifying organizations decrease the stringency of their standard and focus their efforts on marketing in order to increase their adoption by firms. This unanticipated outcome is, in part, a result of consumers being unable to distinguish between the certifications in terms of quality.

2

There has been a proliferation of sustainability standards in industries such as coffee, flowers, apparel, and tourism. Sustainability standards refer to certification programs that require that environmental or labor practices be implemented in the production process to protect the natural environment and enhance the welfare of employees (Blackman and Rivera, 2010). For example, the coffee industry has at least seven different programs: Fair Trade, Rainforest Alliance, UTZ certified, Bird Friendly, 4C, CAFÉ Practices (Starbucks) and AAA Sustainability (Nespresso). The proliferation of certification programs has been quite extensive for some products, with over 40 certifications for textiles, over 30 programs for forest products, and over 100 for food products (Harbaugh, Maxwell and Roussillon, 2010). These standards are developed and promoted by different types of organizations, including non-governmental organizations (NGOs), industry associations, multinational corporations (MNC), and private specialized firms. These organizations have assumed the role of regulating environmental and labor practices, a role that used to belong primarily to government (Mathews, 1997; Matten and Crane, 2005; Scherer et al., 2006). They compete to define the rules of sustainability performance in an industry, appealing to different actors, such as buyers, producers, and governments (Smith and Fischlein, 2010). Sustainability certifications can contribute to improving firms’ environmental and labor performance, which can have important welfare implications for workers and the conservation of natural resources. This social impact can be particularly important for industries in developing countries, where institutions are weak or governments do not have the necessary resources to enforce regulation. Given the increasing amount of goods that firms are importing from the developing world, these certifications may also affect international trade by influencing the transactions between exchange partners from developed and developing countries.

3

Given that buyers and consumers in developed countries cannot observe the environmental and labor standards used to produce goods in developing countries, certifications can be used to reveal information about otherwise hidden organizational attributes and behaviors (King, Lenox, and Terlaak, 2005). However, buyers and consumers are often unsure of the exact quality of a standard—they are unlikely to know its true impact on social welfare or how difficult or easy it is for producers to implement its requirements (Harbaugh et al., 2010). Moreover, even after consuming a certified product, they will not “truly know” its environmental and labor attributes. What buyers and consumers are exposed to are merely the certifying organizations’ promotional strategies and marketing efforts. Given the asymmetric information problem that exchange partners face regarding the environmental and labor attributes of a good, much of the self-regulation literature has analyzed these certification programs as signals to communicate unobservable characteristics (King and Toffel, 2007). Furthermore, scholars have studied why these certifications arise, which firms are more likely to adopt them, and their effectiveness in influencing firm behavior (Khanna, 2001; King, Lenox and Terlaak, 2005; Lyon and Maxwell, 2002). However, most of the empirical research has focused on firms’ decision to adopt a single program—often ISO9000 or ISO14000 (Berchicci and King, 2007)—rather than on the choice between multiple competing certifications. Furthermore, these studies have typically focused only on industry self-regulation in developed countries, leaving a dearth of understanding about certification adoption in developing countries, where these institutions could have significant social impact due to the lack of government resources to enforce environmental and labor regulation (Blackman, 2008; Blackman, Uribe, van Hoof, and Lyon, 2008).

4

In this study, I analyze how producers choose among multiple certification programs to signal their environmental and social practices to exchange partners. Signaling theory would predict that under conditions of asymmetric information and in the presence of multiple competing certifications, producers self-select into a certification that matches the current level of social responsibility of their practices (Spence, 1973). Once producers have emitted that signal, we would expect buyers and final consumers to differentiate among signals and to reward products exhibiting a more stringent certification with higher sales and/or a price premium. In fact, it is precisely this mechanism that underlies our expectation that sustainability certifications will lead to increased efficiency in these markets and an improvement in social welfare among producers and their communities (Berchicci and King, 2007). In the case of environmental and labor attributes, however, consumers and buyers do not get to observe, after purchase, the stringency of the certification programs or how demanding they are of producers. Thus, producers may not have the necessary incentives to self-select into the certification that most appropriately communicate their products’ unobservable attributes to exchange partners. Besides the stringency of the certification program, other attributes, such as the certifying organization’s promotional strategy, play an important role in producers’ decision to adopt a standard. As Smith and Fischlein (2010) point out, in the presence of competing certification programs, the one offering the most efficient environmental or social quality may not necessarily garner the largest acceptance in the marketplace. It is this misalignment between commercial value and program stringency that raises the primary research question explored here: How do producers choose among multiple competing sustainability certifications? Multiple theories may be brought to bear on this question, including signaling theory and related work in self-regulation; theories from the research on technological innovation that

5

predict the adoption of technology standards; and theories about the role of social and institutional influences on the diffusion of practices. In this paper, I review the most relevant aspects of these theories to develop hypotheses about factors that may influence how producers choose among competing sustainability certifications. I then present an exploratory study that combines both quantitative and qualitative data to evaluate which factors are the most influential in the adoption of sustainability certifications. The empirical context of this paper is the cut-flower industry. I focus on the South American countries of Colombia and Ecuador, which are two of the most important flower producers in the world, with almost all of their production being exported to markets in developed countries. Indeed, more than 80 percent of the flowers imported into the U.S. are grown in these countries. This industry provides an appropriate setting because it faces important environmental and labor challenges and can have a significant social impact by improving the working conditions of employees—most of them females from rural areas—and preserving natural resources. The cut-flower industry is also interesting because most of its production centers are located in developing countries, and eleven different certification programs operate in this industry worldwide. To address the above research question, I first conducted extensive fieldwork that encompassed semi-structured interviews of participants in the cut-flower industry, including certifying organizations, buyers, and flower farmers. I also built a unique database that includes information on the whole population of exporting firms from the Colombian and Ecuadorian cutflower industry. The database also includes the characteristics of the certification programs and the information on which firms have based their adoption decisions. I validated the findings of my fieldwork through econometric testing using this data.

6

The insights from this study suggest that producers’ decisions about which certifications to adopt are influenced by multiple factors, including those suggested by self-regulation and signaling theory, network externalities, and institutional forces. Understanding how producers choose a certification(s) requires a comprehensive theoretical analysis. Furthermore, the evidence suggests that the processes that drive the adoption of sustainability certifications do not necessarily drive the industry toward a standard that is optimal for social welfare; nor do they necessarily cause the certifications to become better over time. Instead, the process seems to be vulnerable to adverse-selection problems that can lead to a “race to the bottom.”

FACTORS INFLUENCING THE ADOPTION OF SUSTAINABILITY CERTIFICATIONS As noted previously, multiple theories may be brought to bear on the question of how producers will choose among competing sustainability standards. One of the bodies of research that has tackled this question most directly is the work on self-regulation. This line of work attempts to predict when firms will voluntarily adopt standards that one might otherwise expect to be imposed by regulatory bodies. It typically does not, however, attempt to explain how firms choose among standards. The literature on innovation and network effects focuses on this latter question much more directly. Though this body of work is most closely associated with technology standards, the logic of how factors such as timing of entry, installed base, and promotion influence adoption is also readily applicable to the domain of sustainability certifications. Finally, I will also draw from work on social contagion and bandwagon effects to explore the role that institutional forces play in certification adoption.

Self-Regulation and Signaling Theory

7

Unlike a “command and control” approach in which standards are specified by law, the certifications studied in this paper are not imposed by government regulation. Firms participate in these programs voluntarily. This voluntary type of norm has been called self-regulatory institutions and includes codes of conduct, trade-association-sponsored standards, and management-certification programs (Berchicci and King, 2007). Private specialized firms, NGOs, industry associations and corporations are some of the organizations that have developed and promote these institutions. The literature on self-regulatory institutions has identified multiple motivations for firms’ participation in these non-mandatory programs (e.g., self-regulation may help firms preempt tougher regulation, respond to stakeholders’ demands, or protect themselves from other firms’ “spillover” harms) (Barnett and King, 2008; Khanna, 2001; Lyon and Maxwell, 2002; Rees, 1997; Rice, 2001). However, the primary motive considered in this literature is the use of certifications as signals to reduce asymmetric information between exchange partners (King, Lenox, and Terlaak, 2005; King and Toffel, 2007). Certification programs can help firms communicate their environmental and labor practices to their customers (King, Lenox and Barnett, 2002). These practices are usually hidden attributes of a good or service, as stakeholders may never know whether they have been deceived since they cannot determine by inspection whether the good was produced in a socially responsible way (Reinhardt, 2002). For firms to credibly communicate the unobserved attributes of these “credence goods,” they often need to set up inspection and certification mechanisms (Darnall and Carmin, 2005). In economics, there is a long history of research on signaling in contexts in which asymmetric information exists among exchange partners, including the market for used cars (Akelford, 1970), the job market (Spence, 1973) and the quality of products (Milgrom and Roberts, 1986). In the presence of asymmetric information, a party can send a signal that will reveal some

8

piece of relevant information to the other party. For example, in Spence’s 1973 job-market signaling model, potential employees send a signal about their skill level to the employer by acquiring certain education credentials. A critical assumption of most signaling models is that it is differentially costly for good- and bad-type partners to emit the signal. The cost of obtaining the same credentials is strictly lower for the good-type partners than it is for the bad-type partners. For example, good-type employees believe that they deserve higher pay for their higher productivity, and, thus, they have the incentive to invest in the signal—i.e., obtain education credentials—to separate themselves from the bad-type employees. The difference in cost structures between the two types of employees is of key importance to the value of the signal. In general, two conditions have to apply for signaling to work: a) The payoff to goodtype senders from signaling outweighs the payoff from not signaling; and b) bad-type senders receive higher payoffs from not signaling than from signaling. These conditions generally hold best when consumers are available to evaluate the quality of the good post-purchase, and firms are vulnerable to consumer sanction. It is not obvious, however, that these conditions always apply to the case of sustainability certifications. In such a case, consumers usually do not know the labor or environmental repercussions of a product, even after purchasing it. Consumers are poorly equipped (and poorly incentivized) to understand the stringency of a sustainability standard. The standards’ requirements tend to be complex and technical, and they combine multiple performance dimensions that make them difficult to interpret and distinguish (Smith and Fischlein, 2010). Any standard may, thus, emit a positive signal, causing producers to benefit from adopting even weak standards. The stringency of a certification program has two key components: 1) the monitoring mechanism; and 2) the set of rules, principles or guidelines—i.e., the environmental and labor

9

practices that the firm must implement to be awarded the certification (Gereffi, Garcia-Johnson, and Sasser, 2001). Based on who conducts the monitoring and produces the guidelines, certifications can be classified from first- to fourth-party programs, ranging from self-report to industry associations, external party—NGO or private specialized firms—and governments, respectively (Gereffi, Garcia-Johnson, Sasser, 2001). For the purpose of this paper, I focus only on second- and third-party certifications: programs developed and promoted by industry associations (second-party) or by NGOs and private specialized firms (third-party). Both of these types require the firm to go through an audit process to be awarded the certification; what varies is the type of organization that developed the program and the structure of the monitoring mechanisms. Industry certifications can be considered less strict than external-party certifications, as the guidelines are developed by the same firms that must comply with them. The auditing in this case can suffer from conflict of interest, as the association’s staff members or an auditing firm that the association hires are responsible for inspecting industry members. The industry association’s monitoring mechanism is also less likely to enforce compliance or to expel the noncompliers. After all, the certified firms are members of the industry association and are paying their membership fees. As a representative of a certifying organization noted, “[T]he problem with the industry association certification is that it is a certification for producers by producers. In other words, it is the producers themselves certifying that they are doing something good.” (Nicklain, 2009) Thus, firms might perceive industry certifications as less threatening to apply for than those from NGOs or private specialized firms. The degree of independence of the auditor is essential to ensure an unbiased certification and to decrease potential conflict of interest that could lead to the certification of undeserving

10

firms (Seddon, 1997; Swift et al., 2000).2 These auditing characteristics influence whether firms implement the certification practices in a substantive or a symbolic way (Christmann and Taylor, 2006). Given the above arguments, I hypothesize that firms are more likely to adopt certifications with less-independent auditing and monitoring. Hypothesis 1: Producers are more likely to adopt certifications with less-independent auditing and monitoring mechanisms. The other key component of stringency is the set of environmental and labor guidelines included in a certification’s technical norm. For example, everything else equal, a certification that requires a firm to provide employees with three weeks of vacation time can be considered more demanding or difficult to implement than one that requires only two weeks. Likewise, there is variation among the environmental requirements. For instance, under the item handling and storage of agrochemicals, some certifications may require only the separation and labeling of substances when stored, while others have requirements for storage facilities regarding ventilation, shelves, and construction materials. Most flower farmers interviewed perceived Fair Trade as the most demanding certification available in the industry. For example, according to Fair Trade’s requirements for free association, the firm has to allow outside groups (unions or specialized lawyers) to train employees on their rights and responsibilities, and this training must take place during working hours. Thus, everything else equal, Fair Trade is more difficult to implement and requires a greater investment than if the firm could choose its own time and method for informing employees about their rights. In contrast, Rainforest Alliance requires only that the firm

2

Authors have also raised concerns about auditor qualification and the periodic nature of audits on the effectiveness of certifications (Van der Wiele and Brown, 1997; Stenzel, 2000; Swift et al., 2000; O’Rourke, 2002; Boiral, 2003b, 2005; Yeung and Mok, 2005), though, in general, these concerns have been less emphasized than auditor independence.

11

recognize in practice and in writing its employees’ free-association and collective-bargaining rights. In sum, the difficulty of the technical norms varies among certifications and, accordingly, influences the adoption decision. As a Colombian flower farmer noted: I am adopting the Rainforest Alliance certification and not Florverde because the former is easier to award if the farm is just starting to implement sustainability practices (Maya, 2009). Hypothesis 2: Producers are more likely to adopt certifications with less-demanding technical norms. I am assuming that the most-demanding certifications are the ones that could have the highest social impact, as compared with the minimum that might be achieved through government regulation. Given the lack of resources available for most governments in developing countries to enforce regulation, some of these self-regulatory programs have an important social impact just by achieving compliance with local regulation. The potential impact is even greater the more demanding the certifications. I am not ignoring this fact; however, I am arguing that programs with the most-demanding technical norms have the most potential for social impact. Notably, nearly all of the prior work on self-regulation has examined the decision on whether or not to have a certification (any certification). In general, these studies have found that firms are more likely to certify if they export their production to foreign markets (Christmann and Taylor, 2001; Corbett, 2006; Corbett and Kirsh, 2001; Guler, Guillen, and MacPherson, 2002; King et al., 2005; Potoski and Pratash, 2004). The literature has also identified other determinants of adoption, including firm size and ownership, industry association membership,

12

type of goods produced, and education level of the firm’s manager (Christmann and Taylor, 2001; Rivera and De Leon, 2005).

Network Externalities and the Diffusion of Standards The technology and innovation literature has extensively studied the competition among standards (Arthur, 1989; Farrell and Saloner, 1985; Katz and Shapiro, 1986; Schilling, 2002, 2003). In this literature, researchers have considered a wide range of factors that influence the standard-adoption decision, including characteristics of the organization that developed or promotes the standard—often called the sponsor organization—and strategies used to deploy the standard. Much of the research exploring factors that drive the adoption of technology standards has focused on network externalities (e.g., Choi, 1994; Cottrell, 1998; Katz and Shapiro, 1986; Khazam and Mowery, 1994; Kristianson, 1998; Shurmer, 1993). Network externalities are positive consumption externalities, whereby the value a user derives from a good increases with the number of other users of the same or similar good. The classic examples of markets demonstrating network-externality effects are those involving physical networks, such as railroads or telecommunications; however, network externalities can also arise in markets that do not have physical networks. For example, a user's benefit from using a good may increase with the number of users of the same good when compatibility is important, such as the compatibility of videotapes with VCRs, or of documents with the particular software program used to create and/or read them. A large installed base attracts the developers of complementary goods, and the availability of complementary goods attracts more users, further increasing the installed base

13

(Brynjolfsson, 1996; Katz & Shapiro, 1986; Shurmer, 1993). This leads to a self-reinforcing cycle that can help entrench a single (or a few) technology standard(s) as dominant. Although the adoption of technology standards and of sustainability certifications have their differences, they also have similarities that allow me to hypothesize that some of the findings in the research on the former can hold in the case of the latter. For example, sustainability certifications reap network-externality effects in that the more producers that adopt a certification, the more likely it is that buyers will recognize the eco-label or that local auditors will be accredited, making it easier for producers to opt for that certification. Sustainability certifications are an archetypal multi-sided market, in which the certification’s sponsor must promote the standard simultaneously to producers, intermediary buyers (such as retailers), and consumers. Convincing one side (e.g., producers) that the certification is gaining popularity with another side (e.g., retailers) is a crucial element of the sponsor’s success; thus, sponsors are highly motivated to increase both the actual and perceived installed base of the certification in order to accelerate its adoption. The most direct routes to having a better actual installed base and/or perceived installed base in sustainability certifications are likely to be timing of entry and promotion. Timing of Entry. Other things being equal, sustainability certifications that enter the market earlier have a better chance of accruing a large installed base. A certification that is developed and adopted earlier than others can preemptively capture producers and buyers. Furthermore, because the standard can be improved over time, and revenues from the standard can be used to further develop and promote it, an early certification standard is likely to have advantages over later entrants. Once producers have adopted a certification, they face switching costs to change to a new certification. Producers are not “locked in” to a single certification—in

14

fact, many adopt multiple certifications—but the cost of keeping an existing certification is lower than the cost of adopting a new one, indicating that, other things being equal, a producer is more likely to keep a previously adopted certification than to abandon it for a new certification. Consistent with this, a growing body of research in the technological-innovation literature indicates that being early (though not necessarily first) to market can improve a firm’s chance of securing the position of dominant design (Arthur, 1989; Garud, Jain and Phelps, 1998; Lilien and Yoon, 1990; Schilling 2002). This leads me to hypothesize that the age of a sustainability certification will be positively related to its adoption. Hypothesis 3: Older certifications are likely to have more producer adoptions than younger certifications. Promotion. A considerable body of research on network externalities suggests that firms can improve the adoption rates of their innovations through their promotion efforts (NagardAssayag and Manceau, 2001; Gowrisankanran and Stavins, 2004; Schilling, 1999, 2002, 2003; Witt, 1997). In directly analogous fashion, the sponsor of a certification can increase both the actual installed base and the perceived installed base through its promotion efforts in the markets to which producers wish to export. Clearly, a certification’s potential to increase sales in a producer’s existing markets or to open new markets should have considerable influence on producers’ adoption decision. I consider three aspects of certification promotion: market overlap, market size, and portfolio breadth. The market domain of certifications tends to be geographically bounded. For instance, the Flower Label Program is known in German-speaking countries, while the flower program of the Rainforest Alliance is positioned in the North American market. The industry association programs are not recognized by buyers or consumers outside of their local countries. A flower

15

farmer said, “Buyers do not know what FlorEcuador—the Ecuadorian industry association program—is about. They usually require us to adopt a certification that is known in the market where they operate” (Henao, 2009). Likewise, the Fair Trade label is well-known around the world, but the market domain for its flower certification is Switzerland, the country with the world’s highest consumption of cut flowers per capita. Market overlap refers to the degree to which a certification is promoted in markets in which the producer currently sells. Firms are more likely to adopt certifications whose market domain overlaps with their exporting destinations. Thus, if a firm exports all its production to North America, it will most likely adopt a certification known in the USA or Canada, rather than one known in Germany. Hypothesis 4: Producers are more likely to adopt certifications with a market domain that overlaps with their exporting destinations. Market size refers to the potential sales in the market domains in which a certification is promoted. Just as developers of complementary goods for technology standards prefer to adopt the standard that has a large installed base because it increases the potential payoff of their development investment (Katz and Shapiro, 1986; Schilling, 2002), producers of cut flowers hope to tap into a large sales base as the payoff to their investment in a sustainability certification. Thus, other things being equal, a certification that is promoted in larger markets is more attractive to producers. Hypothesis 5: Producers are more likely to adopt certifications promoted to a larger aggregate market size. As noted previously, sponsors must simultaneously promote their certification to producers, intermediate buyers, and end consumers. If a sponsor can induce large, powerful

16

retailers to demand its certification, this is an extremely powerful lever to bring about producer adoptions.3 As one Colombian farmer noted, “[O]ur farm got the certification basically because Costco asked for it. Otherwise, we would not have adopted an additional certification. We already had the one from the industry association” (Lemaitre, 2009). One of the more effective ways that certification sponsors create relationships with large, powerful outlets is through the development of a multi-product sustainability brand. For example, Rainforest Alliance was able to convince the supermarket chain Costco to demand that its flower suppliers adopt the Alliance’s multi-product certification. As an auditor of a certification program noted: If I were a supermarket manager and had to choose a certification program to go with, I would invest my marketing money in one that has a wide variety of products: coffee, bananas, cacao, tea, and fruit. By choosing a certification that has a portfolio of products, I would make things easier for my customers as they get confused with all these certifications. Besides, it might also be better for the supermarket to associate themselves with certifications that have more products as more consumers might recognize the label (Nicklain, 2009). An Ecuadorian farmer confirms this argument by saying: “Rainforest Alliance and Fair Trade are certifications that certify different products. They already have a brand and they are able to open commercial channels with supermarkets” (Penaherrera, 2009). Organizations that certify multiple products likely reap some economies of scale and learning economies and, as a result, have more bargaining power over retailers. This, in turn, influences adoption and, thus, the installed base of the brand. Furthermore, because the sustainability brand is leveraged across multiple product categories, it can create greater brand strength, increasing the certification’s

3

Notably, it is more powerful to engage retailers (such as supermarkets) than wholesalers. Given their proximity to final consumers, the former have a stronger interest in certified products, which allows them to leverage a socialresponsibility strategy. A wholesaler at the World Floral Expo 2009 mentioned that “wholesalers are not really interested in certified flowers. These labels are more relevant for growers who export directly to supermarkets” (Garcia, 2009). And an Ecuadorian farmer noted that “supermarkets develop responsible sourcing practices as part of their corporate social responsibility programs. Thus, it is more likely for them to require certifications among their suppliers than wholesalers will” (Penaherrera, 2009).

17

perceived installed base. Thus, I argue that certifications that apply to multiple products will attract more producer adoptions. Hypothesis 6: Producers are more likely to adopt certifications with a broader portfolio of products than more-specialized ones.

Social and institutional forces Finally, adoption of certification could be influenced by institutional forces. First, a long line of work on diffusion provides evidence that individuals and organizations learn about new innovations or practices through their social connections (e.g., Rogers, 1995; Gladwell, 2000). Peers offer a valuable information channel that provides not only exposure to the innovation or practice, but also information about its value or acceptability to the potential adopter. For example, in a study of the cultural and network embeddedness of managers, Davis and Greve (1997) found that firms were more likely to adopt poison pills if they had an interlocking directorate with another firm that had a poison pill. They further showed that the likelihood of a firm adopting golden parachutes was significantly related to its geographic proximity to other firms that used golden parachutes, and they argued that this was likely due to the overlapping social spheres of their top managers. Sometimes, organizations adopt innovations not because of their individual assessments of the innovation’s efficiency or returns, but because they see other organizations adopt the innovation (Wade, 1995; Abrahamson and Rosenkopf, 1993). Both institutional and competitive pressures can trigger firms’ decision to jump on the bandwagon. Under institutional pressures, many firms decide to adopt because they fear appearing different from those that have adopted, while, under competitive pressures, they fear that their performance will decline if many

18

competitors profit from adopting (Abrahamson and Rosenkopf, 1993). Thus, when competitors adopt a specific certification, a bandwagon effect might trigger other firms to choose the same certification. Many theories of adoption assume that the greater the ambiguity, the more social considerations influence the adoption decision (Abrahamson and Rosenkopf, 1993). These pressures are usually bounded to a collective—a group of competitors in which each competitor knows when others in the group have adopted an innovation. The boundaries of these collectives can be set within a geographic area or an industry. Given the above, I argue that firms within a collective or close geographical proximity are likely to choose the same certifications. Trade associations are powerful collectives in developing countries. These associations provide many services to producers that governments in these countries do not have the resources to provide. They also provide a unified voice that gives producers greater bargaining power in their negotiations with powerful foreign buyers. Furthermore, these trade associations often are important sources of information about certifications for local producers, as well as vehicles for promoting social identification among them. Therefore, it is likely that they will play a role in shaping the certification preferences of their members. A final important aspect of trade associations is that they also commonly promote their own certification standards. Thus, I argue that producers that are members of a trade association a) are likely to adopt that trade association’s certification, and b) are likely to adopt certifications that others in their trade association have adopted. Hypothesis 7: Producers that are members of trade associations that promote certification standards are likely to adopt the certification standard of the trade association.

19

Hypothesis 8: Producers that are members of trade associations are likely to adopt certifications that others in their trade association have adopted. Finally, the arguments above suggest that geographic proximity should play a strong role in producers’ certification adoption. As a representative of Rainforest Alliance in Colombia noted: “Many times, farmers come to us asking for the certification because they saw the benefits that their neighbors are having.” Another illustrative quote comes from a Colombian farmer, who, when asked how he finds out about certifications, replied, “We find out about certifications in Medillín [a city in another state] because we have a farm there. About certifications in this area, gossip arrives quickly.” Hypothesis 9: Producers are likely to adopt the same certifications as other producers within close geographical proximity. In the next section, I test my hypotheses and report on other findings using a combination of qualitative and quantitative data on the cut-flower industry in Colombia and Ecuador.

DATA AND METHODS Empirical Setting Floral industry. Before the 1980s, the flower industry was dominated by producers in the developed world. Dutch producers dominated the European market and Californian producers dominated the North American market. The beginning of the 1980s, however, saw a migration of production activity to developing countries, which offered better weather and cheaper labor. Countries such as Colombia, Kenya, Ecuador and Tanzania became important producers of cut flowers, competing strongly with developed countries through lower prices and

20

equal or better quality. Thus, flowers from developing countries started to dominate the international market. Flower producers compete mainly in two dimensions: price and quality. However, the environmental and labor practices used to produce the flowers have become another source of differentiation among flower producers. Flower production has a set of environmental and labor challenges. First, the activity can pollute natural resources such as water, air, and soil through the use of pesticides, fertilizers, and chemicals. Both humans and local wildlife can be harmed by these substances. Second, it is a labor-intensive activity and employs a substantial number of female workers, often from rural areas, including a significant share of female heads- ofhousehold. Thus, the production of flowers has the potential to have significant social impact. The floral industry is extremely fragmented at all levels, with many small, familyoperated companies (Wylie and Salmon, 1995). The three major agents in the floral industry are growers, wholesalers and retailers, with a trend towards eliminating intermediaries between growers and retailers and the establishment of supermarkets as the preferred outlet for consumers to buy their flowers. The typical distribution channel is from growers to distributors located in the growing regions, to geographically dispersed wholesalers, who sell to florists, supermarkets, and other retailers within geographic proximity (Wylie and Walter, 1995). Often, industry participants are involved in more than one of these activities. Floral industry in Colombia and Ecuador. Colombia is the leading cut-flower producer in South America and, excluding the Netherlands, the most important cut-flower exporter in the world. After Colombia, Ecuador is the most important flower exporter in the world. Ecuador is well known throughout the world for its high-quality, large-headed roses.

21

By 2006, Colombia had 7,266 hectares assigned to the harvest of flowers and exported US$1.1 billion worth of flowers, 80 percent of which went to the USA. Ecuador had approximately 3,500 hectares cultivated by 2006 and exported US$470 million worth of flowers to markets such as the United States, Russia, Germany and Canada. At the farm level, the industry employed approximately 60,000 workers in Ecuador and over 111,000 in Colombia in 2006, with roughly 50 and 65 percent, respectively, being women. Certifications in the floral industry. There are eleven certification programs operating in the flower industry worldwide. For the purposes of this paper, I analyzed the adoption of the eight relevant to Colombia and Ecuador. Among these eight certifications, some were developed by local industry associations, while others belong to NGOs or private specialized firms. Expoflores and Asocolflores, the industry associations in Ecuador and Colombia, promote their own certification programs, FlorEcuador and Florverde, respectively. The certifications developed by NGOs and private specialized firms in my sample include Rainforest Alliance, Fair Trade, Flower Label Program, Veriflora, Milieu Project Sierteelt, and Fair Flower Fair Plants. All these certification programs require the implementation of environmental and labor practices.

Data This study uses a combination of original qualitative data from fieldwork and a unique dataset of exporting firms from developing countries. The qualitative data were collected through 70 semi-structured interviews conducted in the spring of 2009 with flower farmers, industry-association representatives, wholesalers and representatives of the certification programs. I conducted most of the interviews during a two-month visit to Colombia and

22

Ecuador. Other interviews were conducted at the World Floral Expo in Miami, Florida in March 2009 and by phone. The dataset includes the whole population of exporting flower firms in Colombia and Ecuador. I collected the data from multiple sources. Regarding flower farms in Colombia, I obtained data from Proexport and the Superintendent of Companies. For Ecuadorian firms, I used data from the Central Bank, Superintendent of Companies, and the Ministry of Agriculture. These institutions shared this information under strict confidentiality agreements. The data on certification programs were collected through the websites of each of the certifying organizations and, in some cases, through multiple e-mail exchanges with the organizations. I conducted interviews with representatives of most of the certifications programs to complement the quantitative data. Measures Dependent variable. Adoption of a certification is operationally defined here as a dichotomous variable (0,1). The data are observed for 2008. Each firm has seven entries, one for each certification program. A firm can adopt zero, one, or multiple certifications simultaneously. Only one firm in the database had adopted all seven certifications, and most of them had adopted none. The average number of certifications among firms with at least one certification is 1.5. The data on which firms are certified were taken from the websites of the certification organizations or provided directly by the organizations. The firms in the sample are exporting flower farms from Colombia and Ecuador. There are a total of 1,241 farms, 205 of which have adopted at least one certification. Independent variables. Explanatory variables (all of which will be described in detail below) can be grouped into two types: certification characteristics and firm characteristics. Certification

23

characteristics include the following variables: difficulty of the technical norm, independence of the audit, portfolio of certifications, age of certification, and aggregate market size. Firm characteristics include total exports, percentage exports to Europe, percentage roses exported, and country. The overlap of a firm’s export with a certification’s market domains and the number of members certified are two variables that concern both a firm and a certification characteristic; however, I discuss these two variables in the section on certification characteristics. Certification characteristics Independence of the audit: The independence of the audit is captured through a dummy variable that takes on 1 for those certifications whose audits are conducted by external organizations, either NGOs or private specialized firms. This dummy is 0 when the audit is conducted by either staff of the industry association or an auditing firm appointed by this organization. Difficulty of the technical norm: The degree of difficulty of the technical norm is measured as a continuous variable between 0 and 1. I constructed this measure based on the assessment by industry players (general managers/owners of farm that have implemented at least three different programs). For a preliminary analysis, this number is assigned based on the extensive number of interviews conducted in the field with general managers/owners, agricultural engineers and human-resource managers.4 [to be developed further] Age of certification: This measures the number of years since the certification was launched until 2008, the year of analysis. For those organizations that have standards for

4

This construct is going to be developed through a survey among managers, agricultural engineers and humanresource staff from flower farms in Colombia and Ecuador, for which they will rank the standards according to their degree of difficulty.

24

multiple products, the year included in this variable is the one in which the standard for cut flowers was launched. Portfolio of Certifications: The portfolio of certifications includes the number of products certified by the organization that manages the program. Overlap exports to certifications’ market domain: This variable is the percentage of exports that the firm sells in the certifications’ market domain. For instance, if the certification is Veriflora, this variable will represent the firm’s percentage of exports sold in the USA and Canada, Veriflora’s market domain. The market domain was established from interviews with the certification organizations and farmers. Aggregate market value: The value (in billions of euros) of the cut-flower market in the countries where the certification is promoted is used to construct this variable. The values are taken from the International Statistics Flowers and Plants Report (2009) published by Union Fleur and the International Association of Horticultural Producers. Number of members with certification: This variable counts the number of industry association members that have adopted each certification. When included in the econometric analysis, this variable takes the corresponding value when the firm is a member of the organization; otherwise, its value is 0. Firm characteristics Total of exports (ln): A firm’s total exports are used in the analysis as the control for firm size. Flower farms in Colombia and Ecuador are considered mainly exporting firms, as they sell most of their production in foreign markets. The values are transformed with a natural logarithm to normalize the skewed distribution of the data.

25

Percentage of exports to Europe: Previous research has shown that firms exporting to Europe are more likely to adopt a certification (Anderson, Daly, and Johnson, 1999). Thus, I control for the firm’s percentage of sales that goes to Europe. Percentage of roses exported: Roses can be considered more of a final product than other types of flowers, such as carnations, chrysanthemums, or gerberas, because they are less likely to be mixed with other flowers to make a bouquet. When a flower becomes part of a bouquet, it loses the value of the certification because the bouquet cannot carry a label unless all the products in it are certified. Previous research has shown that the closer a product is to the final consumers, the more likely it is to get certified (Wall, Weersink, and Swanton, 2001). Country: A dummy variable with value 1 is included for Colombian firms and value 0 for Ecuadorian ones. Both countries have environmental and labor regulations and have been exposed to similar international campaigns by activists or competitors arguing that their flowers were produced with very low labor and environmental standards. However, a very important institutional difference between these two countries is that Asocolflores (the Colombian trade association) reacted to these campaigns by developing its own certification. At the same time, this association strongly urged its members not to adopt other standards in order to avoid entering into a “war of seals” that could harm the local producers. Another important difference between the two countries is that one finds bigger and more diversified firms (in terms of cutflowers types) in Colombia than in Ecuador. Membership: A dummy variable with value 1 is included for firms that are members of the local industry associations—Asocolflores in Colombia and Expoflores in Ecuador—and value 0 if they do not belong to the organization. Model specification

26

In selecting an appropriate multivariate statistical method, I considered three characteristics of the model. First, the dependent variable is a dichotomous variable that represents whether a firm has adopted or not adopted a specific certification. Second, each firm has seven certification options to choose from and can adopt more than one program. Third, the data are observed in 2008. Thus, even though the dataset does not have time-variant variables, we can treat it as panel data, in the sense that there are multiple observations per firm; but in this case, all firm-specific observations correspond to the specific year of analysis, 2008. While a multinomial logit model may seem like a logical alternative (given that firms have multiple options to choose from), such a model would not allow me to investigate how certification characteristics influence their adoption. As firms can adopt more than one certification, such a model would require me to evaluate all possible certification combinations.5 In doing this, I would not be able to obtain results for how the characteristics of an individual certification influence its adoption, as I would have to combine them in the right-hand side of the equation. For this reason, I instead chose to use a fixed effects logistic regression to test the hypotheses of my model. A fixed-effect logistic regression allows me to model the firm’s decision to certify or not certify with the different programs. In such a model, the decision to adopt a specific certification (or certifications) is determined not only by certification characteristics, but also by firm characteristics. The explanatory variables can be classified as one of these two categories. Each firm might have different propensities to adopt a certification based on unobservable certification or firm characteristics. If a firm has seven choices from which to adopt, it is possible that the standard errors will not be independent within the observations of each firm. This would occur,

5

The number of possible combinations of certifications would be 8!, making the computation of the regression intractable.

27

for instance, if some firms adopt certifications owing to better management or owing to each firm’s previous experience with implementing responsible environmental practices. As in a time-variant panel dataset, my data have multiple observations per firm. However, there is no time variation because I observe a single year of data. Instead, I am able to control for firm-specific unobservable characteristics using variation across certifications for a given firm. Equation (1) depicts the fixed-effect model in which firm i chooses whether or not to adopt certification j. yij* = αidij + xij'β +εij, where i = 1,…..n, and j = 1,… m

(1)

yij = 1 if yij*> 0, and 0 otherwise where dij is a dummy variable which takes the value one for firm i and zero otherwise, while xij denotes the certification characteristics. Equation (2) depicts the log likelihood function for the fixed-effects model

(2) where P(.) is the probability of a firm adopting a certification. The fixed-effects binary logit model can also be expressed as Equation (3) e αi + xij'β Prob (yij = 1 | xit) = ---------------1 + e αi + xij'β

(3)

The maximum likelihood estimators are calculated by controlling for unobservable firm characteristics that are stable among the certification options. Given that some of the unobserved firm characteristics (αi) can be correlated with the certification characteristics (xij) and that the data present variation across firms and within firms across certifications, I argue for a fixed-

28

effects—as opposed to a random-effects—model to be used in this case6. The result of the Hausmann test (χ2 = 26.82, p=0.0001) supports the decision to use fixed effects. Given that the likelihood equation for the log likelihood function has no solution if there is no within-firm variation in yij, this model specification will drop firms that have adopted either no certification or all of them. Therefore, from the population of exporting firms in the sample, the MLE estimators are based on the data for firms that have adopted at least one certification. By 2008, there was only one firm that had adopted all seven certifications. This observation is also dropped from the analysis. Under this model specification, the sample is reduced from 1,241 firms to 205. Table 1 presents the results comparing certified versus non-certified firms. The results show that there is a statistical difference between firms that certify and those that do not. Most certified firms are what are considered big firms in this industry and their exports are mainly roses—instead of flowers such as carnations, gerberas, astromelias, or fillers. Firms that are members of the local industry association and operate in Ecuador are also most likely to adopt a certification program.

RESULTS Summary statistics and correlations are presented in Tables 2 and 3. Results for the fixed-effects logistic regression are presented in Table 4. Following standard practice, I report the estimated coefficients and standard errors for this model specification.7 The data were analyzed using STATA 9.0.

6

A random-effects model is estimated as a robustness check. For other model specifications used, I report the standard errors and marginal effects as suggested by Hoetker (2007) and Wieserman and Bowen (2009). I do not report the marginal effects in the case of the fixed-effects logistic regression model, as a constant is needed to calculate the probability of adoption. 7

29

Models 1 to 6 in Table 3 analyze the main effects of a certification’s stringency and a program’s promotion strategy on the probability of a firm adopting that certification. The age of the certification, as well as the number of other association members that have adopted the standard were included. I control for unobservable firm characteristics and number of years since the certification was launched. The MLE maintain their signs along the different models. I focus on discussing the results of models 5 and 6, which include all variables of interest. As predicted, firms are more likely to adopt less-stringent certifications (all else being equal). Hypotheses 1 and 2 explore these programs’ stringency characteristics. The independence of a certification’s monitoring mechanism is found to be significant and negatively related to its likelihood of adoption. The negative estimated coefficient on independence of audit suggests that firms are more likely to adopt industry-association certifications than those of NGOs or private specialized firms since the latter two tend to require independent audits. Thus, hypothesis 1 is supported, showing that firms are more likely to adopt programs the weaker they perceive their monitoring and sanctioning mechanism to be. Another dimension of a certification’s stringency is captured through the difficulty of its technical norm. As predicted in hypothesis 2, there is a significant and negative relationship between the difficulty of the technical norm and a certification’s likelihood of adoption. As shown by the sign of the coefficient, producers are less likely to adopt certifications the more difficult their technical norm is. Thus, hypothesis 2 is supported. Hypothesis 3 predicts that older certifications are more likely to be adopted. Results support this hypothesis, suggesting an early-mover advantage for flower certifications that were launched early. Hypotheses 4 and 5 address the effect of a certification’s promotion strategy on its likelihood of adoption. The former proposes that producers are more likely to adopt

30

certifications whose market domain overlaps with their export destinations. The coefficient for this variable is positive and significant, providing support for the idea that producers are more likely to adopt certifications that are known by buyers and consumers in the markets where they export. Thus, hypothesis 4 is supported. To test hypothesis 5, I consider the cut-flower aggregate market size (billions of euros) in each certification’s market domain.8 In model 5, the coefficient for aggregate market size is positive and significant, providing support for the hypothesis that certifications are more likely to be adopted if they are promoted to larger-sized markets. Given that hypotheses 4 and 5 are supported, it is not surprising to observe most certifying organizations seeking to expand their market domains. Some organizations do this by opening offices in new markets, by forming alliances with other established programs known in those regions or even by changing their names to convey a more global image, such as EurepGAP’s change of name to GlobalGAP in 2007. As hypothesis 6 predicted, producers are more likely to adopt certifications that have a greater portfolio of products. This coefficient is also found to be positive and significant, indicating that producers perceive certifying organizations with a portfolio of products as more likely than specialized organizations to provide commercial benefits. Having standards for multiple programs allows certifying organizations to exploit economies of scope in tasks such as branding, accreditation of auditors, and relationships with retailers. I test hypothesis 7 in model 1 of table 6. Firm membership in the trade association cannot be analyzed in the logistic fixed-effect specification of Table 4 because it is a firm characteristic that shows no variation within firm. Thus, a logistic regression that pools the data, clusters the standard errors at the firm level, and has a dichotomous variable for adoption of the

8

Given the high correlation between the size of the certification’s market domain and the overlap with firm exports, I do not include these two variables simultaneously in the analysis [multicollinearity analysis to be conducted].

31

industry standard as its dependent variable is used to test this hypothesis. Results show that members of the industry association are indeed more likely to adopt the organization’s certification. Hypothesis 8 argues that industry-association members are more likely to adopt certifications that other members in the organization have adopted. I include the number of members that have adopted each certification in the fixed-effects logistic regression (see Table 4, model 7) for those firms that are members of the association—zero otherwise. Results show that the industry associations might indeed be providing a mechanism for these standards to diffuse. Finally, for hypothesis 9, I provide descriptive statistics for the location—state and city—of certified firms in Ecuador according to the cut-flower farm census conducted in 2010 by the Ministry of Agriculture [additional analysis to be conducted]. Log likelihood for the fixed-effects model (Table 4) decreases when increasing the number of variables, and the chi-square statistic is significant through models 1-6. The McFadden R2 measure ranges from 0.30 to 0.56 in models 1 to 4. Caution should be observed when relying on this measure since it attempts to only approximate ordinary least squares regression analysis multiple squared correlation coefficients (R2), and the two cannot be considered close analogs. The above measures provide an overall evaluation of the model. Robustness tests I conduct additional analysis through other model specifications in order to test the robustness of the results. The results of a logistic regression model with clustered errors at the firm level and a random-effects logistic regression are presented in Table 5 (cf models 3 and 4). The latter specification, compared with the fixed-effects model, does not wipe out the

32

independent variables that are time-invariant, allowing me to include the firm characteristics in the analysis. The results of these models support those from the fixed-effects model. Coefficients on the certification-characteristic variables not only have the same sign, but also have similar magnitudes. Firm-characteristic variables that previous research finds to be determinants of adoption were also found significant when included as controls. More specifically, firm size and the percentage of the final product (in this case, roses) exported are positively related to the likelihood of adopting a certification. A firm’s percentage of exports to Europe is not found to be a significant determinant of its likelihood of adopting a certification. Previous studies have found that bilateral trade relations are an important determinant of adoption (Albuquerque, Bronnenberg, and Corbett, 2007; Guler et al., 2002). However, these studies consider one certification across industries. The findings in these models are more consistent with the results in Delmas and Montiel (2008). Those authors find no trade-related factors significant in explaining adoption when analyzing a particular industry. Firms located in Ecuador are more likely to certify than those located in Colombia. Based on the qualitative analysis, the main reason for this result is that the Colombian industry association explicitly encouraged its members not to adopt foreign certifications and, instead, to support the local one in order to avoid what they call the “war of the seals.” In Ecuador, on the other hand, when the first certification programs were introduced, the Ecuadorian industry association had no certification program to compete with them. The logistic regression and the random-effects specifications provide support for hypotheses 1-4 and 6-8. Thus, these results can be considered robust.

33

It might be argued that firms are likely to adopt industry-association certifications for purposes other than signaling exchange partners, such as forestalling regulation. Thus, I analyze the adoption of certifications only by NGO and private specialized firms, using a fixed-effects logistic regression. Results are presented in the second model in Table 5. The findings support hypotheses 2-4 and 8 and confirm their robustness. Hypothesis 1 and 7-9 cannot be tested if we consider only non-industry association certifications. Finally, I ran a two-stage Heckman selection model where the first stage explores the decision to certify (with any program) based on firm characteristics, and the second stage looks at the choices among certifications for those firms that decided to adopt a program. I use a Heckman selection with a discrete choice model—probit model—in both stages. I cluster for standard errors at the firm level to adjust them for intra-group correlation. Results for the second stage support hypotheses 1-4 and 6-8, confirming the robustness of the results. One limitation of the two-stage model specification is that the number of firms relevant for the first-stage calculation is 1,241, with one observation per firm. For the second stage, the relevant number of firms is 205, with seven observations per firm. However, the software used is not able to conduct the analysis using a cross-section form in the first stage and a panel-data form in the second stage. Furthermore, this model specification does not allow me control for firm characteristics—through a fixed-effects specification—in the second stage. As a result, both stages pool the data and just cluster it at the firm level. Thus, even though it is worth looking at the results of this analysis, those obtained through fixed-effects logistic regression are still preferable.

DISCUSSION [to be further developed]

34

In this study, I set out to explore how producers choose among different certification programs operating in an industry. To address this question, I conducted interviews with flower farmers, buyers, certifying organizations and industry representatives. I complemented this with quantitative analysis to validate the findings from the field. I identified a set of certification characteristics that influence producers’ adoption decision: stringency of the certification, promotion strategy of the certifying organization, and time of entry. More specifically, I found that producers are more likely to adopt certifications with a higher potential to influence their sales and less likely to adopt stringent standards. Producers’ decision is also influenced by social and institutional pressures, mainly through their membership in the local industry association. The self-regulation literature argues that certifications can be used to decrease the asymmetric information among exchange partners. However, in the presence of multiple competing standards, the promotion strategy of certifying organizations can distort the value of this signal. This is particularly true in the case of sustainability standards, as they certify attributes that cannot be observed or experienced by buyers and final consumers after purchase. The choice of certification programs not only seeks to convey information about the implementation of environmental and labor standards. It is also driven by a promotion component that has not been considered in the self-regulation literature and that influences the buyers’ perception of the program’s installed base. This perception is likely to influence adoption, as the literature on network externalities and diffusion of innovation has shown. When I classified the standards according to their different dimensions, the most stringent certifications were not necessarily the ones most rewarded in the market. A stringent certification adopted by a producer might not signal anything to a group of buyers or final consumers if the certifying organization does not promote the certification in their region.

35

Likewise, a certification that requires the least-demanding labor and environmental practices might provide more commercial value to producers than one with a highly demanding technical norm. Smith and Fischlein (2010) point this out when they argue that certification with the highest potential for social impact may not necessarily garner the largest acceptance in the marketplace. The finding that producers are more likely to adopt less-stringent and more-promotionoriented certifications has implications for the social outcomes of these programs. For instance, in order to increase their installed base, organizations might focus their resources on promotion, while relaxing their stringency levels. Therefore, there is a serious potential for a “race to the bottom.” Another consequence of having multiple certification programs is that producers in developing countries exporting to various markets or selling to different buyers will face pressures to adopt multiple certifications. This derives from the fact that certifying organizations are trying to “lock in” important industry buyers to require a specific certification program among suppliers. Consumers in different geographical regions are also more likely to recognize those labels that invest heavily in promotion. Thus, an unintended consequence of having multiple competing certification programs could be that they end up “squeezing” exporting firms in developing countries seeking to sell their products in multiple markets or sell to different buyers. That is, producers have to adopt multiple programs and assume the costs of implementation without a guaranteed reward in the marketplace. This study provides useful insights for certifying organizations, buyers, and producers. Certifying organizations interested in increasing their installed base could enter early, invest in promotion, or use an industry association as a mechanism to improve the diffusion of their

36

standard. Buyers and consumers could be more cognizant of the stringency of the certification programs they are requiring or demanding among producers. However, these agents do not have the motivation and incentives to invest their time and resources to sort out this information and provide higher payoffs for firms with more-stringent certifications. Limitations and Future Research This study is not without limitations. First, these results are derived from a cross-section analysis for 2008. Thus, I was not able to analyze the effects of the other variables that have been shown—and that were confirmed by the interviewed farmers—to influence adoption. These variables include the installed base of a certification program at the time of adoption and the certifications that the firm had adopted previously. A panel dataset that includes the date of adoption for each firm would allow this analysis. Similarly, the nature of the data does not allow me to observe if firms drop the certification—i.e., decertify. Second, even though the paper shows that firms are more likely to adopt less-stringent certifications, I do not observe the firms’ environmental and labor performance. These data would facilitate further inquiry on which firms are choosing which types of certifications. The value of these programs to distinguish between firms with different environmental and labor performance could be better assessed with this information. I would like to propose three avenues for future research. First, given the important role that buyers—both retailers and wholesalers—play in producers’ certification choice, it is necessary to also understand how these agents choose among the different options. For instance, why did Unilever decide to require its tea suppliers to certify with Rainforest Alliance certification instead of Fair Trade? Why did Starbucks develop its own certification program for its coffee suppliers instead of requiring a standard that already existed, as Kraft and Sara Lee do?

37

Multinationals’ sourcing practices can have significant social impacts in the countries where suppliers operate. By identifying the different strategies that MNC can use to implement their responsible sourcing practices, we can help predict the diffusion dynamics and social impact of different certification program(s). Future research should also look at adoption of certification programs in a dynamic context. First, firms not only get certified, but also get de-certified. They can be de-certified by the certification organization because of non-compliance, or they can decide not to renew the certification because they believe it is not paying off. Second, certification organizations can modify their standards and monitoring mechanism by decreasing or increasing their demands, which can attract or discourage certain types of firms from adopting the program. Likewise, the promotion strategies of certifying organizations vary over time. Finally, there might be consolidation among certifying organizations that results in significant increases of installed base. For example, the Florverde certification—sponsored by the Colombian industry association—was homologated with GlobalGap (GG)—a European eco-label—so that firms that are awarded the former automatically get the latter. Another avenue for future research is the relationship of these certifications with government regulation. Whether these programs crowd out government intervention should be explored to increase our understanding of their welfare implications.

38

REFERENCES Allen, T. J., Lee, D. M. S., & Tushman, M. L. 1980. R&D performance as a function of internal communication, project-management, and the nature of the work. IEEE Transactions on Engineering Management, 27, 2–12. Anderson, S.W., Daly, D., and Johnson, M.F. (1999) Why Firms Seek ISO 9000 Certification: Regulatory Compliance Advantage? Production and Operations Management, 8(1), 1843. Arthur, W.B. (1989) Competing Technologies, Increasing Returns, and Lock-in by Historical Events, Economic Journal, 99: 116-131. Barnett, M. L. and King, A. A. (2008) Good fences make good neighbors: A longitudinal analysis of an industry self-regulatory institution. Academy of Management Journal, 51(6), 1150-1170. Blackman, A. (2008) Can voluntary environmental regulation work in developing countries? Lessons from case studies. Policy Studies Journal, 36(1), 119-141. Blackman, A. and Rivera, J (2010) ‘The Evidence Base for Environmental and Socioeconomic Impacts of “Sustainable” Certification’, Resource for the Future, Discussion Paper 10-17. Blackman, A., Uribe, E., van Hoof, B. and Lyon, T.P. (2009) Voluntary Environmental Agreements in Developing Countries: The Colombian Experience, Resource for the Future Report. Brynjolfsson, E. (1996). Network externalities in microcomputer software: An econometric analysis of the spreadsheet market. Management Science, 42:1627-1648. Christmann, P. andTaylor, G. (2001) Globalization and the environment: Determinants of firm self-regulation in china. Journal of International Business Studies, 32(3), 439-458. Christmann, P. andTaylor, G. (2006) Firm self-regulation through international certifiable standards: Determinants of symbolic versus substantive implementation. Journal of International Business Studies, 37(6), 863-878. Corbett, C. J. (2006) Global diffusion of iso 9000 certification through supply chains. M&SomManufacturing andService Operations Management, 8(4), 330-350. Corbett, C. J. and Kirsch, D. A. (2001) International diffusion of ISO 14000 certification. Production and Operations Management, 10(3), 327-342. Corbett, C. J., Montes-Sancho, M. J. and Kirsch, D. A. (2005) The financial impact of iso 9000 certification in the united states: An empirical analysis. Management Science, 51(7),

39

1046-1059. David, P.A. (1986) Understanding the Economics of QWERTY: the Necessity of History. In Parker, W.N. (ed.) Economic History and the Modern Economist. Basil Blackwell, Oxford, pp. 30-49. Delmas, M. and Montiel, I. (2008) The Diffusion of Voluntary International Management Standards: Responsible Care, ISO 9000, and ISO 14001 in the Chemical Industry, The Policy Studies Journal, 36(1): 65-93. Delmas, M. and Montiel, I. (2009) Greening the supply chain: When is customer pressure effective ? Journal of Economics andManagement Strategy, 18(1), 171-201. Farrell, J. and Saloner, G. (1985) Standardization, Compatibility, and Innovation, RAND Journal of Economics, 16: 70-83. Fischer, C. and Lyon, T. P. (2008) Competing environmental labels. Resources for the Future Working Paper. Presented at EPA-NCER Workshop on Environmental Behavior and Decision-Making: Corporate Environmental Behavior, U.S. Environmental Protection Agency (EPA) National Center for Environmental Research (NCER), New York, NY, January 15th, 2008. Garcia, S. (2009) Owner and General Manager, Sisapamba S.A. Personal interview. April, 2009. Garud, R., Jain, S. and Phelps, C. (1998). Organizational Linkages and Product Transience: New Strategic Imperatives in Network Fields. Advances in Strategic 15: 205-267. Gereffi, G., Garcia-Johnson, R., and Sasser, E. (2001). ‘The NGO-Industrial Complex.’ Foreign Policy, 125: 56-65. Gladwell, T. (2002). Tipping point: How little things can make a big difference. Boston: Little, Brown. Gowrisankanran, G. and Stavins, J. (2004). Network externalities and technologoy adoption: Lessons from electronic payments. RAND Journal of Economics, 35:260-276. Hamilton, I., Godfrey, F. E., & Linge, J. R. 1979. Industrial systems (vol. 1 of I. Hamilton & J. R. Linge (eds.), Spatial analysis, industry and the industrial environment. New York: Wiley. Harbaugh, R., Maxwell, J.W., and Roussillon, B. (2010) Label Confusion. Working Paper. Presented at Informing Green Markets Conference, Ross School of Business, Michigan University, June 18, 2010. Henao, A. (2009) Sales Manager, Flores de la Montaña. Personal interview. April, 2009. 40

Hoetker, G. (2007) The Use of Logit and Probit Models in Strategic Management Research: Critical Issues. Strategic Management Journal, 28: 331-343. Katz, M. and Shapiro, C. (1985) Technology Adoption in the Presence of Network Externalities, Journal of Political Economy, 94: 822-841. Katz, M. and Shapiro, C. (1986). Technology adoption in the presence of network externalities. Journal of Political Economy, 94:822-841. Katz, R., and Tushman, M. (1979). Communication patterns, project performance, and task characteristics: An empirical evaluation and integration in an R&D setting. Organizational Behavior and Human Performance, 23: 139–162. Khanna, M. (2001) Non-mandatory approaches to environmental protection. Journal of Economic Surveys, 15(3), 291-324. King, A. A. andLenox, M. J. (2000) Industry self-regulation without sanctions: The chemical industry's responsible care program. Academy of Management Journal, 43(4), 698-716. King, A. A., Lenox, M. J. and Terlaak, A. (2005) The strategic use of decentralized institutions: Exploring certification with the iso 14001 management standard. Academy of Management Journal, 48(6), 1091-1106. King, A.A., Lenox, M.J., & Barnett, M. 2002. Strategic responses to the reputation commons problem. In A.J. Hoffman & M.J. Ventresca (eds.), Organizations, Policy and the Natural Environment: Institutional and Strategic Perspectives: 393-406. Stanford, CA: Stanford University Press. Lane, C., and Bachman, R. (1996). The social constitution of trust: Supplier relations in Britain and Germany. Organization Studies, 17: 365–395. Lenox, M., and Nash, J. (2003). ‘Industry self-regulation and adverse selection: A comparison across four trade association programs’. Business Strategy and Environment, 12/6: 343356. Lilien, G.L. and Yoon, E. (1990). The timing of competitive market entry: An exploratory study of new industrial products. Management Science, 36:568-585. Lyon, T.L. and Maxwell, J.W. (2002) "Voluntary Approaches to Environmental Regulation: A Survey." In Maurizio Franzini and Antonio Nicita (eds.), Economic Institutions and Environmental Policy: Post Present and Future, Aldershot, Hampshire, UK: Ashgate Publishing Ltd.

41

Mariotti, S., and Piscitello, L. (1995). Information costs and location of FDIs within the host country: Empirical evidence of Italy. Journal of International Business Studies, 26: 815– 841. Matten D. and Crane, A. (2005) Corporate Citizenship: Toward an Extended Theoretical Conceptualization, Academy of Management Review, 30 (1) 166-179. Mathews, J.T. (1997) Power Shift, Foreign Affairs, January/February: 50-66. Nagard-Assayag, E.L. and Manceau, D. (2001). Modeling the impact of product preannouncements in the context of indirect network externalities. Research in Marketing, 18:203-219. Nicklain, A. (2009). Consultant and auditor. Personal interview. April, 2009. O'Rourke, D. (2003) Outsourcing regulation: Analyzing nongovernmental systems of labor standards and monitoring. Policy Studies Journal, 31(1), 1-29. Potoski, M. andPrakash, A. (2004) Regulatory convergence in nongovernmental regimes? Crossnational adoption of iso 14001 certifications. Journal of Politics, 66(3), 885-905. Rivera, J. (2002) Assessing a voluntary environmental initiative in the developing world: The Costa Rican certification for sustainable tourism. Policy Sciences, 35(4), 333-360. Rivera, J. andDe Leon, P. (2004) Is greener whiter? Voluntary environmental performance of western ski areas. Policy Studies Journal, 32(3), 417-437. Rivera, J., De Leon, P. andKoerber, C. (2006) Is greener whiter yet? The sustainable slopes program after five years. Policy Studies Journal, 34(2), 195-221. Rogers, E. (1995).Diffusion of innovations. Free Press, New York. Salazar, I. (2009) Director, Ecuadorian Spirit and Manager, Roses & Roses. Personal interview. March and April, 2009. Scherer, A.G., Palazzo, G., and Baumann, D. (2006) Global Rules and Private Actors: Towards a New Role of the Transnational Corporation in Global Governance, Business Ethics Quarterly, 16: 505-532. Schilling, M.A. (1999). Winning the standards race: Building installed base and the availability of complementary goods. European Management Journal, 17:265-274. Schilling, M.A. (2002). Technology success and failure in winner-take-all markets: Testing a model of technological lock out. Academy of Management Journal, 45: 387-398. Schilling, M.A. (2002) Technology Success and Failure in Winner-Take-All Markets:

42

The Impact of Learning Orientation, Timing and Network Externalities, Academy of Management Journal, 45 (2): 387-398. Schilling, M.A. (2003) Technological Leapfrogging: Lessons from the U.S. Video Game Console Industry, California Management Review, 45 (3): 6-32. Schilling, M.A. (forthcoming). To Protect or to Diffuse? Appropriability, Network Externalities, and Architectural Control, in A. Gawer (Ed.) Platforms, Markets and Innovation. Northampton, Mass: Edward Elgar. Shurmer, M. (1993). An investigation into sources of network externalities in the packaged PC software market. Information Economics and Policy, 5:231-252. Smith, T.M. and Fischlein, M. (2010) Rival private governance networks: Competing to Define the Rules of Sustainability Performance. Global Environmental Change. Spence, M. (1973). Job Market Signaling. Quarterly Journal of Economics, 87(3): 355–374 Terlaak, A. and King, A. A. (2006) The effect of certification with the iso 9000 quality management standard: A signaling approach. Journal of Economic Behavior andOrganization, 60(4), 579-602. Wade, J. (1995). Dynamics of organizational communities and technological bandwagons: An empirical investigation of community evolution in the microprocessor market. Strategic Management Journal, 16:111-134. Wall, E., Weersink, A., and Swanton, C. (2001) Agriculture and ISO 14000, Food Policy, 26(1), 35-48. Wiersema, M.F. and Bowen, H.P. (2009) The Use of Limited Dependent Variable Techniques in Strategy Research: Issues and Methods. Strategic Management Journal, 30: 679-692. Witt, U. (1997). “Lock-in” vs. “critical masses” – industrial change under network externalities. International Journal of Industrial Organization, 15:753-773. Yeung, G. andMok, V. (2005) What are the impacts of implementing ISOs on the competitiveness of manufacturing industry in china? Journal of World Business, 40(2), 139-157.

43

Table 1: Comparison of certified versus non-certified firms Variable 1. Size (total exports $) 2. Exports Europe (%) 3. Exports roses (%) 4. Colombia (dummy) 5. Membership industry association (dummy)

Certified (n=205)

Non-certified (n=1036)

4004192 (466741) 0.116 (0.011) 0.716 (0.027) 0.434 (0.035)

780774 (69057) 0.124 (0.011) 0.557 (0.014) 0.646 (0.015)

0.844 (0.025)

0.137 (0.011)

p-values 0.000 0.673 0.000 0.000 0.000

( ) standard deviation

44

Table 2: Summary statistics

Firm characteristics Firm size (total exports) Exports USA Exports Europe Certification programs FlorVerde / FlorEcuador Milieu Project Sierteelt Veriflora Rainforest Alliance Fair Trade Flower Label Program Fair Flowers Fair Plants # of certifications # of certified firms Certification characteristics FlorVerde / FlorEcuador Milieu Project Sierteelt Veriflora Rainforest Alliance Fair Trade Flower Label Program Fair Flowers Fair Plants

Colombia n=758 $ % exports mean mean 1,430,374 (4,188,869) 1,119,790 70.71 (3,856,893) (39.37) 148,685 13.25 (423,879) (29.14) # adoptions 65 54.17 1 0.83 5 4.17 46 38.33 2 1.67 0 0.00 1 0.83 120 100.00 89 Difficulty of Independency technical norm audit 0.60 0/0 0.40 1 0.80 1 0.65 1 1.00 1 0.75 1 0.90

1

Ecuador n=483 $ % exports mean mean 1,129,436 (2,363,471) 812,780 70.21 (1,932,786) (27.77) 148,430 10.57 (569,120) (18.07) # adoptions % 91 46.67 12 6.15 15 7.69 18 9.23 8 4.10 47 21.54 4 2.05 195 100.00 116 Portfolio Market domain certifications 1/1 Colombia/Ecuador 2 Netherlands 2 USA and Canada 34 USA 20 Switzerland 1 Germany, Austria, Switzerland 2 Sweden, Denmark, Norway, Germany, Austria, Netherlands

Aggregate market size (1) 0/0 0.896 7.176 6.180 0.616 3.817

Time of entry organization 1998/2001 2005 2005 2006 2002 1997

4.994

2005

(1) in billions of Euros

45

Table 3: Correlation Table Description

Mean

s.d.

Min

1. Size 2. Exports Europe

Natural log of total exports Percentage of a firm's exports to Europe

11.75 0.12

3 0.25

-1.43 17.99 1 0 1 -0.08

3. Exports roses

Percentage of firm's exports of roses

0.58

0.45

0

1

0.14

-0.13

1

4. Colombia

Whether a firm is located in Colombia or Ecuador

0.61

0.49

0

1

-0.13

0.05

-0.67

1

5. Age of certification Difference between 2008 and the year when certification was launched

6.26

3.26

3

12

-0.01

0

-0.04

0.06

1

6. Independence (third-party certification)

Whether a certification is from a third-party or from the industry association

0.86

0.35

0

1

0

0

0

0

-0.45

1

7. Difficulty of technical norm

How difficult would it be for an average farm in the industry to implement the certification's requirements

0.75

0.18

0.4

1

-0.01

0

-0.06

0.1

0.25

-0.01

1

8. Portfolio of products

Number of product lines that the certifying organization certifies

8.86

12.08

1

34

0

0

0

0

-0.36

0.27

0.08

1

9. Market domain

Percentage of firm's exports to the certification's market domain

0.22

0.38

0

1

0.02

-0.1

-0.01

0

-0.48

0.24

-0.08

0.4

1

10. Aggregate market size

3.38 Flower's market size (billions of euros) of certification's market domain

2.68

0

7.18

0

0

0

0

-0.45

0.52

0.09

0.19

0.68

1

11. Membership

Firm membership to industry 0.25 association

0.44

0

1

0.47

0.02

0.1

-0.18 -0.01

0

-0.02

0

0.01

0

12. # members certified with same program

Number of member firms that have adopted same certification

25.24

0

86

0.02

-0.01

0.12

-0.18

18.9

Max

1

2

3

4

5

6

7

8

9

10

11

12

1

0.36

46

1

-0.86 -0.13 -0.01 -0.05 -0.32 0.03

1

Table 4: Fixed effects logistic regression model

Age of certification Independence

1

2

3

4

5

6

0.050* (0.025) -2.431*** (0.186)

0.078** (0.026) -2.311*** (0.187) -1.623*** (0.441)

0.225*** (0.036) -2.585*** (0.192) -3.091*** (0.567) 0.064*** (0.008)

0.462*** (0.051) -3.118*** (0.225) -4.170*** (0.842) 0.032*** (0.008) 4.066*** (0.482)

0.253*** -0.033 -3.512*** -0.277 -3.420*** -0.671 0.051*** -0.008

0.293*** -0.052 -0.849* -0.358 -2.105* -0.869 0.014 -0.009 3.148*** -0.468

Difficulty of technical norm Portfolio of products Overlap export market domain Aggregate market size

0.273*** -0.053

# members with certification Number of observations Log Likelihood LR Chi Square 2

McFadden's R

1428 -347 298***

1428 -340 311***

1428 -306 306***

1428 -253 485***

1428 -289 412***

0.052*** -0.007 1428 -214 563***

0.300

0.314

0.382

0.490

0.416

0.568

( ) standard deviation * p < .05 ** p < .01 *** p < .001

47

Table 5: Analysis through multiple model specifications

constant firm characteristics Size

% of exports Europe

% of exports roses

Colombia

Membership industry assoc.

certification characteristics Age of certification

Logit model (clustered se)(1) -12.265*** -1.893

Fixed effects model (2)

0.608*** -0.134 [0.006] -0.196 -0.682 [-0.002] 0.148 -0.466 [0.002] -0.671 -0.384 [-0.008] 3.677*** -0.425 [0.139] 0.428*** -0.049

Independence

Difficulty of technical norm

-3.704*** -0.891

Portfolio of products

0.032*** -0.008

Market domain

3.661*** -0.466

# members with certification

constant (2nd stage)

Logit model (clustered se) -17.224*** -1.461

Random effects model -19.471*** -1.587

Heckman probit (3) -6.852*** -0.954

0.769*** -0.083 [0.001] 0.363 -0.462 [0.001] 0.897** -0.311 [0.002] -0.443 -0.266 [-0.001] 0.536* -0.244 [0.001]

0.874*** -0.091 [0.001] 0.35 -0.618 [0.000] 1.025** -0.397 [0.001] -0.455 -0.326 [-0.000] 0.53 -0.321 [0.000]

0.376*** -0.069 [0.000] -0.193 -0.32 [0.000] 0.487* -0.202 [0.000] -0.157 -0.175 [0.000] 1.389*** -0.135 [0.000]

0.292*** -0.045 [0.001] -0.752* -0.299 [-0.002] -1.122 -0.838 [-0.002] 0.021** -0.007 [0.000] 2.911*** -0.391 [0.005] 0.042*** -0.005 [0.000]

0.321*** -0.048 [0.000] -0.859* -0.35 [-0.001] -1.458 -0.838 [-0.001] 0.022** -0.008 [0.000] 3.253*** -0.42 [0.002] 0.050*** -0.006 [0.000]

0.127*** -0.022 [0.019] -0.818*** -0.201 [-0.180] -1.025** -0.335 [-0.156] 0.009* -0.004 [0.001] 1.322*** -0.181 [0.201] 0.024*** -0.003 [0.004] -1.222** -0.385 8687 -2529 317*** 7252 1435

Number of observations 8687 1428 8687 8687 Log Likelihood -1637 -253 -709 -688 LR Chi Square 152*** 485*** 738*** 390*** Censored observations Uncensored observation ( ) standard deviation, [ ] marginal effects, * p < .05, ** p < .01, *** p < .001 (1) The dependent variable in this model only considers adoption from an industry association program. (2) The dependent variable in this model only considers adoption of non-industry association programs. Independence and # members with certification drop out of the analysis they are not relevant if only considering non-industry certifications. (3) First stage's dependent variable is adoption (of any certification) and firm characteristics. Second stage's dependent variable is choice of certification and certification characteristics

48

Table 6: Geographic location of certified firms in Ecuador, 2010 [A graph with firms’ location and certifications will replace this table in order to identify clustering among certified firms] # farms in province 18

3

9

2 82

7

State AZUAY

# farms in city

City

3 4 1 7 1 2

CUENCA GUALACEO ONA PAUTE SAN FERNANDO SANTA ISABEL

2 1

BIBLIAN DELEG

CAÑAR

CARCHI 1 5 1 2 CHIMBORAZO 2 COTOPAXI 70 6 6 GUAYAS 1 1

BOLIVAR ESPEJO MIRA MONTUFAR RIOBAMBA LATACUNGA PUJILI SALCEDO GRAL. ANT. ELIZALDE GUAYAQUIL

FlorEcuador 3 0 1 0 2 0 0 2 2 0 1 0 1 0 0 0 0 19 15 1 3 1

Fair Trade 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 1 1 0

0 0

0 0

FFP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 1 1 0

Rainforest Alliance 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 3 1 1 0

0 0

0 0

49

Veriflora 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 1 1 0

Flower Label Program 5 0 2 0 1 1 1 1 1 0 0 0 0 0 0 0 0 18 16 1 1 0

MPS 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 1 1 0

0 0

0 0

0 0

3 1 1

MILAGRO NARANJAL SAN JACINTO

1 0 0 0 0 0 0 0 0 0 0 0 23 IMBABURA 4 1 0 2 4 ANTONIO ANTE 0 0 0 0 8 COTACACHI 2 0 0 2 1 IBARRA 0 0 0 0 5 OTAVALO 1 1 0 0 5 URCUQUI 1 0 0 0 228 PICHINCHA 80 3 4 18 62 CAYAMBE 28 2 2 6 13 MEJIA 6 0 0 1 90 PEDRO MONCAYO 23 1 2 6 62 QUITO 23 0 0 5 1 RUMINAHUI 0 0 0 0 1 SANTA ELENA 0 0 0 0 1 SANTA ELENA 0 0 0 0 3 TUNGURAHUA 1 0 0 0 2 AMBATO 1 0 0 0 1 PILLARO 0 0 0 0 376 TOTAL 111 7 7 25 Note: A more systematic analysis of this data as well as a graphic representation of firm locations will be conducted.

50

0 0 0 0 0 0 0 0 0 12 4 1 5 2 0 0 0 0 0 0 14

0 0 0 3 0 0 0 3 0 22 9 2 9 2 0 0 0 0 0 0 49

0 0 0 0 0 0 0 0 0 10 6 0 4 0 0 0 0 0 0 0 13