Supplier development to improve supplier sustainable performance: a test on a dairy supply chain of a developing country Open Access

As good corporate citizens and to gain a competitive advantage, manufacturers focus on improving productivity and performance across their supply chain (SC) (Kumar and Goswami, 2019; Carter and Rogers, 2008). Within the realm of supply chain management (SCM), it has been cogently argued that a manufacturing firm can gain a competitive advantage over its competitors by developing the performance of its core suppliers—a concept known as supplier development (SD) (Krause et al., 2009; Wagner, 2009). Here, one assumption is that when performing without the buyer’s support, suppliers will not have optimal capability to generate customer value—the customer being the buyer—required for the SC to function optimally (Krause et al., 2009). Another assumption is that, while SD programs are initiated by the buyer for its strategic gain, the suppliers' improved performance in terms of quality, delivery, and other supply parameters would improve the supplier’s business performance. The concept of sustainable supplier development (SSD) extends traditional SD by incorporating programs aimed at improving suppliers' economic, environmental, and social outcomes, in addition to traditional performance parameters such as quality and delivery. SSD encourages manufacturers—as focal companies—to adopt a triple-bottom-line (TBL) approach to supplier performance improvement expectations (Seuring and Müller, 2008; Carter and Rogers, 2008; Bai and Satir, 2022; Pedroso et al., 2021). In this study “sustainable performance” refers to a farmer’s performance across these three dimensions. Bai and Satir (2022, p. 1) defined SSD as “any initiative taken by the buying firm to improve their supplier sustainability capabilities to meet two or more elements of the triple bottom line (TBL) of multiple stakeholders along the supply chain (suppliers, buying firms, customers, etc.).”

Despite substantial theory development on SSD recently, what is not known well is how SSD applies to an agri-food SC, particularly in a developing economy. Farmers supply their produce to a processor, who in effect becomes the buyer. Thus, in an agri-food context, SSD translates to processor-led farmer development (FD). Specifically, this study aims to develop a theoretical model and empirically evaluate it to demonstrate how SSD initiatives of a milk processor (i.e. FD) improve the sustainable performance of dairy farmers in a specific context.

In a non-perishable goods SC, when SSD serves as a manufacturing strategy, such as in lean manufacturing, supplier selection plays a crucial role. This is because the buying firm can select the optimal supplier (or few suppliers) for purchasing a given component or raw material, by considering multiple criteria such as quality of goods supplied, product reliability, delivery reliability, and the carbon footprint (Krause et al., 2007; Benton et al., 2020). Under these circumstances, strategically important suppliers would receive more direct support from the buyer (Wagner, 2009; Krause et al., 2007); such situations result in maintaining a strong, mutually reinforcing collaborative relationship between the two parties to ensure an uninterrupted supply of high-quality goods and services (Benton et al., 2020; Krause et al., 2007). In contrast, perishable goods SCs, such as agri-food SCs, face unique challenges—perishability, high waste, quality variability, seasonality, and vulnerability to natural events—worsened in developing economies by fragmented SCs, limited resources, and poor support, disproportionately impacting smallholder farmers (Susanty et al., 2019; Yawar and Seuring, 2020).

Among many agents who can be involved in improving the performance of smallholder farmers in a developing economy, privately-owned processors as the focal companies, have an incentive to develop their farmers for mutual benefits when the involvement of other agencies on FD is minimal (Yawar and Seuring, 2018; Mukucha and Chari, 2021). This situation stands partially analogous to SSD in non-perishable goods SC, the major difference being the presence of potentially thousands of suppliers (mostly smallholder farmers) of the same commodity. Given the aim of the study, its central proposition is: “In situations where there are minimal SSD initiatives from other agents, SSD initiatives undertaken by a privately-owned processor can result in improved sustainable performance of farmers in a developing country.” In the study context, FD and SSD overlap significantly as theoretical concepts. A dairy SC of a developing country was specifically chosen as the context to develop and test the theoretical model because dairy SCs in developing countries have received great attention due to issues such as food self-sufficiency, population increase, economic growth, and the resulting high demand (Susanty et al., 2018; Bhat et al., 2022).

The first question that needs an answer before developing and testing a theoretical model that underpins the above central proposition is: what hinders dairy farmer performance—product quality, delivery, and other supply parameters important to the buyer—in a developing country agri-food SC? For the dairy SC under consideration, the main hindrances have been the lack of high-quality inputs (especially silage) for cows, poor dairy management practices, inadequate extension services, and a reluctance to enforce a national milk quality standard. The second question that needs an answer is: what is the theoretical relationship between FD initiatives and the sustainable performance of farmers, under a scenario of SSD? If SSD takes place, given the definition of SSD, when SSD increases, at least two of the three TBL outcomes of the farmers must increase. However, the scope of a business can limit what it can achieve in each TBL dimension (Carter et al., 2020; Bai and Satir, 2022). Although the literature supports the notion that a milk processor's efforts to develop their farmers improves the socioeconomic status of the farmers (Brix-Asala et al., 2021; Yawar and Kauppi, 2018), what is unclear is the link between SSD initiatives (cause) and the TBL performance of the farmers (effect). The present study introduces farmer capability (FC) and the processor-farmer relationship (PFR) as mediators (intervenors) that mediate the relationship between SSD initiatives and the TBL performance of the farmers.

The remainder of this paper is structured as follows: the next section reviews the literature leading to theory development. Section 3 presents the methodology. Section 4 covers the results. Section 5 discusses the findings. Finally, section 6 concludes the study, outlining key implications, study limitations, and suggestions for further research.

The development of the concept of SD has progressed through three stages. During the initial stage, the focus was on providing direct assistance to suppliers—sometimes known as direct SD—to improve the quality of goods supplied through supplier performance improvement. Leenders (1966) was a key figure during that era. During the second stage, which coincided with the quality movement (1980–2000) and was driven by principles advocated by quality pioneers such as Deming (1986), the focus shifted toward fostering long-term buyer-supplier relationships, while retaining the core characteristics and motives of direct SD. The direct SD-indirect SD dichotomy emerged in this era. In indirect SD, the buyer commits only extremely limited resources to improve the supplier's performance. In contrast, in direct SD, the buyers take an active role and commit more resources to enhance suppliers' performance. The work of Krause et al. (2007) and Wagner (2009) was significant during this period. In the third stage (2000 onwards), the focus shifted towards SSD due to parallel theory development on sustainable supply chain management by key authors such as Seuring and Müller (2008), Carter and Rogers (2008). Aligning the TBL framework (introduced by Elkington, 1994) with sustainable supply chain management as a tool for measuring a supplier’s sustainability was a significant development in the third stage through the works of Pagell and Wu (2009), Pedroso et al. (2021), and others. In addition, the works of Yawar and Kauppi (2018) and Yawar and Seuring (2018) were particularly important, as these studies focused on SSD in the agri-food SC of developing economies.

As a profit-optimizing firm, looking through an instrumental stakeholder theory lens (Donaldson and Preston, 1995), the processor needs to sustain an optimum level of SSD and relationships—not too much and not too little—with individual farmers, whose developmental needs may vary (Yawar and Seuring, 2018; Benton et al., 2020). From an agri-food produce perspective, this means the delivery of high-quality produce by the processor in the right quantity at the right time with minimal variability. This justifies why a processor of agri-food produce may develop its farmers to improve their performance, while maintaining good relationships with them.

Several large sample empirical studies have examined the relationship between SD initiatives and suppliers' performance and/or the buyer’s performance (e.g. Benton et al., 2020; Lee et al., 2018; Nagati and Rebolledo, 2013; Wagner, 2009). In general, these studies support the SD initiatives → supplier (or buyer) performance causal links. It is observed that: (1) the number of indicators used to operationalize SD initiatives depends on the context; (2) supplier performance revolves around product quality, delivery, price, and the service aspects; and (3) no agri-food SCs featured in these studies.

Some notable studies that examined the relationship between SD initiatives aimed at improving the supplier’s sustainable performance (e.g. Sancha et al., 2019; Yang and Zhang, 2017) show two research gaps: (1) focusing on only one of the three TBL dimensions in the research design, and (2) ignoring the influence of confounding variables (no use of control variables seem to have been used in the data analysis). The limited number of case studies and longitudinal studies show some insight into the effects of SD initiatives aimed at improving the sustainable performance of the suppliers. Through a case study design, Liu et al. (2018) found that SD initiatives can improve a supplier’s sustainable performance—environmental performance and compliance with social sustainability norms—when sustainability facilitators and inspectors are deployed in the SC, but not otherwise. Through a longitudinal quantitative study Bartos et al. (2024), found that buyer-led training, such as web-based training, assessment sessions, supplier days, and workshops, can lead to sustainable supplier performance (the supplier implementing policies and procedures on business ethics, social and environmental performance). However, the study emphasized that training needs to be well-structured and context-specific to achieve better results.

The phrase ‘buyer-initiated’ emphasizes that the FD is analogous to SD. Farmer training, credit support, providing input materials, evaluation and feedback on both performance and production practices, and ensuring a guaranteed market and a price were found to be commonly implemented FD initiatives in developing countries (Mukucha and Chari, 2021, 2024; Yawar and Kauppi, 2018). Of these, training and financial support appeared as the commonly used FD forms.

Mukucha and Chari (2021, 2024) found that in Zimbabwe, contract farmers in tobacco and cotton—those engaged by merchants to supply produce—outperformed regular farmers in yield, delivery, quality, and target achievement, due to substantial extension services provided by buyers. However, the two studies lacked clarity on the magnitude of performance differences, the control variables being used (if any), and the farmer social and environmental performance outcomes.

Using a case study of the Indian dairy industry, Yawar and Kauppi (2018) found that private dairies implement SSD initiatives like those of cooperative dairies, driven by institutional pressures—a concept grounded in institutional theory (DiMaggio and Powell, 1983). They also found that private dairies implement SSD initiatives for corporate gain while cooperatives do so due to social and legitimacy concerns (see Yawar and Kauppi, 2018, p. 170 for the specific SSD initiatives). Using case studies of a Kenyan dairy and a Ugandan pineapple SC, Brix-Asala et al. (2021) developed a framework showing how bottom-of-the-pyramid farmers—those in the lowest socio-economic tier—can be integrated into formal SCs, improving functional coordination and sustainable performance while resolving paradoxical outcomes for both buyers and farmers.

In developing the research model from the extant literature, three considerations were made: (1) clearly distinguishing traditional supplier performance from sustainable performance based on TBL dimensions, and assigning appropriate labels; (2) tailoring operational definitions of theoretical constructs to the agri-food context; and (3) ensuring a balanced definition of SD that includes initiatives targeting both traditional performance and sustainability outcomes. For instance, supplier training (as a farmer development initiative) should enhance both FC and sustainable farming practices.

Based on the literature, this study considers three categories of FD initiatives: farmer training (FT)–providing training and education to dairy farmers to improve their farming skills; financial support (FS)–a broad category which refers to providing equipment, production input material (e.g. animal feed), affordable loans, a favorable and expeditious payment system for the produce; and evaluation and feedback on farmer’s quality performance (EFFQP)–providing continuous evaluation and feedback on the quality of milk being supplied by the farmer (DeSilva et al., 2023; Yawar and Seuring, 2018; Brix-Asala et al., 2021; Pedroso et al., 2021; Chopde et al., 2019).

For this study, sustainable farmer performance is captured through TBL dimensions which serve as formative constructs (see Diamantopoulos and Winklhofer, 2001; Hair et al., 2022 for details on formative constructs). Consequently, the role of the indicators of each TBL dimension is to form the construct to estimate its score; thus, by default, the indicators of a construct become its causes/elements. This means that for example, while the research model would theorize that FD initiatives would positively impact a farmer’s social performance—through a well-specified causal path mechanism—that does not necessarily mean that the resulting improvement of the farmer’s social performance causes improvements in any specific indicator (e.g. education given to the children) that was used to operationalize the farmer’s social performance (see direction of causal predictive arrows shown in Figure 3 for farmer’s TBL outcomes). Among the extant literature (for example Zanin et al., 2020; DeSilva et al., 2023; Lebacq et al., 2013), Zanin et al. (2020) was the most widely used in this study to operationalize farmer’s TBL performance (with suitable adaptations to account for the social aspects of Sri Lankan farmers). In the study context, economic sustainability (ES) is a collection of financial measures relevant to the farm business, social sustainability (SS) is a collection of social measures (quality of life of the family, a modified version of personal satisfaction and personal socialization referred to in Zanin et al., 2020), and environmental sustainability (EnvS) is a collection of environmentally sustainable farming practices.

In the study context, FC reflects a farmer being able to practice what was learned from formal training, maintaining relationships with extension service providers, delivering milk quantities the processor envisages, and maintaining milk quality specifications (DeSilva et al., 2023). Product quality and delivery performance that FC reflects are two universal manifestations of supplier performance. The other two facets of FC— being able to practice what a farmer learned from formal training and maintaining relationships with extension service providers—are contextualized manifestations of supplier performance reflecting continuous improvement. By labeling the traditional notion of farmer performance as FC, the paper reliably distinguishes it from what a farmer achieves for themselves economically, socially, and environmentally.

In a dairy SC, the processor obtains its primary production from thousands of farmers to optimize the utilization of their plants to meet the customer demand. Hence, the processor needs to manage their large smallholder farmer base to ensure its productivity by establishing a steady supply of quality milk. Thus, relational exchanges in this context—as opposed to close relationships with a selected few critical suppliers, such as in lean manufacturing—require collaborative relationships with a large supplier base for optimal performance (DeSilva et al., 2023; Giraldi et al., 2024). In maintaining collaborative relationships, trust, satisfaction, and commitment were identified as the relevant indicators of the construct ‘buyer-supplier relationship’ (Lees et al., 2020; DeSilva et al., 2023; Moses et al., 2023). In this study, the buyer-supplier relationship is labeled the processor-farmer relationship (PFR). The following subsections cover formulations of the hypotheses.

Processor’s investment in FT specifically for the buyer’s input needs results in asset specificity, ensuring higher buyer and supplier productivity for that specific situation (Mariyono, 2019; Brix-Asala et al., 2021; Yawar and Seuring, 2018). Farmers who have received relevant training are likely to use more food safety and quality practices to ensure milk quality (Korale-Gedara et al., 2023; Kataike et al., 2018; DeSilva et al., 2023) while improving their capacity and productivity (Mariyono, 2019; Susanty et al., 2019; Luther et al., 2018). Therefore:

The processor’s involvement in FS benefits the farmer in several ways (Yawar and Seuring, 2018). FS to purchase machinery and equipment leads to improved FC (Korale-Gedara et al., 2023; DeSilva et al., 2023; Chopde et al., 2019). Farmers who are financially assisted tend to adopt more food safety practices, ensuring the quality of milk (Korale-Gedara et al., 2023). Moreover, industry-specific assistance (e.g. silage for cows and veterinary support) has been shown to benefit dairy farmers by improving their productivity (Brix-Asala et al., 2021). Therefore:

The processor’s involvement in performance monitoring is effective in improving supplier capability (Wagner, 2009). Milk processors test the delivered milk at their collection centers for quality; they provide feedback to address quality issues at the farmer level (Brix-Asala et al., 2021; Wouters et al., 2007). While this feedback system allows farmers to maintain a good rapport with their processor, it also creates a platform to identify milk quality deficiencies and upskilling needs to continue supplying milk to the processor, leading to enhanced capabilities derived from the knowledge and skills gained during the training (DeSilva et al., 2023; Yawar and Seuring, 2018). Therefore:

Long-term buyer-supplier relationships are necessary for business success of both parties (Monczka, 2020; Benton et al., 2020). When a milk processor invests in FT, farmers become more equipped with essential farming skills and knowledge. This, in turn, fosters trust and mutual commitment between the processor and their farmers, thus strengthening the PFR (Yawar and Seuring, 2020; DeSilva et al., 2023; Shukla et al., 2023). Therefore:

FS to improve a farmer’s business enhances a farmer’s trust, commitment, and satisfaction while creating an environment to sustain long-term relationships—on one hand, farmers view their processor as a reliable partner interested in their success, and on the other, it becomes necessary for the processor to continue maintaining the relationship with the farmer to ensure that the investment results in its intended purpose (Yawar and Seuring, 2018; Benton et al., 2020). Additionally, FS can stabilize the milk supply by fostering commitment, as farmers are more likely to maintain a long-term partnership with processors who contribute to their economic resilience. Therefore:

A processor’s feedback on a farmer’s milk quality and through the sharing of information strengthens the tie between the two parties (Wagner, 2009; Yawar and Seuring, 2020). When a processor actively engages in assessing the quality of a farmer’s outputs (i.e. raw milk) and shares tailored and timely feedback, a farmer can gain insights into how they can meet the required product standards and specifications, keeping the farmer interested in sustaining the PFR (Brix-Asala et al., 2021). This collaborative approach builds trust as farmers see processors as committed partners invested in their continuous improvement and success. Additionally, feedback creates transparency and accountability, aligning farmer performance with processor expectations. This alignment creates farmer satisfaction, trust, and long-term commitment towards their processor (DeSilva et al., 2023; Yawar and Seuring, 2018). Therefore:

A more capable supplier will be more knowledgeable about what they need to do to improve their business to meet the processor’s expectations (DeSilva et al., 2023). One such expectation is milk quality. Another expectation is a continuous supply of milk to maintain the smooth and even flow envisaged by the processor. While maintaining a closer relationship with the processor enables the farmer to respond to any operational issue (e.g. nonconformity) more swiftly due to faster information flow, the farmer must have the necessary capability to solve the operational issue(s) to satisfy the processor (Sachitra and Chong, 2018). A capable farmer thus positions themselves as a reliable supplier, reinforcing the processor's confidence in the FC. Consequently, more capable farmers are likely to engage in more productive and collaborative relationships with their milk processors (Shukla et al., 2023). Therefore:

The next two hypotheses, each having three sub-hypotheses, are posited to complete the explanation of SSD.

The more capable the farmer—meaning a farmer who can practice the knowledge gained from training, maintain close relationships with farm extension service providers, provide year-round milk supply in the required quantity, and meet required quality specifications—the more revenue they could earn over costs (DeSilva et al., 2023; Mariyono et al., 2022). Therefore:

A more capable farmer is more able to improve net returns on their investment, resulting in a higher standard of living (Yawar and Seuring, 2018). In addition, the more capable the farmer, the more recognition the farmer receives from fellow farmers as a valuable person who exchanges resources with them (DeSilva et al., 2023; Luther et al., 2018). Therefore:

A farmer who possesses high capability can practice environmentally friendly practices, such as proper waste disposal, the ability to find applications for solid waste, and ways to reduce air pollution to boost environmental performance (Zanin et al., 2020; DeSilva et al., 2023; Korale-Gedara et al., 2023). In a smallholder farmer context, even a simple step such as frequent washing of the cattle shed would reduce air pollution. Therefore:

Relationships can lead to relational rents, meaning above normal profits, and gain a competitive advantage from relational networks both via inimitable resource bundles and relationships (Lees et al., 2020; Benton et al., 2020). Thus, it can be inferred that the processor has an incentive to actively work towards establishing a strong relationship with their farmers; a strong relationship between a processor and the farmer enables a farmer to better identify the processor’s requirements and make a concerted effort to meet the processor’s expectations, which could result in greater economic outcomes to the farmer (DeSilva et al., 2023; Yawar and Seuring, 2020). A stronger PFR may also secure a longer-term purchase contract, greater incentivization for exceeding volume and quality requirements, a better price for milk, reduced income fluctuation, and more help from other parties (e.g. banks, veterinarians) that eventually lead to greater financial outcomes (DeSilva et al., 2023; Lees et al., 2020; Yawar and Seuring, 2020). Therefore:

A stronger PFR improves the quality of life of the farmer and their family, receiving more help from the processor to educate the farmer’s children (e.g. free schoolbooks and uniforms) and community recognition (Yawar and Seuring, 2018). Moreover, farmers who maintain close relationships with their processors are more likely to establish third-party collaborations, further strengthening the farmer’s social network and status within the farming community (DeSilva et al., 2023; Yawar and Seuring, 2018). Therefore:

The PFR contributes to EnvS by facilitating the adoption of environmentally friendly practices and enhancing resource efficiency for the following reasons. A farmer who maintains a strong relationship with their processor is more likely to learn and practice green farming practices as they get more opportunities to learn and reach out for advice (Korale-Gedara et al., 2023; DeSilva et al., 2023). This strong PFR not only helps farmers reduce their environmental footprint but may also bolster the processor’s corporate social responsibility campaign, which may reciprocally benefit the engaged farmer. Therefore:

Hypotheses H1 to H9c collectively represent the explanation and prediction of the SSD phenomenon in the study context, which is diagrammatically shown in Figure 1 as a theoretical model. Three control variables also appear in Figure 1, which are not part of the explanation/prediction of the SSD phenomenon; these control variables serve as predictors in the regression analysis to obtain more precise estimates of the causal effects corresponding to the hypotheses shown above. The middle variables in the above relationships (i.e. FC and PFR) are the mediators.

The theoretical model (Figure 1) specifies only indirect relationships between the three FD initiatives and sustainable farmer performance. Consequently, causation of sustainable farmer performance occurs solely because: an increase (or decrease) of FD initiatives causes FC to increase (or decrease), which in turn causes sustainable farmer performance to increase (or decrease); or, an increase (or decrease) of FD initiatives causes PFR to increase (or decrease), which in turn causes sustainable farmer performance to increase (or decrease). In statistical language, this means that given the specified relationships in the model (Figure 1), supposing direct relationships are specified between the FD initiatives and sustainable farmer performance (in a multiple regression sense, making FD initiatives also as predictors of sustainable farmer performance) for whatever reason, such relationships would be statistically nonsignificant due nonsignificant regression coefficients corresponding to the newly specified direct relationships (see Nitzl et al., 2016, p. 1855 for full details). One can consider testing H10 purely as a robustness check of the theoretical model. To support the null hypothesis that there is no direct relationship between FD initiatives and sustainable farmer performance, the following null hypothesis is specified:

In Sri Lanka the domesticated cattle population amounts to 1,574,918, of which 33% are milking cows that produce approximately 418 million liters of milk (Department of Animal Production and Health (DAPH), 2023). The two districts covered in the fieldwork, Nuwara Eliya (a major district in the upcountry region) and Kurunegala (a major district in the low country region), are the highest milk producing districts in Sri Lanka. Due to differing climatic conditions, the types of cows utilized for milking and the farming methods used vary between the two districts. Due to the cool climate, farmers in Nuwara Eliya primarily utilize exotic high-milk-yielding cows, such as Jersey and Holstein-Friesian breeds. However, unlike in Australia and New Zealand—the countries where the heifers originated—Nuwara Eliya lacks large grazing areas. As a result, farmers rely mostly on commercial dry rations to feed their cows. In contrast, farmers in Kurunegala, who raise local or local-Indian crossbreeds—suited to the hot climate—benefit from the availability of natural forage for their cattle.

Although there are about 21 milk processing companies in Sri Lanka (DAPH, 2023), farmers belonging to one large privately-owned milk processor were selected for this study due to the following three reasons: (1) the processor has its own bank to support their farmers financially (this relates to FS in the theoretical model); (2) the processor has milk collection/chilling centers around the country with a strong presence in Nuwara Eliya and Kurunegala districts; and (3) the processor has a documented FT schedule and deploys its extension officers to provide on-site training and occasionally arrange training sessions through its milk collection centers for farmers who own more than five lactating cows or produce more than 20 liters/day, which is also the farmer inclusion criteria of the study. Participation in the training is quasi-voluntary.

Extant literature was used to identify the indicators of the eight constructs in the theoretical model (see Section 2.4) and were translated into statements ( Appendix 1). Agreement with each statement was sought from the farmers via a five-point scale. In addition, the questionnaire covered a range of farmer demographic questions including the herd size, region and additional training received. The content validity of the questionnaire was established by following three steps: (1) the questionnaire was pretested to four subject/content experts for ratification; (2) the questionnaire was translated into the native languages (Sinhala and Tamil), which were then back-translated by independent translators to ensure equivalence between the Sinhala and Tamil versions of the questionnaire (Fowler, 2013); and (3) the questionnaire was pilot tested using 30 randomly selected farmers. This helped to remove overlapping statements and the reshaping of two statements belonging to SC for contextual relevance; more specifically, the SC indicators identified by Zanin et al. (2020) to cover personal satisfaction and personal socialization (for the Brazilian farming context) were proxied by “the being able to provide education to their children” and “community recognition through interactive farming activities”.

The sample selection process is concisely depicted in Figure 2. The minimum sample size based on the “inverse square root method” (Hair et al., 2022, p. 26) was found to be 275, which is below the 324 responses obtained; this method considers the minimum statistical power sought (=0.80), significance level (=0.05), and the expected minimum standardized regression coefficient in a causal predictive relationship. The latter was taken as 0.15, as smaller regression coefficient has no practical significance.

The study assumes that a time gap of 12-months (on average) between the implementation of the farmer training program and data collection via the survey (the temporal asymmetry) is sufficient for the farmers to realize the benefits that they may have gained from training. The data were collected by four trained research assistants (two for each district) who visited the farms during April–June 2023. Anonymity of the responses was maintained throughout the data collection process to ensure against possible response bias.

Figure 3 depicts the specified statistical model (before conducting the robustness check to test H10) with the estimated standardized regression coefficients. In testing the theoretical model, both regions were analyzed together, as the purpose of this study is to estimate the average effects for the population and generalize the explanation of the phenomenon of SSD.

Partial least squares structural equation modeling (PLS-SEM) technique was adopted to test the theoretical model, mainly because it contained formative constructs, which do not pose problems unlike the traditional, covariance-based SEM (Hair et al., 2019, p. 5). Four of the eight constructs in the theoretical model (Figure 1), namely FS, ES, SS, and EnvS are formative constructs corresponding to indexes rather than scales. The second reason was the exploratory nature of the study, which involves conducting a robustness check—testing H10, which required creating nine more paths—that resulted in an already complex model becoming even more complex (Hair et al., 2019, p. 5). Missing data (just 0.01%) were imputed using the ‘mean replacement’ method included in the software. Since the data were collected via a self-administered survey, two tests—Harman’s single factor test (Podsakoff et al., 2003) and the full collinearity test (Kock, 2015)—were conducted to demonstrate the absence of common method bias (CMB), which is a specific form of response bias (e.g. presence of socially desirable responses) that could be present in self-administered surveys. The full collinearity test has been tailor-made for PLS-SEM as PLS-SEM researchers often incorporate formative constructs in their theorizations (for details, see Kock, 2015), which could lead to false negative results on CMB if a standard test on CMB, such as Harman’s single factor test is used.

In PLS-SEM, the validity of the measurement model is assessed first, followed by an evaluation of the structural model to test the hypothesized relationships between the constructs (Hair et al., 2022). In Figure 3, the measurement model shown encompasses all the relationships between the indicators of the constructs and the constructs themselves, while the structural model represents the hypothesized causal relationships between the constructs. PLS-SEM was conducted using SmartPLS 4 software.

The relevance of each indicator of a reflective construct was examined by examining its outer loading, which is the correlation between the indicator and its assigned construct, when the outer loadings are reported in standardized units (Hair et al., 2022). An outer loading >0.70 is used as an unconditional acceptance criterion of an indicator, but lower values (between 0.40 to 0.70) are acceptable for new scales, provided scale reliability and convergent validity requirements are met (see Hair et al., 2022, p. 118).

Cronbach’s alpha and composite reliability coefficient (ρ_c) were used to demonstrate scale reliability (Table 1). Once indicators are accepted as being relevant and the reliability of the constructs has been established, then the validity of reflective constructs is demonstrated via PLS-SEM by establishing: (1) convergent validity by examining the average variance extracted (AVE) by a construct’s indicators (= mean squares of outer loadings); and (2) discriminant validity by examining the heterotrait-monotrait (HTMT) ratios of correlations between the constructs (Hair et al., 2022).

In the measurement model of this study, only two outer loadings were <0.70 (see Figure 3); however, these outer loadings (FC1 = 0.625; Commitment = 0.678) have not been low enough to threaten scale reliability or convergent validity (see Table 1). Therefore, the two indicators were retained for the subsequent data analysis.

In PLS-SEM, the validity of the formative constructs is established using the criteria prescribed by Hair et al. (2022), which requires the measures of the formative constructs showing no multicollinearity (variance inflation factor<5.0) and the weights of the indicators being significant (p < 0.05), and when not, its correlation with the construct be > 0.50; these conditions were met by the measurement model.

The R² values of the endogenous constructs (Table 2) suggest that the hypothesized model is useful in predicting and explaining the effects of the hypothesized causal relationships. Notably, the R² of FC drops markedly when the control variables are excluded. From Table 3, it becomes clear that the only control variable that makes any practically noticeable impact on an endogenous construct is the farming region (R), which appears to impact FC and EnvS. These results are further reviewed in section 4.2.4.

The results in Table 4 indicate that at a 0.05 significance level, four hypotheses (H3, H4, H7, and H8c) have not been supported by the data, but there is weak support for H3, H7, and H8c (p < 0.10). These unsupported/weakly supported hypotheses do not threaten the overall explanation of the SSD phenomenon represented by the theoretical model (Figure 1), such that the FD initiatives have a positive effect on sustainable farmer performance, through the mediators FC and PFR.

Subject to H10 being supported by the data (section 4.2.5), an FD initiative has only an indirect effect on sustainable farmer performance via a set of mediated relationships (Figure 3). To examine how these mediated relationships in the model have worked, the total indirect effects and their statistical significance were examined. Table 5 depicts the results.

In summarizing the results shown in Table 5, it can be stated that FS and EFFQP do have a significant positive effect on all three TBL dimensions of sustainable farmer performance—through the mediators FC and PFR—but FT has only a weakly significant effect (0.05 < p < 0.10).

Although R is not a theoretical variable, as a control, it has acted as an influential predictor of FC and EnvS (Table 3). The results imply that given the specified model, Kurunegala farmers are 1.258 standardized units (=1.597 in unstandardized/actual units), on average, more capable than Nuwara Eliya farmers, given the 1–5 scale used to collect data. Similarly, results imply that given the specified model, Kurunegala farmers are 0.771 standardized units (=0.679 in unstandardized units), on average, more environmentally sustainable than Nuwara Eliya farmers.

A possible reason for the lower FC returned for Nuwara Eliya farmers could be that the questionnaire justifiably assesses FC based on relative performance (e.g. relative to the milk processor’s expectations regarding the quantity of milk supplied, as detailed in Appendix-1). Consequently, farmers in Nuwara Eliya may have chosen response options that result in lower scores for FC because the exotic cows they farm (e.g. Jersey and Holstein-Friesian breeds) are expected to produce more milk, setting higher benchmarks. In contrast, Kurunegala farmers, who primarily raise local or local-Indian crossbreeds, face lower production targets. Additionally, Kurunegala farmers benefit from lower feeding costs, as they do not regularly purchase commercial dry rations.

The two-way interaction plot shown in Figure 4 suggests that FT has resulted in an improvement of FC for Nuwara Eliya farmers on average, but not for Kurunegala farmers. It could be possible that through FT, Nuwara Eliya farmers learn more about farming practices (technical know-how) on farming exotic cows.

To test H10, nine direct paths between the three FD initiatives (FT, FS, and EFQP) and the three TBL metrics on sustainable farmer performance were created. These paths are shown in red in Figure 5 to focus the reader’s attention on the paths under scrutiny.

Out of the nine direct effects forced into the existing hypotheses to test H10, eight can be removed summarily, as they are nonsignificant (p > 0.05). The FS→ES direct link returned a path coefficient of −0.166 (p = 0.018), which is statistically significant. Due to this negative direct effect, the overall effect of FS on ES (direct effect + total indirect effects) came to 0.097 (p = 0.180), which is nonsignificant. Earlier (without the direct effect) the total effect of FS on ES (=total indirect effect) turned up as 0.208 and significant (p < 0.001), as shown in Table 5. Thus, it is argued that the FS→ES direct link forced into the existing set of hypotheses, for the sake of testing H10 (the robustness check), can also be removed on the grounds of untenability. Thus, it is concluded that the data supports H10.

The results imply that the impact of FT on FC is small (Table 4), which was largely responsible for FT having only a weak effect on sustainable farmer performance (Table 5). The hypothesized theoretical model (Figure 1) supported by the data considers the main effect of FT on FC. However, the FT*Region two-way interaction plot (Figure 4) sheds light on how FT might have worked in the two regions: worked for Nuwara Eliya farmers but not for their Kurunegala counterparts. The literature suggests that knowledge of dairy farming and the use of improved dairy management practices are essential in improving the profit margin of farmers (Kataike et al., 2018; Wouters et al., 2007) and the livelihood of dairy farmers (Wouters et al., 2007; DeSilva et al., 2023). The results of the present study do not necessarily contradict the literature.

The results in Table 4 suggest that if FT is increased by 1 standardized unit, FC is increased only by 0.105 standardized units (i.e. 10.5%) if all other factors remain unchanged. It is suggested that to improve FC, it is not the quantity of FT that matters but its quality and relevance—a suggestion consistent Bartos et al. (2024).

Table 5 shows that the effects of FT on the three TBL dimensions ES, SS, and EnvS are nonsignificant at the 0.05 level. This, coupled with the low impact of FT on FC, has a significant impact on the milk processor because arguably, they invest more in FT than in FS and EFFQP. Thus, it seems that neither the milk processor nor their farmers, in general, achieve much benefit from FT in its current form. The following possible reasons are suggested for the nonsignificant effect of FT on farmer sustainability:

Training thousands of farmers is quite challenging, and relationships can gravitate towards transactional/arm’s-length relationships (Monczka, 2020; Krause et al., 2007). This might also be the reason why data did not support the FT→PFR link (H4). Although the study covered farmers who receive training (according to the milk processor’s register), the translation of FT into action can be impaired by barriers, such as insufficient capital to invest in implementing the needed changes (Luther et al., 2018; Rogers, 1962); moreover, FT→FC improvement is associated with a very delayed trajectory, and the 12-month time lag considered may have been inadequate (Luther et al., 2018; Susanty et al., 2018; Bartos et al., 2024).

As regards focused training—tailoring training to suit the training needs of a specific farmer group—the literature refers to the establishment of farmer training centers/field schools and mobile training units to provide non-formal education as an efficient method of reaching out to rural farmers (Chopde et al., 2019; Mariyono, 2019; Mariyono et al., 2022; Bartos et al., 2024). These training centers may also provide an opportunity to enhance the relational capability of farmers and make the relationships more collaborative as opposed to transactional (Mariyono et al., 2022).

The literature (e.g. Korale-Gedara et al., 2023; Pedroso et al., 2021; DeSilva et al., 2023) also refers to FT on the adoption of environmentally friendly farming practices, which the study did not support (the nonsignificant effect of FT on EnvS). Since the price being paid to a farmer for supplying milk is directly linked with the quality of the milk (Korale-Gedara et al., 2023), the training can include imparting knowledge on food safety and quality (Susanty et al., 2018).

The regression coefficient corresponding to the FS→FC path is 0.174, which is much higher than the FT→FC link (=0.105) and the EFFQP→FC link (=0.078, and nonsignificant) implying that of the three FD initiatives, FS has been the most effective (also evident from the f² effect size reported in Table 4) in impacting FC positively. The reason for FS to emerge as the most important influencer of FC could be that what farmers need the most is FS, to improve their capability. The results also show that of the three FD initiatives, FS is the most important influencer/driver of a farmer’s SS (Table 5). A similar finding was reported by Susanty et al. (2019) in the Indonesian dairy farming context. Also, FS enables a farmer to implement the necessary quality and productivity improvement interventions and expand the business to be socially and economically better off (Yawar and Seuring, 2018; Yawar and Kauppi, 2018; Mukucha and Chari, 2024). Brix-Asala et al. (2021) reasoned that FS provided by processors in a developing economy enables smallholder farmers to overcome their hurdles in using the credit to improve their business gradually by being able to purchase equipment or machinery to improve their productivity/capability.

Korale-Gedara et al. (2023) found that in the Sri Lankan context, financially assisted dairy farmers tend to adopt safer practices to ensure that the quality of milk being supplied meets food safety and quality standards. For example, they found that FS to build or renovate cattle sheds improved milk storage capacity, milk hygiene, and effluent/waste disposal. Similarly, Chari et al. (2023) argued that FS, by way of providing farming equipment, has a positive effect on farmers' ES. The findings of this paper do not contradict the findings of Korale-Gedara et al. (2023) and Chari et al. (2023). Judging by the regression coefficients in the mediated paths (Figure 3), real gains of FS on sustainable farmer performance seem come through PFR, rather than through FC, which is something that is not highlighted much in the literature.

Table 5 also shows that FS influences EnvS (effect = 0.110) more positively than FT and EFFQP. The literature suggests, supporting farmers financially to promote environmentally friendly farming practices such as home gardening, action to reduce methane gas emissions, establishing domestic biogas plants, and implementing an effective wastewater management system has been useful in reducing the carbon footprint in agriculture (Bhat et al., 2022; Susanty et al., 2019).

The EFFQP→FC relationship returned a nonsignificant effect (p = 0.056), which can be interpreted as a weak effect. Since FC reflects farmer productivity in terms of milk volume, quality, and milk supply continuity, the weak link could mean that the milk processor has not been receiving the benefit that they expect from EFFQP. The SC literature argues that supplier evaluation and feedback improve suppliers' capability (Wagner, 2009; Benton et al., 2020; Monczka, 2020) which should, in turn improve sustainable supplier performance. The agri-food literature suggests that EFFQP is quite common in that field (Yawar and Kauppi, 2018), but the feedback seemed to be coming from dedicated extension service providers rather than from processors, who can only offer extension services to a limited extent.

Although EFFQP seems to have a weak impact on FC, it has a positive impact on sustainable farmer performance due to PFR taking over the mediator role from FC—the mediation paths EFFQP→PFR→ES and the EFFQP→PFR→SS are associated with higher regression coefficients (Table 4).

This study aimed to develop and test a theoretical model to ascertain to what extent FD initiatives implemented by a milk processor result in improved TBL outcomes of dairy farmers (Figure 1). FC (farmer’s quality, delivery, and continuous improvement performance) and PFR are both important to the milk processor because improved FC and PFR benefit them in maintaining a continuous flow of milk in their production line by being able to purchase good quality milk year-round in higher volumes through farmers who are likely to remain with them. The study found that FT has only a small effect on FC and does not seem to affect PFR, which may not be a sufficient return on the milk processor’s investment. These findings (FT’s weak influence on FC and PFR, the mediators) mean that FT is incapable of transmitting a significant impact on sustainable farmer performance. However, the study found that the remaining two FD initiatives (FS and EFFQP) do positively impact all three sustainable farmer performance dimensions.

There are several theoretical implications. The study found that: (1) the concept of SSD is applicable to agri-food SCs; (2) in its current form, FT does not influence sustainable farmer performance due to its weak effect on FC; and (3) in the study context, FS is the most significant driver of both FC and sustainable farmer performance.

There are several practical implications of the study too. It found that: (1) the type of FT provided to farmers must consider local production factors—while Nuwara Eliya farmers seemed to benefit, Kurunegala farmers did not; (2) feedback on milk quality needs to be modified by the processor to meaningfully enhance FC; and (3) FS has a relatively strong positive impact on SS.

The weak links between the FD initiative and FC have policy implications. This finding implies that FD should not be left almost entirely in the hands of the privately-owned processor because it is quite challenging to improve the capability of a large number of farmers efficiently by implementing a multitude of highly tailored programs for farmers who are likely to be heterogeneous. A privately owned milk processor will be more cognizant of the returns for their investments. The findings suggest that the government may need to be more involved in developing dairy farmers (e.g. through a private-public partnership or by providing more incentives for private milk processors who actively engage in FD). Governments and regulatory bodies can also improve FC by developing and enforcing milk quality standards and environmental standards to promote cleaner production. These measures would force farmers to self-improve to stay in business; having such farmers would also benefit milk processors because it is much easier to support farmers who self-improve. This said, the governments and regulatory bodies must gradually enforce the standards, allowing farmers to adapt to the new norms.

The study has some limitations. Firstly, the responses are based on the perception of the respondents (e.g. the judgments on farmer sustainability have been made with reference to the goals being set by the individual farmers. A farmer may feel that they are economically and socially well-off, but they may not be as well-off as they perceive). The same can be said about the judgment on FC (the traditional notion of supplier performance). Secondly, although anonymity was assured during data collection, and a robust statistical test was conducted to eliminate CMB, the quasi-voluntary training provided and the role of the milk-processor as the main buyer has the potential to influence farmers in providing socially desirable responses (a post hoc statistical test on CMB cannot fully clear the presence of socially desirable responses). Thirdly, the theoretical model has been tested in a reasonably favorable environment for theory testing (e.g. the milk processor has its bank to support its farmers, the inclusion criteria being used for the survey, and the non-existence of other major players that are strictly involved in FD), which means that potential confounding variables remained dormant by choice (the research context), but the effects of confounding variables cannot be fully eliminated. For these reasons, the generalizability of the findings of this study should be made with some caution.

It is suggested that the study be repeated in other agri-food contexts, ideally incorporating more objective judgments on FC and sustainable farmer performance (e.g. replace self-judgments on performance with more verifiable measures). Another attractive proposition might be to depart from SSD theory testing to a more direct approach of examining the relationship between the drivers FT, FS, and EFFQP—including their two-way interactions—and farmer performance. Yet another possibility might be to conduct a longitudinal study to examine how tailored FD programs work over a period that can handle temporal asymmetry more reliably. Finally, it is useful to collect quantitative and qualitative data from the processors to get their perspective, such as their expectations on FD.