Supply chain performance is defined as the overall improvement in business operational measures of an individual and the whole supply chain as a result of opportunities created by trust (Odongo et al., 2016; Gera et al., 2022).

Trust is considered important for competitiveness in supplier–buyer relationships through a reduction in transaction cost (Mottaleb and Rahut, 2018; Mbapila et al., 2019). In agribusiness, transaction cost influences operational measures such as financial performance and sales quota. A change in financial security and cash flow from agribusiness defines the financial performance of business actors (Wahdan and Emam, 2017; Martins et al., 2019). Similarly, a change in the quantity of specific sales goal explain whether or not sales quota is achieved by supply chain actors (Good and Stone, 1991).

Transaction cost is defined as expenses incurred in market exchange. These are widely suggested to include the cost of discovering market prices, storage, and transportation (Dyer and Singh, 1998; Ali et al., 2017). In this study, transaction cost was measured by information sharing, uncertainty, and transaction frequency. Information sharing is defined as the extent to which production, storage, and market (demand and supply) information are regularly and accurately shared (Dyer and Chu, 2003; Ghosh and Fedorowicz, 2008). Information sharing is suggested to optimize the benefits of supply chain relationship when trust is built (Walker et al., 2018). Uncertainty is defined as unquantifiable technological risk arising from supply, demand, technology, and price that eventually impact the overall cost, quality, and cycle time (Khan et al., 2018; Gokarn and Kuthambalayan, 2019). According to Van Der Vorst (2000), uncertainty is the inability of business partners to predict accurately the impact of decisions on performance. Fynes et al. (2008) argue that uncertainties are eliminated by trust in a business relationship. Transaction frequency refers to a buyer's total purchase frequency from a specific seller in a business relationship within a particular time period (Chang et al., 2012; Zhao et al., 2020). Trust should influence the expectation of exchange partners in a business relationship to scale up transaction frequency (Chang et al., 2012). It is hypothesized that:

H_1a: Trust will be positively related to information sharing.

H_1b: Trust will be negatively related to uncertainty.

H_1c: Trust will be positively related to transaction frequency.

Transaction costs are regarded as the cost incurred on access to complete information about all the market conditions by buyers and suppliers, uncertainty, and transaction frequency (Fischer, 2013; Jraisat et al., 2013). This cost in a contractual arrangement is bounded by rationality (Simon, 1957). Transaction costs are minimized by taking different forms of information sharing, having joint problem solving, and safeguarding against opportunistic behaviors (Williamson, 1981; Walker et al., 2018). In the e-supply chain, transaction costs are mainly environmental and technological uncertainty (Puška et al., 2022). Transaction costs such as uncertainty and transaction frequency were found to influence buyer–seller relationships (Zhao et al., 2020). When critical information is shared timely and regularly, it enables supply chain members to become efficient in negotiation and subsequently increases business returns (Walker et al., 2018). Furthermore, information sharing plays a primary role in outsourcing, providing benefits of collaboration and alliances (Min et al., 2005; Gajdić et al., 2023). Information sharing has been suggested to have a positive influence on supply chain performance in previous studies (Baihaqi and Sohal, 2013; Odongo et al., 2016; Owot et al., 2022). This study hypothesized that:

H_2a: Information sharing has a positive effect on supply chain performance.

H_2b: Uncertainty has a negative effect on supply chain performance.

H_2c: Transaction frequency has a positive effect on supply chain performance.

The influence of transaction cost on the relationship between trust and supply chain performance is well established in previous research (Odongo et al., 2016; Owot et al., 2022). Information sharing is viewed as a critical element for strengthening the achievement of common goals when trust is built by coordination and joint planning (Khan et al., 2018; Agarwal, 2019). Exchange partners who share complete information are considered to have an enabling environment for improvement in financial security by all exchange partners (Odongo et al., 2016). The existing literature has found information sharing a construct that lies in the heart of agribusiness relationship, influencing each exchange partner's trust toward one another (Khan et al., 2018), and supply chain performance in terms of competitiveness. Dominic and Theuvsen (2015) and Negi and Anand (2019) found that provision of market information by exchange partners brings confidence and trust and link them to markets with better outcomes of farmer-trader relations. Markets full of uncertainty create opportunism that eventually impacts the overall cost, quality, and cycle time (Khan et al., 2018; Gokarn and Kuthambalayan, 2019). Exchange partners who experience uncertainty tend to mistrust and predict inaccurately the impacts of decision on performance (Van Der Vorst, 2000; Fynes et al., 2008). No known study has assessed the mediating influence of transaction frequency on buyer–supplier relationships. Against this background, this study hypothesizes that:

H_3a: Information sharing positively mediates the trust-supply chain relationships.

H_3b: Uncertainty negatively mediates the trust-supply chain relationships.

H_3c: Transaction frequency positively mediates the trust-supply chain relationships.

Study respondents consist of literate, semi-literate, and illiterate farmers and traders dealing in tomatoe and soybean agribusiness. They were drawn from Lango and Acholi sub regions in Northern Uganda within the districts of Oyam, Kole, Gulu, Nwoya, and Omoro. Whereas, the selection of the two crops followed from the group of high-value crops (fresh and dry), contributing to income and livelihood through commercialization (Owot et al., 2022), the selection of the two regions was based on the fact that one of the districts from each of this region was elevated to city status in 2018. The emergence of the two cities has created a new level of aggregate demand for fresh and dry commodities.

In this cross-sectional survey study, data were collected between September and November 2019 from farmers and traders using a structured questionnaire. Approximately 400 respondents were contacted to participate in the interviews. However, there were four respondents who chose not to complete the interviews on business emergencies. Hence, 396 questionnaires from 203 farmers and 193 traders were fully filled and qualified for analysis (Table 1).

Sampling began with identifying districts and sub-counties in which the respondents were located using two-stage sampling procedures, where in the first stage, five districts were selected, and in the second stage, 15 sub-counties got identified. After identifying the respondents' location, a purposive sampling technique was used to identify farmers participating in agribusiness in the two crops, specifically those who nominated the traders. Snowballing was used to follow and interview the nominated traders in the sub-counties, identified on the same questions given to farmers. The farmers and traders were selected based on their age (at least 18 years old), business age of at least 1 year of business relationships in soybean and tomatoe agribusinesses, and should have been engaged in an informal or contractual agribusiness relationship. These respondents constituted both the unit of inquiry and the unit of analysis. Farmers who met the inclusion criteria were purposively identified and asked to nominate traders with whom they are in a business relationship in a snowball sampling approach.

Roscoe (1975) suggests that for a study of an unknown population, the sample sizes have to be more than 300 and <500, applications for an utmost research, and it is believed to be appropriate for most behavioral research. Therefore, using Roscoe's rule of thumb, at a 5% level of significance, a sample of 203 farmers and 193 traders was selected to minimize sampling error and obtain statistical convergence. Table 1 indicates the category of respondents, and the sampling techniques used based on their relevance.

The variables in this study were operationalized. Transaction cost attributes were conceptualized based on the dimensions of information sharing, transaction frequency, and uncertainty as adopted from previous scholars (Chang et al., 2012; Khan et al., 2018). To describe information sharing, eight items were selected. For uncertainty, eight items were chosen to capture the construct. Concerning transaction frequency, this construct was represented by six selected items. In addition, trust was conceptualized as benevolence, integrity, and competence (Sekhon et al., 2014; Xue et al., 2018; Franklin and Marshall, 2019). Seven, six, and four items were selected to measure benevolence, integrity, and competence, respectively. The dimensions of financial performance, sales quota, commitment, and satisfaction were adapted to measure supply chain performance (Bunte, 2006; Fearne et al., 2012; Chou and Chen, 2018; Qian et al., 2020). Therefore, in this study, the measures developed and used by scholars above were adapted to measure constructs of the outcome variables. Five, four, and seven items were selected to measure financial performance, sales quota, commitment, and satisfaction, respectively. Measurements of all items were anchored onto a five-point Likert scale starting from strongly agree (5), agree (4), not sure (3), and disagree (2) to strongly disagree (1).

The results of composite reliability of the constructs provided by Cronbach's alpha meet the minimum threshold for adequate reliability (Fornell and Larcker, 1981). Furthermore, the study conducted a confirmatory factor analysis to determine the validity of the latent constructs. Accordingly, the measurement model provided a reasonably good fit (χ² = 32.996, degrees of freedom = 24 and probability level = 0.104; CFI = 0.992; TLI = 0.989; IFI = 0.992; and RMSEA = 0.031) (see Figure 1). In addition, the study constructs met conditions recommended by Gerbing and Anderson (1988) that all factor loadings should be >0.50 for convergent validity. Discriminant validity was assessed through a comparison of variance between the constructs and average variance extracted (AVE) for each individual construct (Fornell and Larcker, 1981). The conditions for discriminant validity recommended by Gerbing and Anderson (1988), for the diagonal elements, representing the square roots of the AVE for each construct to be greater than each of the off-diagonal elements in the rows and columns corresponding to it were met. The results demonstrate sufficient discriminant validity between factors.

Data were analyzed in SPSS version 23. Upon collection, data were captured in SPSS for preliminary analysis. These included missing data screening and checking for outliers, normality, multicollinearity, and homogeneity parametric assumptions. Frequencies run did not show that missing values were a problem. In addition, the box plots did not reveal the existence of outliers. Concerning the normality of data, the histogram was bell-shaped, and most observations on the P-P plots fell along a straight line. Multicollinearity was tested using the variance inflation factor (VIF) and tolerance level, which provided values within acceptable limits (VIF < 4 and tolerance > 0.2) to conclude that multicollinearity was not a problem. Homogeneity parametric assumptions were checked using Levante's test (Field, 2005). The results revealed that Levante's test for all variables was not significant at P > 0.05, which indicates that variances were stable at all levels. Hair et al. (2013) argued that the presence of both outliers and missing data may affect multivariate analysis when poorly managed. With most of the preconditions for multivariate analysis met, the key hypothesis of mediation was tested using bootstrapping structural equation modeling in Amos based on Baron and Kenny (1986) and Rashid et al. (2022).

The results in Table 2 revealed that 62% of the farmers and traders were male while 38% were female. Of these respondents, 51% were farmers while approximately 49% were traders. Furthermore, the results also showed that the majority (38.4%) of the respondents were in the 26–35 age range, while 22.7% were in the 36–45 age range. On the aspect of education status, the findings revealed that 46.2% were illiterate and 42.9% were semi-literate. Moreover, the results indicated that 40.2% of the respondents were in agribusiness for a period of 6–10 years, followed by those who were in business for 5 years or less at 31.6%.

To model the relationship between trust and transaction cost, a regression model was run to determine the influence on each transaction cost attribute (information sharing, uncertainty, and transaction frequency). Table 3 presents that the path from trust to information sharing was positive and statistically significant (b = 0.57, S.E = 0.04, P ≤ 0.001), providing support for H_1a.

The effects of information sharing, transaction frequency, and uncertainty on supply chain performance were found to be positive for the first two constructs and negative, respectively (see Table 4). The influence of information sharing (b = 0.7284, S.E = 0.1440, p < 0.01), uncertainty (b = −0.3741, S.E = 0.3028, p < 0.1), and transaction frequency (b = 0.3741, S.E = 0.1855, p < 0.05) on supply chain performance (SCP) was significant. Thus, hypotheses H_2a, H_2b, and H_2c were supported.

To understand the mediation effects of transaction cost attributes, i.e., information sharing, uncertainty, and transaction frequency on the trust-supply chain performance relationships, the test of mediation was performed using non-parametric bootstrapping. The results show that only information sharing (0.26, 0.64) mediated the relationship between trust and supply chain performance (SCP). This provides support for hypothesis H_3a (see Table 5).