The marketing field has broadened the reach of transaction cost economics (TCE) to highlight research challenges and opportunities to be addressed. Williamson and Ghani (2012) recall Erin Anderson’s early recognition of TCE’s implications for marketing inquiry in the 1980s and beyond in her seminal work that combined the two fields. More recently, various studies have continued to acknowledge the applicability of TCE in marketing (Cheng & Lee 2011; Gatignon & Gatignon 2010; John & Reve 2010; Mukherjee, Banerjee & Bandyopadhyay 2012; Ndoro, Mudhara & Chimonyo 2015; Seggie 2012; Varadarajan 2019; Yang & Su 2014). Yet one aspect of marketing that has received scant attention in this literature is the applicability of TCE in online retail.

The Internet has directly affected retailing practices and thus customers’ e-buyer behaviour. It is especially the way in which customers interact with and respond to retail offerings that have been profoundly modified (Mulky, Sarkar & Mukherjee 2019). With Internet penetration in South Africa in 2019 at 81%, more consumers have access to online retail stores (Gilbert 2019). Consumers are also moving towards online shopping because of the convenience it offers, the variety of channels available (e.g. websites, apps and purchasing through social media) and consumers’ ability to search for products and services quickly and easily (Mandal 2020). Moreover, the growth of the Internet through smartphones has increased the use of e-commerce amongst rural consumers (Han & Li 2019). Significant growth in online retailing, especially in emerging countries such as South Africa in recent years, provides promising commercial potential. In this regard, Euromonitor (2019) predicted that digital commerce in South Africa would achieve a 15% total value of the compound annual growth rate (9% constant with 2018 prices) to reach South African Rand (ZAR) 120.5 billion by 2023. Staffa et al. (2019) predicted a 9.9% growth from 2019 to 2023, which is 1% above the predicted global growth in retail e-commerce sales. Nisar and Prabhakar (2017) argued that such an increase in e-commerce would result in customers increasingly preferring to shop in the virtual marketplace. Many authors recognise the advantageous nature of e-buying for both the buyer and the retailer (Kumar & Anjaly 2017; Lin & Sun 2009; Rose, Hair & Clark 2011; Wu et al. 2014). For the retailer, this can be the catalyst that attracts new customers from new markets – those who are not physically in close proximity to the retailer – and e-buying provides a multitude of contact point opportunities for e-buyer satisfaction and increased loyalty. From a customer perspective, e-buyers can visit online stores in the comfort of their own home and in their own time. Yet Wu et al. (2014) concur with Bolton, Kannan and Bramlett (2000) and Tsai and Huang (2007) that e-buyers and online retailers alike often face challenges in acquiring and retaining customers, especially in online business-to-consumer (e-B2C) online retailing. These challenges can range from establishing trust and addressing data privacy concerns from an e-buyer perspective to integrating and coordinating omni-channel brand experiences from an online retailer point of view. The key to long-term success from an online retailer perspective is to give e-buyers satisfaction and, ultimately, to win their loyalty (Nisar & Prabhakar 2017; Sfenrianto, Wijaya & Wang 2018; Tran, Pham & Le 2019).

This article forms part of a larger study, based on the work of Kumar and Anjaly (2017), which examined the management of touchpoints in the service delivery process that act as differentiators between a one-time purchase and long-time loyalty in order to measure customers’ post-purchase experience with a brand. However, the primary objective of this article is to describe the influence of TCE theory variables (specifically, uncertainty and asset specificity) on buyer satisfaction and loyalty in an e-B2C milieu. The secondary research objectives include:

This article has considerable theoretical and practical importance. Firstly, its most significant contribution lies in the exploration of e-buyer behaviour from a TCE theory perspective. In general, this study contributes to the generalisability of TCE and provides an insightful lens through which to view future research on TCE’s impact on e-satisfaction and e-loyalty. Secondly, previous research in this regard has traditionally involved Western samples (Teo & Yu 2005). By collecting data from South Africa, this research examines the applicability of the TCE model in the marketing research context of a non-Western and emerging market. In addition, as suggested by Hossain (2019), most businesses take aspects of TCE and its effect on satisfaction and loyalty for granted, even though it has not been empirically validated. This study therefore aims to provide empirical evidence of the effects.

This article has been organised such that the following section provides a theoretical framework for the study, focusing through the theoretical lens of the TCE and its application and consequences in e-B2C retail and marketing, leading to the formulation of the hypotheses for the study. This is followed by a description of the research design and method and the data analysis and findings. Thereafter, the research implications, limitations and suggestions for future research are presented.

Transaction cost economics, developed by Williamson in 1975, originates from the ‘new institutional economics paradigm’ and is usually researched from an economics and supply chain (i.e. B2B) perspective (Ketokivi & Mahoney 2020; Kim & Li 2009). Although Tsay et al. (2018) suggested that the theorem can be used in many other fields because of its dynamic nature, for the purposes of this study, the TCE theory has been applied in a marketing context.

This decision is supported by Ludorf (2016), Hossain (2019) and Cuypers et al. (2021), who argue that TCE can and should be more frequently applied to theories in contemporary marketing, as it examines individual transactions between buyers and sellers and should be expanded from a B2B to a B2C perspective and from a customer’s perspective.

Liang and Huang (1998), Teo and Yu (2005) and Hossain (2019) argued that TCE reveals profit amongst economic actors in either traditional or online markets and that minimising transaction costs as the rationale for using TCE to explain e-buyer transaction behaviour is based firmly in the transaction at the heart of e-buyer behaviour. Considering this, the study focuses on the effects of TCE from a customer’s perspective in an online retailer context.

As discussed above, TCE theory was initially developed to explain the selection of ‘market exchange and internal organization among firms’ (Khan 2010:21). However, TCE has been employed in a variety of buyer perspective transaction settings (Bhattacharya, Singh & Nand 2015; Che et al. 2015; Tang & Lin 2019). The literature indicates that, whereas actual costs are used in a firm’s (a retailer’s) valuation of costs in the value exchange, it is perceived costs that are more applicable in a consumer buyer value exchange (Liao & Cheung 2001; Rao et al. 2016). Thus, when a consumer is about to engage in a potential e-transaction, the cost benefits (or losses) are considered relative to other e-transactions. Williamson (1996:36) described three dimensions that have been ‘especially instructive to the study of commercial transactions’: the frequency with which transactions recur, the uncertainty to which they are subject and the condition of asset specificity. Earlier studies (Che et al. 2015; Khan 2010; Liang & Huang 1998) infer that for e-transactions, uncertainty and asset specificity affect customer behaviour.

Table 1 operationalises the three dimensions of TCE in an e-B2C retail context.

Considering the above, together with the aim to link TCE to a marketing perspective, the following section describes each construct measured in the study and their linkages to the theory.

For the purposes of this study, the TCE elements of uncertainty, asset specificity and frequency of transactions are related to the marketing concepts of price and risk reduction, special (customised) offers and loyalty.

As this study was quantitative and descriptive in nature, a positivistic research approach was used. This approach was used as it is most suitable when researching human behaviour (Burns & Burns 2008), when it is objective in nature (Rahi, Yasin & Alnaser 2017:1) and when the researcher is independent (Wahyuni 2012:70), as is the case in this study. As this study was based on the previous studies of Kumar and Anjaly (2017), Rose et al. (2012) and Park and Kim (2003), as the existing literature about the subject was available, and as the research questions could be empirically tested, a deductive approach was most suitable (Saunders, Lewis & Thornhill 2016:148).

As this study aimed to describe the behaviour of consumers in terms of the influence of pricing and costing elements on their online purchase behaviour and the relationships between the constructs, a quantitative, descriptive research method was employed (Saunders et al. 2016:165; Quinlan et al. 2019). A non-probability convenience sampling method, using a self-administered, face-to-face survey method, enabled the researchers to employ a deductive research approach (Rahi et al. 2017:2).

The population for the study included any person living in South Africa who had purchased from an online retailer such as Superbalist, Amazon, the Google App store or any retailer’s website (e.g. Woolworths online). From this population, the sample elements included respondents living in Gauteng who met the inclusion criteria of being 18 years and older and who had purchased online in the 6 months preceding the data collection. Gauteng was selected as it includes five major metros and is seen as the economic hub of South Africa, with a large number of people having access to the Internet and earning higher incomes that enable them to purchase online (City of Johannesburg 2019).

The questionnaire used in this study was adopted from the previous studies of Kumar and Anjaly (2017), Rose et al. (2012) and Park and Kim (2003). Although these studies included aspects measuring elements such as website design and quality, only the items in the questionnaire relating to the elements of cost and pricing and their effects on satisfaction and loyalty were used for the purposes of this study. The questionnaire included two sections. Section A measured the demographic and behavioural aspects such as age and how often they purchase online. This section included nominal and ordinal scales. Section B of the questionnaire measured aspects of pricing and risk (uncertainty) and exclusive offers (asset specificity) in purchasing online (nine items), satisfaction (seven items) and loyalty (four items). An unlabeled Likert-type scale with end points of 1 = strongly disagree and 5 = strongly agree was used in Section B.

Ethical clearance was provided by the relevant institutional committees before the data were collected. Trained fieldworkers were used to distribute questionnaires to respondents who met the inclusion criteria. Before providing the questionnaire to potential respondents, the fieldworkers provided them with a description of the study, and they were informed about their rights (e.g. respondents could withdraw from the study at any time without any negative consequences, and no payment would be given for participating in the study). Respondents were assured of their anonymity and confidentiality, as no names or contact details were requested. Respondents were approached using a convenience sampling method in public areas where permission to distribute questionnaires was not needed. After data cleaning, 355 questionnaires were retained for further statistical analysis.

The data were analysed using Statistical Package for the Social Sciences (SPSS) and Analysis of Moment Structures (AMOS) version 25. Various steps were included in the data analysis. First, the data were cleaned and checked for normality using skewness and kurtosis. The data were found to be normally distributed, as it fell within the parameters of the absolute values of -2 and +2 (skewness) and -7 and +7 (kurtosis). As the data were normally distributed, further parametric testing could be done. Second, descriptive analysis was conducted in which the means and standard deviations of the data were determined. Finally, an exploratory factor analysis (EFA), a confirmatory factor analysis (CFA) and structural equation modelling (SEM) were conducted (see the ‘Results’ section for more details on the factor loadings, reliability, validity and model fit). Both an EFA and a CFA were conducted, as this research combined three previous studies, and the uncertainty and cost specificity elements of the questionnaire had not previously been tested in a South African context. The authors wanted to determine whether there was a further underlying factor structure amongst the items before confirming the structure.

Before conducting the EFA and the CFA, the data were analysed to determine whether it was suitable for those analyses. Pallant (2016) suggested that data must meet at least two criteria before an EFA, or a CFA can be conducted. The first criterion of enough data points was met, as 18 subjects for each scale item were available, which exceeds the suggested 7–10 subjects per scale item (Pallant 2016). Second, the Kaiser-Meyer-Olkin (KMO) value was 0.922 (above the suggested value of 0.6), and Barlett’s test of sphericity was significant (p ≤ 0.000). These results showed that the data met the criteria for sampling adequacy. The principal components method with Varimax rotation was used for the EFA, whilst the maximum likelihood method was used for the CFA, as suggested by Fabrigar et al. (1999:277), as this method allows for the significance testing and the determination of goodness of fit indices to be done. When conducting the analysis, only Eigen values above one were extracted, with coefficients of 0.3 and less being excluded. The results of the EFA and the CFA and the reliability and validity of the analysis are presented in the following section, together with the results of the SEM.

Ethical clearance was obtained from the University of Johannesburg, College of Business and Economics, School of Consumer Intelligence and Information Systems Ethics Committee (Sub-committee of CBEREC) – 2018MM018.

Table 2 depicts the results of the demographic and behavioural profile of the respondents for characteristics such as age, gender and Internet purchase activity. The table reports the frequency (F) and percentage (%) for each variable.

The results in Table 2 indicate that most respondents were between the ages of 18 and 29 (76.6%), female (51.3%), accessed the Internet via their smartphone (53.5%) and mainly purchased online every 3 months (22.5%). In addition to the demographics reported in Table 2, respondents were also asked to write down from which online store they mainly purchased from. Analysis of this indicates that the majority of respondents purchased from Takealot (24.5%), followed by Superbalist (16.9%) and then Pick n Pay online (7.9%).

As the study combined previous studies and these have not been tested in a South African context before, an EFA was conducted. The results of the EFA indicated that the data could be further reduced into four factors. The first factor (three items) was labelled ‘price and risk reduction’ as the items measured in each linked to the ‘uncertainty’ element of the TCE, whilst the second factor (eight items) was labelled ‘exclusive offers’, which linked to the ‘asset specificity’ element of the TCE. The final two factors were labelled ‘satisfaction’ (seven items) and ‘loyalty’ (four items) and all included the Frequency of Transaction aspects of the TCE. These four factors explained 73.0% of the variance. After conducting the EFA, a CFA was conducted to confirm the structure.

As indicated in research design section above, the data were firstly analysed to determine whether it was suitable to conduct the CFA. The data met all the criteria of minimum sample size, the KMO being above 0.6 and Bartlett’s test of sphericity was significant indicating sampling adequacy (Muthen & Muthen 2002). The CFA was conducted using the maximum likelihood method as suggested by Fabrigar et al. (1999) in order to determine goodness of fit indices. The CFA resulted in a four-factor solution, but in order to improve the model fit, eight items were removed. One price reduction item (uncertainty reduction) and three exclusive offers items (asset specificity) had to be removed. Three ‘satisfaction’ items and one ‘loyalty’ item were also removed (frequency of transactions). For the analysis, only Eigen values above one were extracted and any coefficient below 0.3 was removed. To ensure that the results of the CFA were valid and reliable, the construct validity – through factor loadings, average variance extracted (AVE), composite reliability (CR) and Cronbach’s alpha (CA) as shown in Table 3 – was calculated, as well as the discriminant validity, as depicted in Table 4.

According to Fornell and Larcker (1981), the factor loadings in Table 3 should be ≥0.5, the AVE should be ≥0.5, that assesses convergent validity composite reliability and CA should be ≥0.7 (but in the social sciences, a value of 0.6 is acceptable) to determine reliability and internal consistency (Pallant 2016). The results of Table 3 therefore indicate that the criteria for reliability and validity were met.

Discriminant validity, as shown in Table 4, is calculated by determining the square root of the AVE for each factor. According to Fornell and Larcker (1981), this value (shown in bold) should be greater than the correlations’ values. As shown in Tables 3 and 4, all parameters for reliability and validity were met. The high correlations may indicate a level of multicollinearity although this was tested throughout the analysis. According to Grewal, Cote and Baumgartner (2004:526), multicollinearity in SEM only becomes an issue when the correlations are above 0.9; the CR is below 0.7 and a small sample size is used. None of these three are valid for this study as the results show indicating that multicollinearity was not a problem in this study.

The results of the SEM indicate the relationships between the variables and give the goodness of fit indices. Table 5 provides the model fit summary. For a model fit to be seen as good or great, the CMIN/DF should be ≤ 3, comparative fit index (CFI) > 0.90, TLI > 0.95, root mean square error of approximation (RMSEA) < 0.05 and PCLOSE > 0.05 (Hair et al. 2010). All the parameters are met or exceeded, indicating that the results show a good model fit.

Figure 1 provides the results of the SEM for the relationships between the factors and the strength of those factors (beta values).

The results from the factor analysis and the SEM, as shown in Figure 1, indicate that there are positive significant relationships between price and risk reduction (uncertainty reduction) and satisfaction (p ≤ 0.001), exclusive offers (asses specificity) and satisfaction (p ≤ 0.001) and satisfaction and loyalty (p ≤ 0.001) (frequency of transactions). Price and risk reduction especially are strong predictors of satisfaction (B = 0.71), but satisfaction is the strongest predictor of loyalty (B = 0.83). Based on these findings, H1, H2 and H3 can all be accepted. This indicates that respondents require online retailers to reduce the uncertainty when purchasing online and to increase the asset specificity of the purchase in order to increase the frequency (and loyalty) of transactions.

From the results of the study, it can be noted that the TCE can be extended to other fields in management and should be reconsidered in new contexts. The findings show that although the TCE is usually researched in a B2B setting, its relevance can be extended to a B2C perspective. Businesses should therefore consider the TCE as an alternative way of developing marketing strategies from an e-retailer perspective as shown in this study. With many businesses moving towards omni-channel retailing because of the changes in the current economic environment and the increased use of technology, it is suggested that B2C omni-channel retailers reconsider their business model to include the findings of this study. The results of the study show that online retailers must redevelop their marketing strategies to also focus on the uncertainty and asset specificity elements linked to customers’ perceptions of the business. This requires that all marketing mix elements need to be redesigned with the focus on reducing uncertainty and providing customers with value through asset specificity of special (customised) offers. Furthermore, to redesigning the marketing strategy, it will be essential for B2C online retailers to focus on the promotions and relationship marketing efforts and clearly communicate on an individual level with each customer, how uncertainties can be reduced during the transaction (e.g. have clear and easy return policies, provide alternative payment methods and ensure that personal and credit card information is kept confidential). Customers want to be seen as important to the online retailer and want to be offered exclusive offers (i.e. offers that are customised to the individual customer based on their previous purchases). Online retailers also need to ensure that customers are satisfied before they will increase their spending with the online retailer. This is essential in order to create repeat purchases through relationship marketing (frequency of transactions in the TCE) and to generate more profit for the online retailer that is essential for the future sustainability of the online retailer.

The limitations of the study include that a convenience sampling method was used to collect the data. The data were also only collected in one province of South Africa, and, in addition, the demographic profile of the respondents was skewed towards respondents under the age of 30. Therefore, future research should consider a quota sampling method to ensure a spread of ages and to identify whether the results would be different based on generational theory. The study also did not consider the difference in terms of the type of online purchase made and whether that could influence the type of cost factor involved. The study also did not consider the full extent of the TCE and all various aspects of asset specificity.

The main recommendations for this study include that online retailers should: