Content area
Purpose
This study investigates augmented reality (AR) retailing and attempts to develop a profound understanding of consumer decision-making processes in AR-enabled e-retailing.
Design/methodology/approach
The study is grounded in rich informational cues and information processing mechanisms by incorporating the elaboration likelihood model (ELM) and trust transfer theory. This study employs a mixed analytic method that incorporates structural equation modeling (SEM) and fuzzy-set qualitative comparative analysis (fsQCA) to provide a complete picture of individual information process mechanisms in AR retailing under the tenet of ELM.
Findings
The SEM analysis results confirm the relationships between the central and peripheral route factors, information processing outcomes and eventual behavioral intentions. Moreover, all configurations revealed by the fsQCA include both central and peripheral factors. Hence, the dual routes proposed in the ELM are verified by using two distinct analytical approaches.
Originality/value
This study is pioneering in validating and contextualizing ELM theory in AR retailing. In addition, this study offers a methodological paradigm by demonstrating the application of multi-analysis in exploring consumers’ information process mechanisms in AR retailing, which offers a holistic and comprehensive view to understand consumers’ decision-making mechanisms.
1. Introduction
Augmented reality (AR) has become a central part of online retailing’s repertoire, promising an evolution for e-commerce. AR technology integrates the real world with the virtual (computer-generated) objects that appear to co-exist in the same space as the real world (Kowalczuk et al., 2021). The global AR market is expected to maintain an average market growth of 46.3% from 2020 to 2026, meaning that the market size will reach $89bn in 2026 (KBV, 2021). Moreover, it is estimated that 20% of e-commerce sales will be generated from AR retailing within 5 years (Thinkmobiles, 2021). Hence, e-commerce giants (such as Amazon and Alibaba) have incorporated AR technology into their marketing strategies. By integrating AR technology into online shopping channels, consumers can online experience products themselves in real-time or in their immediate real surroundings (Smink et al., 2020). The application of AR in online retailing helps customers gain a better understanding of product information in a specific setting. Therefore, consumers are 90% more likely to buy a product if they see it with AR content, indicating a high conversion rate (Motti, 2021). Moreover, AR retailing addresses the achievement of organizational goals and offers opportunities to explore how consumers process product information in a novel field (Sun et al., 2022).
Through consolidating reality and virtuality, AR establishes a highly informed shopping scenario. For example, accurate product information can be obtained since virtual products can be presented as real and seamlessly integrated into the actual physical environment. Moreover, functions such as zoom, rotation and movement allow users to view products in more detail (Kowalczuk et al., 2021). Product information can be presented in a variety of information formats, such as images, videos and animations, allowing consumers to obtain multiple informational cues about the products (Barhorst et al., 2021). These additional informational cues will enhance consumers' understanding of the advantages of AR retailing, attract consumers to the novel effects of AR and bring consumers to light the quality of augmentation (Javornik et al., 2021). As a result, their positive perceptions toward products and AR retailing will further result in consumers’ purchase and continuance intention in AR retailing (Chiu et al., 2021; Qin et al., 2021a; Whang et al., 2021). Therefore, investigating the consumer decision-making process grounded on rich informational cues related to products and AR technology in shopping scenarios will provide a profound understanding of AR retailing. Whereas existing studies have not fully addressed this critical issue.
While researchers have attempted to understand the phenomenon of AR retailing from several theoretical perspectives, investigations remain at an early stage. By applying theories such as flow theory (Barhorst et al., 2021), the technology acceptance model (TAM) (Goebert and Greenhalgh, 2020) and the stimuli-organism-response model (S-O-R) (Daassi and Debbabi, 2021; Hsu et al., 2021; Seong and Hong, 2022), existing studies have mainly focused on the technological characteristics of AR and the benefits of AR in influencing consumers’ motivations, attitudes and behaviors (Kowalczuk et al., 2021). As a result, the exploration of the consumers’ information processing components, mechanisms and information processing consequences in the context of AR retailing remains limited in the literature. Hence, concepts such as critical factors, nature, patterns, decision routes, outcomes and consequential behaviors of consumers’ information processing in AR retailing remain underexplored. Accordingly, three research gaps remain in the literature. First, despite the relevance and importance of informational cues that have been widely highlighted, the critical informational cues in AR retailing remain unclear. Second, despite AR retailing creating an informed shopping context, few studies have addressed how consumers process the information in this phenomenon. Third, there is a lack of knowledge regarding how these critical informational cues and information processing outcomes can be combined differently to predict consequential behavioral decisions.
To address these research gaps, we developed a research model based on the elaboration likelihood model (ELM) which has been applied to explain how people process information and how their beliefs or attitudes toward information are changed. Importantly, to present a more holistic and comprehensive perspective, we applied a mixed analytical approach by employing both structural equation modeling (SEM) and fuzzy-set qualitative comparative analysis (fsQCA) based on two distinct principles.
This research contributes to the literature in two ways. First, this study represents an early attempt to adopt ELM and offers a theoretical explanation for the different information processing mechanisms when recipients are exposed to various informational cues in AR retailing. The ELM provides theoretical support for identifying the critical antecedents of cardinal behavioral decisions in AR retailing, including informational cues and information-processing outcomes. This study also adds pertinent knowledge to the literature by contextualizing informational cues and processing outcomes under the framework of ELM in AR retailing. Moreover, ELM suggests that the identified informational cues should exert influences on behavioral intentions through the dual routes of information processing: central and peripheral. From the perspective of the information processing mechanism, this study also offers a unique and in-depth understanding of how these central and peripheral routes are operationalized to facilitate eventual decisions surrounding purchasing and continuing in AR retailing.
Second, to unveil the complexity of the information processing and decision-making mechanisms in a realistic situation holistically, this study employs mixed analytic methods incorporating both SEM and fsQCA. These two methods complement each other by providing evidence from distinct principles and offering more comprehensive results (Xie and Tsai, 2021). Most studies apply the one-size-fits-all assumptions proposed in the ELM by examining predetermined associations in the research model (Bhattacherjee and Clive, 2006; Chang et al., 2020). However, this widely adopted approach understates the causal complexity of what drives behaviors since such a narrow scope overlooks other vital combinations of factors when forming a decision process. Researchers attribute the shortcomings of this approach to the mismatch between studying causally complex relationships using conventional symmetric regression methods (such as SEM) (Du and Kim, 2021). This is because it is difficult to comprehensively determine how different combinations of variables cause an outcome by merely adopting SEM (Ragin, 2018). While asymmetric configurational theorizing methods (such as fsQCA) provide suitable approaches that incorporate complex complementarities and nonlinear relationships among constructs. Moreover, fsQCA can provide a unique perspective when interpreting the complexity of the decision-making process since it addresses potential configurations beyond the predetermined associations in theories (Pappas et al., 2020).
To conclude, the mixed-method approach can provide an in-depth understanding of ELM and AR retailing. The application of SEM allows the examination of the predetermined relationships in the research model that was developed based on ELM, extending the explanatory power of ELM in AR retailing (Hung et al., 2020). Moreover, fsQCA can address the complexity of consumers’ decision-making process by revealing all possible configurations of critical factors without any predetermined assumptions. In particular, when all the configurations resulting from fsQCA still include both the central and peripheral factors, the dual-routes process proposed in ELM is verified with two distinct analytic approaches. Therefore, in addition to the associations proposed in ELM, configurations beyond predetermined relationships provide a more complete picture of individual information process mechanisms in AR retailing under the tenet of ELM.
2. Literature review and theoretical framework
2.1 Literature in augmented reality
We searched the relevant literature using an approach incorporating a systematic review and a “cited reference search” in the Web of Science bibliographic databases, which is the largest database of research literature. Using the keyword “augmented reality” combined with several other keywords (such as “retailing” and “marketing”). We collected academic articles published in the recent 3 years (2020–2022) which are included in the core collection of the databases in the business field. To ensure the quality of the papers, scrutinized screening and initial assessment were conducted by reading the abstracts of the literature published in reputable journals. After assessing their relevance to the topic, 37 papers were selected and reviewed, as shown in Appendix 1.
One of the main streams of existing studies has focused on exploring the features of AR shopping. Among the features investigated, informativeness was widely suggested as one of the key values of AR retailing (Chiu et al., 2021; Riar et al., 2022; Sun et al., 2022; Yuan et al., 2021). Researchers have suggested that a variety of AR features are designed to empower customers with rich information regarding a product. For example, the interactivity of AR enables consumers to explore and inspect products thoroughly by rotating and zooming in or out (Park and Yoo, 2020; Poushneh, 2021), while virtual presence allows consumers to better experience product usage effects by moving virtual products into diverse real surroundings (Javornik et al., 2021; Qin et al., 2021a; Sun et al., 2022). In this way, consumers can acquire sufficient details about a product and a variety of contextual information. The vividness of AR technology facilitates clear presentations and the aesthetic design of visual products to enable consumers to acquire high-quality information about the products (Nikhashemi et al., 2021; Whang et al., 2021). To conclude, it has been widely indicated that AR creates an informed environment (Barhorst et al., 2021; Sun et al., 2022). Accordingly, our investigation focused on the informational cues provided in AR retailing that could address the common understanding of AR in the previous literature.
Existing studies have attempted to explore AR retailing from different theoretical perspectives, such as the cognition-affect-conation (Qin et al., 2021a), telepresence (Whang et al., 2021) and experience economy theories (Jung et al., 2021). For example, highlighting the antecedents that influence consumers’ attitudes and behaviors, Qin et al. (2021b) employed the S-O-R framework to explore consumers’ AR shopping experiences. Similarly, by employing the TAM framework, Chiang et al. (2021) highlighted the influence of the usefulness and ease of use of AR on consumers’ behavioral intentions. Through the lens of flow theory, Barhorst et al. (2021) examined the unique capabilities of AR in facilitating consumers’ state of flow. Although existing studies have offered useful knowledge regarding AR retailing by applying different theories, little research has adopted an information-processing perspective to explore how behavioral decisions are made in AR retailing. Considering informativeness is one cardinal value of AR retailing, the theories focus on information processing, meaning that ELM could be more effective in unveiling the underlying mechanism of decision-making in AR retailing (Chang et al., 2020).
Regarding research methods, prior studies have mainly employed situational or laboratory experiments and questionnaire surveys, and most studies have employed a single research design and analytical technique (such as SEM or linear regression analysis) (Alimamy and Gnoth, 2022; Chiang et al., 2021; Jiang et al., 2021; Sun et al., 2022; Whang et al., 2021; Yuan et al., 2021). Therefore, existing studies only examined the correlations that were assumed to exist between the concepts embedded in a predetermined conceptual model. However, these predetermined associations only can provide one possibility for interpreting the phenomenon of interest, since different combinations of factors cannot be explored using SEM or linear regression analyses (Pappas and Woodside, 2021). In addition, it might be difficult to produce comprehensive and robust findings using a single research design (Duarte and Pinho, 2019). Given that AR retailing is an emerging field that has not been fully investigated, a mixed-method approach was selected to explore this novel phenomenon. This would address both the correlations and different configurations of factors at the case level and could provide robust and richer findings to reveal the underlying solutions that result in behavioral decisions in AR retailing.
2.2 Elaboration likelihood model (ELM)
The ELM provides a framework to describe how recipients process information that affects their attitudes or behaviors (Petty and Cacioppo, 1986). According to ELM, the information process is conducted through dual routes: central and peripheral routes. These dual routes are distinguished based on the depth of recipients’ cognition endeavored in information processing and the extent of “elaboration” (Bhattacherjee and Clive, 2006). The central route is a cognitive-based process with more cognitive endeavors; individuals tend to ponder and scrutinize the received information and comprehensively consider related issues rationally before decision-making (Ham et al., 2019; Thomas et al., 2019). It usually relates to the assessment of the information quality (Bhattacherjee and Clive, 2006) and establishes a cognitive-based perception of products. Thus, information quality is most frequently adopted as a central route factor (Chang et al., 2020; Thomas et al., 2019). In contrast, individuals who process information through the peripheral route exert less cognitive effort. Instead of focusing on information quality, they tend to rely on heuristic cues and emotional aspects of the environment with less elaboration (Li, 2013). Source credibility is most frequently adopted as a peripheral route factor in current studies (Bhattacherjee and Clive, 2006; Chang et al., 2020; Thomas et al., 2019).
The ELM has been successfully applied in previous studies about online retailing to explain how decisions are influenced by messages through distinct information processing routes (Li et al., 2023). These include the determinants of review credibility and their influence on consumers’ purchase intention (Thomas et al., 2019), the diffusion of health misinformation on social media (Zhao et al., 2021), attitudes toward posts in Facebook’s secondhand marketplace (Chang et al., 2020), and the key antecedents of crowdlending investor satisfaction (Ribeiro-Navarrete et al., 2021). Researchers have suggested that the ELM can offer new insights into unaddressed areas, such as the constantly changing e-commerce environments facilitated by novel technologies (Barhorst et al., 2021). This is because the ELM addresses how diverse context-based messages can play different roles in shaping distinct users’ elaboration paths (Bhattacherjee and Clive, 2006). Accordingly, the underlying mechanism of the ELM can provide a unique and contextual interpretation of consumers’ information processes in AR retailing.
The unique setting of AR can facilitate the success of the two previously mentioned processing routes. On the one hand, consumers have to devote time, effort and cognition to comprehensively evaluate the product information displayed by AR before believing the information is trustworthy (Alimamy and Gnoth, 2022; Fan et al., 2020). Moreover, AR retailing has unique advantages, such as a variety of display formats and real-time interactions with virtual products in physical surroundings (Whang et al., 2021). Thus, by facilitating cognitive-based routes, AR retailing empowers consumers to obtain information and judge its quality more effectively, develop positive attitudes toward a product and make behavioral decisions. On the other hand, consumers employ peripheral routes to process heuristic cues in AR retailing. They exert less cognitive effort when developing impressions or personal preferences regarding AR retailing based on their experiences and feelings. The innovation of AR retailing can create superior experiences, pleasant shopping scenarios and appealing special effects (Daassi and Debbabi, 2021; Huang et al., 2019). Thus, by enriching suggestive cues in the environment, AR retailing facilitates consumers taking the peripheral route and eventually leads to consumers’ continuing in this novel shopping scenario.
Existing studies have not applied the ELM as the dominant tenet to investigate AR retailing, although some researchers have attempted to employ the ELM as a supplemental theory to explain partial relationships proposed in their research models. For example, Barhorst et al. (2021) took flow theory as the main perspective and merely examined the direct association between the state of flow and elaboration outcomes in AR settings. Similarly, Chen et al. (2021b) partially applied the ELM in their research model and proposed that diagnosticity and arousal are two cues that directly result in AR users’ impulsive buying behavior. Consequently, researchers have neither explored how to contextualize rich informational cues based on AR features nor have they investigated the information processing mechanism under the effects of these informational cues in AR retailing. In this study, we endeavor to employ the ELM as the dominant theoretical perspective. This study aims to explore how consumers process and react to the presented information in AR retailing. Thus, it is suitable to apply ELM to understand consumers’ information influencing and processing route. This will offer a unique and in-depth understanding of how the central and peripheral routes are operationalized to facilitate eventual decisions on whether to purchase products and continue use in AR retailing.
2.3 Trust transfer theory and elaboration likelihood model
Trust transfer theory focuses on the overall process of how an individual’s trust is transferred from one object to another object (Stewart, 2003). The approach of trust transfer theory concentrates on the general trust as a holistic measure to grasp the core of the trust transfer mechanism and outcome, rather than delving trust into the intricate components that define trustees’ characteristics (i.e. honesty, benevolence and competence).
Online shopping consumers will ground their product choices on less information than in-store consumers who can sensorially experience products in the store (Kowalczuk et al., 2021). In particular, AR retailing integrates unfamiliar AR technology into the uncertain environment of online shopping (Sun et al., 2022). Thus, exploring how to improve consumers' trust in product information and AR retailing has become an imperative issue. Prior studies have suggested that trust in online shopping is established through cognitive reasoning and affective perception (Shen et al., 2019). These two approaches comply naturally with the underlying assumption of the two routes in the ELM (Bhattacherjee and Clive, 2006). Specifically, consumers need to put more time and effort into evaluating, scrutinizing and analyzing product-related information to establish cognitive-based trust in products (Chen et al., 2021a; Gwebu et al., 2014). In other words, trust in product information can be established as the outcome of the central route with cognition endeavored. Trust in product information refers to consumers’ belief that the product information provided in AR retailing will satisfy their expectations, such as appearance and usage effects (Chen et al., 2020). The peripheral cues in AR retailing can be utilized by consumers to develop affective-based trust when they are not able to elaborate on product information carefully, since they may lack time, energy, or knowledge (Yang et al., 2019). For example, the popularity of AR, the appealing effects of augmented images and the superior shopping experience could induce consumers’ affective feelings and generate their trust in AR retailing. Thus, trust in AR retailing refers to consumers’ belief that it is trustworthy without any cheating or misguiding (Chen et al., 2020).
Trust transfer theory can be adopted to explain how the outcomes of the peripheral route may be associated with the outcome of the central route. Trust transfer is a cognitive process that explains how trust in an unfamiliar object can be derived from a trusted object due to associations between the two objects (Stewart, 2003). When unknown objects and the trusted entity are correlated or interplayed, the consumer is inclined to trust the targeted object based on the existence of perceived relational bonds with the trusted object (Shen et al., 2019). Among these relational bonds, prior studies in e-commerce have proved that faith in a credible source that provides product information can be transferred to trust in the product (Chen et al., 2020; Shao and Yin, 2019; Zhao et al., 2019). Moreover, it is apparent that consumers’ trust in the new technology incorporated in e-commerce is a prerequisite for encouraging faith in shopping online (Alimamy and Gnoth, 2022). Similarly, consumers often learn about AR technology directly and intuitively through the peripheral route which requires less effort. Hence, it is easier for them to become familiar with and generate trust in AR technology. Finally, since consumers rely on AR retailing to display information (Qin et al., 2021b), trust in product information can be transferred from their trust in AR retailing.
3. Research design: a three-staged approach
The research approach is designed to answer the main research question of “how the consumers’ decision-making process is influenced in AR retailing?”. Developing a research model integrating ELM and trust transfer theory, this study applies a three-staged research design and a robust analytical approach by incorporating SEM and fsQCA to reveal systematically the complexity of consumers’ information processing and decision-making process in AR retailing. These three research stages are conducted to achieve three sub-research purposes using three research approaches and data analysis methods. As shown in Figure 1, we have presented a general framework to explicitly present the structure of the proposed methods in answering the sub-research questions and their roles in completing the whole research. We will elaborate on each stage in the following three sub-sections.
3.1 Stage 1: the development of conceptual framework
In stage 1, the research purpose is to answer the sub-research question 1 “What are the critical factors that influence consumers’ behavioral decisions in AR retailing?”. Thus, based on a systematic literature review regarding AR, ELM and trust transfer theory, this study develops a conceptual framework and classifies the potential factors based on its conceptual nature that determines the underlying AR retailing consumers’ decision-making mechanisms.
Based on a systematic literature review, individuals’ information process routes can be explained by the ELM framework through central and peripheral routes (Bhattacherjee and Clive, 2006). In addition, through integrating trust transfer theory, the establishment of consumers’ trust in online shopping contexts complies naturally with the underlying assumption of the two routes in the ELM (Shen et al., 2019). Specifically, consumers’ trust in information can be established as the outcome of the central route with cognition endeavored while trust in information source is facilitated by affective factors through the peripheral route (Chen et al., 2020).
The conceptual framework establishes a holistic perspective to map the identified critical factors in AR retailing, as shown in Figure 2. The developed conceptual framework explores how the ELM framework can be integrated with trust transfer theory and offers a solution regarding how the combination of the two theories can be contextualized in the context of AR retailing.
3.2 Stage 2: research model development and SEM analysis
In stage 2, the research purpose addresses the sub-research question 2 of “What are the consumers’ information-processing mechanisms in AR retailing?”. Identifying the critical factors and mapping them according to the conceptual framework developed in stage 1, the research model is developed in stage 2. The SEM analysis of the research model explores the complex associations between the variables using empirical data. The results not only verify and contextualize the conceptual model proposed in stage 1, but also offer specific explanations toward the associations between the factors, and thus unveil consumers’ underlying information-processing mechanisms in AR retailing.
As shown in Figure 3, the research model is developed to explore the effects of central factors on trust in information (product information) and the effects of peripheral factors on trust in information sources (AR retailing). Moreover, based on the trust transfer theory, the association between trust in information and trust in information sources is examined. At last, this study examines how these factors influence the consumers’ purchase intention and continuance intention in AR retailing. As a result, SEM-based quantitative analysis verifies the research model with empirical evidence and verifies the proposed hypotheses.
3.3 Stage 3: configuration analysis (fsQCA technique)
In stage 3, the research purpose is to answer the sub-research question 3 “What configurations of critical factors effectively determine consumers’ behavioral decisions in AR retailing?”. In Stage 3, this study unveils the complex causality behind consumers’ purchase intention and continuance intention in AR retailing. Taking a step forward in the research field of AR retailing, this study employs the fsQCA to build a configurational model of the antecedents leading to purchase intention and continuance intention in AR retailing. Specifically, incorporating all the central and peripheral route factors, trust in product information, and trust in AR retailing, this study explores more combinations of antecedents that lead to purchase intention and continuance intention rather than only one combination of the pre-determined model. The hybrid analytical approach of SEM and fsQCA is shown in Figure 4.
Considering a variable-oriented technique, SEM tries to estimate the net effect of predictors on dependent variables. Relying on the principles of additive effects, linearity and nonfinality, SEM treats independent variables as competing to explain the variation in the dependent variables (Woodside, 2017). While fsQCA is a hybrid technique combining the elements of qualitative and quantitative approaches, highlighting the combinatorial effects and focusing on set-theoretic associations rather than correlational connections (Ragin, 2018). This method assumes asymmetry between independent and dependent variables and can capture multiple paths that lead to the same outcome. FsQCA offers comprehensive combinations of variables as predictors of the outcome (Du and Kim, 2021). These combinations are usually not identified if the analysis only focuses on the main effects (Pappas and Woodside, 2021; Li et al., 2020). In fact, the real-life consumption environment is complex, and the potential outcome is determined by the combinations of several heuristic cues embedded in the consumption environment. Several pioneers endeavored to employ a hybrid analytical approach incorporating fsQCA analytical techniques to further compensate and validate the SEM analysis (Duarte and Pinho, 2019). It has been indicated that a comparison between the findings analyzed by SEM and fsQCA could not only draw a scientific and robust conclusion toward consistent results but also complement each other by offering distinct interpretations of the empirical data and research model (Ribeiro-Navarrete et al., 2021).
Diagrammatizing the research approach, we present Figure 5 to explain the detailed approaches for each stage, including the specific research method, the proposed research subjects and the expected results and contributions.
4. Research model and hypotheses development
4.1 Effect of the central route factors
As suggested in ELM, information quality is suitable and precise to represent the central route factors in online markets as it requires careful elaboration from the receivers to estimate the reliability of the message (Bhattacherjee and Clive, 2006; Chang et al., 2020). Information quality is represented from four perspectives, namely information completeness, accuracy, format and currency (Nelson et al., 2005). These four dimensions of information quality are contextualized and enhanced in AR retailing and further facilitate the outcome of consumers’ cognition-based information process routes.
Information completeness refers to the extent of sufficient depth and breadth of information (Filieri and McLeay, 2014). The information is perceived as complete when consumers can access necessary, sufficient and ample product-related information (Gao et al., 2021). By allowing the consumers to interact with virtual products intuitively using 360-degree rotation, zooming in and out, and observing the virtual products in physical surroundings (Kowalczuk et al., 2021), AR retailing empowers the consumer to observe the product and usage effects intuitively and check the details of products as real (Daassi and Debbabi, 2021). Thus, information completeness is essential for consumers to make a comprehensive evaluation regarding the trustworthiness of the product information. Hence, following the above notion, we hypothesize:
Information accuracy is the degree to which information is correct, unambiguous, meaningful, believable and consistent (Nelson et al., 2005). Accurate and precise information can enhance information quality by reducing the potential risks due to fake and misleading online information (Filieri and McLeay, 2014). The augmented world can closely mimic real-world sensations; thus, AR retailing can display virtual products like real ones, such as the appearance, shape, color and usage effects. A natural and authentic experience is created to facilitate the consumers’ trust in the product information in AR retailing (Whang et al., 2021). In addition, since consumers can evaluate the actual information without the interference of ambiguous interpretations (Gao et al., 2021), accurate information can enhance consumers’ perceived credibility and trustworthiness toward online products (Thomas et al., 2019). Hence, we hypothesize:
Following the notion of Nelson et al. (2005), we define the information currency as the degree to which the information presented in AR retailing is up to date and reflects the current state of the relevant products. The latest products and novel shopping concepts are promoted to consumers in AR retailing, and the consumers can experience the use effect of new products in time without leaving home (Sun et al., 2022). Thomas et al. (2019) suggested that information currency is very useful for consumers to make a timely estimation of product choice, and further enhances a positive attitude and the perceived credibility toward the information provided. Several studies reveal that timely information has a significant positive effect on users' trust in various contexts. For example, the current online review enables consumers to evaluate the current state of products and has a positive effect on consumers’ attitudes (Filieri et al., 2018). Hence, we proposed the following hypothesis:
Information format refers to the degree to which information is presented as understandable, representational and interpretable to the buyer and thus aids during shopping. This construct highlights how the consumer devotes cognitive effort and ponders deeply over whether the format presenting the information may facilitate his understanding of the products. AR retailing technology greatly enriches forms of displaying product information, such as 3-D interface, dynamic animation, sound, video and graphics (Barhorst et al., 2021). As indicated by media richness theory, richer media forms reduce uncertainty and fuzziness perceived by consumers and influence their decision-making process more easily (Daft and Lengel, 1986). Existing studies have provided ample evidence to support that the formats of presented ways positively influenced the comprehensibility of information and importantly changed users’ trust in the information (Diwanji and Cortese, 2021). Hence, the rich information formats can encourage consumers to believe that the product information presented is trustworthy and reliable in AR retailing (Kowalczuk et al., 2021). Hence, we hypothesize:
4.2 Effect of the peripheral route factors
Ohanian (1990) indicated that the positive characteristics of the information sender also exert a critical role in consumers’ attitudes and behaviors and enhance the recipients’ acceptance of the information. Thus, consumers are influenced by the sources of product information in the process of establishing their perceptions of the environment. Source credibility has been widely acknowledged as the peripheral cue in existing ELM studies (Chang et al., 2020; Gao et al., 2021; Thomas et al., 2019). Since the product information is displayed by AR technology in AR retailing, the credibility of AR technology should be studied as the source credibility. Source credibility indicates how much the consumers perceive the AR technology to be believable, credible and trustworthy (Sussman and Siegal, 2003). Several factors are proposed to represent different aspects of source credibility, such as attractiveness, reputation and expertise (Hu et al., 2019).
The attractiveness of AR retailing refers to the extent to which recipients perceive AR retailing as attractive and appealing (Gao et al., 2021). Prior studies suggest that the attractive features in online settings often exert a positive influence on consumers’ trust-building process, such as attractive advertising ways and attractive design of selling stores (Bonnin, 2020). AR retailing enables an attractive shopping experience blending real, virtual and fantasy (Whang et al., 2021). The attractiveness of AR retailing is deemed a significant peripheral route factor and helps people naturally develop a favorable impression regarding the trustworthiness of activities or technologies that appeal to them (Bonnin, 2020). Hence, we hypothesize:
Popularity means being widely recognized, preferred and adopted by consumers. As a typical peripheral cue, the popularity of AR retailing is likely to generate the “herd effect”. The extent of popularity and preference by the majority will become signals to encourage individuals to follow the trend and generate trust (Chang et al., 2020). As a heuristic cue for individuals, the popularity of AR indicates that the public believes AR technology helps assist online shopping (Oyman et al., 2022). Hence, the popularity of AR retailing is very likely to enhance consumers' confidence and trust (Alimamy and Gnoth, 2022). Hence, we hypothesize:
In our study, the expertise of AR retailing can be contextualized into two aspects. One is the relative advantage that manifests the expertise of AR technology-empowered shopping activities (Hinsch et al., 2020), while the other one is the augmented quality that represents the expertise of AR technology. Relative advantage is the degree to which an individual perceives using AR shopping to be better than traditional shopping ways (Rogers, 2003). If a service can better fulfill consumers' performance requirements than others, consumers are more likely to trust it (Alimamy and Gnoth, 2022). Compared to traditional online shopping, AR retailing has a variety of advantages to better fulfill the consumers’ needs and thus boost shopping confidence (Kowalczuk et al., 2021), such as the advantages of presenting products more intuitively as real (Song et al., 2020). Jiang et al. (2021) indicated that AR had the advantage of offering a more enriched shopping experience that can enhance the consumers’ perceptions of its reliability and trustworthiness. Hence, we hypothesize:
Augmented quality represents the core ability of AR technology, it refers to the ability to elaborate the perceptual quality of an augmentation experience as if products will appear in real surroundings (Nikhashemi et al., 2021). For example, AR combines virtual and reality and thus enables consumers to feel that the virtual clothes are wearing on their physical body or that a digital sofa appears in their living room (Barhorst et al., 2021). Higher augmented quality of AR retailing provides consumers with a more realistic shopping experience and leads to a lower discrepancy between the virtual and actual product experience, which results in a more positive attitude toward AR retailing (Daassi and Debbabi, 2021). Hence, we hypothesize:
4.3 Effect of trust
As stated by trust transfer theory (Stewart, 2003), trust in an unknown target can be transferred from a known and familiar target when they are related and associated with each other. A known or trusted target for consumers will be what they feel familiar with or can directly interact with (Chen et al., 2020). Zhao et al. (2019) demonstrated that consumers’ trust in the sellers can be transferred to the brand recommended by the seller. Shen et al. (2019) also indicated that consumers’ trust in a recommended product is often derived from their trust in their friends who recommend it. In the same vein, consumers can form faith in an unfamiliar product that is recommended and presented by AR retailing, since they have perceived AR retailing as trustworthy. Hence, we hypothesize:
By eliminating the perceived risk and uncertainty, trust can enhance consumers’ shopping confidence and thus promote consumers’ purchase intention (Barhorst et al., 2021). Existing studies have demonstrated the critical role of trust in promoting consumers’ purchase intention (Chen et al., 2020; Zhao et al., 2019). Trust in the product information resulting from the central route and the trust in AR retailing deriving from the peripheral route can affect consumers' purchase intention. Thus, we propose:
Continuance intention refers to the degree to which consumers intend to use AR retailing continuously in the future (Qin et al., 2021a). While trust is an influential factor in reducing uncertainty and enhancing technology usage (Alimamy and Gnoth, 2022). Trust improves the understanding and reliability between services and customers, and thus inherently links to consumers’ values, needs and interests (Chen et al., 2020). Thus, trust is essential for relationship building that results in consumers’ willingness to develop and maintain the relationship with AR retailing (Arghashi and Yuksel, 2022). Following this notion, Chen et al. (2020) indicated that two types of trust including trust in product information and trust in streamers can jointly promote consumers’ behavioral intentions. In the same vein, trust in product information and trust in AR can jointly impact consumers’ continuance intention of using AR retailing. Hence, we hypothesize:
5. Measurements and data collection
5.1 Measurements
We adapted items measuring the corresponding constructs based on prior literature. Slight modifications are made to better suit our research scope. The items for measuring information completeness and information accuracy were adapted from the work of Luo et al. (2013) and Chen (2010), respectively. The items of information currency and information format are extracted from the work of Nelson et al. (2005) and Huang et al. (2019) respectively. Attractiveness and popularity of AR were measured using the scale from Yang and Lee (2019) and Milman et al. (2020), respectively. Relative advantage and augmented quality were obtained from Johnson et al. (2018) and Lee et al. (2021), respectively. In addition, trust in product information and trust in AR are both modified based on the study of Chen et al. (2021a). The items of purchase intention and continuance intention were adapted from the studies by Whang et al. (2021) and Chiu et al. (2021) respectively. A five-point Likert scale anchored from 1 (completely disagree) to 5 (completely agree) was employed. Appendix 2 presents all constructs along with descriptive analysis and factor loadings.
To ensure the content validity of the questionnaire, we conducted a group discussion and a small-scale pilot test. On the one hand, we invited nine experts including AR practitioners and professors who have experience in AR retailing to review the questionnaire, such as the wording, expression and understandability. The content validity ratio is calculated and the results are acceptable and above the required minimum criterion of 0.56 (Lawshe, 1975). We modified some questions according to experts’ advice to ensure the questionnaire quality. On the other hand, we collected 81 valid responses in the pilot test to further ensure the suitability of our questionnaire. At last, the formal questionnaire includes a motivation letter, a confidentiality statement regarding respondents’ information, and items measuring the constructs.
5.2 Data collection
In China, the dominant e-commerce platforms are keen on incorporating AR to facilitate consumers’ experiences, such as Alibaba and Jingdong. These dominant e-commerce platforms adopt a similar strategy and scenario to apply AR-based online shopping. For example, this platform allows consumers to try virtual make-up on their faces, virtual clothes on their bodies, and furniture in their rooms. Consumers can scan their faces with mobile phones’ cameras, and click different virtual lipsticks to see their effects.
We employed a leading web-survey website (https://www.sojump. com/) in China to distribute questionnaires from May 1st to May 20th, 2022. The website has 194 million users and has collected 15.63 billion questionnaires. To ensure that the respondents can better understand our research context, we have provided many pictures at the beginning of the questionnaire. These pictures presented the common AR scenarios while using these dominant e-commerce platforms. To ensure the respondents had experienced AR retailing, after the consumer browsed the variety of images, we proposed the screening question: “Did you have experience with AR-based online shopping?”. Only the consumers who provided affirmative answers can continually answer the questionnaire. Parental permission and participant consent were obtained before obtaining the information of the children.
In addition, we screened out invalid questionnaires based on three criteria. First, the respondents did not completely answer the questionnaire items. Second, the respondents gave the same answer to all the questions. Third, the respondents did not pass the attention check questions, such as the repeated and opposing questions. Finally, 546 valid questionnaires were collected and applied for the formal data analysis. As indicated by Faul et al. (2009), G*Power is applied to estimate the minimum required sample size. With a 0.05 alpha level and 0.95 power, 121 is required and our sample is largely surpassed. In addition, our sample size is over ten times the number of the formative indicators used to measure the constructs and the number of structural paths directed in the structural model, which implies our sample size has met the threshold and sufficient statistical power while detecting relational effects (Hair et al., 2017).
The respondents’ demographic information is shown in Appendix 3. In terms of gender, 51.6% of the respondents were male (n = 282) and 48.4% were female (n = 264), which is in line with the existing studies (Qin et al., 2021a, b; Hinsch et al., 2020; Bonnin, 2020; Song et al., 2020). The slightly higher percentages of male users seem in line with the observation of the phenomenon. AR is a novel technology which has not yet been fully popularized among the general population. Existing investigations have shown that more males are more likely to try new technologies (Davis, 1989). Among the participants, 39.6% are aged between 18 and 24 years, while 29.1% of them are 25–30 years old, and 85.7% of the respondents had obtained a bachelor’s degree or above, which indicates that the respondents were primarily young and well-educated. In addition, 33.3% of the respondents have experienced AR retailing for about half a year to 1 year, and 45.4% of them have the experience for 1–2 years. The distribution of the respondents is similar to prior studies (Arghashi and Yuksel, 2022; Jiang et al., 2021).
6. Data analysis
This study applied a mixed analytical approach. Incorporating two analytical techniques of SEM and fsQCA, this study takes a two-phase process in data analysis. Employing R-Studio with package composite-based structural equation modeling (cSEM) (Rademaker and Schuberth, 2020), this study first validates the measurement model and examines the hypotheses in the structural model using SEM. Second, we analyzed the data using fsQCA 3.0 to generate insights into configurations of variables in an asymmetric way. A collection of results produced by distinct techniques validates the predetermined associations and provides alternative configurations of the constructs identified in ELM, which enriches the understanding of how consumers process information in AR retailing.
6.1 PLS-SEM
6.1.1 Measurement model
PLS-SEM is used as the primary analytical tool for the quantitative data analysis stage by using R-Studio with cSEM (Rademaker and Schuberth, 2020). The measurement quality of the proposed model was tested through construct reliability, convergent validity and discriminant validity (Li et al., 2022). Incorporating two analytical techniques of SEM and fsQCA, this study takes a two-phase process in data analysis. Employing R-Studio with package cSEM (Rademaker and Schuberth, 2020), this study first validates the measurement model and examines the hypotheses in the structural model using SEM. Second, we analyzed the data using fsQCA 3.0 to generate insights into configurations of variables in an asymmetric way. A collection of results produced by distinct techniques validates the predetermined associations and provides alternative configurations of the constructs identified in ELM, which enriches the understanding of how consumers process information in AR retailing (Hung et al., 2020).
Cronbach’s alpha and composite reliability (CR) were examined to ensure the reliability of the measurement model. All the values exceed 0.7, which indicates a satisfactory reliability level (Fornell and Larcker, 1981). In addition, for establishing convergent validity, factor loadings and the average variance extracted (AVE) were calculated. All the factor loadings of each construct are above the thresholds of 0.7 (Hair et al., 2010) and the values of AVE are above 0.50 (Chin, 1998). The results showed sufficient convergent validity. All these results are presented in Appendix 4.
Moreover, this study employed two methods to ensure discriminant validity. First, it is assessed by the Fornell- Larcker criterion. We present the results in Appendix 5, which show the intercorrelations of all the constructs are lower than the square roots of the AVE values (Fornell and Larcker, 1981). Second, the discriminant validity was also tested through the heterotrait-monotrait (HTMT) ratio of correlations approach (Henseler et al., 2015). The obtained values shown in Appendix 6 are all below the cut-off of 0.85. Both two methods suggest great discriminant validity and all constructs are valid and reliable in our study.
6.1.2 Common method bias
We applied two methods to estimate the potential common method bias (CMB). Firstly, following the guidelines of Kock (2015), we examined the full collinearity VIF scores for each item of our research model’s constructs. The results show that the VIF scores ranged from 1.377 to 2.17 (see Appendix 4), all of which fall significantly below the threshold of 3.3. This indicates that CMB is not a main concern in this study (Kock, 2015).
Secondly, we applied the marker variable approach as suggested by Williams et al. (2010). As shown in Appendix 7, the correlations between the marker variable and other variables of interest are from 0.043 to 0.132, all below the recommended threshold (0.3) (Charterina et al., 2018). This reflects that there is no potential threat to CMB. In addition, the marker variable was used as an exogenous variable to estimate whether the significance of the paths will be affected by the method factor (Wang et al., 2016). The results suggest that the significant paths in the baseline model remain significant in the method factor model, indicating that CMB is not a threat in this study.
6.1.3 Structural model
We estimate the structural model and present the path coefficients, statistical significance and explanatory power of the constructs in Appendix 8. Our empirical analysis supports all the proposed hypotheses. Specifically, regarding the central route, information completeness (β = 0.240; p < 0.001) and information accuracy (β = 0.273; p < 0.001) are the key determinants of trust in product information, followed by information format (β = 0.191; p < 0.001). Whereas the effect of information currency only exerts marginal effect on trust in product information (β = 0.113; p < 0.05). In the peripheral route, the popularity of AR (β = 0.284; p < 0.001) is the most influential variable on trust in AR retailing, followed by the attractiveness of AR (β = 0.248; p < 0.001), relative advantage (β = 0.240; p < 0.001) and augmented quality (β = 0.158; p < 0.01). In addition, trust in AR retailing exerts a significant influence on trust in product information (β = 0.159; p < 0.001). Moreover, the influence of trust in information and trust in AR retailing on purchase intention and continuance intention were both supported by the empirical data (H10a: β = 0.386; p < 0.001; H10b: β = 0.247; p < 0.01; H11a: β = 0.134; p < 0.001; H11b: β = 0.509; p < 0.001). Overall, trust in product information, trust in AR retailing, purchase intention and continuance intention were explained by 52.0%, 52.8%, 49.2% and 33.3% variances respectively. The results of the structural model analysis are shown in Figure 6.
As suggested by prior research (Shao et al., 2023; Duarte and Pinho, 2019), this study adopted the basic demographic variables including consumer’s gender, age and education as control variables in our research model. Specifically, the age variable has a −0.059 effect (p-value = 0.396) and 0.012 effect (p-value = 0.884) on consumers’ purchase intention and continuance intention, respectively. The variable gender has a 0.009 effect (p-value = 0.890) on purchase intention and a −0.007 effect (p-value = 0.925) on continuance intention. In addition, education has a −0.062 effect (p-value = 0.49) and a −0.017 effect (p-value = 0.848), respectively. Regarding the experience with AR retailing, the effect on purchase intention is −0.014 (p-value = 0.912) and the effect on continuance intention is −0.041 (p-value = 0.764). Thus, the analysis results show that control variables including age, gender, education and experience with AR retailing exert non-significant roles in consumers’ purchase and continuance intention.
6.2 fsQCA
6.2.1 Analysis process
The fsQCA analysis process consists of several steps, including calibrating the data by fsQCA software, conducting the analysis of the necessary conditions and obtaining the configurations by sorting the truth table by frequency and consistency.
Data calibration, a critical step in fsQCA, transforms factors into fuzzy sets, with a value range of 0–1 (Ragin, 2018). Before calibrating the measures for constructs, the standard score of each construct was obtained to ensure all constructs were measured unidimensional (Anderson and Gerbing, 1988). Three qualitative thresholds of 95th, 50th and 5th percentiles were applied to calibrate all the values to full membership, intermediate membership and full non-membership respectively (Pappas and Woodside, 2021), using the quartiles method of calibrating fuzzy sets (Xie and Tsai, 2021). The three anchors of each construct were identified and presented in Appendix 9. In addition, considering it is difficult to analyze the conditions in the cases exactly on 0.5, the analysis will drop these cases (i.e. intermediate-set membership) in fsQCA (Ragin, 2008). Thus, after calibration, this study followed the principles of Fiss (2011) and endeavored to avoid the cases exactly on 0.5 by adding 0.001 to the values that were exactly 0.50.
Following established qualitative comparative analysis (QCA) practices and indications of prior studies, a necessary conditions analysis was conducted with a consistency benchmark of 0.90 (Ragin, 2008), where a consistency score over 0.9 indicates the condition is necessary for explaining outcomes (Ragin, 2008). Necessary conditions analysis both for the presence of purchase and continuance intention and for their negation is conducted. The results showed that there is no single factor that can necessarily influence consumers’ behavioral decisions. The results are shown in Appendix 10.
In fsQCA, the resulting configurations are obtained by first computing a truth table of 2k rows with k number of predictors, where each row represents every possible combination of predictors. Then, depending on the frequency and consistency thresholds the table is sorted. Following the instruction of Fiss (2011), the frequency threshold should be 3 when the sample size is over 150, and the consistency threshold should be 0.8 (Pappas and Woodside, 2021). Proportional reduction in inconsistency referred to as PRI consistency is an alternate measure of consistency (Pappas and Woodside, 2021). The score threshold of PRI consistency is set as 0.7 (Du and Kim, 2021) so that a configuration does not simultaneously occur both in the case of the presence of the outcome and the negation of the outcome.
fsQCA offers three sets of solutions for researchers, including complex, parsimonious and intermediate solutions. Researchers widely suggest using parsimonious and intermediate solutions to better interpret the results since the interpretation should both address the generalization and details (Pappas and Woodside, 2021). While all the possible combinations of conditions are presented in the complex solution, thus it may be insufficient in providing a concise interpretation. Following the prior notions, the intermediate solution is presented in this paper which is an optimal trade-off between the presentation of the complex and parsimonious solution. With the intermediate solution, both core and peripheral conditions can be identified (Fiss, 2011). Conditions that are part of both intermediate solutions and parsimonious solutions are considered the core conditions whereas conditions present only in intermediate solutions are called peripheral conditions (Ragin, 2008).
6.2.2 Analysis results
Because of the two distinct outcomes in our study, we first test the sufficient configurations for purchase intention and then test for continuance intention. Consistency depicts the degree to which all configurations together consistently result in the outcome, and coverage evaluates the share of observations explained by the solution (Ragin, 2008). The recommended threshold for overall solution coverage is 0.45 (Duarte and Pinho, 2019) and overall solution consistency is 0.8 (Woodside, 2017). The overall solution consistency and coverage values for purchase intention are 0.94 and 0.53 respectively. As for continuance intention, the overall consistency of 0.91 and overall coverage of 0.53 was obtained from the solution. Thus, all the values are greater than the recommended values. We report our results using the conventional notations: core conditions are present ● or absence/negation ⊗, peripheral conditions are present ● or absent ⊗, and a blank space depicts the presence or absence of conditions that do not matter (Fiss, 2011).
Table 1 illustrates eleven configurations sufficient for purchase intention. We summarize the 11 paths into four categories. In Category 1, solutions 1a and 1b incorporate all the central and peripheral factors, including four factors related to information quality and four factors related to source credibility, while two factors related to information quality are core factors in both solutions. The results reveal that the solutions including all the informational cues identified in ELM are one useful approach to predicting purchase intention.
In Category 2, the common ground in solutions of 2a to 2e is the inclusion of factors related to information quality in different combinations and implies that two factors related to information quality are core factors. These five solutions unveil another approach to interpreting purchase intention and suggest that the significance of factors related to source credibility may vary across different combinations.
In Category 3, solutions 3a and 3b simultaneously highlight information completeness and accuracy are core factors, while other factors may play peripheral effects in predicting purchase intention and their significance in combinations may vary.
In Category 4, different from solutions 3a and 3b, solutions 4a and 4b suggest that all the other factors play a peripheral effect in making purchase decisions, while these two solutions need only one core factor represented by either information accuracy or information completeness.
The fsQCA results demonstrate that both the factors related to information quality and source credibility are necessary for predicting purchase intention in all solutions, while information quality plays a leading role. Only the factors (information completeness and information accuracy) related to information quality are the core factors in both parsimonious solutions and intermediate solutions.
fsQCA produced nine solutions contributing to continuance intention, see Table 2. Interestingly, the core factors determining continuance intention are different from the core factors (information quality) predicting purchase intention to a great extent. The factors related to source credibility play a significant role in determining continuance intention in these nine solutions. These results unveil that the solutions and critical factors in predicting different behaviors show great differences. We summarize the solutions into four categories. In Category 1, solutions 1a and 1b incorporate all the central and peripheral factors, including four factors related to information quality and four factors related to source credibility, while two factors (e.g. attractiveness and relative advantage of AR retailing) related to source credibility are core factors in both solutions. The results reveal that the solutions including all the informational cues identified in ELM can be one approach to predicting continuance intention.
In Category 2, the common ground in solutions of 2a to 2c is the inclusion of factors related to source credibility in different combinations and implies that two factors related to source credibility are core factors in these 3 solutions. These three solutions unveil another approach to interpreting continuance intention and suggest that the significance of factors related to information quality may vary in different combinations.
In Category 3, the common ground in solutions of 3a to 3c is that the factors related to information quality cannot provide sufficient interpretations without the core factors related to source credibility. Specifically speaking, information format, augmented quality and trust in AR retailing are three core factors in these three solutions, and two of these indispensable factors are closely related to source credibility.
In Category 4, solution 4 provides stronger evidence of the significance of factors related to source credibility. The solution suggests that no factors related to information quality are core factors while only two factors related to source credibility are indispensable and cardinal in predicting continuance intention.
These fsQCA results demonstrate two major conclusions. First, both the factors related to information quality and source credibility are necessary for predicting continuance intention in all solutions. Second, factors related to source credibility play a leading role in determining continuance intention. All the core factors are related to source credibility in two of the three parsimonious solutions (Attractiveness*Relative advantage and Augmented quality*Relative advantage*Trust in AR retailing). Only one factor related to information quality is identified as the core factor in one solution, while the rest of the core factors are related to source credibility (Information format*Augmented quality*Trust in AR retailing). Moreover, when Solutions 1a to 2c identify all the factors related to source credibility as the necessary factors, none of the factors related to information quality is identified as an indispensable factor.
6.2.3 Robustness checks
Following the notion of previous studies, some additional checks are conducted in this study for the robustness of our sufficiency analyses (Du and Kim, 2021; Lewellyn, 2018). First, the analysis is repeated by applying a higher consistency threshold of 0.85 (compared to the raw consistency benchmark of 0.8 used in our models). The result showed the same subset of the initial solutions. This change produced the same configuration results and confirmed a perfect subset of the initial solutions (Schneider and Wagemann, 2010). Second, we reran our sufficiency analysis with a higher proportional reduction in the inconsistency of PRI is greater than or equal to 0.75 (Du and Kim, 2021), compared to the original PRI is greater than or equal to 0.70. The solutions remained similar. It also showed the same configuration results as the initial, which further confirmed the robustness of our study. Third, as indicated by Lewellyn (2018), we recalibrated the data by applying the values of the 75th and 25th percentiles. We obtained similar sufficient configurations with these modified calibration settings as well. The results of the various robustness checks all show that the findings of our study are robust.
7. Discussion and implications
7.1 Discussion
By employing a mixed analytical approach that combines PLS-SEM with fsQCA, this study examines the hypotheses proposed in the research model and investigates how all antecedents may form a variety of configurations to interpret consumers’ behavioral intentions in AR retailing. A mixed analytical technique can produce interesting results for discussion. The results of this study can elaborate on the discussion in four ways. First, we discuss how the results of PLS-SEM and fsQCA explain purchase decisions. Second, we focus on the determinants of continuance. Third, the findings of trust in product information are illustrated, followed by a discussion regarding trust in AR retailing. Finally, we present the findings on associations based on trust transfer.
7.1.1 Discussion on purchase intention
This study aims to reveal the mechanisms for determining purchase and continuance intention. Regarding the determinants of purchase intention, rich results are provided by both the SEM and fsQCA. Based on the results of SEM, trust in both product information (β = 0.386; p < 0.001) and AR retailing (β = 0.247; p < 0.001) exert significant effects on purchase intention. The fsQCA confirms the same findings by identifying these two types of trust as the necessary factors in 9 of the 11 combinations as presented in solutions 2a to 4b. In addition, trust in product information is the outcome of the central route and is determined by factors that reflect information quality. In comparison, trust in AR retailing is the outcome of the peripheral route and is determined by factors pertaining to source credibility. These results imply that purchase intention is jointly determined by dual routes that include outcomes and informational cues. The conclusions comply with the basic assumptions of the ELM, while the fsQCA results support the same conclusion and reveal that both central and peripheral route factors are necessary for predicting purchase intention in all the solutions identified. To conclude, both the SEM and fsQCA results imply that neither the central route nor the peripheral route alone is sufficient to engender purchase intention.
The SEM results suggest that the effect of trust in product information on purchase intention is stronger than the effect of trust on AR retailing. Prior studies have indicated that consumers often make rational shopping decisions based on their evaluations of product information quality (Gao et al., 2021), thus, the trustworthiness of product information plays a critical role in consumers’ purchasing intentions. It can also be implied that the factors in the central route are more important in influencing purchase intention. The findings produced by fsQCA confirm the SEM results and provide richer and more comprehensive interpretations of the mechanism of purchase intentions.
First, the fsQCA results highlight the significance of factors that reflect information quality in the central route, especially the completeness and accuracy of information. These results are in accordance with prior studies, which suggested that higher information quality can effectively reduce the problem of information asymmetry, enhance consumers’ shopping confidence and promote purchasing intention. The fsQCA results also suggest that information completeness and information accuracy are the only core factors in parsimonious and intermediate solutions. Second, the fsQCA results complement the SEM results with new and richer findings. On the one hand, fsQCA provides solutions that show different combinations of related factors to explain purchase intention in AR retailing, while SEM only shows the results of the associations proposed in the predetermined research model. On the other hand, fsQCA offers the potential to identify indispensable factors, helping to remove any unnecessary factors in different solutions. For example, Solution 2a reveals that AR attractiveness should not be included, and popularity is not necessary for this specific solution. In other words, the fsQCA provides a different principle for simplifying and optimizing the SEM results and offers a rich interpretation of the purchase decision process in AR retailing.
7.1.2 Discussion on continuance intention
By integrating the results generated from the SEM and fsQCA, this study provides a holistic review to interpret continuance intention in AR retailing. First, although continuance intention is determined by the factors in both the central and peripheral routes, the peripheral route plays a leading role. These results are consistent with the SEM results, suggesting that trust in AR retailing (β = 0.509; p < 0.001) as the outcome of the peripheral route exerts a much stronger effect on continuance than the effect of trust in product information (β = 0.134; p < 0.01). These results reveal that the solutions and critical factors for predicting purchase and continuance intentions exhibit significant differences, which was verified by the results of both the SEM and fsQCA. Second, fsQCA provided 9 solutions that used different combinations of related factors to explain continuance intention in AR retailing. Similarly, several indispensable factors and unnecessary factors were identified in different solutions by the fsQCA. For example, Solution 2b reveals that information format and completeness are not necessary for this specific solution. Other examples are the different roles of popularity in SEM and the several solutions in fsQCA. In contrast to the SEM results, popularity was not identified as an indispensable factor in Solutions 3b and 3c, which provides a different understanding of the role of popularity in different solutions. We conjecture that popularity reflects the objective status of AR retailing, while people mainly rely on their subjective estimations to form a solution leading to continuance intention. This is exemplified by the significance of augmented quality in the solutions since it reflects peoples’ subjective perceptions of the degree to which the virtual products are being augmented in their personalized physical setting.
7.1.3 Discussion on information processing outcomes (trust)
The SEM analysis part of the research model also provided detailed results regarding the associations between informational cues and information processing outcomes. The results of this study confirm the significant effects of the four factors reflecting information quality on trust in information, which is explained by the 52.0% variances. These results are consistent with observations of the phenomenon and the conclusions of prior studies (Gao et al., 2021; Tam et al., 2020; Yang, 2021). In particular, the significance of the four factors may be due to the features of AR. Moreover, AR retailing can display product information more effectively and enable consumers to easily perceive and evaluate every dimension of a product’s information (Kowalczuk et al., 2021). In addition, compared to information completeness (β = 0.240; p < 0.001) and accuracy (β = 0.273; p < 0.001), information currency (β = 0.113; p < 0.05) exerts a weaker effect on trust in product information. The fsQCA also highlights information completeness and accuracy are core factors in the solutions. Previous studies have confirmed the effect of information quality on the formation of trust in the central route (Chen et al., 2020). This study provides more detailed findings and indicates that different dimensions of information quality may exert distinct effects on trust in product information displayed by AR retailing. To unveil the effects of peripheral route factors on trust in AR retailing, we contextualize source credibility as attractiveness, popularity, relative advantage and the augmented quality of AR retailing. The results verify the significant effects of the four factors reflecting source credibility on trust in AR retailing, which is explained by the 52.8% variances. The results also highlight the significant effects of these four factors on people’s positive attitudes in AR settings. These findings imply that all contextualized dimensions of source credibility provide ample heuristic cues and help consumers perceive AR retailing as reliable and trustworthy, with less effort and elaboration. In addition, compared to augmented quality (β = 0.158; p < 0.01), the other three factors produce similar effects on trust in AR retailing (β = 0.248, p < 0.001; and β = 0.240, p < 0.001). Few studies have explored the different effects of peripheral factors on trust in AR retailing (Arghashi and Yuksel, 2022). In comparison, this study provides rich findings with the suggestion that consumers believe heuristic cues derived from attractiveness, popularity and relative advantage are more powerful in facilitating the formation of trust in AR retailing.
Finally, the SEM analysis confirms the significant effect of trust in AR retailing on trust in product information. Previous studies have observed trust transfer in online contexts (Chen et al., 2020; Shen et al., 2019), while this study provides empirical evidence to reveal how the different types of trust are established and transferred in AR retailing through two routes proposed in the ELM. The findings reveal that trust in AR retailing resulting from the peripheral route can be transferred to trust in product information generated from the central route, confirming the interrelationship between these two routes.
7.2 Theoretical implications
Based on the ELM, we identified factors related to information quality and source credibility that form dual distinct paths in explaining two behavioral intentions: purchase and continuance. Trust transfer theory was integrated into the research model to explain the outcomes of dual routes and the trust-building process in more detail, and the interrelationships between these different trusts were verified. Importantly, a mixed analytical approach was employed that integrated SEM and fsQCA in a novel context. This study provides theoretical implications for the literature on ELM and AR retailing from the perspectives of theory validation and analytical methods.
Existing studies have not applied the ELM as the dominant tenet to investigate AR retailing (Barhorst et al., 2021, Chen et al., 2021b). Researchers have neither explored how to contextualize rich informational cues based on AR features nor have they investigated the information processing mechanism under the effects of these informational cues in AR retailing. Thus, this study is pioneering in validating and contextualizing ELM theory in AR retailing by applying a mixed analytical approach.
First, we verified the basic underlying assumptions of the ELM using two distinct research methods. On the one hand, we verified the relationships between the central and peripheral route factors, information processing outcomes and eventual behaviors. Thus, the underlying assumption in the ELM about the relationships between constructs was verified by the SEM results. On the other hand, we verified the assumptions of the ELM using the fsQCA results by indicating that neither the central nor peripheral route alone is sufficient to make behavioral decisions. In other words, all the configurations revealed by fsQCA included both central and peripheral factors, meaning that the underlying assumption of the dual routes process proposed in the ELM was verified. The ELM theory has mainly been examined using a single research design in the previous literature. This study is advanced by verifying the indispensable role of both parallel routes and the relationships of their components using cross-validated results. Thus, we confirm the validity and reliability of the ELM with rich and in-depth theoretical understandings.
Second, the application of a mixed analytical method offered a holistic and comprehensive view to understand consumers’ decision-making mechanisms in more detail. Most studies apply the one-size-fits-all assumptions proposed in the ELM by examining predetermined associations in the research model (Bhattacherjee and Clive, 2006; Chang et al., 2020), which understates the causal complexity of what drives behaviors and overlooks the combinations of factors when forming a decision process (Du and Kim, 2021). In this study, the SEM allowed examination of the correlations proposed in the research model, which revealed a possible underlying mechanism proposed by the ELM to predict behavioral intentions in AR retailing. Configuring the factors identified based on the ELM, the fsQCA compensated for the SEM results by enhancing asymmetric thinking concerning the data and increasing the understanding of both the phenomenon of AR retailing and the ELM theory (Barhorst et al., 2021; Chen et al., 2021b). Beyond the predetermined associations in the SEM model, the fsQCA offered 11 and 9 solutions for interpreting purchase and continuance intention, respectively. In addition, by providing parsimonious solutions, the fsQCA offers a different approach to optimizing the SEM results. Moreover, by identifying the indispensable factors and removing any unnecessary ones from different solutions, the fsQCA revealed a concise yet cardinal approach that results in behavioral decisions in AR retailing (Barhorst et al., 2021; Daassi and Debbabi, 2021; Kowalczuk et al., 2021; Sun et al., 2022).
Several theoretical implications are provided in this paper, based on a detailed examination of the research model developed on ELM in AR retailing. Existing studies have mainly focused on the technological characteristics of AR and the benefits of AR in influencing consumers’ motivations, attitudes and behaviors (Daassi and Debbabi, 2021; Hsu et al., 2021; Seong and Hong, 2022; Kowalczuk et al., 2021; Barhorst et al., 2021). No one has explored the consumers’ information-processing mechanism. Thus, incorporating ELM, this study expands the explanatory power of ELM in AR retailing in two ways. First, this study is one of the first to apply ELM as the main theoretical perspective for studying continuance and purchase intention in AR retailing. Moreover, we carefully examined and contextualized the informational cues under the tenet of ELM. The factors related to source credibility were identified, contextualized and mapped as peripheral route factors to explain behavioral intentions in AR retailing.
Second, in this study, we integrated ELM and trust transfer theory to examine the outcomes of central and peripheral routes, as well as their interrelationships in AR retailing. Accordingly, this study offers a richer and more in-depth understanding of how trust in product information and AR retailing was established through distinct information processing routes in AR retailing. Scant research has addressed the issue of trust in AR retailing, and previous studies investigating trust have not examined how diverse informational cues can exert different influences on trust formation (Alimamy and Gnoth, 2022; Arghashi and Yuksel, 2022).
7.3 Practical implications
Several practical and valuable implications for retailers and AR developers are offered in this study. First, both the SEM and fsQCA results have heightened the joint effects of informational cues in central and peripheral routes in determining the purchase and continuance intention of AR retailing. Hence, this study contributes to both the economic and commercial practitioners. Commercial practitioners should allocate the necessary resources to offer rich informational cues in the AR retailing environment and improve consumers’ perceptions of information quality and AR credibility.
The analytic approach in this study offers several demonstrations from an educational and teaching perspective. This study employed several different methods to offer solid and comprehensive results regarding the critical factors that facilitate consumer purchase and continuance intention, such as SEM and fsQCA. Practitioners can acquire knowledge of how to identify the most critical elements to facilitate their marketing and managerial strategies using these scientific methods. These rigorous analytic methods can offer useful implications to guide them to effectively design and operate AR retailing to facilitate consumers’ better process-related information. For example, the parsimonious solutions identified using the advanced method of fsQCA suggest that App developers can enhance information completeness by improving the functions of AR to facilitate consumers to observe more details and try on the virtual products to see their actual effects, such as adjusting, rotating and zooming in and out the product on multiple shopping scenarios. To facilitate information accuracy, AR developers should ensure that the virtual products are presented to customers like the real ones in the physical environment. The size, color and use effect of the products should be precisely displayed to customers.
Similarly, the research results also offer the practitioners the core factors to enhance consumers’ continuance intention in AR retailing, including information format, attractiveness, relative advantage and augmented quality. To stimulate consumers’ interest in AR retailing, managers should provide a vivid and well-designed interface, and further improve the quality of image and display, such as the sharpness of the videos and pictures, atheistic interface design and eye-catching elements. In addition, AR designers should increase the quality of augmented reality and enable the consumer to experience the seamless integration of virtual products and physical surroundings and an immersive shopping process (Erdmann et al., 2023). Considering the development of AR is still in its infancy, improving augmented quality is an imperative task for AR developers. Practitioners should also enable consumers with a rich shopping experience and sufficient information for decisions, and thus make them believe that AR retailing has advantages to facilitate their customer journey.
We intend to address the implications to public policy and society. Though the development of AR retailing is at an early stage, our study results verify its influences as a novel technology and unveil its potential to impact society, since it empowers a common individual in processing shopping information in a new way and facilitates people’s understanding and application of new technologies to enhance the efficacy and enrich their daily life. Thereby, public policy should be made to facilitate the business application of the new technology. Preferential policies, tax deductions, exemptions, financial subsidies, etc. should be provided to research institutions and commercial companies that focus on the application of AR retailing.
7.4 Limitations and future research
Utilizing mixed methods, this study endeavors to provide reliable and comprehensive results to interpret decision-making. Whereas this research is not without limitations. First, this study endeavors to explore the representative factors influencing consumers’ information processing mechanism in AR retailing. This study did not produce a complete list of all the possible antecedents. For example, this study merely applied the dimensions of information quality and source credibility as the central and peripheral route factors. Future studies may explore other potential factors in influencing consumers’ information processing via central and peripheral routes. In addition, as a significant positive perception toward AR retailing, future studies can explore the role of satisfaction in consumer’s information processing mechanism. Moreover, since the research focus of this study is not to explore the distinct effects of differing usage experiences, this study only employs the “number of years” as a control variable to represent the user’s experience of AR retailing. However, it is acknowledged that user experience of AR retailing can be measured by multiple metrics, such as usage frequency, duration, contexts, etc. Future studies are encouraged to apply other different usage experiences as the control variables to further explore their effects.
Second, future research may employ different research designs or data analytical approaches. For example, future studies can combine experimental studies and the eye-tracking technique to examine users’ actual behaviors rather than the behavioral intention while experiencing AR retailing. In addition, the mediating role of trust in influencing the information cues and consumers’ further decision-making can be further explored. Moreover, the moderating analysis can be further incorporated to explore the interfering influence of the type of product, individuals' experience and preferences, previous brand familiarity, motivation and ability in information processing.
The third limitation is based on our empirical data. Our data is collected in China. Empirical data collected from other cultures or regions can be used to compare with the results in this study. With the development of AR retailing, consumers’ attitudes and behaviors may change over a period of time. Future studies may apply a longitudinal research design to capture temporal effects and explore the reasons behind the change in attitude or behaviors. In addition, the empirical data is collected from consumers who have experienced AR retailing. We did not explore a specific kind of AR retailing scenario or explore the differences in consumers’ perceptions of different AR platforms. Future studies can further explore the differences in various factors influencing consumers’ decision-making process, such as the role of the different AR shopping scenarios and different e-commerce shopping platforms.
Lastly, although our study provides valuable insights by applying trust as the first-order construct, future studies can delve into the multidimensional aspects of trust such as honesty, benevolence and competence to scrutinize the information processing and trust formation outcomes in AR retailing.
7.5 Conclusions
Applying a three-staged research design with an integrated model based on the theories of ELM and trust transfer, this study aims to investigate consumer information-processing and decision-making mechanisms in AR retailing. SEM analysis confirms the hypothesized relationships between the central route factors (information completeness, accuracy, currency and format) and peripheral route factors (AR attractiveness, popularity, relative advantage and augmented quality), information processing outcomes (trust in product information and trust in AR retailing), and eventual behavioral intentions (purchase intention and continuance intention). Both the SEM and fsQCA results imply that neither the central route nor the peripheral route alone is sufficient to engender consumers’ behavioral intentions and comply with the basic assumptions of the ELM. This study is pioneering in validating and contextualizing ELM theory in AR retailing. In addition, this study offers a methodological paradigm by demonstrating the application of multi-analysis in exploring consumers’ information process mechanisms in AR retailing, which offers a holistic and comprehensive view to understand consumers’ decision-making mechanisms. This study offers practical implications and directions for future studies addressing the importance of exploring other factors influencing AR retailing, the application of multiple research methods, the moderating and mediating analysis of potential factors and the investigation of AR retailing in other cultures and countries.
This work was supported by the Nature Science Foundation of Shaanxi (2023-JC-QN-0794).
Figure 1
The framework of research design
[Figure omitted. See PDF]
Figure 2
Conceptual framework
[Figure omitted. See PDF]
Figure 3
Research model
[Figure omitted. See PDF]
Figure 4
The hybrid analytical approach of SEM and fsQCA
[Figure omitted. See PDF]
Figure 5
Research design
[Figure omitted. See PDF]
Figure 6
The results of SEM
[Figure omitted. See PDF]
Table 1
Configurations that lead to purchase intention
| configuration | 1a | 1b | 2a | 2b | 2c | 2d | 2e | 3a | 3b | 4a | 4b |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Information completeness | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Information accuracy | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Information currency | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Information format | ● | ● | ● | ● | ● | ● | ● | ⊗ | ● | ● | |
| AR attractiveness | ● | ● | ⊗ | ● | ● | ● | ⊗ | ● | ● | ● | ● |
| AR popularity | ● | ● | ● | ● | ● | ● | ● | ● | ● | ||
| Relative advantage | ● | ● | ● | ● | ● | ● | ● | ⊗ | ● | ● | |
| Augmented quality | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Trust in AR | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Trust in information | ● | ● | ● | ● | ● | ● | ● | ● | ● | ● | |
| Raw coverage | 0.36 | 0.36 | 0.21 | 0.27 | 0.37 | 0.37 | 0.27 | 0.39 | 0.20 | 0.40 | 0.38 |
| Unique coverage | 0.04 | 0.01 | 0.01 | 0.03 | 0.01 | 0.01 | 0.04 | 0.02 | 0.04 | 0.02 | 0.02 |
| Consistency | 0.98 | 0.98 | 0.97 | 0.98 | 0.98 | 0.98 | 0.98 | 0.98 | 0.98 | 0.98 | 0.99 |
| solution coverage | 0.53 | ||||||||||
| solution consistency | 0.94 | ||||||||||
Note(s): ●: core conditions are present; ⊗: core conditions are present; ●: peripheral conditions are present; ⊗ peripheral conditions are absent; Blank space: the presence or absence of conditions that do not matter
Source(s): Authors' own creation/work
Table 2
Configurations that lead to continuance intention
| Configuration | 1a | 1b | 2a | 2b | 2c | 3a | 3b | 3c | 4 |
|---|---|---|---|---|---|---|---|---|---|
| Information completeness | ● | ● | ● | ● | ● | ● | ● | ||
| Information accuracy | ● | ● | ● | ● | ● | ● | ● | ● | |
| Information currency | ● | ● | ● | ● | ● | ● | ● | ● | |
| Information format | ● | ● | ● | ● | ● | ● | ● | ||
| Attractiveness | ● | ● | ● | ● | ● | ● | ⊗ | ● | |
| Popularity | ● | ● | ● | ● | ● | ● | ● | ||
| Relative advantage | ● | ● | ● | ● | ● | ● | ● | ● | |
| Augmented quality | ● | ● | ● | ● | ● | ● | ● | ● | |
| Trust in AR | ● | ● | ● | ● | ● | ● | ● | ● | |
| Trust in information | ● | ● | ● | ● | ● | ● | ● | ● | |
| Raw coverage | 0.36 | 0.36 | 0.39 | 0.40 | 0.37 | 0.39 | 0.37 | 0.21 | 0.39 |
| Unique coverage | 0.01 | 0.01 | 0.01 | 0.03 | 0.02 | 0.01 | 0.02 | 0.01 | 0.02 |
| Consistency | 0.94 | 0.94 | 0.93 | 0.93 | 0.94 | 0.93 | 0.92 | 0.96 | 0.93 |
| Solution coverage | 0.53 | ||||||||
| Solution consistency | 0.91 | ||||||||
Note(s): ●: core conditions are present; ⊗: core conditions are present; ●: peripheral conditions are present; ⊗ peripheral conditions are absent; Blank space: the presence or absence of conditions that do not matter
Source(s): Authors' own creation/work
© Emerald Publishing Limited.