2.1 O2O Commerce

O2O commerce refers to “services that match providers to consumers through an Internet-based platform or smartphone application” [21]. As an upgrade of traditional e-commerce models, it completely changes the competition situation in traditional markets [23]. A core platform that connects everything delivers intangible product/service information to users [21]. Offline merchants launched on the platform leverage the merchant-provided information and customer-posted contents to engage prospective customers and induce their offline purchase [43]. O2O platforms and offline merchants do not fight with each other, but work together to serve consumers [43]. By doing this, O2O commerce grows business opportunities for traditional industry as well as improves feeling and value of customer experience.

O2O business model can be performed as online to offline or offline to online commerce [21]. From a customer perspective, the former model means researching products/services online first and then experiencing them in physical stores offline. The later model means visiting physical store first and then getting virtual services online [49]. From a firm perspective, the former model means “find customers online and bring them into real-world stores” [46]. The later model means exhibiting products/services offline and meeting customers’ spontaneous needs online [21]. In reality, most customers carry out extensive information searches online first and then purchase products/services at the real-world stores [13]. In line with this dominated O2O commerce direction, this study focuses on the online to offline e-commerce.

2.2 Trust Transfer

Trust transfer is “a cognitive process that may arise from one familiar context to another new context or from one trusted entity to another unknown entity” [33]. Three actors are involved in the trust transfer process, that is, the trustor, the trustee, and a third entity [50]. The trustor is an individual who judges whether to trust others; the trustee is another individual who is assessed by the trustor based on his/her trustworthiness, and the third entity act as the broker in the trust transfer process [9]. The underlying logic of trust transfer mechanism is that “when the trustor trusts in the third person and there is a close relationship between the trustee and the third person, the trustor's trust in the third person will be transferred to the trustee” [50]. Here, the third entity, the familiar context or trusted entity is called as the source of trust transfer and the trustee, the new context or unknown entity, is called as the target of trust transfer [50], [61]. Perceived relationship between the trustee and trust source plays a crucial role in the trust transfer and trust formation process [9], [34]. There are two main types of relationship band linking trusted target and trust source namely similarity and business tie [47], [50]. Similarity “captures the internal relationship between source and target, such that they share certain same features which make people have same perception about them” [50]. Business ties “capture the external relationship between source and target, such that they may not be same in nature but share certain external cues” [50].

E-commerce research has examined four types of trust transfer process (TTP) [30]. The two intra-channel trust transfer processes (offline to offline TTP and online to online TTP) rely perceived similarity to link two similar channels, while the other two inter-channel trust transfer processes (offline to online TTP and online to offline TTP) rely business tie to link online and offline channels. Since the intra-channel trust transfers have no relationship with the current research, we do not consider them in this study. The existing inter-channel trust transfer literature focuses on the retail industry, in which the online and offline businesses are dual-channel competitive businesses, namely the competition between online stores and traditional stores [7]. Online stores and traditional stores are operated by the same business entities. Thus, online trust and offline trust involves only one trustee (vendor). The two types of trust are in parallel and can function independently. Trust transfer in O2O e-commerce is inherently different from the above inter-channel trust transfer. First, this trust transfer process involves two different parties (O2O platforms and offline merchants). Second, Trust in platforms and trust in merchants collaborate serially as they function in the same shopping process. Considering all these, we draw upon the trust transfer theory to take a new look at the effect of trust transfer from O2O platforms to offline merchants in our study.

2.3 Online Contents

Online contents refer to the information components on an online platform, either textual or graphic [25]. Given the important role of online contents on customer intention and behavior, scholars have paid much attention to it. Overall, their studies generally fall into three main streams. The first stream focuses on the business or information value of online contents [44], the second stream focuses on the trustworthiness of online contents [3], and the third stream studies customers’ willingness to contribute online contents based on psychological lens [55].

Besides the research focuses, each stream has its own research methodology. Specifically, text mining, machine learning, and econometric model are mainly used to examine online content related issues. In particular, econometric model is more suitable for predicting perceived trust, sales, and behavior intention. Thus, we use econometric model to understand the effects of online content on purchase intention. Following Bao, Li, Shen and Hou [4], information effectiveness is adopted to measure content quality. This study also denotes content quantity (content size) as another important factor affecting consumer trust and purchase intention. It is worth to mention that online contents consist of use-generated product reviews and numerical ratings as well as merchant-provided product descriptions in the O2O commerce context. To examine the effects of online content size, we naturally need study several variables, such as the average length of product reviews (textual reviews), the length of product descriptions, the number of product reviews, the number of numerical ratings, and the number of online pictures. In reality, there is no obvious difference in the length among different product descriptions as merchants generally release product information according to specific schemas and norms. Meanwhile, the number of numerical ratings is equal to the number of product reviews for most popular O2O platforms. Thus, this study uses the total number of product reviews and online pictures as well as the average length of online reviews to measure online content size.

3.1 Effects of Trust Transfer

Trust can be transferred from a known entity to an unknown entity because of the associations between them [34]. Within the O2O commerce context, O2O platform acts as online promotional channels to drive offline sales [43]. Following the trust transfer theory, we argue that customer trust may transfer across O2O platform (the third party) and offline merchants (the trustee) because the interrelationship between them. Indeed, consumers interacting with a web platform make strong inferences about the attributes of the offline merchants from what they experience on the platform. If he/she accrues a favorable experience with the platform, he/she is more likely to trust the merchants tapped on the platform [9]. In addition, popular social media enabled O2O platforms generally provide fair norms and standards to avoid other members (e.g., merchants on the platforms) to engage in information fraud behaviors. Therefore, when an individual trusts an O2O platform, he/she will more likely to trust the merchants on the platform, because he/she believes that the merchants would comply with these norms and standards [34]. Furthermore, O2O platforms have the responsibilities to monitor the merchants and ensure the online information reliable. Any dishonest behavior may incur severe punishments, such as monetary penalties, cooperation relationship taken away, or feedback product reviews clearance. In this respect, we can infer that when customers build up strong trust toward an O2O platform, they tend to view any unfamiliar merchant related to the platform as a trustworthy transaction partner because any of its opportunistic behaviors will be punished by the platform. Given this, we hypothesize:

H1: In O2O e-commerce, trust can be transferred from an O2O platform to the merchants on the platform.

3.2 Effects of online contents

Since the effectiveness and the size of online contents both affect customer attitude and purchase intention and there is no evidence to conclude that they will produce the same effects, we examine their effects separately.

3.3 Trust and Purchase Intention

Positive trust attitude toward a vendor helps alleviate perceived transaction risks, which in turn promote transaction in traditional e-commerce [62]. In O2O commerce, customers’ consumption decision largely depends on their perceptions whether offline merchants can provide them with consistent products/services as claimed online. When customers believe that an unfamiliar merchant will behave in a trustworthy and expected way by reading reputation information online, they may generate purchase intention [43]. Therefore, we argue that the relation logic between trust and purchase intention is also applicable to the O2O commerce model. Accordingly, we hypothesize:

H6: In O2O e-commerce, trust in merchants is positively associated with purchase intention.

Combining the effects of trust transfer and online contents on purchase intention in O2O e-commerce, we present the research model with hypotheses in Figure 1. Additionally, it has been confirmed that trust in O2O platforms has a direct positive influence on purchase intention [1]. Thus, though the confirmed relationship is not hypothesized, we still include it in Figure.1 to ensure the model completeness.

Figure 1: Proposed research model 

4.1 Measurement Development

We developed a questionnaire survey containing two parts. Part 1 included questions regarding gender, age, education, O2O transaction experience, the number of user-provided reviews, the number of online pictures, and the average length of top reviews. Part 2 included all of the questions measuring the six constructs on a 7-point Likert scale, as shown in Appendix 1. Wherever possible, we adapted existing validated items from prior literatures. Perceived effectiveness of product reviews (four items) [4], [41], perceived effective of numerical ratings (three items) [4], trust in platforms (three items) [12], trust in merchants (four items) [62]and purchase intention (four items) [24], [62]. As no appropriate scale can be adapted from prior research, we developed new items measuring perceived effectiveness of online product descriptions based on the definition of the construct and focus-group interviews. Meanwhile, taking the items for measuring perceived effective of numerical ratings for references and corresponding to the O2O practical business, we carefully worded the new item. Adapting ideas from Park, Lee and Han [40], we classified 10 reviews as a small number of reviews while 100 reviews as a large number of reviews. The same principle was used to classify the picture number. Regarding the average length of top reviews, we conducted data scraping technology to get the general distribution. Data scraped from (Site 2) consisted user reviews on destinations, restaurants, education, etc. The detailed statistics of the dataset showed that average word counts per review were 95.6 Chinese characters. Based on this information, we grouped the top reviews into three subgroups. The top reviews that 50% of them with word count exceed 150 were viewed as long reviews, while 50% with word count less than 50 were viewed as short reviews. The others were viewed as medium reviews.

4.2 Sample and Data Collection

We collected data through (Site 3), a well-known professional questionnaire website in Chinese language. In order to publicize our online survey, we distributed the URL address of our questionnaire to the most widespread social network sites in China, such as QQ, WeChat, and Sina Weibo. The respondents were asked to visit (Site 2), one of the most popular and largest O2O platforms in China. After reading the contents online about a specific unfamiliar merchant who operates tourism destination in rural China, they were invited to fill out the questionnaires. To improve the response rate, we awarded respondents with Wechat Red Bags or QQ Red Bags.

In sum, we received 316 responses. After discarding the invalid responses (giving all the items the same value), we ultimately obtained 255 valid samples for data analysis. Among these 255 respondents, 42.19 % were male and 57.81 were female. The respondents were relatively young: 62.89% aged between 21 and 30, and 26.95% under 21. In terms of education, most of the respondents were well educated: 72.46 % had a bachelor’s degree, and 19.07 % had a master’s degree or higher. The sample represents general customers in China’s O2O market that are quite young but well educated (iResearch, 2017). In addition, the respondents were quite inexperienced, with more than 40% having less than one-year O2O experience, and more than 30% using O2O model less than once within the latest year.

5.1 Measurement Validation

First, we evaluated the factor reliability and internal-consistency reliability of principal constructs. The values of Cronbach’s alpha, as shown in Table 1, rang from 0.86 to 0.91, indicating excellent internal-consistency reliability. The composite reliabilities for all the constructs are far above the 0.7 cut-off point, exhibiting adequate construct reliability. Additionally, all items have loadings greater than 0.6. Next, we assessed the convergent and discriminant validity by inspecting the average variance extracted (AVEs) and correlation coefficient. Table 1 indicates that the AVEs fall between 0.7051 and 0.8085, all above the criterion of 0.5, demonstrating adequate convergent validity. The discriminant validity was assessed by comparing the square root of each AVE with every possible correlation coefficient among constructs and the result shown in Table 1 was found to be well achieved. Meanwhile, we looked at the cross loadings presented in Table 2; the results indicate that the loading of each indicator on its own construct is greater than its loadings on any other latent constructs, also suggesting that the constructs are distinct.

Given several relatively high correlation coefficients among latent constructs, we performed a multicollinearity test by examining the variance inflation factors (VIFs). There are two types of VIFs: outer VIF and inner VIF. Any VIF value higher than 5.0 implies that potential multicollinearity problem will threat the validity of empirical research [19]. In our study, the highest outer VIF was 2.098 and the highest inner VIF was 2.554, both lower than the cut off level of 5. Thus, we contend that multicollinearity pose no significant threat to our research.

Table 1: Reliability, validity, and correlations for constructs 

Notes: RSFL: Range of factor loadings; α: Cronbach’s alpha; CR: Composite reliability; AVE: Average variance extracted

Table 2: Loadings and cross-loadings of measurement ítems 

5.2Testing the Structural Model

After checking validity and reliability, we tested the structural equation model using PLS algorithm. Figure 2 reports the path coefficients along with explanatory power (R2). The model explains 60.4% of the variance for trust in merchants, and 57.3% of the variance for purchase intention. A bootstrapping procedure with 2000 replications was applied to estimate the statistical significance of the regression weights.

Figure 2.  Results of structural model 

According to Figure 2, perceived effectiveness of product reviews (β=0.187, p<0.05) and numerical ratings (β=0.166, p<0.05) are important determinants of trust in merchants. By contrast, perceived effectiveness of product reviews acts as a more important precedent. Contrary to our predictions, perceived effectiveness of online product descriptions is found to have no significant predictive power for trust in merchants. These results support H2a and H2b, while reject H2c. Regarding the effect of trust transfer, as confirmed, trust in O2O platforms can been transferred to trust in merchants (β=0.399, p<0.01), supporting hypothesis H1. Furthermore, trust in merchants (β=0.327, p<0.01) shows strong positive impact on purchase intention, supporting H6.

5.3 Testing the Effects of Online Content Size

We performed several MANOVA analyses to test the hypotheses regarding online content size. Given the high reliability of factor measures, we believed that the factor scores of trust in merchants and purchase intention represented the information contained in the two constructs. Thus, we used SPSS 19 to finish the MANOVA tests and multiple comparisons based on the factor scores.

5.3.3 Effects of Picture Number

The MANOVA test also suggests a statistically significant picture number effects on trust in merchants (F=16.45, df=2, p=0.00) and purchase intention (F=25.55, df=2, p=0.00). The following ANOVA analysis (see in Table 3) reveals that customers who have browsed a large number of pictures are more likely to generate trust toward merchants and purchase intention than those who have browsed a medium or small number of pictures. In addition, customers who have browsed a medium number of pictures are also more likely to generate trust and purchase intention than those who have browsed a small number of pictures. Therefore, H5a and H5b are supported.

Table 3: Multiple comparison of the review lengths, review number and picture number 

6.1 Discussion of Findings

First, this study suggests trust can be transferred from O2O platforms to offline merchants, which in turn enhances customer purchase intention. Our result extends prior finding in Chen, Huang, Davison and Hua [9] that positive perception and attitude toward a C2C platform can be automatically transferred to any seller on the platform and then drive offline sale. Meanwhile, the result implies that in a complex system, such as O2O system, the involved parties are not isolate but interacts with each other. Our study reconfirms the direct effect of trust in platform. It is not very costly and challenging to achieve an adequate system quality

website today. Nevertheless, how to provide a reliable and reputable O2O platform is still need to be considered as trust in platform can affect purchase intention directly and indirectly.

Second, regarding the three proposed compondents of online contents, only perceived effectiveness of product reviews and numerical ratings are found to have significant effects on trust in merchants, perceived effectiveness of online descriptions is found to have not. Our empirical result confirms the conclusion that user-generated reviews at (Site 2) impact customer choice of merchants [53]. In contrast to perceived effectiveness of product reviews, perceived effectiveness of numerical ratings has weaker impact. This finding confirms the argument proposed by Pavlov and Dimoka [42] that text reviews represent a more imperative aspect than crude numerical ratings to faciliate purchase by building customer trust toward offline merchants. A possible explanation is that noisy and biased ratings are widely found in practical web-based rating systems [15]. In addition, the 5 stars rating mechanism in O2O website is not adequate for filtering out untrusted merchants. Contrary to our expection, perceived effectiveness of online production descriptions does not play a positive role in building trust toward merchants. The plausable explanations may be that (Site 2) pays much attention to stimulating customer review generation, while pays little attention to encouraging merchant descriptions provision. Many merchants on (Site 2), such as farm tourism merchants, upload very simple information or just release operating hours. As a consequence, limited online descriptions are inadequate for building customer trust toward merchants in our context.

Third, our findings suggest that online content size has significant effects on both offline trust and purchase intention. Specifically, long reviews increse more customer trust and purhchase intention than medium and short reviews. This result confirms the fact that positive ratings are much more frequent than negative ratings on the present O2O platform, and this is also the scenario in rural tourism O2O business. Meanwhile, our result help understand the positive and statistically significant relationship between average length of reviews and sales at (Site 4) [10]. In accordance with our hypothesis, as the number of review grows, customers are more likely to place confidence on merchants’ trustworthiness and more willingness to make purchase decisions. This empirical result offers evidence for the suggestion that the effect of review volume on sale is significant and positive for average review rating above a certain anchor point [56]. Regarding the effect of the number of online pictures, the result supports our hypothesis that an increase in picture number would enhance customer trust as well as purchase intention. This result is interesting and extend prior findings that the addition of pictures to textual reviews influences consumer attitude toward merchants and behavioral intention to experience the products/services [26].

6.2 Implications for Theory

Although the focus of this study, the determinants of purchase intention is not new to the traditional e-commerce literatures, relevant studies concerning the effects of trust transfer and online contents on purchase intention in the context of O2O commerce are rare. This study provides a fine-grained insight into purchase intention under the O2O commerce environment.

First, it extends extant research by employing the trust transfer across two totally different entities in the same transaction process, a trust transfer is quite different from extant inter-channel trust transfer, in which the trustee only shifts from offline vendor to e-vendor or from e-vendor to offline vendor. Up to now, this trust transfer mechanism has received less research attention. More specifically, following the online to offline commerce logic, we focus on the relationship direction trust in platform to trust in merchant, a relationship direction fit common transaction practices in O2O commerce. Based on trust transfer theory, we confirm that trust can transfer from O2O platforms to offline merchants.

Second, this study examines the effects of product reviews, numerical ratings, and online product descriptions separately. This extends previous research that view online content as a single cue and pay little attention to the effect of merchant-provided product descriptions. Results confirm that the three components contribute differently to customer trust toward merchants. Though we found insignificant association between perceived effectiveness of online product description and merchant trust building, future research can choose other O2O platforms, for example (Site 5), as research context to investigate the relationship again.

Third, this study investigates the effects of online contents from two important dimensions, namely content effectiveness and content size. Regarding the effects of content size, this study adds to extant literature by examining the effect of picture number on trust in merchants and purchase intention. It also examines the relationship between review length, review number and their dependent variables. Though our datasets support the positive effect of review length, it opens up interesting directions for future research. For example, we can re-examine the negative relationship that an increase in review length can significantly decrease customers’ trust toward merchant or purchase intention when the merchant has a low review balance.

Finally, review number or review length is often examined as control variables in SEM models with PLS algorithm. By contrast, MANOVA tests used in this study may produce more detailed and accurate results.

6.3 Implications for Practice

With our study, we address several practical implications for practitioners to lead a better O2O outcome in the highly competitive O2O markets.

First, since trust transfer from O2O platforms to offline merchants is found to be an important driver of purchase intention in the O2O commerce context, O2O platform vendors should set fair norms and standards in their service level agreements for related merchants or fellow customers. The more important work for O2O platforms is to supervise online content generation. Once dishonest merchants are reported or malicious competitors are detected, O2O platform must punish them severely. On the other side, offline merchants can choose to tap on a reliable and reputable O2O platform, from which customer trust toward the platform can be transferred. Meanwhile, merchant trust can be built up quickly but at low cost.

Second, the research results confirm that prospective customers use perceived effectiveness of online contents as the major signal to evaluate merchants’ trustworthiness. Our results also suggest that potential customers show doubt on online reviews and pay less attention to the feedback ratings. This requires platform providers and computer experts take technical measures (e.g., logistic regression, text mining, social network analysis, etc) to detect and control online review manipulation strategies. At the same time, they should improve present numerical rating mechanism or develop a more scientific, accurate, and effective feedback rating system. Furthermore, designers are recommended to develop a certain kind of prosecution service, which can help customers easily report any user-posted contents that are fraud and fake, or with inappropriate content, irrelevant content, and advertisements.

Third, now that the research validates the effects of online content size on merchant trust and purchase intention. O2O platforms and offline merchants should undertake appropriate promotional endeavors that could motivate consumers to contribute more online reviews and pictures for their O2O business. For example, coupons, price discounts, free samples, or points are available and effective promotions.

6.4 Limitations and Future Research

This work is not free from limitations. First, although online to offline commerce model is the dominant O2O model, the less frequent but growing O2O model, called offline to online commerce model is also threatening tradition business. Future research needs to examine purchase intention under this commerce context with a reverse path. Second, other factors such as price and competitive location may also affect customer purchase intention in the O2O commerce [20]. Since it is not appropriate and impossible to include all possible elements in a model for that will make the model to be a pointless ramble, this study only considers the two settings: trust transfer and online content. However, we acknowledge that these critical elements need to be identified and examined in future research. Third, this study focuses on the effects of positive online contents. In fact, negative information may have stronger and more influential effects on customers’ purchase decisions. Future research will conduct a quasi-experiential design where participants are assigned to read different online contents and then investigate the different effects of positive, neutral, and negative online contents. Finally, this study selects online tourism as the research context, but the generalization of our findings to other O2O context (e.g., catering and entertainment) requires additional research. In addition, the effects of online contents may vary according to product types. Thus, the comparation of different types of products (e.g.,destination product, food product and entertainment products) could be an interesting topic in future work.