[page 20↓]

Multi-channel consumer perceptions

The distribution of products via multiple sales channels — often referred to as multi-channel retailing — is the norm today. According to Silverstein, Sirkin and Stanger [2002] multi-channel retailers in the United States (US) increased their online market share from 52% in 1999 to 67% in 2001 — in contrast to Internet-only retailers, who lost market share respectively. In 2003, multi-channel players comprised 43 of the top 50 e-retailers, versus 42 in 2001, 40 in 2000 and only 27 in 1999 [Gallo and McAlister, 2003]. For some pure Internet retailers a development towards multi-channel retailing can be observed.1 The increasing prevalence of multi-channel retailing raises the question how the presence of multiple sales channels may influence consumer perceptions of an e-shop and willingness to buy online respectively. In particular, we are interested in whether an effect between the perception of physical stores and trust in an e-shop can be measured. Trust is an important antecedent of willingness to buy [Bhattacherjee, 2002 Gefen, 2000 Koufaris and Hampton-Sosa, 2002 Pavlou, 2003]. Moreover, we are interested in the effect of consumers’ perceived privacy on trust in an e-shop. The results motivate our further research about multi-channel retailing (Chapter 3) and privacy (Chapters 4, 5 and 6).

This chapter is organized as follows. Section 2.1 presents related work. Hypotheses are proposed in Section 2.2 that constitute the basis for the proposed structural equation model. Section 2.3 concentrates on the used methodology. Results are presented in Section 2.4. Section 2.5 discusses the implications and Section 2.6 concludes the chapter with limitations and further work.

2.1  Related work

A number of surveys suggest that the Internet has a distinct influence on offline sales. In a series of studies conducted by the research consultancy Forrester and Shop.org, retailers claimed that about 24% of their offline sales in 2003 were influenced by the Web, which is an increase from 15% in 2002 [Shop.org and Forrester Research, 2004]. A further study estimates that about half of the 60 million consumers in Europe with an Internet connection bought products offline after having investigated prices and details online [Markillie, 2004]. A study by Doyle et al. [2003] analyzed the influence of store perception on online sales. 64.7% of Internet users in 2002 claimed to sometimes or often look at [page 21↓]traditional retail stores and then buy online – up from 50.3% in 2001. The surveys indicate that there are distinct cross-channel effects between online and offline retailing.

Theoretical contributions discuss advantages of multi-channel retailing and demand further empirical work to analyze how the use of multiple channels affects a firm and its customers [Gallaugher, 2002 Goersch, 2003 Gulati and Garino, 2000 Steinfield, 2002 Stone, et al., 2002].

For Internet-only retailers, numerous multivariate models suggest how the perception of certain variables influences consumers’ willingness to buy online. A overview of these studies has been provided in Grabner-Kräutner and Kaluscha [2003]. A large number of these studies found that trust is one of the most decisive antecedents of consumers’ purchase intentions at Internet-only retailers [Grabner-Kräuter and Kaluscha, 2003]. Doney and Cannon [1997] label trust even as an order qualifier for purchase decisions. The studies explore a number of antecedents and consequences of consumer trust in online merchants:

Jarvenpaa, Tractinsky and Vitale [2000] developed an Internet trust model that tested the influence of the two independent variables perceived size and perceived reputation on customers’ evaluation of trust in a Web site. The model showed that perceived reputation had a much stronger effect on trust in comparison to perceived size. The study was validated in a cross-cultural study by Jarvenpaa [1999] and in a study by Heijden, Verhagen and Creemers [2001]. Moreover, the model suggested that trust has a direct influence on attitude towards the e-shop and perceived risk, which again have an influence on the willingness to buy.

Chellappa [2001] hypothesized relationships among the independent variables perceived privacy and perceived security and the dependent variable consumer trust and received significant support in an empirical evaluation. Further aspects of privacy and its influence on trust have been tested by Belanger, Hiller, and Smith [2002]. Recent work has identified privacy as one of the main requirements for successful e-commerce [Ackerman, et al., 1999 Cranor, et al., 1999 Teltzrow and Kobsa, 2004].

However, none of the reviewed studies explore antecedents of trust in a multi-channel retailer.

2.2  Hypotheses

We are particularly interested in variables influencing trust and willingness to buy in a multi-channel context. From the described models for Internet-only retailers, we used the repeatedly cross-validated antecedents of trust, perceived reputation and perceived size [page 22↓]as suggested in the literature [Doney and Cannon, 1997 Heijden,et al., 2001 Jarvenpaa, 1999 Jarvenpaa,et al., 2000] to analyze effects on trust and willingness to buy in a multi-channel setting. In contrast to models dealing with Internet-only retailers, we analyze how perceived reputation and size of physical stores influence trust in an e-shop. Moreover, we test the influence of privacy on trust as proposed in [Chellappa, 2001]. We are particularly interested in the strengths of the relationships when the three antecedents of trustreputation of stores, size of stores and privacy – are measured simultaneously.

As the hypotheses are related to previous studies, we will just briefly discuss the hypotheses of our model and point out our modifications and new research aspects. For a more elaborate discussion of the underlying theory we refer to the original publications.

Jarvenpaa et al. [2000] use the concept of trust in the sense of beliefs about trust-relevant characteristics of the Internet merchant. In two empirical studies the authors found support for a significant influence of perceived size on trust at Internet-only retailers. According to Doney and Cannon [1997] size also turned out to significantly influence trust in traditional buyer-seller relationships. Large companies indicate existing expertise and resources, which may encourage trust. A large store network indicates continuity as stores may not “vanish overnight” [Goersch, 2003]. In a multi-channel context, we assume that the consumer perception of a retailer’s physical store presence may also have a positive influence on the perception of consumer trust in the same merchant’s e-store. Thus, we hypothesize:

H1: A consumer’s trust in an Internet shop is positively related to the perceived size of its store network.

Reputation is defined as the extent to which buyers believe a company is honest and concerned about its customers [Ganesan, 1994]. Consumers may have more trust in a retailer with high reputation because a trustworthy retailer is less likely to jeopardize reputational assets [Jarvenpaa,et al., 2000]. Several empirical studies support the hypothesis that the reputation of an e-shop has a strong influence on consumer trust in that shop [De Ruyter, et al., 2001 Heijden,et al., 2001 Jarvenpaa, 1999 Jarvenpaa,et al., 2000]. A study of traditional buyer-seller relationships also provided support that reputation is an important antecedent of trust [Doney and Cannon, 1997]. We assume that the effects observed for a single sales channel may also prove true for the influence of perceived reputation of physical stores on trust in the same retailer’s e-shop.

H2: A consumer’s trust in an Internet shop is positively related to the perceived reputation of its store network.


[page 23↓]

Concerns regarding online privacy have increased considerably and are a major impediment to e-commerce [Teltzrow and Kobsa, 2004]. Consumer privacy concerns are particularly elevated on the Internet. A measurement scale for perceived privacy towards an e-shop has been suggested by Chellappa (2001) where privacy has been described as the anticipation of how data is collected and used by a marketer. The author also found empirical support that perceived online privacy towards an e-shop is significantly related to consumer trust. We are interested in replicating this effect in a multi-channel setting.

H3: A consumer’s trust in an e-shop is positively related to the perceived privacy at the e-shop.

Trust is closely related to risk [Hawes, et al., 1989]. Jarvenpaa et al. [2000] refer to risk perception as the “trustor’s belief about likelihoods of gains and losses”. The hypothesis has been confirmed that the more people trust an e-shop, the lower the perceived risk perception [Heijden,et al., 2001 Jarvenpaa, 1999 Jarvenpaa,et al., 2000]. We also test this hypothesis in our study.

H4: Perceived risk at an e-shop is negatively influenced by consumer trust in an e-shop.

The theory of planned behavior suggests that a consumer is more willing to buy from an Internet store which is perceived as low risk [Ajzen, 1991]. The trust-oriented model by Jarvenpaa (2000) suggests that consumers’ willingness to buy is influenced by perceived risk and attitude towards an e-shop. In the studies of Bhattacherjee [2002] and Gefen [2000], a direct influence of trust on willingness to buy has been suggested. However, Bhattacherjee [2002] states that a large proportion of variance was left unexplained, which suggests that there may be other predictors that are missing in the model. We analyzed the causal relationships between risk, and purchase intention tested by Jarvenpaa et al. [2000].

H5: The lower the consumer’s perceived risk associated with buying from an e-shop, the more favorable are the consumer’s purchase intentions towards shopping at that e-shop.

2.3  Methodology

We introduce the methodical approach to test the above hypotheses. The retailer, the questionnaire, respondents’ demographics and the statistical method to develop our model are presented.

2.3.1  The retailer

The above hypotheses will be tested using a survey of visitors of a large multi-channel retail Web site. The company’s retail site considers itself the first fully integrated multi- [page 24↓] channel shop in Europe. The retailer operates an e-shop and a network of more than 6,000 stores in over 10 European countries. The company sells more than 10,000 consumer electronics products both online and offline. The offered product assortment appeals to a variety of consumer types including bargain shoppers and quality-oriented high-end buyers.

The retail site uses a typical online privacy statement that can be accessed through a link on each page of the site.

A questionnaire could be accessed via a rotating banner on the retail site. The banner announcing the survey was kept online for five months from 1st of March to 30th of July 2004. The banner announced the survey and offered an optional raffle (cf. Figure 7-1 of the Appendix). All participants who left their e-mail address automatically participated in the raffle of three digital cameras.

2.3.2 Questionnaire

The answers on the online questionnaire were measured using a Likert scale ranging from 1 to 5, with 1 indicating an attribute was “very weak / unlikely” and 5 “very strong / likely” [Likert, 1932]. Demographic information included age, gender, Internet experience, e-mail address and questions about previous visits and purchases both online and offline.

Scales were constructed on the basis of past literature as shown in Table 7-1 of the Appendix. For each item of the constructs perceived size and perceived reputation the term “this Web site” was replaced with “this retailer’s physical store network” to emphasize the offline context. For the remaining items we used the term “this e-shop” to draw a clear distinction between online and offline presence.

Some modifications of the scale suggested by Jarvenpaa [1999 2000] were adapted from Heijden et al. [2001]. For the construct willingness to buy, we changed the time horizons “three months” and “the next year” to the broader terms “short term” and “the longer term”. For each construct we used only three items to keep the questionnaire as short as possible, which was a requirement from the cooperation partner. We also modified two items of the risk scale suggested by Jarvenpaa [1999 2000] after a pre-test with department faculty. The item “How would you characterize the decision to buy a product through this Web site?” with answers ranging from “a very negative situation” to “a very positive situation” was changed into “How would you characterize the risk to purchase at this e-shop?” with a scale ranging from “very low risk” to “very high risk”. We also introduced a new item to measure consumer perceptions of the store network size: “This retailer’s stores are spread all over the country”. Members of the faculty staff and students [page 25↓] reviewed a preliminary version of the measurement instrument with respect to precision and clearness. A pre-test of 30 participants showed satisfactory results for Cronbach’s Alphas [cf. Cronbach, 1951].

2.3.3 Pre-processing and respondents’ demographics

Records of 266 respondents were eliminated from a total of 1314 due to missing data, in which duplicated e-mail addresses occurred (41 entries) or text fields belonged apparently to the same participant. 1048 complete answer sets are used for modeling.

The user demographics of our sample is predominantly male and between 30-50 years old (cf. Figure 2-1 ). 73% of the respondents in our sample are male (n=770) and 26% female (n=278). Thus, it reflects the gender gap that still predominates Internet usage in Europe [Hupprich and Fan, 2004]. Most of the users in our sample are experienced in using the Internet (cf. Figure 2-2).

Figure 2-1: Age distribution in respondent sample


[page 26↓]

Figure 2-2: Internet experience in respondent sample

Moreover, participants were asked in the questionnaire about their channel experience prior to their actual visit. For each of the four incidents “purchased at e-shop”, “purchased at store”, “visited e-shop” and “visited store”, participants were asked to answer if and how often they had visited the e-shop or store and if and how often they had purchased in the e-shop or in-store. The answers are depicted in Table 2-1.

Table 2-1: Prior experiences with the retailer’s e-shop and stores

 

visited e-shop

visited store

purchased at e-shop

purchased at store

no visit

300

337

818

425

1-2 times

243

274

168

320

3-5 times

101

111

26

85

> 5 times

388

315

20

200

no answer

16

11

16

18

Total

1048

1048

1048

1048

It is interesting that more than 605 participants claimed they had purchased at least once at the store and only 214 claimed to have purchased at the e-shop. Moreover, 200 claimed that they had purchased more than five times at a retail store. In contrast, the number of people who visited the store at least once was almost equal to the number of visitors who visited the e-shop at least once. These numbers remarkably point out the [page 27↓]importance of physical stores to the online audience.

2.3.4 Factor analysis and structural modeling

We used cross-validation and divided the sample of 1048 records into two sub-samples
n 1 =n 2 = 524 using simple random sampling. A confirmatory factor analysis (oblimin rotation) [Jennrich and Sampson, 1966] is performed on sample 1. This analysis was intended to confirm the hypothesized scales in terms of the discovery of six factors that each make up the employed scales.

If a plausible factor structure could be identified, it would be desirable to quantify the effect of perceived size, perceived reputation of stores, and privacy onto trust, willingness to buy, and risk perception. Factors are latent (not directly observable) variables. Linear structural modeling is used here as it allows the simultaneous mapping of relationships between several latent and non-observable variables within a single multi-equation model [Jöreskog and Sörbom, 1979 Jöreskog and Sörbom, 1996].

The variables of the questionnaire have ordinal scales. Model specification and parameter estimation is based on SIMPLIS [Jöreskog and Sörbom, 1996] and LISREL 8.54 [Jöreskog and Sörbom, 1996], and uses only sample 1 units. The model parameters are estimated by weighted least squares algorithm [Jöreskog and Sörbom, 1996]. Model structures were learned and the parameter estimated in an explorative and iterative way. Then the induced model is tested in the following phase on sample 2 in order to guarantee unbiased measures of goodness of fit.

2.4  Results

Firstly, we present a factor analysis, secondly evidences derived from the model, and finally we close with remarks on privacy and trust of respondents.

2.4.1 Factor analysis

The factor analysis included all items from Table 7-1 of the Appendix.The “eigenvalue > 1” - criterion leads to an initial five-factor model. However, a strong evident decline in the scree-plot after the sixth factor demanded a rotation with six factors. The extraction with principal component analysis (PCA), and oblimin rotation (delta = 0°) resulted in 74% explained variance. The first factor has a relatively high fraction of the overall variance, i.e. 33.9%. After rotation, all factors had eigenvalues above 2.

Four factors displayed medium intercorrelations (see Table 2-2), which underlines the necessity of an (oblimin) rotation. The pattern matrix of the rotated solution can be found in Table 7-2 and the factor loading in Table 7-1of the Appendix.


[page 28↓]

Table 2-2: Factor inter-correlation matrix

 

I

II

III

IV

V

VI

I

1.00

.02

.31

.42

.37

.39

II

.02

1.00

-.08

.07

.12

-.06

III

.31

-.08

1.00

.25

.20

-.27

IV

.42

.07

.25

1.00

.19

-.25

V

.37

.12

.20

.19

1.00

-.19

VI

-.39

-.06

.27

-.25

-.19

1.00

All factors include three items each with high factor loading above .6, except for the last factor, cf. -.52, -.58 and -.76. All items that load a factor have the same scale. T he factors allow testing of models of causal influence between factors based on linear structural modeling. The medium factor correlation between factors I and III, I and IV, I and V, and I and VI already indicate that such influences exist.

2.4.2 Linear structural models

To test our main five hypotheses, the six factors identified above are inserted into a linear structural model as proposed in Section 2.2. Linear structural models allow the testing of hypotheses about causal influences between latent (not directly observable) variables. As factors, as identified in the previous section, are latent variables (constructs that influence groups of items), hypotheses about their influence on each other can be tested. In linear structural models, factors are displayed as circles. The items that are influenced by these factors are displayed as boxes. Causal influences are displayed as pointed arrows with path coefficients (between -1 and 1) that indicate the strength of the causal relation. Correlations are displayed as bi-directional arrows. By quantifying the influence of the factors on the items, the model may confirm the factor analysis from the previous section.

The models were developed with the SIMPLIS command language [Jöreskog and Sörbom, 1996] and LISREL 8.54 [Jöreskog and Sörbom, 2003]. Due to the fact that ordinal questionnaire data was used, the weighted least squares algorithm for polychoric correlations was employed, including the asymptotic covariance matrices [Jöreskog and Sörbom, 1996].

However, stable parameter estimates of the model could not be determined after 30 iterations. Consequently, the model is reduced to a simpler one, which tried to capture the [page 29↓] effect of different factors on trust . A model is iteratively searched, which includes the factors perceived size (PS), perceived reputation (PR), and privacy (PRI). The underlying assumption of this model is that these three factors determine trust (TR). This model produced stable parameter estimates and after incorporating a series of modification indices supplied by the LISREL software, reac hed optimal fit indices. The completed model for sample 1 with all standard errors, factor loadings, and path coefficients is depicted in Figure 2-3.

Figure 2-3: Linear structural model for the influence of perceived size (PS), perceived reputation (PR), privacy (PRI) on trust (TR) for sample 1 (N=524)

All path coefficients are significant on the 5% level using a t-test. Goodness of fit statistics gives a Chi square value of 96.17 with 48 degrees of freedom, leading to a p-value of 0.00005 2 . Since the Chi square fit index in linear structural models is highly dependent on the sample size [Byrne, 1998] and tends to underestimate the model fit in larger samples, further fit indices are considered for model assessment. The Root Mean Square Error of Approximation (RMSEA) of 0.044 leads to a p-value for Test of Close Fit of .778, which [page 30↓] indicates a good model fit. A Goodness-of-Fit Index (GFI) of 0.99, an Adjusted Goodness-of-Fit Index (AGFI) of 0.99, a Parsimony Normed Fit Index (PNFI) of 0.721 and a Parsimony Goodness of Fit Index (PGFI) of 0.612 supports a good overall model fit. Refer to Jöreskog and Sörbom [2003] for detailed information on fit indices.

These above measures may be biased since the model is induced from the same sample. An unbiased test of the model can be achieved by applying it to the second sample that remained untouched so far (see Figure 2-4).

Figure 2‑2.4: Linear structural model for the influence of perceived size (PS), perceived reputation (PR), privacy (PRI) on trust (TR) for sample 2 (N=524)

The model for sample 2 gives a Chi-square value of 97.31 with 48 degrees of freedom, leading to a p-value of 0.00003. This RMSEA-value of 0.044 leads to a p-value for Test of Close Fit of .758, a PGFI of .611, a PNFI of .719 and an AGFI of .996. In summary, the se measures point out a good model fit with path coefficients in the same range as in the previous model, cf. Figure 2-3. The relevant path coefficients and fit indices for the two sub-samples as well as for the full sample are summarized in Table 2-3. All path coefficients in the samples are significant on the 5% level except the coefficient PSà TR in the second sub sample. However, the coefficient is significant in the full sample.


[page 31↓]

Table 2-3: Relevant path coefficients and fit indices for sub samples and entire sample

Sample

N

Path PS è TR

Path PR è TR

Path PRI è TR

Χ2

df

P

RMSEA

P (Cl. Fit)

1st

524

0.17*

0.41*

0.46*

96.17

48

0.00005

0.044

0.778

2nd

524

0.04*

0.47*

0.47*

97.31

48

0.00003

0.044

0.758

Full

1048

0.11*

0.42*

0.46*

106.80

48

0

0.034

0.999

With regard to Section 2.2, the findings support hypotheses 1-3. Hypothesis 4 assuming a negative influence of trust on risk and hypothesis 5 assuming an influence of perceived risk on trust have not been fully confirmed with the conservative methodical approach presented above. Further work will analyze the mediation path between trust, risk and willingness to buy in more detail.

2.5  Discussion and implications

The results indicate that perceived online privacy has the highest influence on trust relative to the two variables perceived size of the store network and reputation of the store network. This result has been confirmed in two random samples each with a high P-value. Though surveys indicate that privacy is crucial to successful e-commerce [Teltzrow and Kobsa, 2004], very few of the monthly site visitors accessed the retailer’s privacy statement, which is a typical phenomenon at retail sites. Kohavi [2001] indicates that less than 0.5% of all users read privacy policies. As a consequence, retailers should place clear and readily available privacy explanations on their Web sites in order to increase consumer trust. An efficient privacy communication design will be discussed in Chapter 6.

Moreover, the results confirm a strong effect of perceived store reputation on trust in the e-shop. A small effect of perceived store size on trust is observed. Thus, our study confirms the existence of cross-channel effects between stores and Web site. Jarvenpaa [2000] has shown that reputation and size are important antecedents of trustat Internet-only retailers. Her speculation that the presence of physical stores might increase consumers’ trust in a seller’s Internet store can be supported with our results. It can be assumed that cumulative effects between consumers’ perceptions of online and offline reputation and size exist. This could be an explanation as to why consumers prefer multi-channel retailers that now dominate more than two-thirds of the total online market (Silverstein et al. 2002). Thus, retailers’ multi-channel strategies should increasingly promote trust-building measures between different sales channels. This could include in-store advertising of the Web site, detailed online information about offline stores, better [page 32↓]multi-channel service integration or the placement of in-store kiosks, where consumers can order online when products are out-of-stock. Further studies should explore if there are cumulative effects between the perceived reputation and size of the e-shop on trust in the e-shop as indicated by Heijden et al. [2001] and Jarvenpaa [1999 2000] and the observed influence of perceived store sizeand reputation on trust in the e-shop. Therefore, a larger sample of consumers is required for discriminating between three groups of visitors: “familiar with the Web site only”, “familiar with stores only”, and “familiar with both channels”.

An interesting improvement of our study is a further analysis of the variables trust, risk and willingness to buy. Several authors have suggested a direct influence of trust on willingness to buy on the Internet [Bhattacherjee, 2002 Gefen, 2000 Koufaris and Hampton-Sosa, 2002 Pavlou, 2003]. The relationship between trust and success of relationship marketing is also well-known in traditional marketing theory [Berry, 1995 Morgan and Hunt, 1994]. In further work we will test if the construct perceived risk may function as a mediator between trust and willingness to buy. A mediator hypothesis between trust and future intentions also has been suggested in Garbarino and Johnson [1999]. The authors found that a model where satisfaction has been added as a mediating path between trust and commitment significantly improves the model fit compared to a model suggesting a direct influence of trust on future intentions.

2.6  Limitations

Participants in this study were online consumers. Thus, the sample differs positively from many other empirical studies that primarily use students as a sample of online consumer population [Grabner-Kräuter and Kaluscha, 2003]. However, a limit of external validity within our sample could have occurred through the self-selection of online participants. Other problems of online questionnaires could be reduced: repeated entries could be widely eliminated as most participants provided demographic information and e-mail addresses to participate in the raffle. The use of a rotating banner added randomness to the selection of participants. Only about every sixth visitor saw the banner on the retailer’s home page. Moreover, we explicitly asked participants to provide only honest answers.

The types of products may influence a user’s willingness to buy [Jarvenpaa,et al., 2000], which has not been further considered in this study. The results of Jarvenpaa et al. suggest that perceived size and reputation may influence trust differently depending on the type of products offered. The product sector of consumer electronics tends to be highly suitable for multi-channel retailing [Omwando, 2002]. It could be that the observed effects are less significant for less Internet-suitable product portfolios. A deeper discussion [page 33↓]of product characteristics in multi-channel retailing can be found in the thesis by Goersch [2003]. Critique also concerns the definition of measurement scales [Grabner-Kräuter and Kaluscha, 2003]. We used scales that have been successfully applied in studies of Internet-only retailing. The scales included relatively few items per construct due to the retailer’s request to keep our survey as short as possible. Though the results returned good factor confirmation scores, scaling needs more attention in further studies.


Footnotes and Endnotes

1 The largest e-retailer Amazon.com, for example, features products and services from merchants with physical retail stores since 2002.

2 Note that in linear structural models, the model hypothesis is that the empirical parameter matrix and the model matrix are not different, thus the p-value has to be as high as possible and not below 0.05.



© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.
DiML DTD Version 4.0Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML generated:
14.08.2006