Can you imagine? The effect of online
product representation mode on search and
experience good preference
By
2
Can you imagine? The effect of online
product representation mode on search and
experience good preference
ByAnna Margot van Prooijen June 2019
Master Thesis
MSc Marketing Intelligence University of Groningen Faculty of Economics and Business
First supervisor: Dr. H. Risselada Second supervisor: Dr. P.S van Eck
Reigerstraat 18A 9714EW Groningen
(06)30097339
3
ABSTRACT
Purpose: The purpose of this study was to investigate the influence of product representation
mode on product preference in the e-commerce sector. The present paper aims to study the impact of product representation modes (i.e. solid background, context background, 360 degrees rotation, and Augmented Reality) on consumer preferences for shopping on the Internet across different product types (i.e. search, and experience). Additionally, the author studied whether or not brand familiarity and product price moderates this relation.
Methodology: A quantitative research and explanatory research approach were conducted and was
carried out in 2019. The respondents (n=676) were recruited through a survey on Amazon’s Mechanical Turk. A Choice-Based Conjoint (CBC) analysis was conducted to measure consumers’ preference. Moreover, Latent Class Analysis was conducted to explore segments that differ in preference.
Findings: The findings of the study indicate a significant effect of product representation mode,
brand familiarity, product price, and delivery costs on product preference. These effects differ across search and experience goods. Moreover, brand familiarity positively (negatively) moderates the relation of online product representation mode on search (experience) product preference. These findings are useful for managers to provide the customer with an appropriate online product representation mode under different consumer characteristics and product type conditions.
Originality/value: The current study examined the role of different product types on online
product representation mode, finding that product representation mode is more important for search goods compared to experience goods.
4
PREFACE
Before you lies the thesis “Can you imagine? The effect of online product representation mode on online search and experience good preference.” Seven years ago I could not imagine that I would move from The Hague all the way to Groningen and receive a MSc degree in Marketing. This thesis was written to finalize my study Master Marketing Intelligence at the University of Groningen.
I would like to thank my supervisor Hans Risselada for his guidance and feedback during this process. In addition, I would like to thank Peter van Eck as my second supervisor. Moreover, I would also like to thank my friends, fellow students, and boyfriend for their moral support. Finally, yet importantly, I would like to thank my parents, who have always helped me during my journey to achieve my ambitious goals. Now it is time to bring my knowledge into practice.
5
TABLE OF CONTENT
ABSTRACT 3 PREFACE 4 1. INTRODUCTION 7 2. THEORETICAL FRAMEWORK 10 2.1 Conceptual model 10 2.2 Risk perception 112.3 Online product representation mode 12
2.6 Product type 14 2.4 Brand familiarity 15 2.5 Product price 17 2.7 Covariates 19 3. METHODOLOGY 20 3.1 Research method 20
3.1.1 Choice-Based Conjoint analysis 20
3.1.2 Attributes and levels 21
3.1.3 Choice design 22 3.1.4 Covariates 23 3.2 Data collection 24 3.3 Plan of analysis 25 4. RESULTS 28 4.1 Descriptive analysis 28 4.1.1 Television sample 28 4.1.2 Sofa sample 28 4.2 Familiarity check 30 4.3 Factor analysis 31
4.3.1 Covariate level of innovativeness 31
4.3.2 Covariate online shopping experience 31
4.4 Choice-based conjoint analysis 31
6
4.4.2 Main effect results 32
4.4.3 Relative importance 34
4.4.4 Predictive validity 34
4.4.5 Main effect combined model 35
4.4.5 Moderating effects and covariates combined model 36
4.4.6 Combined model validity 38
4.5 Overview hypotheses 39
4.6 Latent Class Analysis 40
5. DISCUSSION 43 5.1 Theoretical implications 43 5.1.1 Main effects 43 5.1.2 Moderating effects 44 5.2 Managerial Implications 45 6. LIMITATIONS 47 REFERENCES 50 APPENDICES 60 R OUTPUT 67
7
1. INTRODUCTION
The next big thing in e-commerce is Augmented Reality (AR). By 2022 the global revenue of the consumer mobile AR app market will have become more than 15000 million dollars (Statista, 2019). This is an enormous increase by 2036% compared to 2016 ($725.4 million). AR technology consolidates the physical real-time world with digital information by using a camera on an electronic device. The AR feature has already been applied in various industries such as video games and healthcare. Nowadays, major retail brands, such as Amazon, IKEA, and L’Oréal, have introduced their own AR shopping app. AR is increasingly becoming a vital factor in e-commerce by enhancing the shopping experience.
8
Several studies in this field focused on the relation of AR characteristics and consumer attitude and behavior (Brito, Stoyanova, & Coelho, 2016; Yim, Chu, & Sauer, 2017; Poushneh & Vasquez-Parraga, 2017; Yim & Park, 2018). According to Yim et al. (2017) AR “provides effective communication benefits by generating greater novelty, immersion, enjoyment, and usefulness, resulting in positive attitudes toward medium and purchase intention, compared to the web-based product presentations.” Poushneh & Vasquez-Parraga (2017) points out that AR positively impacts user satisfaction and willingness to buy. Moreover, it encourages the intention to recommend and share the brand experience (Brito et al., 2016).
To date, there has been a lack of genuine knowledge about the effectiveness of AR for different product types or brands. Previous work has been limited to one product type or brand; Ray-Ban sunglasses (Poushneh & Vasquez-Parraga, 2017; Verhagen et al., 2016; Yim & Park, 2018; Yim et al., 2017), Converse shoes (Brito et al., 2016) and Tissot watches (Yim et al., 2017). Brito et al. (2016) propose that further work needs to be carried out to establish whether different product types and brand choice affects how customers buy products online. Furthermore, Yoo and Kim (2017) call for further investigation of search and experience goods for understanding the effectiveness of online product presentation. This draws the attention to focus on brand familiarity and product type in relation to product representation mode. It is reasonable to expect differences in choice utilities. Hence, these aspects are important to investigate, because these effects could have seriously different managerial implications of the effectiveness of online product representation modes.
9
What is the impact of Augmented Reality on product preference and what is the role of product type, product price and brand familiarity?
This study seeks to address to what degree AR is an effective tool in e-commerce, and investigating the role of brand familiarity, product type, and product price. To examine the objective of this study, the survey data is obtained by using Amazon’s Mechanical Turk. This quantitative primary research contains a Choice-Based Conjoint (CBC) analysis and helps to measure consumers’ preference. Moreover, Latent Class analysis (LCA) is conducted to account for homogeneous discrete segments that differ in preference.
10
2. THEORETICAL FRAMEWORK
In the following chapter, the conceptual model of this study will be discussed. Secondly, online product representation mode literature is reviewed, followed by the existing academic literature about risk perception. Furthermore, the moderating role of product type and brand familiarity on the effect of visual product experience on product preference are discussed. Finally, the control variables are described.
2.1 Conceptual model
Figure 1 shows the conceptual model of this research, which investigates the effect of online product representation mode on consumers’ choice utility. The product type, product price, and brand familiarity are considered as moderators on the relation between the type of online product representation mode and choice utility.
Figure 1 Conceptual model
Product type: search good versus experience good
Choice utility Online product representation mode
b-11
2.2 Risk perception
Perceived risk is the perception of consumers about the uncertainty and unfavorable outcome of buying a product or service (Dowling & Stealin, 1994). A major problem in e-commerce is the uncertainty felt about an item due to the indirect product experience. The customer cannot hold the product and evaluate if the product truly meets the expectations. For example, if a consumer is considering buying an unknown makeup brand online for a special occasion, the risk perception of purchasing the item could crop up because she is uncertain about how the makeup will look like and as a consequence, she is afraid that she has to return the unsatisfactory purchase. This uncertainty is an element of perceived consumer risk (Weathers, Sharma, & Wood, 2007). Perceived risk plays a fundamental role in the willingness to buy products online (Barnes, Bauer, Neumann, & Huber, 2007; Forsythe & Shi, 2003). Showrooming is strongly associated with this risk perception, meaning that “consumers gather information offline but purchase the product online” (Verhoef, 2017). Consumers try to reduce perceived uncertainty by gathering more information offline. Hence, they try to get a better impression of the product. A better price online positively influences the showrooming phenomena.
12
2.3 Online product representation mode
Online product representation mode is an important factor in e-commerce because this aids consumers to access the product-related information (Li, Wei, Tayi, & Tan, 2016). From an imaginary perspective, Yoo and Kim (2014) questioned whether or not the concreteness of picture presentation impacts the behavioral intention to share the experience with others and purchase the product in the online retailing apparel industry. Exposing the online shopper to relevant consumption background pictures, compared to solid backgrounds, results in better online product experience due to the mental imagery. Consequently, positive emotions are evoked which eventually stimulates behavioral intention. Hence, mental imagery can be considered as an influential underlying source of the valued online product experience (Overmars & Poels, 2015; Schlosser, 2003; Yoo & Kim, 2014). In the case of zoom functions, Schlosser’s (2003) findings support the importance of product representation mode with the idea that the way of representing products affects the degree of mental imagery processing. She shows that interactive product representation results in more vivid mental images and hence higher purchase intention in comparison with passive product representation.
More advanced image interactivity is Augmented Reality (AR). This technology gives consumers the opportunity to understand information about the real world that is not directly accessible (De Paolis & Bourdot, 2018). The conventional interface has a fixed screen space, but with the introduction of AR, the user can explore objects from different angles. This blends virtual objects with the real world and results in an engaging experience (Apple, n.d.). A practical example of AR in the retail sector is the IKEA Place app that lets consumers virtually place IKEA products in their space. Consumers can see true to scale how their furniture looks and fit in their homes. Hence, consumers are better able to find out if the product matches and fits with the rest in the room (Dasey, n.d.).
13
product. Hongshuang, Sanjay & Kannan (2019) conclude that the better the quality of the digital sampling of popular content, the greater the impact on sales. Initial work in this field focused primarily on the attitude effect of product sampling. Instead of directly purchasing the product, product sampling before purchasing does positively enhance the perception and favorable tendency towards the brand (Bettinger, Dawson, & Wales, 1979; Hamm, Perry, & Wynn, 1969). Furthermore, direct product experience generates a higher emotional feeling state in comparison with the indirect product experience of product advertising (Marks & Kamins, 1988; Smith & Swinyard, 1983). Consumers can experience the product and determine whether it fully satisfies their requirements. However, exposing consumers to advertising after direct sampling has little effect on attitudes. They should be exposed to advertisements before the product samples to change the attitude and raise the purchase intention (Marks & Kamins, 1988). Besides attitude, in-store sampling strongly affects actual purchase behavior (Heiman, McWilliams, Shen, & Zilberman, 2001). In the long-run, Bawa and Shoemaker (2004) quantified in a field experiment that free samples boost sales up to 52 weeks because of “larger potential for acceleration of purchases, greater retention of customers after trial, and a higher purchase probability among those who would not have tried the brand without a free sample.” Recent findings support this conclusion and added that firms should supply free samples in the early stage for an optimal sampling effect of product introduction (Han & Zongming, 2017).
14
purchase intention, as well as the exploratory behavior due to the curiosity about the product (Beck & Crié, 2018). The increased purchase intention by using AR is in line with the previously mentioned findings, that a more tangible product representation mode leads to higher purchase intention in the online retailing apparel industry (Schlosser, 2003). The four journals discussed above, are all lacking external validity due to the young adult participant group. The results do not apply to older age categories. However, the results of Kim and Forsythe (2008), which are based on a random sample from U.S. database panel of a commercial survey provider (n=491), are consistent with previously mentioned findings of purchase intention.
In the e-commerce sector, the level of perceived risk is regulated by the information on the website (Montoya-Weiss et al., 2003). Visual product representation mode compensates for the missing tangibility involvement online, whereby consumers become more certain about product performance (Peck & Childers, 2003). In other words, risk perception is an underlying factor related to the online product representation mode. The product tangibility improves from static products to 360 degrees product rotation, followed by AR (Verhagen, Vonkeman & van Dolen, 2016). As tangibility positively influences the risk perception, consumers can better evaluate the product attributes and the product usage situation with the use of AR while shopping online. The tangibility will increase the perceived certainty. Hence, the following hypothesis is constructed:
H1a: Tangibility of online product representation mode has a positive effect on consumers’ choice utility for a search good.
H1b: Tangibility of online product representation mode has a positive effect on consumers’ choice utility for an experience good.
2.6 Product type
15
the attribute evaluation. Search goods are dominated by product attributes for which full knowledge can be acquired preceding the purchase, as experience goods can only be evaluated after purchase and use of the product (Nelson, 1970; Klein, 1998; Weathers & Makienko, 2006). The product features and specifications are sufficient enough for judging the quality of search goods. The quality of experience goods is more challenging to judge than search goods (Hsieh, Chiu, & Chiang, 2005). Therefore, relying on the recommendations of other consumers is more common in case of experience goods in comparison with experience goods. Hence, for experience goods, it is very important to allow consumers to personally examine the good by giving the opportunity to touch the product (Peck & Childers, 2000). In the e-commerce sector, this tangible experience is often missing. As a result, the consumer is less confident to judge the product quality. Hence, buying experience goods is riskier than buying search goods online. This perceived uncertainty can be overcome by offering a substitute for the lacking direct experience when shopping online. One way to reduce this uncertainty for experience goods is by presenting pictures because it enhances the vivid information (Weathers & Makienko, 2006). More preferably, AR increases the quality of the product representation even more, because of an increased level of vividness (Yim et al., 2017). “Vividness refers to the ability of a technology to produce a sensorially rich mediated environment” (Steuer, 1992:80). Higher vivid product representation modes positively affect the attitude and purchase intention (Coyle & Thorson, 2001; Park et al., 2005). Consequently, the AR experience results in positive consumers evaluation (Yim et al., 2017). In contrast with experience goods websites, the goal of search good websites is to present feature information of the product. In the former case, the virtual experience should be delivered to serve reliable information to the customers (Mazaheri, Richard & Laroche, 2012). Therefore, AR might be a successful information source for experience goods. Hence, it is expected that consumers have a higher online product presentation mode preference in case of experience goods, compared with search goods. Based on the above analysis it is stated that:
H1c: Online product representation mode on choice utility will have a greater impact in case of an experience good in comparison with a search good.
2.4 Brand familiarity
16
them from those of competitors” (Kotler, 1997:443). Brands activate consumers brand knowledge by associations, memories, and feelings, and influence their purchase decision and behavior. However, when consumers have no knowledge about the brand, no feelings, experiences, memories or beliefs about the brand will be created. The degree to which consumers feel that the brand is unfamiliar to them affects brand performance (Ghosh, Chakraborty & Bunch Ghosh, 1995). High perceived risk leads to the preference for familiar products, according to Campbell and Goodstein (2001). This risk aversion eliminates the willingness to try novel products. For this reason, perceived uncertainty is affected by brand familiarity. The more prior brand knowledge and experience, the less the consumer needs to process and evaluate the product attributes (Bettman & Park, 1980). The reduced need for seeking additional information reduces time loss risk. If the consumer is less uncertain about the brand, the consumer is more likely to prefer the brand (Ghosh, et al., 1995). Hence, customer brand experience a priori plays a vital role in product preference. The following hypothesis is stated:
H2a: Brand familiarity has a positive effect on consumers’ choice utility for a search good. H2b: Brand familiarity has a positive effect on consumers’ choice utility for an experience good.
17
effect of brand familiarity. Previous research did not distinguish between different brands, namely; Converse shoes (Brito et al., 2016) and Ray-Ban sunglasses (Poushneh & Vasquez-Parraga, 2017; Yim & Park, 2018; Yim et al., 2017) and Tissot watches (Yim et al., 2017). Research is lacking on how brand familiarity moderates the effect of product representation mode on product preference. AR enhances a higher product-related information source compared with less tangible product presentation modes. As mentioned before, brand unfamiliarity increases the level of uncertainty. Hence, it is expected when consumers are unfamiliar with the brand, a more tangible product representation mode is preferred.
H3a: Online product representation mode on consumers’ choice utility will have a greater impact in the case of brand unfamiliarity in comparison with brand familiarity for a search
good.
H3b: Online product representation mode on consumers’ choice utility will have a greater impact in the case of brand unfamiliarity in comparison with brand familiarity for an experience
good.
This moderation effect might differ across product type. As mentioned before, experience goods are perceived as riskier to be bought online than search goods. The experience product attributes are less easy to judge (Hsieh et. al., 2005). In other words, brand unfamiliarity and experience goods are both perceived as more risky conditions when buying the product online. It is expected that the moderation effect of brand familiarity will have a greater impact on experience goods. Hence, the following hypothesis is formulated:
H3c: The brand familiarity moderation on the relation of product representation mode on choice utility will have a greater impact in case of an experience good in comparison with a search
good. 2.5 Product price
18
1991). Product price positively affects quality, but negatively affect the product’s value-for-money and willingness-to-pay for durable goods (Dodds et al., 1991). A study by Sweeney, Soutar, and Johnson (1999) claimed that the perceived relative price affects the perceived product quality and result in a certain amount of perceived financial risk. As the product price increases, the chance of losing money due to a poor purchase decision increases as well (Verhoef, 2017). Hence, the following hypothesis is stated:
H4a: Product price has a negative (linear) effect on consumers’ choice utility for a search good. H4b: Product price has a negative (linear) effect on consumers’ choice utility for an experience
good.
As mentioned before, a vivid product representation online and product sampling decreases the uncertainty of poor decision making. On the contrary to product price, which will increase the perceived risk if the price increases. As the AR feature is a modern version of product sampling and a vivid product representation mode, it is expected that a relatively higher product price will positively influence the relation of online product representation mode on choice utility. Therefore, a higher product price might leads to a consumer higher preference for e-commerce that has an AR feature. Moreover, the moderation effect of price on the relationship of online product representation mode on choice utility may vary between product type. As price is an important purchase driver for search good (Hsieh et. al., 2005), it is expected that consumer has a lower preference for having a vivid product representation mode than in case of an experience good.
H5a: Product price positively moderates the effect of online product representation mode on consumers’ choice utility for a search good.
H5b: Product price positively moderates the effect of online product representation mode on consumers’ choice utility for an experience good.
H5c: The product price moderation on the relation of product representation mode on choice utility will have a greater impact in case of an experience good in comparison with a search
19
2.7 Covariates
20
3. METHODOLOGY
In this chapter, the research method is discussed, followed by the data collection method. Lastly, the plan for data analysis is described.
3.1 Research method
3.1.1 Choice-Based Conjoint analysis
The main objective of this study to examine consumers’ preference in the setting of online product shopping. To test the conceptual framework, a choice-based conjoint (CBC) analysis is selected. This method was chosen because it is one of the most popular ways to measure consumer preference (Eggers, Sattler, Teichert, & Völckner, 2018). The experimental design with all attributes and their levels is displayed in table 1. Within this study, different types of product representation modes are displayed to consumers through a survey. In addition, the moderating effect of brand familiarity, product type, and product price is taken into account. The variable delivery cost variable has been added to the CBC analysis to create a more realistic online buyer’s market scenario. Hence, CBC gives valuable insights due to the realistic representation of online shopping.
Table 1 Attributes and levels
Product type: Search good (television)
Attribute Level 1 Level 2 Level 3 Level 4 Specification
Online product representation mode Solid background product picture Context background product picture 360 degrees product rotation Augmented Reality part-worth
Brand familiarity Samsung Sony Xinoa Atrumai part-worth Product price $319 $349 $379 $409 linear Delivery cost Free $9 $15 $29 linear Product type: experience good (sofa)
Attribute Level 1 Level 2 Level 3 Level 4 Specification
Online product representation mode Solid background product picture Context background product picture 360 degrees product rotation Augmented Reality part-worth
21
3.1.2 Attributes and levels
As shown in table 1, consumers’ choice utility will be accessed through the specified attributes of the search and experience good and their levels. Based on previous research, furniture is considered as an experience good, because the quality of an experience good is unknown until the purchase (Nakayama, Sutcliffe, & Wan, 2010; Weathers & Makienko, 2006). Therefore, a sofa is considered as experience product in this study. Television is treated as a search good because the attributes can easily be inspected prior to the purchase based on their technological aspects. Hence, this research contains two separate surveys based on the product type. The allocation of the product type is randomly assigned to the participant. All the other variables are kept constant among both surveys. It is assumed that consumers will choose the product alternative with the highest utility. All attributes are effect-coded, which means that the utility levels are comparable across the attribute when estimating the part-worth utilities.
3.1.2.1 Online product representation mode attribute
The survey starts with an introduction about online product representation modes by pictures and text (Appendix A). In the conjoint study, the levels of the attribute product representation mode are presented by the same images as in the introduction (Figure 2). The IKEA STRANDMON wing chair and Samsung UE40NU7190 are used as product pictures, but the brand name is not shown. The use of realistic pictures to describe the levels of product representation mode will improve the accuracy and precision of this study (Eggers, Hauser & Selove, 2016). The online product representation mode levels are chosen based on previous research; solid background pictures, consumption background pictures, 360 degrees rotation and AR (Brito et al., 2016; Verhagen et al., 2016; Yoo & Kim, 2014).
3.1.2.2 Brand familiarity attribute
22
brands and the non-existing unfamiliar brands are Hausmann and SEAT&SIT. The familiar (non-fiction) brands in search good context are 'Samsung' and 'Sony' and Xinoa' and 'Atrumai' as unfamiliar (fictional) brands.
3.1.2.3 Product price attribute
Four product price levels are presented to the participant, based on actual prices of both products. The IKEA STRANDMON wing chair costs $249 in the USA, and $199 in Europa, and $272 in India. Furthermore, Samsung UE40NU7190 costs $379 in the USA and $402 in Europa. The price in India is unknown. A lower price is also taken into account because several Internet shops have discounted this television. The price is ranged from a relatively lower and higher priced product in the current market. The price increases each level with thirty dollars.
3.1.2.4 Delivery cost attribute
The attribute delivery cost is divided into four levels: free, $9, $15 and $29 dollar. These levels are actual delivery costs of IKEA, Samsung, Sony, and Walmart. Other actual delivery prices are higher than $29 dollar, which would have led to a too broad level margin between the delivery costs.
3.1.3 Choice design
The experimental design is fractional factorial design with eleven randomly assigned choice sets with three alternatives. There are 256 stimuli in total since the four attributes (product representation mode, product price, brand familiarity, and delivery cost) each have four levels
23
This properly designed experiment is efficient because each attribute level is uncorrelated (orthogonality). Furthermore, the attribute levels are balanced, which prevents respondents from getting the impression that one attribute is more important than another. Moreover, there is minimal overlap between the alternatives in the choice design and the levels are equally attractive. (Eggers et al., 2018).
After the introduction, the respondent is asked to select the most preferred option in each choice situation (Figure 2). Sequently, the respondent will be shown the dual response no choice option with the question of whether he or she would actually choose their preferred option if it was available. This means that the no-choice option is separately estimated. The dual response CBC procedure increases the predictive accuracy by capturing realistically consumer behavior (Wlömert & Eggers, 2016).
3.1.4 Covariates
24
Figure 2 Example choice design sofa
3.2 Data collection
The software program Preference Lab is used to analyze the data. An advantage of this software is that it automatically controls for validity criteria as discussed earlier. The surveys were pre-tested under Dutch respondents (n=6) to make sure all questions are clearly stated or solving any defects (Bolton, 1993). Small adjustments have been made to the introduction of product representation mode text to make the concept as clear as possible. Moreover, the product representation mode attribute levels are changes from text into pictures, because participants could better remember what the meaning of each level product representation mode was.
25
amount of demographic diverse respondents. MTurk is as reliable and valid as traditional methods (Behrend, Sharek, Meade & Wiebe, 2011; Buhrmester, Kwang & Gosling, 2011; Sprouse, 2011). A representative sample is especially important for managerial implications. These benefits far outweigh the disadvantages with regard to convenience sampling. Furthermore, the intention question “Please select the answer that states “I provide honest answers to surveys” is added to the survey to filter respondents who were not engaged enough when filling in the survey.
The LCA for each product type needs at least 200 respondents (Hair, Black, Babin & Anderson, 2014). The target populations are consumers who are older than eighteen years old. Moreover, the sample needs to have a diverse age range, because the level of innovativeness depends on age. This aspect is especially important because most of the previous AR research is lacking wide age distribution. Furthermore, the participants do not have to be familiar with AR, because, in reality, this is the same situation when shopping online.
3.3 Plan of analysis
The two survey datasets are gathered from Preference Lab. All samples were thoroughly checked for outliers. To ensure data quality, respondents will be rejected who answered wrong the intention question (Appendix B). Moreover, the completion time of the survey is measured, to account for the validity rate of the paid panel. The average completed survey time to complete is 342 seconds for television scenario, and 354 seconds for the sofa scenario. A response time lower than the standard deviation of each dataset is considered as a too low behavioral engagement and is excluded from the dataset. Furthermore, false data points will be checked. Based on the Worker IDs of the participants, repeated participation will be rejected for further data analysis to ensure internal validity (Cheung, Burns, Sinclair & Sliter, 2017). Lastly, the datasets will be checked for missing data points and dealt with them appropriately, for example imputing the mean. See section 4.1.1 for further discussion.
26
Moreover, the relative importance of the attributes product representation mode, product price, delivery price, and brand familiarity are calculated.
After building the best model for both product types, a combined model is created. The two datasets are merged in excel, which allows comparing the impact of product type (0 = sofa, 1 = television) on consumer preferences. In addition to the main effects, new variables are created to measure the moderating effect. These moderators are included as additional attributes in the model. Furthermore, the covariates ‘online shopping experience’ and ‘level of innovativeness’ are checked if they can be reduced into one factor by conducting Principal Component Analysis (PCA) in SPSS.
Three models are build based on the combined dataset. Model 3 contains the estimated main effects. In model 4, the moderators and covariates are included in addition to the main effects. Lastly, the significant moderator or covariate are excluded by stepwise deleting the most non-significant variable. This has resulted in model 5 with the main effect, non-significant moderators, and significant covariates. The best model fit of these three models is calculated by the information criteria BIC, and CAIC. These two information criteria lead to a parsimonious model selection because they penalize adding more parameters to the model (Leeflang, Bijmolt, Pauwels & Wieringa, 2015). Moreover, the Likelihood Ratio (LR) Test is used to test if the estimated model has a significantly better fit than the null model or previous models.
To analyze each CBC data, the multinomial logit (MNL) model is used to calculate the effects for the attribute levels. This aggregate-level model is based on the Random Utility Model (RUT), which argues that the choices are based on the overall utilities of alternatives (U) of a consumer n for the product i (Eggers, 2011; Eggers et al., 2018). This is a latent construct that contains a systematic component (V) and an error component (ε). Hence, the equation is:
27
The systematic utility is developed by the sum of the part-worth utilities for consumer n for the product i. Hence, the following utility equation is applied:
𝑉𝑛𝑖 = ∑𝑘=1𝑘 𝛽𝑛𝑘𝑥𝑖𝑘 (2)
with
k = (1,...,K) number of attributes
x = dummy indicating the specific attribute level of product i β = part-worth utility of consumer n for attribute k.
A drawback of the MNL is that it neglects the preference heterogeneity across consumers. Therefore, LCA is conducted which assumes homogeneous discrete segments that differ in preference and estimates the true distribution of preference. The statistical software R is used to investigate the segment levels. The segments are defined by the attributes as well as possible significant covariates age, gender, household income, online shopping experience, and AR experience. The optimal number of segments can be found by the information criteria BIC and CAIC. The probability (𝑠𝑚𝑛) that consumer (n) belongs to each segment is:
𝑃𝑟𝑜𝑏𝑛𝑖(𝑖|𝐽) = ∑𝑚=1𝑀 𝑠𝑚𝑛( 𝑒𝑥𝑝(𝛽𝑚𝑋𝑖
28
4. RESULTS
4.1 Descriptive analysis4.1.1 Television sample
Both datasets have no false data points and all respondents completed the survey. Moreover, the household income variable observations ‘prefer not to say’ are transformed into the average household income level of the sample (μ = 3.5, n=11). A total of 330 respondents were recruited for this television survey. After excluding the respondents (n=58) with a wrong answer on the intention question, the elapsed fill-in time has been analyzed (μ = 355, SD = 211). Based on the range of the average response time, respondents are excluded if the elapsed time to fill in the survey is outside the minimum range of 144 seconds (n=32). Hence, television dataset contains 240 respondents. Table 2 lists the descriptive statistics of both samples. In this study, slightly more females (53.8%) than males (45.4%) participated. The dataset has a wide age distribution between the age range of 18 and 68 years old. The majority has a bachelor degree (49.6%) or higher (20.5%) and has a yearly household income of $50,000 to $75,000 (20.4%). Regarding the AR experience, 57.3% has heard about AR but has never tried it before while online shopping and 23.4% has tried AR before in this context. Figure 2 shows that more than half of the respondents are from the USA (55.4%), followed by India (25.4%).
4.1.2 Sofa sample
29
Table 2 Descriptive Statistics
Sample tv (n=231) Sample sofa (n=271) Gender Female Male Unknown
Age (in years)
Mean Range St. Dev 18-24 25-29 30-34 35-39 40-44 45-49 50-54 55+
Household Income (yearly)
Under $10,000 $10,000 to $20,000 $20,000 to $30,000 $30,000 to $40,000 $40,000 to $50,000 $50,000 to $75,000 $75,000 to $100,000 More than $100,000 Prefer not to say
Education
Less than a high school diploma High school degree or equivalent Some college, no degree
Associate degree Bachelor’s degree Master’s degree Professional degree Doctorate AR experience
Never heard about it.
Heard about it, but never tried before while online shopping. Heard about it and already tried it while online shopping.
Never heard about the term Augmented Reality, but I have already tried it while online shopping.
30
4.2 Familiarity check
Table 3 shows that some respondents are familiar with the fictional television brands Xinoa (17.5%) and Atrumai (13.75%). According to table 4, this percentage of familiarity is even higher for the fictional sofa brands Hausmann (27.7%) and SEAT&SIT (20.3%). Despite this familiarity, all respondents are still taken into account of the data analysis because in (top) journals brand familiarity is commonly manipulated by fictional brands (Campbell & Keller, 2003; Lange & Dahlén, 2003).
Table 3 Brand familiarity television (n=240) Brand
Familiarity Samsung Sony Xinoa Atrumai
Familiar 227 (94.6%) 228 (95.0%) 42 (17.5%) 33 (13.8%) Unfamiliar 7 (2.9%) 8 (3.3%) 165 (68.8%) 166 (69.2%) Not sure 6 (2.5%) 4 (1.7%) 33 (13.8%) 42 (17.5%)
Table 4 Brand familiarity sofa (n=271) Brand
Familiarity IKEA Walmart Hausmann SEAT&SIT
Familiar 230 (84.9%) 249 (91.9%) 75 (27.7%) 55 (20.3%) Unfamiliar 27 (10.0%) 15 (5.5%) 155 (57.2%) 176 (64.9%) Not sure 14 (5.2%) 7 (2.6%) 41 (15.1%) 40 (14.8%)
31
4.3 Factor analysis
4.3.1 Covariate level of innovativeness
In both surveys, the level of innovativeness is measured with four items based on Likert-scale measurement, according to Agarwal and Prasad (1998). Several steps have been conducted to see if the set of items can be combined into factors. Firstly, the Cronbach’s Alpha is bigger than 0.6 (television: α = 0.601 & sofa: α = 0.695), which is satisfactory enough for internal consistency reliability (Malhotra, 2010:319). In addition, Kaiser-Meyer-Olkin (KMO) Measure of sampling adequacy (>0.05) and Bartlett’s Test of Sphericity (p=0.000) shows that a Principal Component Analysis (PCA) is appropriate. The communality of item ‘3. In general, I am hesitant to try out technologies’ is too low (< 0.04), which indicates that this item has little in common with the other three variables. Hence, factor analysis is appropriate for the three items ‘1. If I heard about a new technology, I would look for ways to experiment with it.’, ‘2. Among my peers, I am usually the first to try out new technologies.’ And ‘4. I like to experiment with new technologies.’ Appendix C shows that these three high-loaded items (loading >0.5) can be merged into one factor. Moreover, if item 3 is excluded, the Cronbach’s Alpha will be improved to α = 0.874 (television) and α = 0.877 (sofa). In sum, item 3 is excluded and item 1,2 and 4 are merged into one factor which will be used as innovativeness covariate.
4.3.2 Covariate online shopping experience
The three items of the covariate online shopping experience are based on Poushneh and Vasquez-Parraga (2017). According to Appendix C, the internal consistency of these items are good (television: Cronbach’s α = 0.877 & sofa: Cronbach’s α = 0.840) and factor analysis is appropriate (KMO >0.05, Bartlett’s Test of Sphericity p=0.000, and communalities >0.4). Hence, the three items are merged into one factor.
4.4 Choice-based conjoint analysis
4.4.1 Type of relationship between factor levels
32
price seems to have a linear trend (Appendix D). The Likelihood Ratio test shows that there are no significant differences between the part-worth and the product price and delivery cost linear models. However, the model becomes more parsimonious by replacing the two attributes into a linear utility function. Hence, in the following section, the main effect model product price and delivery price are linear attributes.
Table 5 Main model comparison based on information criteria
Television Sofa Part-worth Product Price Linear Delivery Price linear Product Price & Delivery linear Part-worth Product Price linear Delivery Price linear Product Price & Delivery linear LL(0) -3659.817 -3659.817 -3659.817 -3659.817 -4132.543 -4132.543 -4132.543 -4132.543 LL(m) -3247.526 -3247.589 -3248.116 -3248.184 -3704.8 -3715.133 -3705.9 -3716.216 DF 227 229 229 231 258 260 260 262 Parameters 13 11 11 9 13 11 11 9 R2 0.113 0.113 0.112 0.112 0.104 0.101 0.103 0.101 Adj. R2 0.109 0.110 0.110 0.110 0.100 0.098 0.101 0.098 BIC 6566.3 6555.465 6556.519 6545.694 7482.428 7491.889 7473.423 7483.019 CIAC 6579.3 6566.465 6567.519 6554.694 7495.428 7502.889 7484.423 7492.019 LR Test P(Chisq= 605.21; df=10) <0.000 *** 1 P(Chisq= 0.1242; df=2) =0.9402 P(chisq= 1.1785; df=2) =0.5552 P(chisq= 1.316; df=4) =0.8592 P(Chisq= 630.32; df=10) <0.000 *** 1 P(Chiq= 20.634; df=2) < 0.000 ***2 P(Chiq= 2.142; df=2)= 0.3432 P(Chiq= 22.889; df=4)< 0.0002 1 compared to the null model, 2 compared to the part-worth model
4.4.2 Main effect results
33
sum, when shopping for television online, the respondents prefer a familiar brand with a solid background picture representation mode, for a relatively lower product price and no delivery cost.
Table 6 Model 1 main effect model (television)
Attribute Utility SE z-value Pr(>|z|) Rel. imp.
Product representation mode
Solid background 0.138 0.040 3.523 0.000 *** 10.8% Context background -0.147 0.042 -3.559 0.000 *** 360 degrees 0.003 0.041 0.063 0.950 AR 0.007 0.041 0.171 0.864 Brand familiarity Samsung 0.507 0.038 13.537 0.000 *** 37.4% Sony 0.414 0.038 10.886 0.000 *** Xinoa -0.491 0.047 -10.540 0.000 *** Atrumai -0.432 0.0457 -9.449 0.000 *** Product price Price (linear) -0.273 0.021 -12.776 0.000 *** 30.7% Delivery cost
Delivery cost (linear) -0.186 0.091 -19.063 0.000 *** 21.2%
No choice option -1.724 0.090 -19.063 0.000 *** -
*** p < 0.000, * p<0.05
Secondly, the main effects of the sofa scenario are analyzed (Table 7). There was no significant difference detected between the product representation modes. It only reveals that AR has a significant negative effect (β=-0.077, p<0.000) on product preference. Therefore, H1b is rejected. Furthermore, sofa preference is significantly negatively affected by SEAT&SIT (β= -0.128, p<0.001) and positively by affected IKEA (β= 0.075, p<0.05). Remarkably, the two other brands are not significant. Hence, H2b is partially supported because the effect of brand familiarity only holds for one familiar and one unfamiliar brand. Moreover, when the price increases, the sofa preference significantly decreases (β=-0.464, p<0.000). Hence, H4b is supported. Overall, on average the participants are more likely to prefer none of the televisions or sofas because the no choice utility has the highest value (p<0.000).
Table 7 Model 2 main effect model (sofa)
Attribute Utility SE z-value Pr(>|z|) Rel. imp.
Product representation mode
Solid background -0.018 0.038 -0.479 0.632 6.3% Context background 0.039 0.038 1.038 0.299
34 Brand familiarity IKEA 0.075 0.038 1.991 0.046 * 9.4% Walmart 0.027 0.038 0.706 0.480 Hausmann 0.026 0.038 0.692 0.489 SEAT&SIT -0.128 0.040 -3.232 0.001 ** Product price Price (linear) -0.464 0.021 -22.526 0.000 *** 61.0% Delivery cost
Delivery cost (linear) -0.168 0.020 -8.471 0.000 *** 23.3%
No choice option -1.724 0.084 -25.432 0.000 *** *** p<0.000, ** p<0.001, * p<0.05
4.4.3 Relative importance
When buying a television online, brand familiarity has the most relative important preference (37.4%) relatively to the other attributes, because it has the largest impact on the utilities. This differs from the sofa dataset where the price (61%) is relatively most important, followed by delivery costs (23.3%), and brand familiarity (9.4%). In both surveys, the product representation mode is considered as least important (Table 6 and 7).
4.4.4 Predictive validity
The holdout choice set in the survey is used as a benchmark for predictive validity (Appendix E). Predictive validity shows how good the estimates predict the actual preference of the participants. The other eleven choice sets are used for estimation. The difference between the predicted and actual preference for each option in the holdout choice set is measured by the Mean Absolute Error (MAE). The MAE is calculated by dividing the sum of absolute errors (predicted – observed errors) by the number of alternatives (Table 8). An MAE value of 0 indicates a perfect model prediction. The MAE value of model 1 is about 5.7%, indicating that model 1 is missing actual television preference by about 5.7%. The model accuracy is lower for model 2 because the predicted sofa preference differs from the actual preference by 7.2%.
Table 8 Predictive validity
Model 1 (television) Model 2 (sofa)
Option 1 Option 2 Option 3 Option 1 Option 2 Option 3
Predicted 54.0% 31.5% 14.4% 59.3% 13.3% 27.3%
Observed 61.3% 32.9% 5.8% 60.5% 22.9% 16.6%
Abs. difference 7.3% 1.4% 8.6% 1.2% 9.6% 10.7%
35
4.4.5 Main effect combined model
In this section, the two models are combined. A new attribute for the two product type (0=sofa, 1=television) and a dummy variable is created, because search and experience goods have to be relative to each other to compare their effects. Firstly, the differences between the main effects are discussed. In the next paragraph, the moderating effects are compared.
The significance and sign of the product representation mode levels differ across the two product types. Firstly, the product representation mode solid background influences choice utility positively in case of buying a television online (β= 0.138, p<0.001), but the context background has a stronger and negative choice utility (β= -0.147, p<0.000). Secondly, a 360 degrees product rotation has no significant effect on both scenarios. Thirdly, the AR has a significant positive influence on the television choice utility (β= 0.006, p<0.1), but has a significant negative influence on the sofa choice utility (β= -0.079, p<0.1). Concerning H1c, the strongest product representation mode preference across the two product types is the television context background (β= -0.147, p<0.000). Moreover, the relative importance of product representation mode (Table 6 and 7) is greater in the television scenario (10.3%), compared with the sofa scenario (6.3%). Hence, H1c is rejected because the effect of online product representation mode on choice utility has not a greater impact in case of an experience good in comparison with a search good.
Table 9 Model 3 main effect combined model
Television Sofa
Attributes Utility SE Utility SE
Product representation mode
Solid background 0.138 ** 0.055 -0.018 0.038 0.038 0.038 0.039 Context background -0.147 *** 0.056 0.039 360-degrees 0.003 0.056 0.057 AR 0.006 * 0.040 -0.079 * Brand familiarity
Samsung & IKEA 0.507 *** 0.053 0.075 * 0.038 0.038 0.038 0.045 Sony & Walmart 0.414 *** 0.054 0.027
Xinoa & Hausmann -0.491 *** 0.060 0.026 Atrumai & SEAT&SIT -0.431 *** 0.041 -0.128 **
Product price (linear) -0.273 *** 0.030 -0.464 *** 0.021 Delivery cost (linear) -0.186 *** 0.029 -0.168 0.020
36
4.4.5 Moderating effects and covariates combined model
It is measured if the moderators and covariates improve the model fit of the combined model 4 by stepwise excluding the highest non-significant variable. More details on this are given in the next section. Table 10 presents the significant moderating and covariates effect of model 5, which is used to discuss the moderation and covariates effect. Table 10 demonstrates that the level of innovativeness positively affects the product representation mode preference when shopping a sofa (β=0.077, p<0.000). In the television case, the level of innovativeness does not affect the results. Furthermore, the covariate consumer´s yearly household income significantly negatively moderates the relationship between the product representation mode and consumer choice utility (tv: β= -0.005, p<0.000, sofa: β= -0.043, p<0.000). With respect to the moderating effects, price does not influence the relation of product representation mode on product preference. Hence, H5a,b,c are rejected.
Secondly, the moderation effect of brand familiarity on the relation of the product representation modes context background and AR have a significant effect. This moderation effect is different across the two product types. A familiar brand positively affects this relationship in case of a television context background (Sony: β=0.051, p<0.05), but negatively affects the relationship in case of a sofa context background (Walmart: β= -0.134, p<0.05). On the contrary, the preference for a context background picture is negatively affected by an unfamiliar television brand (Xinoa: β= -0.130, p<0.05) and positively affect by an unfamiliar sofa brand (Hausmann: β=0.103, p<0.1). Regarding the moderation effect of brand familiarity on the relationship of product representation mode on choice utility, AR has the highest television utility in case of brand unfamiliarity (β=0.048, p<0.05) and the highest sofa utility in case of brand familiarity (β=0.135, p<0.05). In sum, except the moderating effect of the familiar brand Walmart, the product representation modes context background and AR have a greater impact on television and sofa choice utility in the case of brand unfamiliarity in comparison with brand familiarity. However, in both scenarios, this effect
*** p<0.000, ** p<0.001, * p<0.05, . p<0.1
This table presents the utilities of the main effect (sofa) and the main + interaction effect (television). For example, main effect of solid background sofa utility is -0.018. If you add the interaction effect of 0.156 up to the main effect, the solid background television utility is 0.138
37
only applies to one familiar (Walmart & Sony) and one unfamiliar brand (Hausmann & Xinoa). In addition, brand familiarity does not moderate the relation of the other two levels solid background picture and 360 degrees rotation on choice utility. Hence, H3a and H3b are partially supported. Moreover, the moderation effect of brand familiarity on the relation of the context background picture and AR product representation mode choice utility has, except the unfamiliar brand Xinoa, a greater impact on the sofa scenario compared to television scenario. Hence, H3c is partially supported.
Table 10 Model 5 main and significant moderating and covariates effects combined model Utility television SE Utility sofa SE
Product representation mode
Solid background 0.639 ** 0.119 0.249 0.136 ** ** 0.084 Context background 0.020 . 0.068 0.046 360 degrees rotation -0.164 ** 0.066 -0.028 0.045 AR -0.494 0.121 -0.356 *** 0.085 Brand familiarity
IKEA & Samsung 0.512 * 0.053 0.077 *** 0.038
Walmart & Sony 0.362 . 0.054 0.023 *** 0.038
Hausmann & Xinoa -0.496 * 0.060 0.027 *** 0.038
SEAT&SIT & Atrumai -0.432 *** 0.061 -0.132 *** 0.040
Product price (linear) -0.664 *** 0.030 -0.469 *** 0.021
Delivery cost (linear) -0.189 0.029 -0.169 *** 0.020
Non option -0.931 *** 0.208 -1.737 *** 0.148
Interactions
Context background x Sony & Walmart 0.051 * 0.060 -0.134 * 0.061 Context background x Xinoa & Hausmann -0.130 * 0.093 0.103 . 0.088 AR x Sony & Walmart 0.032 . 0.060 0.135 * 0.061 AR x Xinoa & Hausmann 0.048 * 0.093 -0.185 * 0.088
Covariates
Yearly household income -0.005 *** 0.010 -0.043 *** 0.007 Level of innovativeness 0.087 0.015 0.077 *** 0.011
*** p<0.000, ** p<0.001, * p<0.05, . p<0.1
This table presents the utilities of the main effect (sofa) and the main + interaction effect (television). For example, the main effect of solid background sofa utility is 0.249. If you add the interaction effect of 0.390 up to the main effect, the solid background television utility is 0.639.
38
4.4.6 Combined model validity
First of all, the combined main effects model is significantly better than the null model (p<0.000). Moreover, the Likelihood Ratio Test in table 11 shows that the estimated combined model 4 with moderators and covariates has a significantly better fit compared with the main effects model. The estimations of model 4 can be found in Appendix E. Model 5 has not a significantly better model fit than model 4. Although the fit of model 5 is not significantly better than model 4, this model is more preferred because it is simpler due to the fewer parameters. Moreover, model 5 has the lowest the information criteria BIC (13599.69) and CIAC (13573.69) and has a better goodness-of-fit of the MNL model after taking the different number of independent variables into account (Adjusted R2 = 0.114). Hence, model 5 has the best balance between a good fit and parsimony.
Table 11 Model validity combined model Combined model 3:
Main effect
Combined model 4: Main, moderators, and
covariates effect
Combined model 5: Main, significant moderators and
significant covariates effect
LL(0) -7792.361 -7792.361 -7792.361 LL(m) -6964.445 -6863.347 -6880.918 DF 493 457 485 Parameters 18 54 26 Pseudo R2 0.106 0.119 0.117 Adjusted R2 0.104 0.112 0.114 BIC 14041.14 14063.46 13599.69 CIAC 14059.14 14117.46 13573.69 Likelihood Ratio Test P(Chiq=1214.4; df=15)<0.000***1 P(Chiq=202.19; df=36) < 0.000***2 P(Chiq=35.140; df=28) =0.1663 *** p<0.000
39
4.5 Overview hypotheses
Table 12 Overview hypotheses
Hypothesis Supported?
H1a Tangibility of online product representation mode leads to a higher consumers’ choice utility for a search good.
❌ H1b Tangibility of online product representation mode leads to a higher consumers’ choice
utility for an experience good.
❌ H1c Online product representation mode on choice utility will have a greater impact in case of
an experience good in comparison with a search good.
❌ H2a Brand familiarity has a positive effect on consumers’ choice utility for a search good. ✔️ H2b Brand familiarity has a positive effect on consumers’ choice utility for an experience good. Partially
supported H3a Online product representation mode on consumers’ choice utility will have a greater
impact in case of brand unfamiliarity in comparison with brand familiarity for a search good.
Partially supported H3b Online product representation mode on consumers’ choice utility will have a greater
impact in case of brand unfamiliarity in comparison with brand familiarity for an experience good.
Partially supported H3c The brand familiarity moderation on the relation of product representation mode on choice
utility will have a greater impact in case of an experience good in comparison with a search good.
Partially supported H4a Product price has a negative (linear) effect on consumers’ choice utility for a search good. ✔️ H4b Product price has a negative (linear) effect on consumers’ choice utility for an experience
good.
✔️
H5a Product price positively moderates the effect of online product representation mode on consumers’ choice utility for a search good.
❌ H5b Product price positively moderates the effect of online product representation mode on
consumers’ choice utility for an experience good.
❌ H5c The product price moderation on the relation of product representation mode on choice
utility will have a greater impact in case of an experience good in comparison with a search good.
40
4.6 Latent Class Analysis
To account for preference heterogeneity across consumers, an explorative LCA for several numbers of segments models are estimated, because the optimal number of segments is unclear prior to the analysis. In this analysis, consumers belong to discrete segments that differ in preferences.
4.6.1 Number of segments
The best fit for the number of segments is measured according to the information criteria, and the classification error (Eggers, et al.). The explorative segmentation is based on MNL model 3 with the attribute levels of product representation mode, brand familiarity, product price, and delivery cost (Table 1). Previously, product price and delivery price were considered as a linear function. The LCA the segments are estimated by the part-worth utilities (Eggers, et al., 2018:38). Moreover, the significant covariates level of innovativeness and yearly household income of model 5 is included for the segmentation. The covariate yearly household income is significant (p<0.000) for both product types. Nevertheless, the level of innovativeness is not significant for both product types. Therefore, it is first tested if the model fit increases by excluding the level of innovativeness. The value of BIC and CIAC are constantly lower when both covariates are included (Table 12). Moreover, the goodness of fit of the model with the covariates yearly household income and level of innovativeness is better, because the (adjusted) R2 is higher than the model with only the covariate yearly household income. Hence, the covariates yearly household income and level of innovativeness are included for the LCA. After deciding to run the LCA on model 3, which includes both covariates yearly household income and level of innovativeness, the segment-level is estimated. The largest feasible number of classes is four. Table 12 shows that the BIC and CIAC have the lowest value for four latent classes. For this reason, the optimal number of classes is four. Note that the model fit adjusted R2 increased from 0.114 to 0.291 and the classification error
41
Table 12 Overview LCA model fit
Yearly household income Yearly household income + Level of innovativeness
Number of classes
BIC CIAC Class.
Error
R2 Adj R2 BIC CIAC Class.
Error R2 Adj R2 1 14048.20 14077.20 0.000 0.110 0.106 13939.67 13970.67 0.000 0.118 0.114 2 12498.82 12557.82 0.012 0.222 0.214 12475.08 12538.08 0.012 0.225 0.217 3 11815.10 11904.10 0.030 0.277 0.266 11798.62 11893.62 0.035 0.281 0.269 4 12224.83 12319.83 0.041 0.254 0.241 11749.43 11876.43 0.057 0.312 0.291
4.6.2 Four segment analysis
42 51% 8% 26% 6% 19% 35% 51% 6% 26% 24% 9% 80% 3% 33% 14% 8% S E G M E N T 1 S E G M E N T 2 S E G M E N T 3 S E G M E N T 4
Product representation mode Brand familiairity Product price Delivery cost
Table 13 Overview segments
Segment 1 (17.3%) ‘Outside good’
Segment 2 (33.1%) ‘Delivery service’
Segment 3 (29.1%) ‘Familiarity & Solid’
Segment 4 (20.5%) ‘Price sensitive’
Utility tv Utility sofa Utility tv Utility sofa Utility tv Utility sofa Utility tv Utility sofa
Product representation mode
Solid background 1.371 *** -0.161 0.191 0.165 0.977 0.683 ** 0.086 -0.041 Context background 0.582 * 0.049 -0.164 ** 0.273 * -0.116 0.178 -0.194 0.064 360 degrees rotation -0.552 . -0.047 -0.003 -0.241 * -0.246 * 0.199 . 0.358 0.053
AR -1.401 0.159 -0.024 -0.197 -0.615 -1.06 -0.250 -0.076
Brand familiarity
Samsung & IKEA 0.415 0.216 0.113 -0.062 1.592 *** 0.084 0.459 0.366 ** Sony & Walmart 0.512 0.128 0.468 . 0.270 *** 1.348 *** -0.075 0.056 -0.225 Xinoa & Hausmann -0.393 . 0.105 -0.133 -0.033 -1.579 *** 0.095 -0.222 -0.007 Atrumai & SEAT&SIT -0.534 -0.449 -0.448 -0.175 -1.361 -0.104 -0.293 -0.134
Product price $319 & $199 0.685 ** 1.378 *** 0.148 *** 0.576 *** 0.198 * -0.130 2.693 3.141 *** $349 & $229 0.471 0.146 -0.369 0.088 -0.371 * -0.083 1.310 1.032 *** $379 & $259 -0.435 -0.277 -0.053 -0.151 * 0.094 0.110 1.807 -1.396 *** $409 & $289 -0.721 -1.247 0.274 -0.513 0.079 0.103 -5.810 -2.777 Delivery cost Free 0.068 0.416 * 0.334 . 0.548 *** 0.342 ** -0.049 0.400 ** 1.003 *** $9 0.093 0.179 -0.084 * 0.142 * 0.206 0.043 0.229 -0.039 $15 -0.084 -0.474 * -0.544 * -0.317 *** -0.092 0.047 -0.141 . -0.084 $29 -0.077 -0.121 0.294 -0.373 -0.546 -0.041 -0.448 -0.880 Non-option 3.679 *** 2.272 *** -2.150 -1.834 *** -0.280 -1.130 ** -1.421 -1.777 ** Covariates Household income -0.324 *** -0.162 *** 0.043 0.059 ** -0.013 -0.061 ** 0.089 -0.000 Level of innovativeness 0.164 0.128 *** -0.068 -0.030 0.100 0.082 ** -0.086 0.030 *** p<0.000, ** p<0.001, * p<0.05, p<0.1 Grey = significant
Figure 3 Importance attribute television per segment Figure 4 Importance attribute sofa per segment
7% 17% 77% 2% 15% 15% 9% 7% 58% 37% 11% 77% 20% 31% 43% 14% S E G M E N T 1 S E G M E N T 2 S E G M E N T 3 S E G M E N T 4
43
5. DISCUSSION
Previous studies tried to explore how the level of interactive product representation affects purchase intention when shopping online. This study has gone some way towards enhancing the understanding of the effect of AR on product preference under various consumer characteristics and product type conditions. The conjoint analysis was used to test the effect of product representation mode on search and experience product preference. This chapter discusses the theoretical implications, managerial implications, and limitations.
5.1 Theoretical implications 5.1.1 Main effects
44
Secondly, online product representation mode has a greater choice utility impact. The relative attribute importance in case of a search product is higher in comparison with an experience product. This is contradicting with the idea that experience goods are more risky to buy online compared to search goods. A more vivid product presentation reduces this risk perception (Weathers & Makienko, 2006). A possible explanation could be that in this study the television product price is higher than the sofa. Therefore, the risk perception is higher and product representation mode is more important (Montoya-Weiss et al., 2003).
Thirdly, the impact of brand familiarity on product preference is confirmed. For search and experience good it holds that brand familiarity has a positive effect on product preference and unfamiliarity has a negative effect on product preference. Consequently, consumers who are familiar with the brand are more likely to prefer the product. This is consistent with Nepomuceno, Laroche, & Richard (2014), who argue that brand familiarity decreases the perceived risk of buying online. Moreover, consumers shopping a search good on the Internet consider brand familiarity as four times more relatively important compared to consumers who are shopping for an experience good online.
Fourthly, product preference is affected by product price. Search and experience good preference decrease if the price increases. This concurs well with Sweeney, Soutar, and Johnson (1999) and Dodds et al. (1991). Another price element while shopping online is the delivery cost. Although this variable is used as a filler in the conjoint analysis to create a realistic setting, it negatively influences search and experience good preference as the costs increase.
5.1.2 Moderating effects
