• No results found

Perceived Quality of Smartphone Applications and its Influence on Brand Attitude

N/A
N/A
Protected

Academic year: 2021

Share "Perceived Quality of Smartphone Applications and its Influence on Brand Attitude"

Copied!
67
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Perceived Quality of Smartphone Applications and its

Influence on Brand Attitude

(2)

2

Perceived Quality of Smartphone Applications and its

Influence on Brand Attitude

Master Thesis

University of Groningen

Faculty of Economics and Business Department of Marketing

31 October, 2012

Name: Tom Verhaaf

Student number: s1540831 Address: Kraneweg 89A

(3)

3

Management Summary

Nowadays, an increasing number of companies have a smartphone app available that their (potential) customers can download. These apps are often used as a marketing tool, or as support for other marketing tools. Mobile commerce (m-commerce) is growing rapidly, and an increasing number of consumers prefer their smartphone as device of choice for reading and writing emails, browsing the internet, and doing product research. Despite all this, there is only a small number of recent studies about consumer preferences of mobile applications.

In this study, a review of existing literature was carried out to gain more insight in the structure of the relationship between a brand and its (potential) customers. This gave an indication towards the central role of brand attitude. This is the overall evaluation of a brand, and is based on the set of associations linked to the brand.

This study examines the effect of the perceived quality of a smartphone app on the brand attitude, which is described in terms of five different characteristics: favorability, accessibility, confidence, persistence and resistance. Furthermore, a conjoint analysis was performed in order to discover the extent to which the different attributes of an app determine the perceived quality of an app. The attributes that were chosen based on the literature review are: usefulness, ease of use, enjoyment and the appeal of the visual design. Next, a latent class analysis was performed to discover if there is any segmentation possible based on demographic variables, socio-economic variables and phone usage variables. Finally, several simple regression analyses were performed to test the hypotheses, and to test if the quality of an app has a significant effect on the brand attitude.

The conjoint analysis showed that for the total population, the usefulness of an app is the most important attribute (relative importance of 48,56%). The second most important attribute is the ease of use (relative importance of 35,21%). The enjoyment of an app is the third most important (relative importance of 10,67%). The appeal of the visual design is the least important attribute (relative importance of 5,56%)

(4)

4 which offer a reasonable level of usefulness and is not too difficult to use. The convenience users do not care for the enjoyment that an app provides (relative importance of 3,71%).

(5)

5

Preface

This paper was written as the final part of my master Business Administration with the specializations Marketing Management and Marketing Research. The topic was chosen because it reflects my own interests and it offers possibilities to combine several courses that I have followed during the final phase of my study.

I would like to thank my supervisor, prof. dr. Jaap Wieringa, for all his input and feedback during the writing of this thesis. Our meetings always included some interesting discussions, which encouraged me to keep a critical look towards my own work. Furthermore, I would like to thank dr. Jenny van Doorn for her feedback during the final phase of this thesis.

On a personal level I would like to thank my family and friends for all their interest, support and patience.

(6)

6

Table of Contents

Management Summary ... 3 Preface ... 5 1 Introduction ... 8 1.1 Relevance ... 9 2 Theoretical Framework ... 12 2.1 Brand Attitude ... 12

2.1.1 Attitude formation and change ... 13

2.2 Brand attitude as a driver of customer equity ... 16

2.2.1 Value Equity ... 17

2.2.2 Brand Equity ... 18

2.2.3 Relationship equity ... 19

2.3 Brand attitude as a driver of customer commitment ... 20

2.3.1 Affective commitment ... 20

2.3.2 Calculative commitment ... 21

2.4 Brand attitude as a driver of customer brand engagement ... 21

2.5 Attributes of an app ... 23

2.5.1 Usefulness ... 23

2.5.2 Ease of Use ... 23

2.5.3 Enjoyment ... 24

2.5.4 Appeal of Visual Design ... 24

2.6 Customer Characteristics ... 25

2.7 Conceptual Model and Hypotheses ... 26

3 Research Design ... 28

3.1 Research methods ... 28

3.1.1 Attributes concerning app quality (conjoint analysis) ... 28

3.1.2 Allowing for heterogeneity (latent class analysis) ... 31

3.1.3 Relationship between app quality and brand attitude (regression) ... 32

3.2 Variables and data collection ... 32

3.2.1 CBC... 33

3.2.2 Describing the variables ... 34

4 Results ... 36

4.1 Descriptives ... 36

4.2 Aggregate Model Estimation and Validation ... 37

(7)

7

4.4 Descriptives and Part-worths for each class ... 40

4.4.1 Class 1 ... 43

4.4.2 Class 2 ... 44

4.4.3 Class 3 ... 45

4.4.4 Class 4 ... 46

4.5 Significant differences between segments (ANOVA) ... 47

4.6 Relation Between Perceived Quality of an App and Brand Attitude ... 47

5 Conclusions ... 49

5.1 Total population ... 49

5.2 Describing the segments ... 49

5.2.1 Segment 1: Non-downloaders ... 49

5.2.2 Segment 2: Demanding users ... 50

5.2.3 Segment 3: Entertainment users ... 51

5.2.4 Segment 4: Convenience users ... 52

5.3 Effect of perceived quality of an app on brand attitude ... 53

5.4 Limitations & suggestions for further research ... 54

6 Literature ... 56

7 Appendices ... 63

Appendix A: Survey question with a random conjoint choice task ... 63

Appendix B: Utilities and segment sizes (2-5 class solutions) ... 63

(8)

8

1 Introduction

Nowadays, more and more companies have a smartphone- or tablet-app available that their (potential) customers can download. These apps are often used as a marketing tool, or as support for other marketing tools, since they offer the capability for ongoing brand awareness and engagement and to use the full potential of a mobile handset as a marketing channel (Chiem, Arriola, Browers, Gross, Limman, Nguyen, Sembodo, Song & Seal, 2010).

Many people view their mobile phone as a necessity (Mort & Drennan, 2007), and the many available options to personalize their mobile device are an acknowledgement of the increasing integration with one’s personal identity and image (Pura, 2003; Wilska, 2003; Mort & Drennan, 2007). Using a smartphone with an internet connection enables consumers to undertake decision-making tasks at virtually any location. However, the technical limitations of a smartphone in comparison with a desktop computer, such as the smaller screen size or limited input possibilities, make these tasks relative difficult due to the higher cognitive cost associated with processing the information (Maity, 2010). This is the reason why the right combination of attribute levels is arguably even more important than in E-commerce channels. Until now, one of the reasons why M-commerce has not lived up to its expectations is the confusion of the development of new technology with actual buyers, or users in this case, benefits (Shugan, 2004). App designers pay more attention to the technical possibilities that an app can offer, but too little attention to the needs of the users.

According to Wells, Valacich & Hess (2011), a website can serve as a signal of product quality. When consumers have limited information about a product, they use other cues to make an assessment of the product quality (Kirmani & Rao, 2000). Consumers use both attributes that are inherent to a product (intrinsic cues) and attributes that are not inherent to a product (extrinsic cues) to assess product quality (Richardson, Dick & Jain, 1994; Wells, Valacich & Hess, 2011). A cue that can be used to convey information about unobservable product quality to a (potential) buyer in a credible manner is a signal (Rao, Qu & Ruekert, 1999). At the moment, technological capabilities limited the ability of a seller to convey intrinsic product attributes (Grewal, Iyer & Levi, 2004). Because extrinsic cues are more influential when they are more readily available than intrinsic cues, a website may be a powerful signal for assessing product quality (Dawar & Parker, 1994; Wells, Valacich & Hess, 2011). It is assumable that this holds for a smartphone app as well.

(9)

9 Alford, 2001; Holehonnur, Raymond, Hopkins & Fine, 2009; MacInnes, Priester, Eisengerich & Iacobucci, 2010), customer commitment (Ahluwalia, 2000; Fullerton, 2003) and customer brand engagement (Sprott, Czellar & Spangenberg, 2009). Data concerning the change in brand attitude as a result of using an app, as well as the perceived quality of the app, will be collected. This data can be used to gain more insight in the influence of the perceived quality of an app on the brand attitude. Therefore, the subject of this research is the effect of the design of a smartphone app, in terms of attributes, on the brand attitude of the user. This results in the following problem statement:

“What influence do the different attributes of an app have on the perceived quality of the app, and the brand attitude of the user?

To answer the problem statement, the following research questions need to be answered:  What is the structure of the relationship between a brand and the consumer?  What is the role of brand attitude within this relationship?

 How can an app influence the brand attitude?  Which attributes of an app are relevant?

 What is the optimal combination of attribute levels?

In order to answer these questions, existing literature will be used to construct a framework on which the hypotheses will be based. Next, the research method, data collection and the plan of analysis will be addressed. Hereafter, the empirical data will be discussed, and the analyses will be performed, after which the conclusions will be drawn, and recommendations will be given.

1.1 Relevance

(10)

10 doubled within the last year (42% in august 2011). When looking at a global scale, smartphones already outsold PC’s in 2011 (table 1.1), which is an indication towards the rapidly growing potential of M-Commerce. Please note that pads/tablets are categorized as a client PC, but they run on the same operating systems as smartphones (iOS/Android). Since this is the only category of client PCs that shows substantial growth, the future for PCs looks even worse than this data indicates.

Category Q4 2011 (millions of units) Growth Q4 2011/Q4 2010 Full year 2011 (millions of units) Growth 2011/2010 Smartphones 158.5 56.6% 487.7 62.7% Total client PCs 120.2 16.3% 414.6 14.8% Pads 26.5 186.2% 63.2 274.2% Netbooks 6.7 -32.4% 29.4 -25.3% Notebooks 57.9 7.3% 209.6 7.5% Desktops 29.1 -3.6% 112.4 2.3%

Table 1.1: Level of global smartphone and PC sales in 2011 (source: Canalys 2012)

Another indication of the increasing relevancy of smartphones and the potential of M-commerce is the role of smartphones as device of choice for reading and writing emails, browsing the internet, and doing product research. A third of all smartphone owners prefers using their smartphone even when a PC is nearby (Chappuis, Gaffey & Parvizi, 2011). Furthermore, recent research by InMobi (2012) showed that 48% of all consumers let a M-commerce channel influence their purchase decisions, from awareness to the actual purchase of a product or service, in some way. This means that the influence of M-commerce is stronger than the influence of E-commerce (47%) or television (46%).

Many of the earlier studies concerning consumer preferences of mobile applications are 5-10 years old. During this period, the development of apps experienced a rapid growth due to the increased technical possibilities such as larger screens and better connectivity, the massive increase of smartphone usage, and the introduction of tablets.

(11)
(12)

12

2 Theoretical Framework

In this chapter, a theoretical framework will be developed based on existing literature. The main goal of this framework is to gain more insight in the relationship between a company and its (potential) customers, and the role of brand attitude in this. Brand attitude is a central concept within the relationship between a brand and its customers. It is linked to different constructs, such as customer equity (Aron, Aron & Smollan, 1992; Faircloth, Capella & Alford, 2001; Holehonnur, Raymond, Hopkins & Fine, 2009; MacInnes, Priester, Eisengerich & Iacobucci, 2010), customer commitment (Ahluwalia, 2000; Fullerton, 2003) and customer brand engagement (Sprott, Czellar & Spangenberg, 2009). The first part of this chapter will explain why brand attitude is such an important concept. Since there is already an abundance of literature on this subject, this will only be explained through a literature review. The second part of the chapter discusses how brand attitude is influenced by attributes of a smartphone app and customer characteristics. Finally, the hypotheses and the conceptual model will be presented.

2.1 Brand Attitude

Because of their limited cognitive abilities, consumers form attitudes. These attitudes serve as a heuristic that reflects the affection towards an object or brand (Faircloth, Capella & Alford, 2001). Instead of using all available information to evaluate alternatives, consumers use the available attitudes. These brand attitudes can be defined as “the overall evaluation of a brand with respect to its perceived ability to meet a currently relevant motivation” (Percy & Rossiter, 1992). In marketing, attitudes are important because they influence our thoughts, feelings and behavior (Hoyer & MacInnes, 2008)

A widely accepted model regarding brand attitude is the expectancy-value model by Fishbein & Ajzen (1975). According to the expectancy-value model, brand attitude is a multiplicative function of the beliefs a consumer has concerning a non-product-related attributes and symbolic benefits, and a judgment of how good or bad those beliefs are (Rossiter & Percy, 1987; Keller, 1993). In other words: brand attitude is based on the attributes and benefits a consumer associates with a brand, and a evaluation of the positivity or negativity of these attributes and benefits.

(13)

13 set. The inert set consists of brands which are evaluated neutrally, or brands that have yet to be evaluated because there is lack of information. Finally, the inept set consists of brands that are evaluated negatively (Narayana & Markin, 1975). This is illustrated in figure 2.1.

Figure 2.1: Conceptualization of consumer behavior and brand performance (adapted from Narayana & Markin, 1975)

The brands in the evoked set are the only brands that a consumer considers when making a purchase decision. Hence, a positive brand attitude is required for a brand to be considered when a purchase decision is being made.

According to Hoyer & MacInnes (2008), five different characteristics can be used to describe an attitude towards something (e.g. an object, a person, or a brand). These characteristics are:

1. Favorability; the degree to which we (dis)like something 2. Accessibility; how easily an attitude can be remembered 3. Confidence; how strongly we hold an attitude

4. Persistence; how long an attitude lasts

5. Resistance; how difficult it is to change an attitude

2.1.1 Attitude formation and change

(14)

14 MacInnes, 2008). The level of MAO depends on many different factors (table 2.1), and determines the process of attitude formation.

Motivation: the willingness to engage in goal related activities

• Involvement • Personal relevance

• Consistency with self-concept • Values

• Needs

• Personal goals • Perceived risk

• Inconsistency with current attitudes Ability: the extent of availability of resources needed

to make an outcome happen

• Knowledge and experience • Cognitive style

• Intelligence, education and age Opportunity: factors that may impede one’s ability to

act

• Time • Distraction

• Amount, repetition and control of information

Table 2.1: Factors that influence motivation, ability and opportunity (Hoyer & MacInnes, 2008)

Motivation reflects the inner state of arousal that makes an individual engage in goal-relevant behaviors, effortful information processing and detailed decision making. A lack of ability or opportunity may restrict individuals from achieving their goals (Hoyer & MacInnes, 2008). In other words; insufficient ability and opportunity can be factors that limit the amount of effort that one can put in engaging in goal related activities, even though motivation is high enough.

(15)

15 Figure 2.2: Elaboration Likelihood Model (Petty, Cacioppo and Schumann, 1983)

The second route is called the central route, and is relevant if a consumer uses extensive thinking to process information (high MAO). In these situations, attitudes are based on a careful analysis of the available information (Hoyer & MacInnes, 2008). If the consumer perceives this information as persuasive, a positive attitude towards the brand will be the result (Petty, Cacioppo & Schumann, 1983). Attitudes that are formed through the central route are strong, and resistant to change (Hoyer & MacInnes, 2008).

Table 2.2 gives an overview of possible sources of attitude formation and change. These sources are categorized based on MAO (high/low), and on cognitions versus affection.

Cognitions (thoughts) Affect (feelings/emotions) High MAO

(Central-Route Processing)

• Direct or imagined experience • Reasoning by analogy or category • Values-driven attitudes

• Social identity-based attitude generation

• Analytical attitude construction

• Emotional processing • Affective response • Attitude toward an ad Low MAO (Peripheral- Route Processing) • Simple beliefs • Unconscious influences • The environment

• The mere exposure effect • Classical conditioning • Attitude toward an ad • Mood

Table 2.2: Sources of attitude formation and change, based on MAO (Hoyer & MacInnes, 2008)

(16)

16 attitude is strong when it is resistant to change and persistent over time (Krosnick, Boninger, Chuang, Berent & Carnot, 1993; Pomerantz, Chaiken & Tordesillas, 1995). They are more resistant to social influences because of selective cognitive processing, in which an individual tends to prefer attitude-congruent information and shows reduced susceptibility to counterattitudinal messages (Pomerantz, Chaiken & Tordesillas, 1995). In other words: people tend to hear what they want to hear.

2.2 Brand attitude as a driver of customer equity

Customer equity is a concept that combines customer value management, brand management and relationship management (Rust, Zeithaml & Lemon, 2000; Vogel, Evanschitzky & Ramaseshan, 2008). It can be defined as the total of the discounted lifetime values summed over all of the firm’s current and potential customers (Rust, Lemon & Zeithaml, 2004). In the current marketing environment, customer equity is a key strategic asset that must be measured, managed and maximized (Rust, Zeithaml & Lemon, 2000; Kumar & George, 2007; Vogel, Evanschitzky & Ramaseshan, 2008). Figure 2.3 shows a model that illustrates the process from a marketing investment to the return on this investment, and the role of customer equity in this process (Rust, Lemon & Zeithaml, 2004).

Figure 2.3: Return on Marketing Investment (Rust, Lemon & Zeithaml, 2004)

(17)

17 Rust & Zeithaml, 2001). Brand attitude can be a driver for both brand equity (Faircloth, Capella & Alford, 2001; Holehonnur, Raymond, Hopkins & Fine, 2009) and relationship equity (Aron, Aron & Smollan, 1992; Park, MacInnes, Priester, Eisingerich & Iacobucci, 2010). Figure 2.4 provides a better overview of the relation between brand attitude and the customer equity value drivers (value equity, brand equity and relationship equity), and functions as a framework for the rest of this paragraph. The dotted line indicates the moment of a purchase decision.

Figure 2.4: Relationship between brand attitude and the value drivers of customer equity

The relationship between brand attitude, brand equity and relationship equity is a loop. Each time a product or service of a certain brand is used by a consumer, the brand attitude will change based on the evaluation of the product or service.

2.2.1 Value Equity

(18)

18 Even though the app in this study is used as a marketing instrument, and not as a separate product or service, value equity is relevant. Since value equity is based on the utility of a product or service, and this utility is based on perceptions, it is possible that usage of an app can have an effect on these perceptions through the brand attitude, and thus influence the utility of other products or services.

2.2.2 Brand Equity

When compared to value equity, brand equity is more subjective and emotional (Vogel, Evanschitzky & Ramaseshan, 2008). Kotler (1991) defines a brand as “a name, term, sign, symbol, or design, or combination of them which is intended to identify the goods and services of one seller or group of sellers and to differentiate them from those of competitors”. There is no agreed definition of brand equity (Du Plessis, 2005). Generally, brand equity can be defined as the value of the marketing effects which are uniquely attributable to the brand, meaning that certain results of marketing instruments would not have occurred if a product or service would have had a different name (Keller, 1993). For example, Apple products often have lower technical specifications than other, comparable products. But still, a lot of consumers buy Apple products because they believe that they buy a product that is easy to use and has great integration with other products.

Brand equity can be viewed from three different perspectives: the customer mindset, product market outcomes and financial market outcomes (Ye & van Raaij, 2004). Measures of the customer mindset include brand attitude, brand awareness, brand associations, brand loyalty and brand attachment (Keller, 1993; Ye & van Raaij, 2004). Product market outcomes reflect the performance of a brand in the marketplace (Ye & van Raaij, 2004). Financial market outcomes measures the value of a brand as a financial asset (Ailawadi, Lehmann & Neslin, 2003; Ye & van Raaij, 2004).

Brand equity is largely based on consumers brand knowledge, which in its turn consists of two components, namely brand awareness and brand image (Keller, 1993). Brand awareness is related to the concept of a consumer being aware or unaware that a brand exists (Narayana & Markin, 1975). Since it can be assumed that an individual who is the user of an app of a certain brand is aware of the existence of the focal brand, brand awareness will not be part of this study.

(19)

19 that the price has an influence on the consumers’ perception of the quality, and therefore the value equity. Benefits are the different values that a consumer attaches to the attributes of a product or service (Keller, 1993). Park, Jaworski & MacInnis (1986) categorize benefits in three different groups, based on the underlying consumer motivations. These categories are functional benefits, symbolic benefits and experiential benefits. Both the functional benefits and the experiential benefits are usually related to product-related attributes. Functional benefits are the intrinsic advantages that are often connected to basic motivations, like physiological or safety needs (Keller, 1993). Experiential benefits are related to the feeling one gets when using a product or service, and can provide sensory pleasure, or cognitive stimulation. The remaining category (the symbolic benefits) is because they related to non-product-related attributes, like fulfilling the need for social approval and expressing oneself (Solomon, 1983).

To illustrate how attributes and benefits relate to each other, and how a single attribute can translate into multiple benefits, we present the example of a car with sporty suspension in table 2.3.

Attribute: Result: Translates into (Benefits): Sporty suspension Ability to corner at

higher speeds • Shorter braking

distance

Safety (functional benefit) Pleasure/Fun (experiential benefit)

Prestige (symbolic benefit) Table 2.3: Example of the relationship between attributes and benefits

This example shows that brand attitude has no influence on the attributes of a product, since they are physical aspects. However, it is possible that consumers associate a brand with certain benefits (e.g. BMW cars as fun to drive), or that evaluation of the attributes differs, based on the attitude towards a brand. This is supported by Faircloth, Capella & Alford (2001), whose research shows that brand equity can be created and nurtured by the attitude toward a brand, through the creation of positive brand associations.

2.2.3 Relationship equity

(20)

20 2001; Rust, Lemon & Zeithaml, 2001). If customers feel they are treated well and believe that a brand cares about them, relationship equity is high, leading to the customers trusting the brand (Vogel, Evanschitzky & Ramaseshan, 2008). As a result of this, loyalty towards a brand increases (Hennig-Thurau, Gwinner & Gremler, 2008). Another path to loyalty and increasing relationship equity lies in the theory of self-expansion and brand-self connection. According to these theories, consumers tend to incorporate brands into one’s conception of “self” , and develop an emotional connection with these brands through formation of a favorable brand attitude (Aron, Aron & Smollan, 1992). Eventually, they will develop a sense of oneness with the brand, and may show sadness and anxiety when separated from “their” brand, or pride from displaying the brand-self connection (Park, MacInnes, Priester, Eisingerich & Iacobucci, 2010). A good example of this is Harley Davidson and H.O.G. (Harley Owners Group). Consumers can develop a connection to a brand on a identity basis, of because the brand is relevant for their personal goals, concerns or projects. (Mittal, 2006).

2.3 Brand attitude as a driver of customer commitment

Commitment is a central construct in relationship marketing, that can be defined as a desire to maintain a valued relationship (Moorman, Zaltman & Deshpande, 1992; Fullerton, 2003). In addition to this definition, the constraints that keep the relationship intact have to be considered as well (Bendapudi & Berry, 1997). Commitment is one of the building blocks of customer relationship, and a important driver of customer loyalty and relational intentions (Fullerton, 2003; Bansal, Irving & Taylor, 2004). Besides commitment, one of the most prominent loyalty drivers in marketing literature is satisfaction (Gustafsson, Johnson & Roos, 2005). The main difference between commitment and satisfaction is that commitment is looking at the customer’s intentions (forward looking), and satisfaction is looking at the past performance (backward looking). Because of this, commitment is more suitable for predicting behavioral intentions (Gustafsson, Johnson & Roos, 2005).

Earlier study on customer commitment make a distinction between affective commitment and calculative commitment (Johnson, Hermann & Huber, 2006). Affective commitment is based on emotional factors, like psychological attachment to a brand (Garbarino & Johnson, 1999), while calculative commitment is based on rational aspects (Johnson, Hermann & Huber, 2006).

2.3.1 Affective commitment

(21)

21 affective commitment when they like or trust a brand (Fullerton, 2003). Within the customer commitment construct, affective commitment represents the attitudinal component (Ahluwalia, 2000; Fullerton, 2003). Because of this, affective commitment determines the quality of the relationship between a customer and a brand (Moorman, Zaltman & Deshpane, 1992; Bendapudi & Berry, 1997). Affective commitment has a significant positive effect on advocacy intentions, and a significant negative effect on switching intentions. Furthermore, affective commitment is a stronger driver of customer loyalty than calculative commitment (Fullerton, 2003). Because of the strong relationship with brand attitude, it is more relevant for this study than calculative commitment. 2.3.2 Calculative commitment

Calculative commitment, or continuance commitment, is built on switching costs, dependence and lack of choice (Fullerton, 2003; Iglesias, Singh & Batista-Foguet, 2011). A consumer may be dependent on a relationship with a brand because the outcomes of the relationship exceeds the consumers own performance standard, or the outcomes are perceived as better than the available alternatives (Anderson & Narus, 1990). This can cause customers to be “loyal” to a brand, even when the customer satisfaction is low (Gustafsson, Johnson & Roos, 2005). Therefore, instead of determining the quality of the relationship with a brand, calculative commitment determines the stability of the relationship (Bendapudi & Berry, 1997). However, calculative commitment can lead to affective commitment though the means of self-motivation motives (Bobocel & Meyer, 1994). The dependent party, the consumer in this case, may try to justify the commitment to themselves by showing dedication to the relationship with the party that they are dependent on. This improves the quality of the relationship, meaning that the affective commitment of the consumer to the brand increases.

2.4 Brand attitude as a driver of customer brand engagement

Engagement is a concept that has been applied in many different academic disciplines, like psychology, organizational behavior and sociology (Hollebeek, 2011). Because of this, the definition of engagement is content-specific (Little & Little, 2006). What these definitions have in common, is that engagement is viewed as “an individual-specific, motivational, and context-dependent variable emerging from two-way interactions between relevant engagement subject(s) and object(s)” (Hollebeek, 2011).

(22)

22 commitment. Bowden (2009) developed a framework which shows the relationship between the relevant constructs (figure 2.5). Earlier study showed that consumers engage with brands that are important as a part of their self-concept. This brand engagement in self-concept (BESC) has a positive, significant relationship with brand attitude (Sprott, Czellar & Spangenberg, 2009)

Figure 2.5: Framework for the Process of Engagement (Bowden, 2009)

Bowden’s framework shows that a positive evaluation can lead to affective commitment for new customers. This happens through the positive change in attitude that takes place when customer delight occurs. Customer delight is defined as “the combination of high pleasure via joy and elation, combined with unexpected levels of arousal or surprise” (Bowden, 2009). According to the expectancy-value theory by Fishbein & Ajzen (1975), this meeting or exceeding of expectations can lead to positive change in brand attitude.

(23)

23

2.5 Attributes of an app

Since there is very little literature about the attributes of a smartphone app, most of the theory in this section is based on other literature, for example on website quality. Website quality can be a powerful signal to assess product quality through the formation of associations and attitudes (Dawar & Parker, 1994; Kirmani & Rao, 2000; Wells, Valacich & Hess, 2011). There is reason that this holds for apps as well. Davis (1986) proposed a model based on the theory of reasoned action (TRA) and theory of planned behavior (TBA) (Davis, 1989; Venkatesh, 1999). This Technology Acceptance Model (TAM) explains an individual’s intention to use an information system, based on its perceived usefulness and perceived ease of use (Davis, 1986; Mathieson, 1991; Wu & Wang, 2004; Bruner & Kumar, 2005). Another attribute that is relevant for this study is enjoyment, since it is recognized to be a robust construct to capture ones affective reactions towards an environment (Parboteeah, 2009). Finally, earlier study showed that the visual design, and the level in which it appeals to the user, has a significant effect on the perceived usefulness, the ease of use and the enjoyment (Cyr, Head & Ivanov, 2006).

2.5.1 Usefulness

The perceived usefulness in the TAM deals with the degree to which an individual believes that a technology or system will help to complete a certain task (Davis, 1986; Bruner & Kumar, 2005). Therefore, the perceived usefulness of an app is task-dependent; it can be regarded as highly useful for a certain task, but the same individual can view the same app as being not useful at all for another task. Since people have a greater intention to engage in tasks toward which they have a more positive attitude (Childers, Carr, Peck & Carson, 2001), developing an app that is suited for these tasks may increase overall usage and perceived usefulness of the app.

2.5.2 Ease of Use

(24)

24 Because the optimal level of stimulation differs between individuals, finding the optimal overall level of stimulation that an app offers can be difficult.

2.5.3 Enjoyment

According to Bruner & Kumar (2005), hedonic components, such as enjoyment, play an important role in the adoption of new technology in a consumer setting. A higher level of enjoyment that is associated with the use of a technology lead to a more favorable attitude towards said technology (Sheppard, Hartwick & Warshaw, 1988). For mobile devices, this role is stronger than utilitarian components, such as the perceived usefulness (Bruner & Kumar, 2005). An explanation for this may lie in the relative novelty of mobile devices in comparison to a desktop computer. In the field of communication, enjoyment has been recognized as a construct that measures the satisfaction of three intrinsic needs: autonomy, competence and relatedness (Tamborini, Bowman, Eden, Grizzard & Organ, 2010). So instead of just measuring “pleasure-seeking”, enjoyment is about both hedonic and non-hedonic intrinsic needs (Tamborini, Bowman, Eden, Grizzard & Organ, 2010; Tamborini, Grizzard, Bowman, Reinecke, Lewis & Eden, 2011).

2.5.4 Appeal of Visual Design

(25)

25

2.6 Customer Characteristics

Customer preferences vary across different types of customers. For example, earlier study shows that one’s personality traits influence the degree in which one views certain types of apps as important (Lane & Manner, 2012). It is possible that different types of consumers value attributes of an app in a different way as well. Because of this, this study tries to discover if a segmentation with regard to customer preferences can be made, based on the characteristics of the customers.

Segmentation variables can be classified in two ways (Frank, Massy & Wind, 1972; Leeflang, 2003). The first classification splits the variables in general variables and situational variables. General variables can be used regardless of the type of product that the segmentation is for, while situational variables depend on the situation (which the name already implies). The second classification splits the variables in objective variables and subjective variables. This classification is based on the measurement scale. For subjective variables, the researcher is able to choose the measurement scale. For objective variables, the measurement scale is fixed (Leeflang, 2003). An overview of this classification, along with some examples, can be found in table 2.4.

General Variables Situational Variables Objective

variables

• Cultural, demographic & socio-economic variables • Adoption categories • Frequency of use • Brand loyalty • Buying situation • Purpose • Customer values Subjective variables • Personality • Values • Lifestyle • Lifestyle, attitude • Perception, preference • Buying intention, appreciation • Domain-specific values Table 2.4: Classification of segmentation variables (Frank, Massy & Wind, 1972)

(26)

26

2.7 Conceptual Model and Hypotheses

Literature shows that brand attitude is a key link in the relationship between a brand and the consumer. The favorability of the brand attitude is a driver of customer equity, customer commitment and customer brand engagement. Since website quality can be a powerful signal to assess product quality through the formation of associations and attitudes (Dawar & Parker, 1994; Kirmani & Rao, 2000; Wells, Valacich & Hess, 2011), there is reason that this holds for apps as well. The five characteristics that can be used to describe an attitude are: favorability, accessibility, confidence, persistence and resistance (Hoyer & MacInnes, 2008). This results in the following hypotheses, based on the theory on the characteristics of an attitude (paragraph 2.1):

H1: There is a positive relationship between the perceived quality of an app and the favorability of the brand attitude.

H2: There is a positive relationship between the perceived quality of an app and the accessibility of the brand attitude.

H3: There is a positive relationship between the perceived quality of an app and the confidence of the brand attitude.

H4: There is a positive relationship between the perceived quality of an app and the persistence of the brand attitude.

H5: There is a positive relationship between the perceived quality of an app and the resistance of the brand attitude.

The following hypotheses are based on the theory on the attributes of an app (paragraph 2.5): H6: There is a positive relationship between the level of usefulness and the perceived quality of an app.

H7: There is a positive relationship between the level of ease of use and the perceived quality of an app.

H8: There is a positive relationship between the level of enjoyment and the perceived quality of an app.

(27)

27 The conceptual model (figure 2.6) gives an overview of the hypotheses, as well as the influence of the customer characteristics.

Figure 2.6: Conceptual Model

(28)

28

3 Research Design

This chapter deals with the methods that will be used to conduct the research. In the first paragraph, the research methods will be explained. Secondly, the measurement of the variables and the data collection will be addressed.

3.1 Research methods

In order to answer the research questions, two multivariate techniques will be applied: conjoint analysis and latent class analysis. Sawtooth Software SSI Web will be used to conduct the conjoint analysis. For the latent class analysis, Sawtooth Software Latent Class Segmentation Module will be used. The reason that this module will be used is that it can handle the output data of the conjoint analysis without the need to convert it to another format.

3.1.1 Attributes concerning app quality (conjoint analysis)

Conjoint analysis is based on the assumption that consumers evaluate the total value of an object or concept, such as a product, service, or brand, by combining the separate amounts of value that each of the attributes, both tangible and intangible, provides (Hair, Black, Babin & Anderson, 2010). Conjoint analysis provides the researcher with a number of options. These options include defining the optimum combination of attribute levels for an object, or isolating certain groups of customers who place differing importance on the attributes (Hair, Black, Babin & Anderson, 2010).

There are three different conjoint methodologies. The method that should be used depends on the basic characteristics of the proposed research (Hair, Black, Babin & Anderson, 2010). An overview of this can be found in table 3.1.

Characteristic Traditional Conjoint Adaptive/Hybrid Conjoint

Choice-Based Conjoint Upper Limit on

Number of Attributes

9 30 6

Level of Analysis Individual Individual Aggregate or Individual Model Form Additive Additive Additive + Interaction Choice Task Evaluating Full-Profiles

One at a Time

Rating Profile Containing Subsets of

Attributes

Choice Between Sets of Profiles Data Collection Format Any Format Generally

Computer-Based

(29)

29 Since the number of attributes (four) in the proposed design is within the limits of a choice-based conjoint (CBC) approach, this will be the preferred method. This is because it can be estimated at both an aggregate or an individual level, and moreover, it allows inclusion of interaction effects.

Experimental Design

When carrying out a CBC, two different types of designs can be used. The first type is an orthogonal, or fixed, design. In this type of design, all respondents get to see the same version of the questionnaire. The biggest advantage of an orthogonal design is that it offers great efficiency in the measurement of main effects and the pre-specified interaction effects (Sawtooth Software, 2008). The second type of design is a random design, in which each respondent gets to see a unique version of the survey. For designs where attributes have a different number of levels (asymmetric designs), random designs can be more efficient. Another advantage of a random design is that it allows for measurement of all interaction effects, even those which are not specified in advance. Another advantage of using a random design is the reduced impact of order and learning effects (Sawtooth Software, 2012).

Since the loss of efficiency in a random design is quite small (usually 5-10%), random designs are usually favored because of the ease of implementation and the robust characteristics (Sawtooth Software, 2012). Because of this, a random design will be used in this study. Sawtooth Web SSI offers different random design options, which are based on some key principles.

These principles are:

Minimal Overlap: Each attribute level is shown as few times possible in a single task. If an attribute’s number of levels is equal to the number of product concepts in a task, each level is shown exactly once

Level Balance: Each level of an attribute is shown approximately an equal number of times Orthogonality: Attribute levels are chosen independently of other attribute levels, so that

(30)

30 Based on these principles, SSI Web offers four different random design strategies. These can be found in table 3.2, along with their advantages and disadvantages.

Strategy Advantages Disadvantages

Complete Enumeration

• Produces the most nearly orthogonal design for each respondent

• Produces minimal overlap

• May require a very large number of concepts to be evaluated

• Strain on PC may result in long waiting time in between questions

Shortcut Method Requires less processing time than complete enumeration

• Only considers one attribute at a time

Random Method Requires less processing time than complete enumeration

• Only recommended if the goal is the study of interaction effects Balanced

Overlap Method

• Middling strategy between complete enumeration and random method

• Better precision of estimates of interaction terms

• Slightly less efficient with respect to main effects

Table 3.2: Random design strategies (Sawtooth Software, 2005)

In general, Complete Enumeration appears to be the most solid method. It keeps track of the co-occurrence of all attribute levels that are shown to the respondents, resulting in a minimal amount of overlap (Sawtooth Software, 2012). Unfortunately, both of its disadvantages can cause respondents to get bored or irritated, and quit before the entire survey is completed. The Shortcut Method only considers one attribute at a time with respect to the amount of overlap. When two or more attribute levels are tied, a random selection is made. This can result in a higher amount of overlap. The Random Method is only recommended for the study of interaction effects, which is not the case for this study. The Balanced Overlap Method appears to be best all-round method, since it is a middling strategy between complete enumeration and the random method. It allows for an acceptable level of overlap, while limiting the number of concepts to be evaluated. Therefore, this method will be used.

Presentation method

(31)

31 profiles at a time with a limited number of attributes. The last method is the trade-off approach, which compares attributes two at a time by ranking all combinations of levels (Hair, Black, Babin & Anderson, 2010).

The method that will be used in this study is the full-profile method. This is because this method offers the greatest realism. Even greater realism could be achieved by using screenshots of apps, and present them in a market setting, but some of the attributes are impossible to convey by using screenshots (e.g. ease of use). One of the limitations of the full-profile method is information overload. Since all attributes are presented simultaneously, the possibility of information overload exists (Hair, Black, Babin & Anderson, 2010). Because of the limited number of attributes (four), it is expected that this poses no problem for this study.

None option

In the real world, consumers can choose to download none of the presented apps. This can be mimicked by adding a none option to the CBC. Doing this will result in more reliable estimation of part-worths.

Sawtooth Web SSI offers the possibility to add a dual-response none option. This means presenting the none option as a second-stage question. This increases the tendency of the respondents to choose the none option, without causing any loss of information, because respondents are asked to choose among the presented profiles beforehand. The benefits of a dual none option outweigh any disadvantages, such as adding more time to each choice task (Sawtooth Software, 2012). Because of this, a dual-response none option will be included in the survey.

An example of a survey question with a conjoint analysis random task can be found in Appendix A.

3.1.2 Allowing for heterogeneity (latent class analysis)

In order to analyze if there any segmentation within the respondents in the data set, a Latent Class Cluster Analysis will be performed. A Latent Class Analysis has a number of advantages over traditional types of cluster analysis. According to Magidson & Vermunt (2004), these advantages include:

(32)

32 • Variables may be continuous, categorical (nominal or ordinal), or counts or any combination

of these,

• Demographics and other covariates can be used for cluster description.

In this study, the demographics and other variables that deal with customer heterogeneity, such as phone usage variables, will be used for cluster description by adding them as inactive covariates. The optimal number of clusters will be based on the information criteria (AIC(3), BIC, CAIC), as well as the Approximate Weight of Evidence (AWE).

After determining the optimal number of segments, an ANalysis Of VAriance (ANOVA) will be used to check if the segmentation variables are significant

3.1.3 Relationship between app quality and brand attitude (regression)

A common method of assessing an attitude in a survey setting is by using a meta-attitudinal measure (Bassili, 1996; Miller & Peterson, 2004). This basically means asking the respondents to assess his or her own attitudes. In order to test hypotheses H1 through H5, six questions in which the respondent has to indicate to what degree (s)he agrees with a proposition will be included in the survey. The data that will be collected through these questions will be used in a number of linear regression to test is there is a significant effect on the five different characteristics of brand attitude. This will be done for both the total population and the individual segments.

3.2 Variables and data collection

According to Hair, Black, Babin & Anderson (2010), a sample size of 200 provides an acceptable margin of error. When the population is being segmented, a sample size of 50 respondents per segment if sufficient to see how the preferences of the different segments might vary (Hair, Black, Babin & Anderson, 2010). Therefore, the population should meet the following two requirements:

• Total sample size should be larger than 200

• More than 50 respondents should be included for each group

(33)

33 3.2.1 CBC

For the CBC analysis, a symmetric design will be used with four attributes. Each of the attributes will have three levels (low, medium and high). This will make the choice tasks easier to comprehend for the respondents, while it still allows to check for nonlinearity. Furthermore, respondents tend to evaluate an attribute as “more important” as the number of levels increases (Hair, Black, Babin & Anderson, 2010). This problem can be prevented by using the same number of levels for all attributes. Table 3.3 shows an overview of all attributes and attribute levels.

Attribute Attribute Levels Attribute Attribute Levels Usefullness Low • Medium • High Enjoyment Low • Medium • High Ease of Use Low

• Medium • High Appeal of Visual Design • Low • Medium • High Table 3.3: Attributes and attribute levels

Generally, it is recommended to include between twelve and eighteen CBC choice tasks in a survey (Sawtooth Software, 2008). To increase the response rate and the percentage of completed surveys, twelve random choice tasks will be included in the CBC design. Furthermore, three fixed choice tasks will be added to the survey. These will be used to measure the validity of the CBC design.

(34)

34 3.2.2 Describing the variables

Relationship between app quality and brand attitude

Like mentioned in paragraph 3.1.3, the survey will include six questions that will be used to determine if the quality of an app has a significant effect on the five characteristics of brand attitude (favorability, accessibility, confidence, persistence and resistance). Answers to these questions will be on a five-point Likert scale. An overview of all meta-attitudinal questions can be found in table 3.4. Respondents will be asked to answer these questions with the app (of a brand) that they most recently downloaded in mind.

Completely disagree

Disagree Neutral Agree Completely agree

The app is of high quality O O O O O

The app made my attitude towards the brand more positive

O O O O O

The app made it easier for me to remember my attitude towards the brand

O O O O O

The app increased my trust in my attitude towards the brand

O O O O O

The app made my attitude towards the brand more durable

O O O O O

The app made my attitude towards the brand harder to change

O O O O O

Table 3.4: Meta-attitudinal survey questions

(35)

35 Customer characteristics/segmentation variables (latent class analysis)

Table 3.5 shows the different segmentation variables that will be used, as well as the levels for each variable. The classification that is being used by the Dutch municipalities was used for the different levels of education. Variable Levels Demographic variables Age - Gender Male • Female Socio-economic variables

Education Low (No education, lagere school, lager beroepsonderwijs

• Medium (VMBO, HAVO, VWO, MBO) • High (HBO, University)

Usage variables Average number of hours of smartphone usage per day

• Less than one hour

• Between one and two hours • Between two and three hours • More than three hours Operating system iOS (Apple)

(36)

36

4 Results

In this chapter the results of the analyses will be described and discussed. First, the descriptives for the total population will be presented. Secondly, the model validation will be addressed, as well as the model estimation for the aggregate model. Hereafter, the latent class analysis will be discussed in order to determine the optimal number of segments. Next, each class will be described in terms of its descriptives and part-worths. And lastly, the results of the regression analyses will be described in order to answer hypotheses H1 through H5.

4.1 Descriptives

In total, 233 respondents filled in the entire survey (N=233). An overview of the age of the respondents is depicted in figure 4.1 (mean=38,39, standard deviation=11,42).

Figure 4.1: Age of respondents (N=233)

Table 4.1 gives an overview of the frequencies and percentages of the other demographic, socio-economic and usage variables, and their levels.

0 2 4 6 8 10 12 14 16 18 13 23 28 33 38 43 48 53 58 64 Frequency

Age (in years)

(37)

37

Variable Level Frequency Percentage

Sex Male 153 65,7%

Female 80 34,3%

Education Low (No education, lagere school, lager beroepsonderwijs

0 -

Medium (VMBO, HAVO, VWO, MBO)

53 22,7% High (HBO, University) 180 77,3% Average number of hours of

smartphone usage per day

Less than one hour 38 16,3% Between one and two hours 95 40,8% Between two and three hours 56 24,0% More than three hours 44 18,9%

Operating system iOS (Apple) 115 49,4%

Android 82 35,2%

Blackberry 21 9,0%

Other 15 6,4%

Table 4.1: Descriptives for total population (N=233)

4.2 Aggregate Model Estimation and Validation

In order to calculate the part-worths and test the validity of the aggregate model, the Sawtooth Hierarchical Bayes Estimation Module (CBC/HB) was used. The following (standard) settings were used for the estimation; 10.000 iterations before using results (in order to reach convergence), and 10.000 draws to be used for each respondent (to estimate the parameters). This results in a total of 20.000 iterations.

Estimation of the aggregate model produces the parameters in table 4.2.

Variable Part Worth Variable Part Worth

Usefulness Enjoyment

Low -3,61 Low -0,86

Medium 0,53 Medium 0,25

High 3,08 High 0,61

Ease of Use Appeal of Visual

Design

Low -2,76 Low -0,096

Medium 0,67 Medium 0,29

High 2,09 High 0,67

NONE=2,18

Table 4.2: Part-worths for total population

(38)

38 Figure 4.2: Preference functions per attribute

It becomes clear that the preference function of usefulness, ease of use and enjoyment are part-worth models. Appeal of visual design has a vector (or linear) model. Table 4.3 shows the relative importance of attributes for the total population. This can be calculated by dividing the range between the minimum and the maximum part-worth for an attribute by the sum of all ranges.

Part-worth Min Part-worth Max Range Importance

Usefulness -3,61 3,08 6,690 48,56% Ease of Use -2,76 2,09 4,850 35,21% Enjoyment -0,86 0,61 1,470 10,67% Appeal of Visual Design -0,096 0,67 0,766 5,56% SUM=13,776 Table 4.3: Relative importance of attributes for total population

This shows that the usefulness of an app is the most important attributes (48,56%), followed by the ease of use (35,21%). The enjoyment that an app provides and the appeal of the visual design of an app are less important (10,67% and 5,56% respectively).

The output of the CBC/HB estimation includes four different statistics that indicate the goodness of fit. These are:

1. Percent Certainty (Pct Cert) 2. Root Likelihood (RLH) 3. Average Variance

4. Parameter Root Mean Square (RMS) -4 -3 -2 -1 0 1 2 3 4

Low Medium High

Usefulness

Ease of Use

Enjoyment

(39)

39 Since the Average Variance and the Parameter RMS are indirect indicators of fit, they are less suitable (Sawtooth, 2009). Because of this, only the Percent Certainty and the Root Likelihood are used to measure the goodness of fit of the aggregate model in this study.

Percent Certainty indicates how good a solution is in comparison to chance (a value of 0) and a perfect solution (a value of 1), and is calculated based on the Log Likelihood. The aggregate model has an average Percent Certainty of 0,604, indicates that the log likelihood is 60,4% of the way between the value that would be expected by chance and the value for a perfect fit (Sawtooth, 2009).

Like the Percent Certainty, the Root Likelihood is derived from the likelihood of the data. Without any information about the part-worths, the RLH would be 1/k, where k is the number of alternatives per choice task (three). This would result in an expected RLH of 0,333 for the chance model. If the fit were perfect, the RLH would be 1 (Sawtooth, 2009). The actual RLH for the aggregate model is 0,593. This means that the aggregate model is 0,593/0,333=1,781 times, or 78%, better than the chance model.

4.3 Latent Class Analysis

The data that the conjoint choice tasks provided were used in the Sawtooth Latent Class module. Five replications were estimated for the range of 2-7 classes, all with a random starting seed. To determine the optimal number of classes, four different measures of fit will be used. An overview of these measures of fit for the best replication for each number of classes can be found in table 4.4.

Percent Certainty CAIC Chi Square Relative Chi Square 2 Classes 32,75425 6625,93904 3142,25692 165,38194 3 Classes 35,61435 6443,53876 3416,63810 117,81511 4 Classes 36,96656 6405,79631 3546,36144 90,93234 5 Classes 38,28300 6371,48536 3672,65328 74,95211 6 Classes 39,19403 6376,06705 3760,05248 63,72970 7 Classes 39,94186 6396,30531 3831,79512 55,53326 Table 4.4: Measures of Fit for each solution

Like in the CBC/HB estimation, Percent Certainty indicates how good a solution is in comparison to chance (a value of 0) and a perfect solution (a value of 1). Since Percent Certainty generally increases as more classes are included, it is not very useful to base the acceptable number of segments on (Sawtooth Software, 2004). However, it can provide an idea to what extent each of the solutions fits the data.

(40)

40 CAIC = -2 Log Likelihood + (nk + k - 1) x (ln N +1)

where k is the number of groups, n is the number of independent parameters estimated per group, and N is the total number of choice tasks in the data set (Sawtooth, 2004). Lower values are better. Table 4.4 shows that a five class solution has the lowest CAIC.

Chi Square can be used to test whether a solution fits significantly better than the null solution, which is almost always true (Sawtooth, 2004). Because Chi Square tends to increase as the numbers of solution increases, it is not very useful to decide what number of segments to use.

The Relative Chi Square is Chi Square divided by the number of parameters estimated (nk + k - 1), and can be a useful way to decide on the number of segments (Sawtooth, 2004). Like most of the other measures of fit, bigger values are better.

Since both the Percent Certainty and the Chi Square are not very useful to base the number of segments on, they will not be used for this decision. Since the CAIC has the lowest value in case of a five class solution, and the Relative Chi Square is highest for a two class solution, everything from a two class solution to a five class solution has been further looked into. These results can be found in Appendix B.

When looking at the segment sizes for the different solutions, as well as the relative attribute importance, the four class solution appears to be the best. In this solution, the groups from the five class solution that places the highest importance on enjoyment (class 1) and visual design (class 2) are largely combined. This can be seen below (table 4.5). More on this later.

1 2 3 4 5 Total 1 0 6 14 0 4 24 2 2 1 0 22 76 101 3 27 14 0 2 0 43 4 2 3 0 60 0 65 Total 31 24 14 84 80 233

Table 4.5: Tabulation of 4 group vs. 5 group solution

4.4 Descriptives and Part-worths for each class

(41)

41 classes, the respondents were classified based on the probabilities. Each respondent was assigned to the class with the highest probability for said respondent.

Table 4.6 gives an overview of the part-worts for each class. Variable/Level Part-worths

Usefulness Class 1 Class 2 Class 3 Class 4 Low -0,77041 -2,60241 -0,96112 -2,28143 Medium 0,18902 0,25656 0,30953 0,40321 High 0,58138 2,34585 0,65159 1,87822 Ease of Use Low -0,78366 -1,92778 -0,84663 -1,59789 Medium 0,23562 0,35652 0,23296 0,43486 High 0,54804 1,57126 0,61367 1,16303 Enjoyment Low -0,10695 -0,32056 -1,44281 -0,17721 Medium 0,28052 0,12456 0,19112 0,04213 High -0,17357 0,19600 1,25169 0,13508 Appeal of Visual Design

Low -0,16205 -0,43187 -0,69318 -0,78983 Medium -0,04741 0,08211 0,08516 0,40426 High 0,20946 0,34975 0,60802 0,38557 NONE 2,52413 1,91158 0,51666 -0,60184 Table 4.6: Part-worths for each class

A visual representation of the preference functions for all attributes per class are depicted in figure 4.3.

(42)

42 Table 4.7 shows the relative importance of the attributes for each class. For the exact calculations, see Appendix C.

Class 1 Class 2 Class 3 Class 4

Usefulness 38,52% 50,77% 22,81% 49,36% Ease of Use 37,95% 35,90% 20,66% 32,76% Enjoyment 12,94% 5,30% 38,12% 3,71% Appeal of Visual Design 10,59% 8,02% 18,41% 14,17%

Table 4.7: Relative importance of attributes for each class

A visual representation of the relative importance per class is shown in figure 4.4, as well as the relative importance for the total population for comparison.

Figure 4.4: Relative importance of attributes per class and for total population

When looking at the relative importance, class one, two and four appear to be quite similar. However, the part-worth for the none-option differs greatly. The part-worth for the none-option for class one (2,52413) is higher than the part-worth for the optimal app for this class (1,6194). Meaning that respondents that belong to class one prefer not to download any app at all. Class four has a negative part-worth for the none-option, meaning that this respondents in this class are eager to download an app. Class two shows the most resemblance with the total population, regarding the relative importance of the attributes and the part-worth of the none-option.

0% 20% 40% 60% 80% 100%

Appeal of Visual Design Enjoyment

(43)

43 4.4.1 Class 1

Figure 4.5 shows the age of the respondents in class one (Mean=34,00 & Standard deviation=11,959).

Figure 4.5: Age of respondents in class 1 (N=23)

Table 4.8 gives an overview of the descriptives (frequencies and percentages) for class one.

Variable Level Frequency Percentage

Sex Male 10 43,48%

Female 13 56,52%

Education Medium (VMBO, HAVO,

VWO, MBO)

10 43,48% High (HBO, University) 13 56,52% Average number of hours of

smartphone usage per day

Less than one hour 6 26,09% Between one and two

hours

5 21,74% Between two and three

hours

8 34,78% More than three hours 4 17,39%

Operating system iOS (Apple) 6 26,09%

Android 8 34,78%

Blackberry 6 26,09%

Other 3 13,04%

Table 4.8: Descriptives for class 1 (N=23)

When looking at the descriptives of class one, there a couple of things stand out. First, the mean age of the respondents is lower than the mean age of the total population. Second, class one consists of relatively many female respondents (56,52% versus 34,61% of the total population). Thirdly, the percentage of respondents with a medium education is high in comparison with the total population (43,48% versus 22,64%). And finally, the number of respondents running Blackberry OS or an unknown operating system is relatively large, at the expense of respondents running iOS.

(44)

44 4.4.2 Class 2

Figure 4.5 shows the age of the respondents in class two (Mean=39,12 & Standard deviation=11,267).

Figure 4.5: Age of respondents in class 2 (N=107)

Table 4.9 gives an overview of the descriptives (frequencies and percentages) for class two.

Variable Level Frequency Percentage

Sex Male 69 64,49%

Female 38 35,51%

Education Medium (VMBO, HAVO,

VWO, MBO)

20 18,69% High (HBO, University) 87 81,31% Average number of hours of

smartphone usage per day

Less than one hour 17 15,89% Between one and two

hours

54 50,47% Between two and three

hours

22 20,56% More than three hours 14 13,08%

Operating system iOS (Apple) 56 52,34%

Android 33 30,84%

Blackberry 11 10,28%

Other 7 6,54%

Table 4.9: Descriptives for class 2 (N=107)

The respondents in this class do not show any big differences in comparison with the total population. The average number of hours of smartphone usage per day in class two is slightly lower than in the total population.

(45)

45 4.4.3 Class 3

Figure 4.6 shows the age of the respondents in class three (Mean=34,00 & Standard deviation=11,066).

Figure 4.6: Age of respondents in class 3 (N=41)

Table 4.10 gives an overview of the descriptives (frequencies and percentages) for class three.

Variable Level Frequency Percentage

Sex Male 27 65,85%

Female 14 34,15%

Education Medium (VMBO, HAVO,

VWO, MBO)

12 29,27% High (HBO, University) 29 70,73% Average number of hours of

smartphone usage per day

Less than one hour 5 12,20% Between one and two

hours

17 41,46% Between two and three

hours

8 19,51% More than three hours 11 26,83%

Operating system iOS (Apple) 15 36,59%

Android 22 53,66%

Blackberry 3 7,32%

Other 1 2,44%

Table 4.10: Descriptives for class 3 (N=41)

This shows that the typical member of class three is young, in comparison with the total population. The education level of the respondents in class three is slightly lower than the total population (29,27% with a medium leveled education versus 22,64% of the total population). The average number of hours of smartphone usage per day is higher, with 26,83% of the respondents in class three indicating that they use their smartphone for more than three hours on a daily basis.

(46)

46 Furthermore, class three consists of relatively many respondents that are running the Android operating system (53,66%).

4.4.4 Class 4

Figure 4.7 shows the age of the respondents in class four (Mean=41,68 & Standard deviation=10,947).

Figure 4.7: Age of respondents in class 4 (N=62)

Table 4.11 gives an overview of the descriptives (frequencies and percentages) for class four.

Variable Level Frequency Percentage

Sex Male 47 75,81%

Female 15 24,19%

Education Medium (VMBO, HAVO,

VWO, MBO)

11 17,74% High (HBO, University) 51 82,26% Average number of hours of

smartphone usage per day

Less than one hour 10 16,13% Between one and two

hours

19 30,65% Between two and three

hours

18 29,03% More than three hours 15 24,19%

Operating system iOS (Apple) 38 61,29%

Android 19 30,65%

Blackberry 1 1,61%

Other 4 6,45%

Table 4.11: Descriptives for class 4 (N=62)

The descriptives show that class four consists of a relatively large percentage of male respondents (75,81% versus 65,38% of the total population), who show an average number hours of smartphone usage per day which is higher than the total population. Furthermore, a relatively large part of

Referenties

GERELATEERDE DOCUMENTEN

The purpose of this study was to investigate the moderating effect of industry regulations on the relationship between corporate social performance (CSP) and corporate

In the COPE-active group as well as in the control group, none of the patient characteristics measured at baseline were correlated with change in daily physical activity over 7

This thesis examines the anticipated and real social impact of Dakpark in Rotterdam within the context of Nature-Based Solutions and their effects on people and the

Para ver de qué manera ‘Los niños de la furia’ se contrapone al discurso de seguridad de Felipe Calderón, se emplea un acercamiento discursivo de los estereotipos:

On these systems estimation and control using periodic sampling is usually not an op- tion due to the large worst-case execution times of the tasks. Furthermore, the proposed

In the concern of friction reduction, the pillar (Z003) texture has the advantage over Hilbert curve and grooved channel textures in decreasing the friction force under the

Table 3: Top URLs and Hashtags in User Groups By URL Bias Liberal URL Users Conservative URL Users Neutral URL Users.. Top

In this paper, our main contribution is that we present combinations of measurements for error modeling that can be used to estimate the quality of arbitrary GNSS receivers