The Role of Quality in the Freemium Business Model and the Moderating
Effects of Hedonic Values and Frugality: An HB CBC-study on Music
Streaming and Cloud Storage Services
by Nicky Strijker University of Groningen Faculty of Economics and BusinessMSc Marketing Intelligence June 16, 2019 Heymanslaan 56B 9714 GN Groningen n.m.strijker@student.rug.nl S2305380
First supervisor Second supervisor
Prof. dr. H. Risselada prof. dr. P.S. van Eck
h.risselada@rug.nl p.s.van.eck@rug.nl
Nettelbosje 2 Nettelbosje 2
9747 AE Groningen 9747 AE Groningen
1
Preface
This thesis represents my final piece of work for the Master Marketing Intelligence at the University of Groningen. The main topic of this article is the freemium business model, which is becoming increasingly common in today’s society. The growing presence of services based on this business model (e.g. Spotify) in consumer households made it a very interesting topic to study. Before writing this article I wondered what it is that drives consumers from using the free version of a service to a paid (premium) version of a service. In this article I particularly focused on improving the understanding of the role of quality in the freemium business model. I hope the insights of this article inspire other researchers or students to continue research in this area.
I would like to express my gratitude towards H. Risselada, my first supervisor, for his valuable feedback and thoughtful criticism throughout the process, as it inspired me to remain critical. Additionally, I would like to thank my thesis group for the insightful discussions, and my friends and family for their support throughout the process.
I hope you enjoy reading this thesis.
Nicky Strijker
2
Abstract
This study focuses on the role of service quality in the freemium business model. Based on previous freemium and quality literature, the study argues the quality of a freemium service to positively influence consumer preference (utility) for premium subscriptions. With a between-group experimental design the influence of quality was tested in two service categories, music streaming and cloud storage. Choice-based conjoint analysis was used to measure consumer preference for the main attribute levels of the freemium services. The study found the quality of a service to influence the conversion of free users to premium users (free-to-fee conversion) both negatively and positively, however its influence is found to differ across freemium categories. The results suggest an ideal point in preferred service quality to exist, as the highest quality attribute level does not always lead to the highest utility for the premium subscription. In addition, frugality is found to moderate the relation of service quality on consumer utility for a premium subscription within freemium music streaming services, but not in cloud services. No support is found for a possible moderating role of hedonic values. Nonetheless, clustering consumer preferences based on HB-means show both hedonic values and frugality to be of high influence on consumer preference for low and high quality attribute levels in both freemium service categories.
Keywords
freemium, service quality, hedonic values, frugality, CBC-analysis, hierarchical bayes
3
Table of Content
1. Introduction 5
2. Literature Review 7
2.1 Research Model 7
2.2 The Freemium Strategy 8
2.2.1 Freemium in the market 8
2.2.2 Free version 10
2.2.3 Subscription pricing 11
2.3 Quality 11
2.3.1 Music streaming quality determinants 12
2.3.2 Cloud storage quality determinants 13
2.4 Hedonic Consumer Values 15
2.5 Consumer Frugality 15
2.6 Control variables 16
3. Methodology 17
3.1 Research design 17
3.2 Method of Preference Measurement 17
3.3 Estimation Procedure 17
3.4 Study Design 18
3.4.1 Attributes and levels music streaming 18
3.4.2 Attributes and levels cloud storage services 19
3.4.3 Measurement Hedonic Consumer Values 19
3.4.4 Measurement Consumer Frugality 20
3.4.5 Measurement Control Variables 21
3.5 Choice Design 21
3.6 Model Specification 21
3.7 Data Collection 22
4. Results 23
4.1 Data 23
4.2 Study 1. Music Streaming Services 23
4.2.1 Introduction 23
4.2.2 Model comparison and assessment (H1) 24
4
4.2.4 Estimation Results: Main effects 26
4.2.5 Estimation Results: Moderating effects 28
4.2.6 HB-estimation and clustering procedure 29
4.2.7 Cluster analysis 29
4.2.8 Additional note study 1 32
4.2.9 Discussion study 1 32
4.3. Study 2. Cloud Storage Services 34
4.3.1 Introduction 34
4.3.2 Model comparison and assessment (H1) 35
4.3.2 Model comparison (H2-H4) 35
4.3.3 Estimation Results: Main effects 36
4.3.4 Estimation Results: Moderating effects 38
4.3.5 HB-estimation and clustering procedure 39
4.3.6 Cluster analysis 39
4.3.7 General findings 42
4.3.8 Additional note study 2 42
4.3.9 Discussion study 2 42
5. General discussion 45
5.1 Conclusions and findings 45
5.2 Academic and managerial contributions 48
5.3 Limitations and further research directions 49
References 51
Appendices 54
R code: Music streaming data 68
R code: Cloud storage data 96
5
1. Introduction
Subscription-based services are on the rise. 50% of adults in developed countries are predicted to have at least four online-only media subscriptions by the end of 2020 (Deloitte Global, 2017). Moreover, a fifth of all adults is expected to pay for or have access to at least 10 paid-for online media subscriptions on which they on average spend over $1,200 annually in addition to traditional media subscriptions (Deloitte Global, 2017). Nowadays, you can get subscription-based services for almost everything. People pay monthly fees for Hellofresh boxes, wine delivery services, bikes, music and video streaming services, or pay for additional features within free apps. The presence of these services, and the increasing demand for them, makes it seem that these business models are increasingly gaining dominance in today’s society.
One of these modern business models gaining ground is the freemium business model. “Freemium” refers to the strategy in which consumers can get a basic version of a product or service for free and switch to a premium version with additional features by paying a price (Gu, Kannan, & Ma, 2018). The freemium business model has been adopted as the dominant model in many industries like gaming, music streaming, and cloud storage. To illustrate, firms like YouTube, Spotify, and Dropbox provide part of the service for free, and a higher quality service including additional features against a price premium. In the gaming industry firms offer “free-to-play” games in which additional services can be purchased against a price premium. Within the digital PC games market Rietveld (2018) compared the freemium business model with the premium business model, and found that freemium games are played less and generate less revenue. He advises firms operating the freemium business model to focus on reducing costs or value creation.
Previous research mainly focused on the design and profitability of the freemium business model. This type of business model can only exist if the premium service provides the consumer with enough benefits, so that he or she is willing to pay an additional price for it (Shi, Zhang, & Srinivasan, 2019). Indicating that the freemium business model can only emerge if the customer can derive higher utility from the premium service compared to the free service. Freemium can help the firm in expanding market share, but also puts pressure on margins gained from premium users (Shi, Zhang, & Srinivasan, 2019). Although pressure is put on margins, the positive effect of paid streaming offsets the negative effect of free streaming (Wlömert & Papies, 2016). Nevertheless, before revenue can be made on premium customers, the free version is needed in order to attract customers. However, in order to become profitable transforming free customers into premium customers, or so-called “free-to-fee conversion”, is required.
6 might be of influence as well. Also, current research on the freemium business model is relatively limited to the gaming industry, therefore it is not known whether the effects differ across service categories. The influence of quality is expected to differ across categories, as each service is likely to be designed for fulfilling relatively specific consumer needs (e.g. hedonic needs). In addition, it is known that consumer choice is influenced by several effects such as the compromise effect and attraction effect (Gu et al., 2018), indicating that consumers weigh up the pros and cons of choice attributes against each other, and then decide on whether to go for the compromise or the option with the most attractive feature(s) providing them with the highest utility. Based on the advice of Rietveld (2018), this study analyzes whether adaption of the design of the freemium business model based on individual customer quality preferences can lead more consumers to choose the “premium” instead of the “free”. Furthermore, this study expands research on possible drivers of free-to-fee adoption, as research suggests that hedonic values and frugality might drive the extent to which consumers value service quality attributes. Here, frugality refers to those consumers who enjoy saving money rather than they hate spending it (Rick et al., 2008). Given the current research gaps, this paper examines the following research question:
How does service quality in the freemium business model influence consumer preference for a premium subscription (1), and is the relation of service quality on preference for the premium moderated by hedonic consumer values (2) and frugality (3)?
7
2. Literature Review
2.1 Research Model
The aim of this research is to contribute to the understanding of free-to-fee conversion based on consumer characteristics and choices. Figure 2.1 depicts the research model and summarizes the hypothesized relationships. A premium subscription offer is assumed to depend on two main elements, namely a premium price and attributes determining the quality of the service. The quality attributes of a music streaming service are defined as: music limitations (shuffle play/unlimited music/a combination of both), audio ads (yes/no), and streaming mode (online/offline/both). The cloud storage attributes are defined as storage space (250GB-1TB), access to the service (online/offline/both) and version history (30/75/120 days). Version history refers to the number of days an earlier version can be restored.
Within the present study, first the length of a free premium trial period is expected to have a positive influence on the relation between the premium price and consumer choice utility for a premium subscription (H1). Second, the quality of a service is expected to have a positive effect on the choice utility for a premium subscription (H2). Third, hedonic consumer values are expected to positively moderate the relationship of service quality and consumer choice utility for the premium subscription (H3). Additionally, consumer frugality is expected to negatively moderate the relationship of service quality and choice utility for a premium subscription (H4).
Fig. 2.1. Research model.
Premium Price Service Quality Music: H2a-e Cloud: H2f-i Choice Utility Premium Subscription Hedonic Consumer Values Frugality - H2+ H4- H3+ Music Streaming Cloud Storage Service Category
Length Premium Trial
8 The further approach of this chapter is as follows. In the next paragraphs, literature on the freemium business model, the independent variables, and moderating variables is discussed and hypotheses are formulated. Finally, the control variables and their importance for later on segmentation is discussed.
2.2 The Freemium Strategy
Nowadays, many firms have incorporated a freemium strategy. Freemium refers to a strategy in which a firm offers its core product or service for free but charges an additional price for extra product functionalities or premium services (Hamari, Hanner, & Koivisto, 2017; Liu, Au, & Choi, 2014). Thus, the freemium strategy of a firm consists of at least two services: a free-to-use service and a premium paid service. This paragraph firstly discusses the freemium business model as it appears in the market. The two freemium service tiers (free and premium) are discussed in the subsequent paragraphs.
2.2.1 Freemium in the market
There is a great diversity in freemium services currently being offered in the market. It is observed that firms operating the freemium business model can be distinguished based on two main factors. The first is the type of restriction applied to the free version [1]. The second is the type of the freemium service offered in general [2]. Roughly three types of service restrictions [1] are distinguished: feature limitations, usage quotas, and limited support (Cox, 2018).
First, feature limitations refer to those features which can only be accessed against additional payment. This type of restriction is prevalent in the free-to-play games market. To illustrate, in many games (e.g. Candy Crush) the general features are accessible to all, but additional functionalities such as extra lives and features which make it easier to complete a level need to be purchased. Second, usage quotas put a limit on consumer access to a service by limiting access in terms of maximum usage capacity (e.g. storage limits). Dropbox for example offers consumers 2GB storage for free, but additional storage space needs to be paid for. Another example is found in the software industry, where students are offered free but limited licenses of software products (e.g. Microsoft Office), whereas paid licenses contain all functionalities. Third, limitations in user support exist. To illustrate, a software businesses operating the freemium model gives additional (technical) support to premium users, but only limited support to free users.
9
Table 2.1. Overview of freemium categories and services offered in the market.
Freemium category Service example
1. Cloud storage services Dropbox, Google Drive, OneDrive.
2. Free-to-play games Need for Speed, Angry Birds, Candy Crush Saga. 3. Social platforms LinkedIn, Skype.
4. Software Microsoft Office, MailChimp. 5. Streaming services Spotify, Deezer, YouTube Music.
As the main focus of freemium research has been on free-to-play games, this research expands to other product categories. It is decided to expand to the categories music streaming services and cloud streaming services. These are chosen, as it is interesting to examine whether the influence of quality between a service mainly limiting its features (e.g. presence of ads), and a service mainly limiting its service in terms of usage quotas (e.g. limited amount of GBs) differs. In addition, these categories are interesting to analyze, as a music streaming service is more likely to fulfill rather hedonic needs, whereas a cloud storage service is more likely to serve consumer utilitarian needs.
A study in the free-to-play games market showed a positive relation between customer ratings of the free version and the number of downloads of the premium version (Liu, Au, & Choi, 2014), suggesting the quality offered in the free version to be of influence on the sales of the premium version. Nevertheless, it also suggests that adoption of the premium product depends on the extent to which consumers value quality. But personal values with regard to quality are assumed to differ, and might therefore either positively or negatively influence the adoption of the premium version. According to research, a person its value orientation influences consumer preference for products and services, and the extent to which a consumer finds a specific issue important (Vinson, Scott, & Lamont, 1977). This suggests that personal values in many forms can be drivers of consumer decisions, and thus these consumer values might also impact consumers’ move from free to premium.
One type of personal values which might have an influence on consumer preference for the quality of a service are hedonic values. Hedonic values are found to influence consumer behavior and are positively associated with novelty seeking and promotion stimuli (Wang, Chen, Chan, & Zheng, 2000). This suggests consumers high on hedonic values might have a preference for higher quality services, as a higher quality service is likely to give the consumer this novelty ‘element’ or ‘feeling’ the consumer is searching for. Hence, this suggests hedonic values might moderate the relation between the quality of a service and the consumer preference (utility) for a premium subscription (Section 2.4).
10 a consumer would want to spend money on a higher quality premium service or not (Section 2.5). However, first the two freemium service tiers (free and premium), and the importance of quality are discussed. 2.2.2 Free version
A firm pursuing a freemium strategy offers consumers a time-unrestricted free basic version of its service before attempting to convert its freemium customers into paying premium customers. In a study of Datta, Foubert and Van Heerde (2015) on the effects of free-trial acquisition on customer behavior and Customer Lifetime Value (CLV), using household panel data of a digital television service, systematic behavioral differences between regular and free-trial users were found. The CLV of free-trial customers is on average 59% lower, but their usage rates and responsiveness to marketing communications are higher compared to those of regular customers (Datta et al., 2015). Although the CLV of free-trial users is relatively low, they are likely to be responsive to customized premium offers, which indicates the opportunity for target marketing in increasing CLV. In addition, it was found that both timing and usage intensity during the free-trial are key to the effectiveness of the promotions (Foubert & Gijsbrechts, 2016), which also signals opportunities for individualized targeting. In addition, Wang and Özkan-Seely (2018) mention the dispersion effect of a trial period, which refers to the ‘lost profits’ due to not taking into account consumer heterogeneous willingness to pay after the trial period. This indicates, taking into account individual consumer pricing preferences, still a lot could be gained in the design of the freemium business model.
Moreover, it is found that the optimal subscription fee is zero or close to zero (Penmetsa, Gal‐Or, & May, 2015), indicating the zero-priced subscription fee in the basic version of freemium business models to be optimal. According to Foubert and Gijsbrechts (2016) free trials of the premium version enhance customers’ propensity to start using the service, therefore it is expected that the presence of a free-trial period increases consumer purchase intention. However, when no limit is put on the free usage of a service many consumers decide not to go premium (Wagner & Hess, 2013), which suggests that consumers would prefer a long free trial period over a short free period. At the same time it illustrates the problem which firms with a freemium pricing strategy currently face, as substantial amounts of revenue are missed out on when consumers do not go premium. Additionally, research in the software industry finds that a longer trial period and a higher price can together convey a superior quality (Wang & Özkan-Seely, 2018). Although in general the effect of price on utility is assumed to be negative, this finding suggests trial length to be able to ‘change’ the negative effect to a positive effect. This suggests that trial length might also moderate the premium price in the freemium business model, resulting in a higher preference for a premium subscription.
11 Accordingly, the following is hypothesized:
H1. The length of a free premium trial period positively moderates the relationship between the
premium price and the utility for a premium subscription. 2.2.3 Subscription pricing
Danaher (2002) was among the first to study pricing mechanisms and pricing design choices of online information products. He distinguished access-based pricing (subscription pricing) and usage-based pricing (per-use pricing), and found that they have different relative effects on demand and retention. Whereas subscription pricing has a strong effect on retention, per-use pricing has the strongest effect on usage (Danaher, 2002). Within the freemium context studied, firms like Spotify and Dropbox aim for retention and expansion of their premium customer base, as in contrast to free users, these customers are the ones bringing in the money. A subscription-based strategy is found to be the most optimal pricing strategy when consumer demand is elastic (Jain & Kannan, 2002), and consumer heterogeneity is large (Cachon & Feldman, 2011). This clarifies why firms often base their premium versions on a subscription-based pricing strategy, as in general consumers are very heterogeneous in their preferences. Freemium services respond to these individual consumer wants and needs by providing a service which enables usage according to individual preferences. When consumer demand is elastic this enables maximum product consumption (Jain & Kannan, 2002), which is exactly what freemium services have to offer: unlimited access to the respective service everywhere. A big plus of the freemium strategy is that it reaches both “light” and “heavy” streaming users (Cachon & Feldman, 2011), nevertheless with respect to pricing this strategy might not yet benefit optimally from and account for individual streaming preference and usage behavior, as everyone is currently being charged the same price and offered the same benefits.
2.3 Quality
12 limited (Shi et al., 2019). So, paying customers should have more benefits, and thus have access to higher quality features than non-paying customers. However, it could be argued that consumers differ in their preferences for quality features. One person might highly value having unlimited access to music, whereas another does not mind listening to music on a randomized basis at all. Freemium research on quality in the category ‘free-to-play games’ found that quality can explain why people use freemium, but not if they go premium (Hamari et al., 2017), which indicates that there is not yet a full understanding of what drives consumer switching behavior. Although research did find that a firm should offer its basic service with a relatively high quality and features limiting consumer network effects (Shi et al., 2019), so far no research has focused on the extent to which quality features drive consumers from free to premium service adoption. As Shi et al. (2019) pointed out the difference in quality between the low-end product and the high-end product to be of high importance, it is expected that a higher quality product would lead more consumers to opt for the premium. Accordingly, the following is hypothesized:
H2. Service quality has a positive significant influence on consumer choice utility for a premium
subscription.
2.3.1 Music streaming quality determinants
In order to determine the main freemium service quality attributes and levels in the music streaming category, the subscription structures of the most prevalent music streaming providers in the world were compared (Appendix A). Spotify, Deezer, and YouTube Music were found to pursue a freemium strategy, hence the quality attributes have been based on these services. Table 2.2 gives an overview of the main differences in quality observed in the free and premium services. Within this study, the number of users will not be considered as an aspect of quality, because the focus will be on the preferences of the individual consumer in fulfilling own needs. However, there will be controlled for household size.
Table 2.2. Main quality differences observed in music streaming
Free Premium Audio ads Shuffle play Online streaming Unlimited music No audio ads Offline mode
Based on the comparison of all freemium music providers, perceived quality is expected to depend on the extent to which the consumer attaches value to: music limitations, audio ads, and streaming mode.
13 premium version offers no limitations: unlimited access to the music and choice on whether to skip a song or not. Based on rational choice theory, which expects consumer to maximize their own utility, it is expected that consumers prefer to have influence on which music tracks to play and skip whenever they desire to. For the same reason, solely having access to the shuffle play feature is expected to be less preferred than a subscription including unlimited access to music.
Audio ads. Free versions include random advertising between songs which is referred to as ‘audio ads’, whereas the premium versions are completely ad free. Consumers are expected to maximize their own utility (Zey, 2001), hence we expect them to prefer the subscription with no audio ads.
Streaming mode. In the free version music can only be played online (with internet access), whereas the premium version contains a function which enables offline streaming (without internet access). Based on rational choice theory (Zed, 2001) it is expected that utility will be higher for access to both offline and online streaming compared to solely offline streaming. In addition, offline streaming is expected to be preferred over online streaming, as with offline streaming music will be accessible at all times.
So, as according to H2 it is expected that for each of the music streaming quality determinants a higher quality will lead to an increase in consumer choice utility for the premium subscription. Accordingly, for the music streaming service quality attributes and levels, the following sub-hypotheses are formulated:
H2a. A premium subscription including both shuffle play and unlimited music is expected to be
preferred over a premium subscription with unlimited music.
H2b. A premium subscription which includes shuffle play is expected to be less preferred than a
subscription including unlimited music.
H2c. Audio ads have a negative influence on consumer choice utility for a premium subscription. H2d. A premium subscription including both on- and offline streaming is expected to be preferred
over a premium subscription including offline streaming.
H2e. A premium subscription including online streaming is expected to be less preferred than a
premium subscription including offline streaming.
2.3.2 Cloud storage quality determinants
14
Table 2.3. Main quality differences free and premium cloud storage services.
Free Premium
2 GB, 5 GB, 15 GB Online access
Version history (30 days)
50 GB, 100 GB, 200 GB, 1 TB, 2 TB, 6 GB Online and offline access
Version history (30/120 days)
Based on the comparison, perceived quality in the cloud storage category is expected to largely depend on the extent to which a consumer attaches value to: storage space, access to the service, and version history.
Storage space. In the free version storage space is limited to a certain amount of GBs. Premium cloud storage services contain additional GBs storage space. Based on rational choice utility (Zed, 2001) it is expected that a consumer would prefer to have more if he or she has the choice, therefore choice utility is expected to be higher for a higher amount of GBs compared to a lower amount of GBs.
Access to the service. In the free version, the data stored can only be accessed through the internet, whereas the premium version also enables offline access to the service. Therefore, based on rational choice theory (Zed, 2001) the consumer is expected to prefer having both online and offline access to the cloud storage service over solely having offline access to the data stored. In addition, offline access to data stored is expected to be preferred over online access, as offline access enables all time accessibility to the data.
Version history. The version history of a cloud storage service refers to the number of days a consumer can restore an older version of a document. To illustrate, if you would consider some lines of text as unneeded now, but might want to revise your decision later on, the version history function comes in quite handy depending on the number of days that you have to decide on restoring the older version. Also based on rational choice theory (Zed, 2001) it is expected that a consumer would prefer a longer version history over a shorter version history, and thus utility is expected to be higher for a longer version history. As according to H2, it is expected that for each of the cloud storage quality determinants a higher quality will have a positive influence on utility for the premium subscription. Accordingly, the following sub hypotheses are formulated:
H2f. Storage space has a positive significant influence on consumer choice utility for a premium
subscription.
H2g. A subscription including both on- and offline access is expected to be preferred over a
subscription including offline access only.
H2h. A subscription including online access is expected to be less preferred than a subscription
including offline access.
H2i. Version history has a positive significant influence on consumer choice utility for a premium
15 2.4 Hedonic Consumer Values
Hedonic values have often been studied in the context of consumer shopping experiences (Babin, Darden, & Griffin, 1994; Vieira, Santini, & Araujo, 2018), however in this research they are considered in the context of personal hedonic values in general. There is a difference between hedonic values when shopping and the basic hedonic values a consumer inhibits in general. On one hand, consumers might adopt an hedonic focus during shopping as they seek for entertainment and stimulation (Büttner, Florack, & Göritz, 2014), which is being associated with a positive effect on purchase intention, satisfaction, and loyalty (Vieira et al., 2018). While on the other hand, consumers already possess a certain ‘amount’ of hedonic values in their personal system (e.g. even before going shopping), which influences their choices regarding purchases (Tarka & Rutkowski, 2015). Consumers with strong hedonic values might not be fully satisfied with the pleasure the general functional aspect of shopping can give them (Wang, Chen, Chan, & Zheng, 2000), suggesting that these consumers might seek for additional pleasure. This finding suggests that consumers who have strong hedonic values might weigh the extent of quality offered in a subscription heavily in their purchasing decisions, possibly making them respond more strongly towards subscription offers which include relatively higher quality attributes. Hence, hedonic values are expected to positively moderate the relationship between the quality of a service and the preference for the premium subscription. Accordingly, the following is hypothesized:
H3. Hedonic values positively moderate the relationship between the quality offered in a service
and consumer choice utility for a premium subscription.
2.5 Consumer Frugality
16 feel enjoyment of the fact that the service can be used for free without having to pay for it. Based on this assumption it is expected that a frugal consumer is less likely to spend money on a premium service compared to a less frugal consumer.
Accordingly, as the frugal get enjoyment out of saving money, it is assumed that they will attach low value to high quality attributes, and thus be satisfied with relatively lower quality compared to less frugal consumers. Assuming a frugal will always prefer a service with low quality over a high quality service, it is expected that frugality will negatively impact the relation between quality determinants of a service and the choice utility for a premium subscription. Accordingly, the following is hypothesized:
H4. Frugality negatively moderates the relationship between service quality and consumer choice
utility for a premium subscription.
2.6 Control variables
17
3. Methodology
This chapter discusses the research design (3.1), method (3.2), estimation procedure (3.3), and study design (3.4), followed by the choice design (3.5), model specification (3.6), and data collection approach (3.7). 3.1 Research design
In order to satisfy the research objective quantitative research was conducted. A between-group experimental design was used to compare the freemium service categories, assigning the study participants randomly to either the music streaming or the cloud storage condition. The cloud storage sample and music streaming sample consist of respectively 156 and 160 respondents. In order to determine individual consumer preference for freemium service subscriptions the study presented the participants with several choice tasks: one hold-out choice set and 10 choice sets of each 2 premium options and 1 free version option. Participants were asked to pick their most preferred option. As this decision requires a tradeoff (Eggers, Sattler, Teichert, & Völckner, 2018), some level of focus is required by the participant, and hence the surveys started with the choice tasks. Next, the participants were asked to indicate the extent to which they agreed or disagreed with statements measuring the moderator variables hedonic consumer values and frugality. Finally, the survey ended with the general socio-demographic questions.
3.2 Method of Preference Measurement
Conjoint analysis can be used as a basis to identify market segments based on differences in preferences for certain product attributes (Teichert, 2001). Within this study Choice-Based Conjoint (CBC) analysis is applied to analyze individual consumer pricing and music streaming preferences. It is decided to use this approach, as it allows confronting our respondents with several choice tasks and, are asking them to each time select their most preferred option (Eggers et al., 2018). Furthermore, this approach increases realism regarding the decisions in the real-life marketplace and it enables including a no-choice option, which is an alternative option in case none of the options suits personal preferences (Eggers et al., 2018).
3.3 Estimation Procedure
18 3.4 Study Design
3.4.1 Attributes and levels music streaming
The attributes and levels used (table 3.1) have been based on characteristics of the main music streaming services (Appendix A). Next, the choice for these levels will be discussed.
Table 3.1. Music streaming attributes and levels used in the conjoint study
Attribute Level 1 Level 2 Level 3 Level 4
Length premium trial 0 days 30 days 60 days 90 days
Premium price 5,99 8,99 11,99 14,99
Music limitations Shuffle play Unlimited music Shuffle Play & Unlimited music
Audio ads Ads No ads
Streaming mode Online
streaming
Offline streaming
Online & offline streaming
Length premium trial. The length of a free premium trial period is hypothesized to positively influence choice utility for the premium version. In music streaming free trials range from 30 to 90 days (Appendix A). Also accounting for absence of a free trial and an in-between level, the range is set from 0 to 90 days. Premium price. A higher price is hypothesized to negatively influence choice utility for the premium version. Currently, in music streaming prices range from 4,99 (student account) to 14,99 (family account). A regular (individual) account is priced €9.99. To account for some variety in pricing, levels are set with a three Euro interval ranging from €5.99 to €14.99.
19 3.4.2 Attributes and levels cloud storage services
Attributes and levels for the cloud storage category were selected based on the main service providers in the market (Appendix B). The levels as included in the experiment are shown in table 3.2.
Table 3.2. Attributes and levels cloud storage category used in the conjoint study
Attribute Level 1 Level 2 Level 3 Level 4
Length free premium trial
0 days 30 days 60 days 90 days
Price per month 5,99 8,99 11,99 14,99
Storage space 250 GB 500 GB 750 GB 1 TB (1000 GB)
Access Online access Offline access Online & offline access
Version history 30 days 75 days 120 days
Length premium trial. The length of a free premium trial period is hypothesized to positively influence choice utility for the premium version. In practice, trial length ranges from 30 to 90 days (Appendix B). By also accounting for absence of a free trial and an in-between level, the levels are set from 0 to 90 days.
Premium price. Price is hypothesized negatively influence choice utility for the premium version. Currently, prices are quite diverse as they range between €2,00 and €16,58 (Appendix B). In general the following is likely to hold: the more GBs offered, the higher the price. To account for various price levels, the levels have been set to: €5,99; €8,99, €11,99, and €14,99.
Storage space. Storage space was hypothesized to positively influence consumer choice utility for the premium subscription. In practice, free cloud storage versions provide a relatively low amount of GBs (e.g. 2 or 5 GB). In the premium version the amount of GBs offered to consumers differs between 50 GB and 2 TB (Appendix B). To account for some variety, the levels are: 250 GB, 500 GB, 750 GB, and 1 TB. Access. A subscription including both on- and offline access was hypothesized to be preferred over a subscription including solely offline access. In addition, a subscription including online access was hypothesized to be less preferred than a subscription including offline access. Hence, the levels are set to: online access, offline access, and a combination of both online and offline access.
Version history. Version history was hypothesized to positively influence choice utility for the premium subscription. In practice, version history is limited to either 30 days or 120 days. In order to also account for the level in-between, the levels are set to: 30 days, 75 days, and 120 days.
3.4.3 Measurement Hedonic Consumer Values
20 fun” and “Consumption style” are adopted, as they tend to focus on measuring pleasure and entertainment. The advantage of facet measures is that they allow for independent measurement of specific facets like for example the Job Descriptive Index, which measures several facets of satisfaction (Smith, Kendall, & Hulin, 1969; Russel, Spitzmüller, Lin, Stanton, Smith, & Ironson, 2004). In order to examine the suitability of the items factor analysis is used to investigate the factor loadings. The factor loadings are used as weights for the items on the scale, which added up together result in the factor score. Originally the items were measured on a 5-point Likert scale, however as a 7-point Likert scale was found to optimize reliability and result in stronger correlations this longer version of the scale was chosen (Colman, Norris, & Preston, 1997). The items were measured on a 7-point Likert scale ranging from strongly disagree (1) to strongly agree (7).
Table 3.3. Measurement scale hedonic consumer values.
Nr. Facet Item Item
1. Curiosity and openness to change X1 I like to be creative and act upon my feelings X2 I like to explore new things and new aspects of life 2. Entertainment and fun X3 I like to spend my time as fun as possible
X4 I strive for an adventurous and exciting life
3. Consumption style X5 I like to earn more so I can spend it on things I like X6 I find consumption itself an enjoyable experience in life 3.4.4 Measurement Consumer Frugality
In order to measure the moderating effect of frugality, the items of the measurement scale used by Lastovicka, Bettencourt, Hughner and Kuntze (1999) were adopted. The original full scale contained 8 items, however Kasser (2005) showed that the scale can be limited to four items, therefore the reduced scale as it is presented in table 3.4. is used. Factor analysis is used to investigate the factor loadings, which are used to determine the frugality based on summation of the item weights. Originally these items were measured on a 5-point Likert scale, however based on optimizing reliability and getting stronger correlations (Colman et al., 1997), these items were also measured on a 7-point Likert scale ranging from strongly disagree (1) to strongly agree (7).
Table 3.4. Measurement scale frugality Nr. Items
21 3.4.5 Measurement Control Variables
This study controls for multiple variables which are of interest for segmentation. Participants are asked for their gender, age, highest degree obtained, employment status and net monthly income. Additionally, participants are asked to indicate the number of people in their household including themselves, and the frequency with which they stream music or use a cloud service to store data. Furthermore, participants are asked to select the services they currently use. Finally, there is controlled for the type of service used, either the free version, premium version, or none.
3.5 Choice Design
As it was decided to include only the most relevant attribute levels, the number of levels has not been balanced across all attributes. A fractional factorial design is used instead of a full factorial design, as the five attributes and respective levels would result in 288 possible stimuli in the full factorial design of music streaming, and 576 in cloud storage. Exposing the respondents to such a high amount of stimuli would be unreasonable. The choices sets for the fractional factorial design must be chosen based on the efficiency criteria minimal overlap and utility balance, as this helps in reducing the number of choice sets and getting most insight into the individual respondents’ preferences (Eggers et al., 2018; Huber & Zwerina, 1996). The algorithm in the program used for the experiment creates this minimal overlap and utility balance automatically. Effect-coding is used for the response coding, as setting the reference level to -1 instead of 0 (in dummy coding) allows interpretation of the partworth utilities relative to the mean (Eggers et al., 2018; Louviere, Hensher, & Swait, 2000), which facilitates the interpretation.
For the CBC-experiment it is chosen to use the attribute levels without mentioning brand names as we want to avoid and limit the possibility of consumers being prejudiced in their choices. Meissner, Oppewal and Huber (2016) state that two to five stimuli per choice set are most common, but advice to apply a choice set size that is relatively similar to the real purchase situation. As choosing a subscription out of three main freemium service providers would be realistic for both service categories, a choice set size of three is chosen. A choice set includes two premium options and a no-choice option. The no-choice option is included to get insight into consumer choice dynamics, e.g. when is the free version preferred over a premium subscription and vice versa. A hold-out set can be used to measure the predictive validity of a model (Sawtooth, 2000), hence a hold-out choice set is included in the experiment for both categories. An example of a choice set can be observed in Appendix C.
3.6 Model Specification
22 HB CBC analysis approach is used as it accounts for consumer heterogeneity in attribute preferences, and it largely solves the Independent of Irrelevant Alternative (IAA) problems as it applies the first choice rule (Orme, 2000). As consumers tend to choose the option which maximizes their own utility, it is decided to use Random Utility Theory (RUT) for the model specification. This theory states that the overall utility (U) of a consumer (c) for a product (i) (here: service i) is a latent construct which includes a systematic component (V) and an error component (𝜀) (McFadden, 1981; Walker & Ben-Akiva, 2002). In this study, the overall utility (U) relates to the utility for a premium subscription [Eq. 1].
Uci = Vci + 𝛆ci (1)
The systematic utility (V) represents the choices the participants make in the CBC experiment. It associates the preference estimate (𝛃) with the product attribute (X). Here, the product attribute (X) refers to the service attributes, e.g. the price and quality attributes. Each service attribute (X) consist of a specific attribute level (a). Furthermore, s relates to the preferred service option. As mentioned in the estimation procedure, the HB model consists of two layers (Lindley & Smith, 1972). Equation 2, visualizes the first layer which assumes a multivariate normal distribution relating the individual-level utilities to each other (Allenby, Arora, & Ginter, 1995). It refers to the utility (U) of the respondents who preferred service s.
𝑈
𝑠= ∑
𝑠𝑠=1𝛽
𝑎∗ 𝑋
𝑖𝑠+ ε
𝑠 (2)The second layer is characterized by the MNL model and is visualized in equation 3. The outcome can be interpreted as the probability that a consumer chooses a subscription i out of a choice set (S).
𝑝(𝑖|𝑆) =
e(𝛽𝑎 ∗ 𝑋𝑖) ∑𝑠 e(𝛽𝑎 ∗ 𝑋𝑠)𝑠=1
(3)
The formulas were taken as the basis for examination of the hypotheses in both service categories. For examination of the first hypothesis moderating effects between trial length and price were added to the main effects model of each service category. In addition, for the examination of hypotheses 2 to 4, the hypothesized moderating effects of hedonic values and frugality with the respective service quality attribute levels were added to the main effects model.
3.7 Data Collection
23
4. Results
This chapter discusses the results of both freemium studies. First, an introduction to the data is given. Secondly, the results of both studies are discussed in the subsequent paragraphs. For each study an introduction to the data and the estimation results of the model with the best fit are given. Additionally, the results of a cluster analysis done based on hierarchical bayes estimation of the mean individual attribute level utilities are presented, which is followed by some general level additional findings.
4.1 Data
In this study we collected data for two freemium service categories. The data collected in preference lab resulted in two datasets per service category. The first, a general dataset containing the socio-demographic variables and control variables as specified in the second chapter. The second, a dataset which contained the choices participants made in the conjoint study. This choices dataset included data on the attributes and levels as specified in chapter 2. The general and choices datasets were merged based on respondent ID, which resulted in two datasets: one for each service category. For examining the hypotheses 1 to 4, the multinomial logit model was used. In this model, we characterized the dependent variable (DV) as the ‘Selection Dummy’, which represents the choice the consumer makes out of a specific choice set. This choice is based on the extent to which a consumer prefers certain attribute levels, which are the independent variables (IVs) in our study. The IVs used in both studies are the attribute levels as specified in chapter 2. The hold-out choice set is used to validate the models. In addition to the aggregate analysis, we also use cluster analysis to further examine consumer preferences on a more individual level. For this analysis we used the mean utilities per attribute level for each individual customer estimated by the hierarchical bayes algorithm, also taking into consideration the moderating effects of hedonic values and frugality. Additionally, we use several variables (e.g. gender and income) to gain insight in freemium usage characteristics by analyzing individual mean differences in free and premium usage.
4.2 Study 1. Music Streaming Services
4.2.1 Introduction
24 bachelor degree from a university of applied sciences (33.3%) or a university master's degree (29.4%). Of the participants 52.3% uses or has used a premium music streaming service, 40.5% a free version, and 7.2% has not used a music streaming service. To establish an hedonic values factor and a frugality factor, the factorability of the moderating items of both item scales was examined with standard PCA. Investigation of the factor loadings showed item X5 of the hedonic consumer values scale to load below 0.4 after rotation, and therefore this item was removed from the scale. The KMO showed a value of 0.781, which is above the recommended .5 (Kaiser, 1974), indicating the sampling adequacy is adequate. In addition, the Bartlett’s Test of Sphericity was significant (χ2(9) = 140.05, P<.05), indicating the suitability of the items for factor analysis. The nine remaining items of both variables all loaded above 0.4 after rotation, indicating the items all have a sufficient influence on the respective factor. The factor loadings were used as weights and multiplied with the respective item scores, which resulted in the two factors hedonic values and frugality. In the limitations it is stated, that usage of the Principal Component would have been more appropriate.
4.2.2 Model comparison and assessment (H1)
Trial length was hypothesized to positively moderate the relationship between the quality of a service and consumer choice utility for a premium subscription (H1). To examine this hypothesis the moderating effects between the trial length attribute levels and price were added to the main effects model. A likelihood ratio test comparing the partworth and linear model showed that the models did not significantly differ (χ2(8)=11.1, p=0.19), and hence the price attribute was assumed to be linear. The model with trial length and price linear interactions did not show significant effects, hence no support for H1 was found.
Although the fit of the linear model was as good as the fit of the partworth model, it was also tested for the hypothesized interaction effects in the partworth model. Both models can be found in Appendix D. The results show that a trial period of 60 days negatively affects the relation of price (€11.99) on utility for the premium service (UFT60daysx11.99=-0.290, p=0.047). A 30 day trial period is also found to (marginally) positively affect the relation of price (€11.99) on utility for the premium subscription (UFT30daysx11.99=0.272, p=0.052). In addition, a trial period of 60 days is found to positively affect the relation of price (€8.99) on utility for the premium, as marginally significant positive effect was found (FT60daysx8.99=0.259, p=0.058). For a longer trial length no significant results were found. Hence, there is no support for the length of a free premium trial period to positively moderate the relation of the premium price on the preference for the premium subscription. And thus, hypothesis 1 is rejected.
4.2.3 Model comparison (H2-H4)
25 Again, it was first investigated whether price could be added as a linear variable. A Likelihood ratio test showed no significant difference between the models (χ2(2)=4.34, p=0.11), and hence the price attribute was assumed to be linear and added to the models accordingly. To decide on the model with the best fit several models were estimated, and compared on several measures (Table 4.1). The first model (M1) is the main effects model, which means it contains the attribute levels as they were included in the conjoint analysis. The second model (M2) is the full model containing both main and moderating effects. This model served as a basis for model 3. In case no significant effects between the respective moderating variable and the respective quality attribute levels was found, the whole attribute including its levels was dropped. This was done attribute per attribute, which resulted in model 3 (M3).
The results show that all models predicted significantly better than the null model. The log Likelihood (-1380), pseudo R2 (16.0%), and the predictive validity (30.2%) are best for the full model (M2), whereas model 1 and model 3 have respectively the lowest BIC (BIC = 2878.8) and AIC (AIC = 2788.0). Although the different measures are not in line on which model fits the data best, likelihood ratio tests show that model 3 performs significantly better than model 1 (χ2(10)=43.6, p<0.001), but not compared to model 2 (χ2(3)=3.07, p=0.38). Note, that this is due to the fact that the model still includes insignificant effects for certain attribute levels. Exclusion of the insignificant levels would disenable estimation of the respective reference levels. To conclude, as model 3 takes up least degrees of freedom and has a relatively high predictive validity as well (MAE = 30%), this model is chosen as the final model.
Table 4.1. Measure-based model comparison
Model 1 (M1) Model 2 (M2) Model 3 (M3) Model description Main effects only model Full model Reduced model: M1 +
Variables with at least 1 sign. interaction Parameters* 1:9 1:9, (5:9)*10, (5:9)*11. 1:9, (8:9)*10, (5:9)*11. Log Likelihood -1400 -1380 -1380 McFadden R2 0.147 0.160 0.159 AIC 2814.5 2790.9 2788.0 BIC 2878.8 2919.5 2897.3 MAE 29.5% 30.2% 30.0% df 10 20 17
*Parameters: FT 0 days (1), FT 30 days (2), FT 60 days (3), Premium monthly price (4), ML Shuffle (5), ML Unlimited (6), Audio ads no (7), Streaming online (8), Streaming offline (9), Hedonic factor (10), Frugality factor
26 4.2.4 Estimation Results: Main effects
Analysis of hypotheses 2 to 4 is done based on Model 3, whose output is presented in table 4.2. It is observed that the option with the highest utility (Upremium=0.7), the most preferred premium subscription, has a 90 days free trial, costs €5.99 per month, contains unlimited streaming, no audio ads, and allows the consumer to stream both online and offline. The least preferred premium subscription (Upremium=-10.3) has no free trial, only enables online streaming and shuffle play, costs €14.99 per month, and contains audio ads. The no choice option (free version) shows a significant negative effect (Unone=-0.255, p<0.05), indicating that many premium subscription combinations are likely to be preferred over the free music streaming service. Additionally, consumers find music limitations (47.3%) and audio ads (33.0%) most important when deciding on which premium subscription to choose. The monthly premium price (7.3%), a free trial period (6.7%), and the type of streaming (5.7%) are of less importance. Despite its relative unimportance, the absence of a free trial has a significant negative impact on utility for the premium (UFT0days=-0.228, p<0.01), and a 90-day free trial has a significant positive effect on utility for the premium (UFT90days=0.202, p<0.01). Highlighting the value of offering a free trial, and in particular a 90-day free premium trial.
In this study, service quality is hypothesized to have a positive significant influence on consumer choice utility for a premium subscription (H2). In freemium music streaming, the quality of a service was assumed to be based on: the type of music playing limitations present in the service, audio ads (yes/no), and the type of streaming mode. First, having the ability to use both the shuffle play function and the unlimited music streaming function was expected to be preferred over solely unlimited music streaming (H2a). The results support this hypothesis as the combination of both streaming functions (U=0.748, p<0.05) is indeed preferred over solely the shuffle play function (U=-1.957, p<0.001). Additionally, shuffle play was hypothesized to be less preferred than unlimited music streaming (H2b). The results also confirm this hypothesis as unlimited streaming (U=1.210, p<0.01) is indeed preferred over shuffle play (U=-1.957, p<0.001). In addition, it is noted that the unlimited music streaming option (U=1.210, p<0.01), which we characterized as the ‘middle’ quality level, is preferred over the ‘higher’ quality level (UShuffle&Unl=0.748, p<0.05), highlighting the possible presence of consumer high preference for unlimited music streaming.
27 Although for the streaming attribute no significant effects were found, we did find positive significant effects for the relatively higher quality attribute levels unlimited music, the combination of unlimited music and shuffle play, and the absence of audio ads. Which indicates that the quality of the service offered in the subscription does play an important role, as offering a higher quality music streaming service is able to positively influence consumer preference for the premium version, and hence increase the likelihood of a consumer choosing the premium version over the free version. However, as not for all attributes significant effects were found, the hypothesis on service quality positively influencing consumer choice utility for a premium music streaming subscription cannot be fully confirmed.
Table 4.2. Main and moderating effects (based on M3)
Attribute Level Utility Relative importance
Main effects
Free premium trial period
0 days β01 -0.228** (.077) 6.7%
30 days β02 -0.017 (.077)
60 days β03 0.043 (.075)
90 days β04 0.202** (.075)
Price per month Price β05 -0.420*** (.040) 7.3%
Music limitations Shuffle play β06 -1.957*** (.347) 47.3%
Unlimited β07 1.210*** (.311)
Shuffle play & unlimited β08 0.748* (.309)
Audio ads No β09 1.532*** (.219) 33.0%
Yes β10 -1.532*** (.219)
Streaming type Online β11 -0.257 (.448) 5.7%
Offline β12 -0.015 (.448)
Online & offline β13 0.272 (.430) Moderating effects: hedonic values (‘H’)
Streaming type Online * H β14 -0.220* (.103)
Offline * H β15 0.103 (.103)
Online & offline * H β16 0.117 (.100) Moderating effects: Frugality (‘F’)
Music limitations Shuffle * F β17 0.346*** (0.083) Unlimited * F β18 -0.216** (0.076) Shuffle & unlimited * F β19 -0.130 . (0.075) Audio ads Audio ads no * F β20 -0.208*** (.052) Audio ads yes * F β21 0.208*** (.052)
Streaming type Online * F β22 0.192* (.080)
Offline * F β23 -0.078 (.077)
Online & offline * F β24 -0.113 (.074)
None-option Free version cnc -0.255* (.107)
28 4.2.5 Estimation Results: Moderating effects
An overview of the estimated betas for the hypothesized moderating effects and their significance is given in table 4.2. Firstly, hedonic values were hypothesized to increase consumer preference for higher quality service attributes, and therewith expected to increase choice utility for the premium service (H3). The results do not confirm this hypothesis, however the results do show those high on hedonic values to dislike low quality more than those low on hedonic values. To illustrate, hedonic values are found to have a significant negative impact on preference for the relatively lower quality online streaming attribute level (U=-0.220, p<0.05). Which indicates that consumers high in hedonic values are less likely to opt for a subscription only enabling them to stream music while having an internet connection. Furthermore, this finding also supports the notion of service quality to be of importance on consumer choice behavior in the freemium business model.
Secondly, frugality was hypothesized to have a negative effect on the relationship between service quality and consumer utility for the premium version (H4). The more frugal the consumer, the less likely this person was expected to opt for a relatively higher quality service attribute, and therewith also decreasing utility for the premium subscription. The results show an interaction between frugality and the combination of shuffle play and unlimited music to have a marginal significant negative effect on utility for the premium (U=-0.130, p<0.1). This indicates that a frugal consumer is not that likely to choose a subscription including both of these music features, which in this study is referred to as a relatively higher quality music streaming attribute. A similar, but stronger significant negative effect is found for streaming music on an unlimited basis (U=-0.216, p<0.01), supporting the notion of a frugal consumer to be less likely to opt for a higher quality feature. Additionally, a strong positive significant effect is found for frugality and shuffle play (U=0.346, p<0.001), which indicates a frugal consumer is highly likely to choose a subscription including a relatively low quality music feature. A frugal consumer is also more likely to choose a subscription including audio ads (U=0.208, p<0.001) and a streaming service including online streaming only (U=0.192, p<0.05), suggesting the frugal consumer to be more satisfied with a lower quality music streaming service.
29 4.2.6 HB-estimation and clustering procedure
For hierarchical bayes estimation data on 153 respondents from the conjoint survey for music streaming is used. The analysis is done in R with the ‘Bayesm’ package. 40.000 MCMC draws were done, and every 10th draw was kept, indicating a total of 4000 draws per respondent were kept. A large list was created containing 153 lists (list-in-list structure), each corresponding to an individual customer. Nesting was done through the respondent id variable. The list of each customer contained two components, namely a numeric matrix in which the covariates were stored, and a numeric vector of the choice indicators. A hierarchical multinomial logit model with mixture of normals heterogeneity was estimated. The results did converge and the normality assumption was reasonable. The model accounts for individual customer preferences, and enables estimation of individual level coefficients. The means for the none-option and each attribute level were calculated for each individual respondent. A new dataset was created merging the HB estimated means dataset with a dataset containing individual customer specifics on customer id. This new dataset served as a basis for the subsequent preference identification.
A partworth model including variables for which significant moderating effects were found is taken as the basis for clustering individual cloud storage service preferences. The partworth model is chosen as it also enables examining consumer preference for each of the price levels. Ward’s method for hierarchical clustering was used to create groups with minimal within-group variance. Individual cluster membership was labelled through dummy coding.
4.2.7 Cluster analysis
Three segments emerged: the premium likers, the doubters and the premium dislikers. An analysis of variance (ANOVA) on the attribute levels shows significant differences in the cluster means for a free premium trial of 0 and 60 days, all price levels, shuffle play, unlimited music, online streaming and the combination of both on- and offline streaming (p-values<0.05). Significant differences between the clusters were also found in the moderating effects. Both the impact of frugality and hedonic values on online streaming and the combination of on- and offline streaming differs significantly across clusters (p-values<0.01). Frugality is also found to have a different impact across segments on shuffle play (F = 4.993, p = 0.027), unlimited music (F = 23.776, p = 0.000), and both the presence and absence of audio ads (F = 3.914, p = 0.050). Table 4.3 presents an overview of the identified music streaming segments and their attribute level means. A demographic profile of the consumers in each of the segments is given in table 4.4.
Segment 1: the premium likers
30 of 30 days and 90 days are found to positively influence utility for a premium subscription (UFT30days=0.162; UFT90days=0.831). A price of €5.99 leads to the highest utility (Uprice5.99=2.183). The premium likers respond relatively more negative to the lower quality attribute levels shuffle play (UShuffle=-2.611) and the presence of audio ads (UAdsYes=-3.775) compared to the doubters (UShuffle=-1.610; UAdsYes=-1.089) and the premium dislikers (UShuffle=-0.606; UAdsYes=-1.177). In the first segment, having access to both online and offline streaming is most preferred (UStreamingOnOff=0.667). The utility for this attribute level is positively influenced by the presence of strong hedonic values (UStreamingOnOfxH=2.192) and frugality (UStreamingOnOffxF=2.632). Additionally, frugality strongly decreases consumer preference for the lower quality attribute levels shuffle play (UShufflexF=-10.303) and audio ads (UAdsYesxF=-14.903).
Segment 2: the doubters
The majority of the doubters is below the age of 36 (82.6%). Although the doubters slightly prefer a free subscription over a premium subscription (Unone=0.240), an offer including their most preferred quality attributes is likely to positively influence preference for a premium subscription. Similar to the first segment, the attribute levels ‘90 day free premium trial’ and a price of €5.99 lead to the highest utility for a premium music streaming subscription (UFT90days=0.858; Uprice5.99=2.613). The doubters attach relatively more value to being able to stream music on- and offline (UStreamingOnOff=1.045) compared to both segment 1 (UStreamingOnOff=0.667) and segment 3 (UStreamingOnOff=0.295). Compared to the other segments, hedonic values and frugality have the strongest effect on the type of streaming in this segment. To illustrate, hedonic values decrease preference for the lower quality attribute level online streaming (UStreamingOnxH=-3.296), whereas they increase preference for the combination of on- and offline streaming (UStreamingOnOffxH=3.510). Similarly, frugality decreases preference for online streaming (UStreamingOnxF=-3.850) and increases preference for both on- and offline streaming (UStreamingOnxF=3.979).
Segment 3: the premium dislikers
31
Table 4.3. Anova tests, means and standard deviations for the identified music streaming segments.
ANOVA Segment 1 (N = 36) Segment 2 (N = 60) Segment 3 (N = 45)
F Sig. Mean SD Mean SD Mean SD
Main effects
Free trial 0 days 8.445 0.004 -0.949 0.955 -0.582 1.257 -1.168 0.186 Free trial 30 days 2.524 0.114 0.162 0.574 -0.487 0.727 -0.842 0.124 Free trial 60 days 33.213 0.000 -0.044 0.617 0.210 0.754 1.178 0.146 Free trial 90 days 0.020 0.888 0.831 0.948 0.858 1.154 0.833 0.175 Price €5.99 27.948 0.000 2.183 1.407 2.613 1.660 4.497 0.246 Price €8.99 6.949 0.009 0.654 0.605 0.270 0.820 0.592 0.135 Price €11.99 45.049 0.000 -0.800 0.886 -0.871 1.034 -2.303 0.153 Price €14.99 6.823 0.010 -2.037 1.268 -2.011 1.687 -2.786 0.225 ML Shuffle play 11.167 0.001 -2.611 0.798 -1.610 1.082 -0.606 0.174 ML Unlimited music 34.054 0.000 1.482 0.486 1.068 0.746 0.056 0.136 ML Shuffle & Unlimited 0.2988 0.585 1.129 0.681 0.543 0.982 0.550 0.149 Audio ads - No 1.907 0.169 3.775 0.974 1.089 1.069 1.177 0.250 Audio ads - Yes 1.907 0.169 -3.775 0.974 -1.089 1.069 -1.177 0.250 Streaming online 24.992 0.000 -0.531 0.430 -0.982 0.764 -0.436 0.125 Streaming offline 2.903 0.090 -0.136 0.544 -0.063 0.569 0.141 0.077 Streaming on&off 26.039 0.000 0.667 0.642 1.045 0.935 0.295 0.165 Moderating effects Streaming online x H 27.139 0.000 -1.742 1.427 -3.296 2.540 -1.393 0.558 Streaming off x H 2.812 0.096 -0.451 1.773 -0.214 1.867 0.443 0.270 Streaming on&off x H 30.363 0.000 2.192 2.087 3.510 2.904 0.950 0.571 ML Shuffle play x F 4.993 0.027 -10.303 3.631 -5.654 4.024 -2.613 0.755 ML Unlimited music x F 23.776 0.000 5.852 2.207 3.794 2.735 0.258 0.606 ML Shuffle & Unl. x F 1.557 0.214 4.450 2.956 1.860 3.704 2.355 0.610 Audio ads – No x F 3.914 0.050 14.903 4.665 3.855 3.892 5.067 1.065 Audio ads – Yes x F 3.914 0.050 -14.903 4.665 -3.855 3.892 -5.067 1.065 Streaming on x F 20.323 0.000 -2.030 1.706 -3.850 3.114 -1.873 0.555 Streaming off x F 2.212 0.139 -0.603 2.276 -0.129 2.154 0.624 0.361 Streaming on off x F 22.743 0.000 2.632 2.662 3.979 3.479 1.249 0.687
None option (Free) 93.672 0.000 -0.571 4.703 0.240 4.596 12.057 0.819
Interpretation
‘H’ = Hedonic values, ‘F’= Frugality The Premium Likers
The Doubters The Premium Dislikers
Table 4.4. Demographic profile of consumers in each of the segments
