1
Optimal trial strategy
Moving consumers from the free trial version toward the commercial version
Author: Jelle Peter Rolf Completion date: June 23rd, 2014
2
Optimal trial strategy
Moving consumers from the free trial version toward the commercial version
University of Groningen Faculty of Economics and Business
Research paper for Master Marketing Intelligence Final paper
June, 2014
1st supervisor: dr. H. Risselada 2nd supervisor: dr. J.E.M. van Nierop
Author: Jelle Peter Rolf Student number: 2400588 Email: jellerolf@hotmail.com Telephone number: 0638004827
3
Abstract
Offering free trials may have two effects, hence increasing sales and cannibalizing sales simultaneously. This study is aiming at understanding how the design of a trial version can increase the intention of consumers to start using the paid version of software. Subsequently generating more sales and prevent consumers to continue using the trial version. Existing literature about cannibalization, different trial versions, social effects, channel choice, advertising avoidance and quality reduction are discussed in the literature review.
This research tested how the design of the trial version and use of different attributes can influence the consumers intention to purchase the commercial version. To identify the most important attributes of a trial version, a conjoint experiment is performed to identify the relative importance of the attributes. The utilities of the different attributes where used to test how reductions influence consumers’ intentions. Results indicate that the available channel, price and the advertising quantity are the most important attributes of a trial version according to the consumers. Furthermore, the results show that different attributes can be used to change the utility of trial version and increase the intention to start using the paid version. It was expected that the relation between quality of the trial version and the intention to start using the paid version would show an inverted u-shape, this relationship could not be found. However, this research reveals a basis to help managers of music streaming services decide on how to compose the trial version.
Key words: trial versions, optimal trial design, optimal pricing, consumer preference
4
1. Introduction
1.1 Background of the study
With the wide availability of peer to peer websites on the internet, consumers have the ability to illegally share digitalized material for free. For that reason companies who develop digitalized material have concerns about their industries future, they try to find solutions to prevent losing sales by online file sharing and software piracy (Bhattacharjee, Gopal & Sanders, 2003).
Piracy is especially threatening the industries of experience goods, examples of experience goods are music, software and movies. These products generally have features which consumers might not be able to assess exactly (Klein, 1998). Every day 146 million people visit piracy websites, which accounts the astonishing amount of 58 billion visits a year (Go-Gulf.com, 2011). Of all illegally downloaded content on the internet 2,9 percent is music and 95 percent of the music downloaded online is illegal, therefore every year $12,5 billion is lost due to piracy in the music industry (Go-Gulf.com, 2011). The same threatening situation stands for the software industry where the extent of illegally downloaded software was $59 billion in 2010, which accounts for 6,9 percent of all illegally downloaded content on the internet (Go-Gulf.com, 2011).
Producers of software, music and other forms of experience goods cannot determine which consumers will download and how they download the product or service (Cheng et al.,2012). By offering a free trial/trial version enables producers to reduce the risk of consumers downloading the product or service illegally and gives the consumers the opportunity to try it on a legal way. By providing free trials, consumers are able to assess experience goods in the process of consumption and are able to estimate whether their expectations about the functionality are satisfied (Cheng & Lui, 2012).
Providing free trials can be an effective strategy to reduce functionality concerns of the consumer and on the other hand reduce concerns of the companies (Cheng & Lui, 2012). The NPD Group investigated peer to peer music services and found a decline of 26 percent peer to peer sharing activities in 2012 compared to 2011 (The NPD group, 2012). The decline is caused by the increased use of free and legal streaming services. These streaming services proved to be a substitute for file sharing activities (The NPD group, 2012). It be concluded that free trials are effective in reducing piracy for experience goods when performed correctly.
Although it is clear that free trial versions of streaming services are able to reduce piracy, offering free trials may have two effects increasing sales and cannibalizing sales simultaneously (Faugère & Tayi, 2006). Trials can be an useful tool in the early stages of the life
5 of a product or service. Trials have the ability to increase the number of early adopters who promote the product or service by word of mouth, which will lead to future consumers and therefore increasing the install base (Jain et al., 1995). The downside of offering free trials is that they reduce the number of paid purchases, which may affect long term sales (Bawa &
Shoemanker, 2004). Eventually, companies need to secure future business and prevent cannibalization of own profits, therefore companies need to move consumers from the trial version to the paid version.
Still little is known on how to make trial users switch to the paid version and what role the design and functionality of the trial have in this process. Changing the design and functionality of a trail might have different consequences. On the one hand, changing the design might move consumers to the paid version and thus generate more profits. But on the other hand, changing the design might also decrease the perceived value of the trial version. Which might lead to consumers being less triggered to start using the trial version and consequently decreasing the install base. Reviewing existing research on trial versions or optimal trial strategies gives insights on how trials are used and designed.
Cheng & Tang (2010) analysed the optimal design of a free trial strategy and uncover the conditions under which firms should introduce free trials. In their study they make a trade-off between the effect of cannibalization and the network effect. Cheng et al. (2010) reveal that companies should introduce a free trial when there is a strong network intensity instead of charging consumers for a lower quality version. Software trials are offered for a short period of time, therefore they still question if a network effect even could arise from trials (Cheng et al., 2010). Another finding of Cheng et al. (2010) is when companies offer a free trial software with high quality, companies can charge higher prices for the commercial version. Concluding, software companies who offer two versions that differ in quality will achieve more profit compared to software companies who are only offering the commercial version. Cheng et al.
(2012) build a framework to help software firms to decide which kind of free trial they should use given the strength of the network effect, they make distinctions between limited version and time locked version.
Dey, Lahiri and Lui (2013), investigate if general learning is influenced by the design of the software trial. They make a distinction between optimal trial period and price. Dey et al.
(2013) find that a time locked trial is optimal when consumers learn fast. However, they do find that neither the optimal price nor trial period is increasing the speed of learning.
The studies of Cheng et al. (2010) and Dey et al. (2013) are focused on the effects of changing time limits, including all features or a limited amount of features and charging or not
6 charging the consumer and give insights on which trial is more effective in the presence of network effects with respect to cannibalization. Still, these studies are based on offering a trial version under certain network conditions and therefore are limited in understanding how consumers motives and preferences can be used in the design of the trial and generating more profit.
Faugère & Tayi (2006) analysed the optimal design of free software, hence time duration and characteristics of the features included, to achieve future sales. They discuss how a trial should be designed taking into account (re)installation of software, consumer learning and cannibalization. This research gives insights on how cannibalization can occur from trial design, but still lacks to explain which features consumers prefer to use within a trial version. The same stands for explaining how features can impact the consumer to move to the commercial version.
T.Wang, Oh, K.Wang & Yuan (2011) investigate user acceptance of paid information technology in the context of free trial marketing strategy. They find that the perceived ease of use and perceived enjoyment have more influence compared to perceived usefulness when clarifying satisfaction and purchase intention. Wang et al. (2011) and Cheng et al. (2010) state that even though consumer know free trial reduces risk, they will only purchase software when they perceive the costs to be less than the benefits they receive. These researches give insights on motives of consumers to move from trial version to paid software, however they do not chart which features of the trial version are responsible for moving consumers towards or away from the commercial version.
Authors Optimal trial level
Optimal pricing
Conditions for trial strategy
Consumer motives
Consumer Preferences
Object
Faugère &
Tayi (2006)
Reinstallation of
software Cheng &
Tang. (2010)
Limited version free
trial T.Wang, Oh,
K.Wang &
Yuan (2011)
Technology
Acceptance and Limited version free trial
Cheng & Liu (2012)
Limited and time
locked free trial Dey, Lahiri
& Lui (2013)
General learning,
Limited and time locked free trial
This study Optimal trial design
Table 1: Overview of existing literature
7 1.2 Purpose of this study
Existing studies are focused on the distinction between a fully functioning free version with a time lock (time locked) and a free version with limited functions and no time lock (limited version). Therefore existing studies are not giving clear insights on the separate effects of the features included in the trial version. As Wang et al. (2011) and Cheng et al. (2010) stated, consumers will only purchase the commercial version if the benefits exceed the costs of acquiring the commercial version. For that reason this study is focusing on what features should be added or removed from the free trial to give consumer the ability to learning about the product or service. But, prevent consumers to remain using freeware since they have no incentive to use the paid version (free riders). The difference between this study and existing research, is that this study is identifying the preferences of consumers regarding the features of the trial version. Furthermore, this study shows how companies can use these features to make a trial more attractive or less attractive. Making the trial version more attractive to increase the consumer base of the trial version and less attractive to move consumers from the trial version to the paid version.
1.3 Research problem
This study is aiming at understanding how the design of a trial version can increase the intention of consumers to start using the paid version of software. In an experiment this study will conduct the preference of consumers regarding the features of a trial version of music streaming services. So, music streaming services will be central to this study. The main research objective of this study is:
How can the design of the trial version influence the consumers intention to purchase the commercial version?
This study aims to contribute to the literature of trial strategies by giving insights on trial design.
There are no studies available that investigate how features of trial versions can increase the intention to purchase the commercial version. The findings of this study can be important for managers in understanding how to designing trial versions that help consumers learn about the software, yet prevent consumers of the trial version to become free riders. It gives insights on how to move consumers from the free trial version to the paid version. Furthermore, this study will investigate to what extent the utility of trial version can be reduced and subsequently increasing the intention to start using the paid version. The results of this study are useful for managers to understand which attributes can be changed to increase the sales of commercial versions and therefore increase profit by understanding consumers preferences.
8 1.4 Outline of the continuing research
The continuing part of the study is structured as following. In chapter two this study will investigate existing literature about trial strategies and their impact on consumer behaviour.
Besides the literature review, this part will also present the hypotheses and the conceptual model. Chapter 3 will explain the methodological part of the study, thus explain how the experiments will be performed. After the methodological part, a data analysis will be conducted and the results of the experiments will be presented in chapter 4. The conclusions and discussion will be presented in chapter 5, together with the practical and theoretical implications and recommendations for future research.
9
2. Literature review
In this chapter existing literature about the forms of trials, features of trials and the effect of trials will be investigated. First, to get a understanding on the advantages of trials, the effects of offering trials will be investigated. After the effects of offering trials are clear, the different types of existing trials will be investigated to understand which features are used in trial design.
Finally, consumers behaviour will be investigated to understand what the consequences of offering different quality levels can be for providers of the trial.
2.1 Trials
Trials and freeware are used alongside the fully functioning commercial functions and are used to encourage adoption (Haruvy et al., 2001). When introducing trials, they are not solely used for raising awareness and to improve the perception of the consumer about a product. Trials give the consumer a direct experience with the new product or service and therefore it reduces uncertainty and risks (Heiman, McWilliams, Shen & Zilberman, 2001).
Trials are also more effective compared to other promotional methods, since trials provide the opportunity to show a product’s features or benefits. These features and benefits can be valued and therefore they give the ability to overcome adoption risk (Jain et al., 1995;
Cheng & Tang, 2010). Trials are effective in creating brand awareness, improve brand loyalty and expending product categories, since it gives the opportunity to encourage brand switching.
However, trials also have disadvantages since it is very expensive (Bawa & Shoemanker, 2004;
Jain et al.,1995).
For a trial to be effective it should have moderate levels of learning, otherwise trials should not be offered (Dey et al., 2013). Consumers are uncertain about the use of an experience good, trials give consumers the ability to learn about the quality of the product or service, which has a positive effect on sales (Hu, Lui, Bose & Shen, 2010). Consumers perceive that there are different kinds of risk involved in the buying process of an experience good, hence financial and psychological risk (Hu et al., 2010).
A preliminary experience that is similar to the actual product usage, supports consumers to make a deliberate evaluation of innovation that leads to a more established decision behaviour (Wang et al., 2011; Faugére & Tayi, 2006). The experience with a product has the ability to change the perception, attitude and beliefs of the consumer with regard to the usability of the product, the beliefs can be different after actually using the product (Howard an Sheth, 1969;
Festinger, 1957). Offering a trial can consequently result in having positive attitude towards the
10 product and therefore resulting in an enhanced purchase intention and in actual behaviour (Motes & Woodside, 2001; Nord, 1982; Rotschild & Gaidis, 1981).
Hu et al. (2010) demonstrate that providing trials for online digital music reduces the uncertainty about the product and show that products with trials have a higher conversion rate, higher interest of shoppers and better sales compared to products without trial. Consumers reconsider their willingness to pay after trial use and thus allowing companies to raise the price (Dey et al., 2013)
2.2 Cannibalization
Cannibalization is the degree to which one product losses customers at the expense of a new or other product by the same firm (Copulsky,1976). Kerin, Harvey & Rothe (1978) describe cannibalization as negative effects of new introduced products or line extensions on existing products serving current markets. Offering free trials is thus increasing the consumers ability to learn and experience the product, however the downsides of offering free trials can have a big influence on the performance of a company. Faugère & Tayi (2006) state that offering free trials may have two effects, increasing sales as cannibalizing sales simultaneously. According to Cheng & Tang (2010), “Free trial software cannibalizes the demand for full functional commercial products”. When a consumer have fewer requirements on quality and functionality of software, the consumer will hold on to the free trial version with its limitations (Cheng et al, 2010) and therefore cannibalizing the commercial version.
Dey et al. (2013), describe cannibalization with reference to the useful life of the commercial software version. They state that the useful life of a commercial version is reduced by trial versions, this implies that people who would have bought the product or service without using a trial software version, will now try the trial version first and thus shortening the useful life of a commercial version (Dey et al., 2013). As a result, by offering free trials companies risk losing demand to free trial users and thus affecting the firms optimal profit (Cheng & Lui, 2012).
2.3 Network
The presence of social effects and network externalities play a role in the adoption of technology. When adopting technology, the effect of other users can play a significant role in the decision of a consumers, these effects are presented below.
11 Social effects
According to Katona, Zubcsek & Sarvary (2011), network effect relates to the influence that the structure of connection patterns of a person who already adopted a good has on potential adopters. Katona et al. (2011) confirmed that the probability that an individual will adopt is positively influenced by the number of already adopted connections. When an individual’s connections are interconnected, that will positively enhance the probability of adopting. For companies who offer software, it might be difficult to impose consumers to buy the product when there is no sufficient user base and therefore provide free trials (Haruvy et al., 2001).
Companies need to generate enough trials to start the diffusion process (Jain et al., 1995).
When introducing a trial, in the presence of a strong network effect, companies are financially better off by offering a free trial instead of charging for a lower quality (Cheng et al.,2010; Cheng et al., 2012).The positive effects of network effect on cannibalization is researched by Haruvy & Parasad (2001), they find that positive network effects can temper the cannibalization of sales when introducing free trials. However, when there is a large base of people that continues to use the free trial software, this may create an incentive for new users to keep using the free trial instead of buying the commercial version (Faugére et al., 2006).
Network externalities
Network externality is when the value of a product or service increases when there is a large network of users, therefore the value of a product is not solely determined by the product’s benefits (Katona et al., 2011). Katz and Sharipo (1985) describe network externality as, the additional utility consumers obtain from the size of the network. Network externalities can have an influence on the success of software. To establish a network companies can share free software, after the network externality is sufficiently large, companies should courteously trigger consumers to prefer the commercial version over the free software (Haruvy et al., 2001).
Research of Cheng et al. (2010) found that the installed based is influenced by the quality of a product, hence the number of total install base and the number of free trial users are simultaneously increasing with the quality of the product.
The discussion of literature about the social influence on consumers shows that consumers are indeed influenced by others to use a product or service. This research will focus on the effects of network externalities, gained from the insights from the discussion of the literature, it is expected that consumers perceive more value when other people use music streaming services and have the ability to share content with other users.
12 2.4 Different types of trials
Existing theory is describing two types of trials to create a positive network effect, one based on time and one based on quality. Exploring these types of trials offered gives us insights about how trials are already used nowadays. Faugere and Tayi (2006) scale free trial versions based on the proportion of features added to the free trial and the evaluation time of the trial consumers have. Where other studies make a distinction between a fully functioning free version with a time lock (time locked) and a free version with limited functions without a time lock (limited version) (Cheng et al, 2012; Dey et al, 2013; Heiman & Muller,1996).
Time locked
Time locked free trials can be used for reducing the uncertainty of consumers by providing the consumers to learn about the complete functionality of a trial software. As discussed earlier, this can also lead to cannibalization by risking the demand of the commercial version. There is the possibility that short term users will not purchase the commercial version (Cheng et al., 2012). Cheng et al. (2012) analysed that the limited version is outperformed by the time locked free trial in the case of a moderate network effect. When there is an intense network effect, companies can use the limited free trial to utilize the positive network effect (Cheng et al, 2012).
Companies benefit from introducing the time locked free trial when the network effect is not very strong and when consumers preceding expectations are rather low.
Expectations about the functionality of the commercial software version can cause consumers to overestimate the commercial version. Offering a time-locked free trial becomes meaningless when consumers overestimate the commercial version and the time lock trial disappoints the potential consumer (Cheng et al. 2012). Thus companies should consider in which situation they do or do not to offer a free trial, this can have a negative influence on the willingness to pay for the commercial version (Cheng et al. 2012).
Dey et al. (2013) find that time locked may not be optimal, they explain that the trial is optimal when the rate of learning is large whether or not in the attendance of network effects.
Furthermore, they find that a time-locked trial improves the consumers’ willingness to pay, but the consumers have to learn about the software in a rapid way.
Limited version
The advantage of limited version free trial is that it enables companies to benefit from capturing the network effect of both trial users and buyers (Cheng et al, 2012). Negative consequences of the limited version free trial is of course cannibalization.
13 2.6 Price of trial and commercial version
Charging the right price is becoming more important factor within consumers adoption of technology and is an essential part of the trial strategy for companies (Wang et al., 2011). When companies setting prices of commercial versions too high, they risk losing demand for the commercial version to free trial users (Haruvy, 2001). Low prices of the commercial version makes consumers choose the commercial version, since the price is affordable. A consideration is made by the consumer whether to adopt a new technology based on the benefits they receive from the technology, hence the perceived value compared to the price (Wang et al., 2011).
The consumers’ willingness to pay increases when the quality of the software is high. So when the quality of the free trial version is significantly lower compared to the commercial version, companies can charge a higher price for the commercial version (Cheng et al., 2010).
Cheng et al (2012) state that companies can charge higher prices, when the consumers have high expectations about the software. The optimal price of the commercial version should be set to the maximum consumers are willing to pay for the commercial version (Haruvy, 2001).
Consumers have the ability to determine the value of a trial version after using the software which lead to expectations about the usability . Reviewing the literature above, for this study it is expected that consumers are willing to pay higher prices for higher quality. Thus it is expected that offering a trial with higher quality increases the willingness to pay, since the difference with the lower quality trial is increasing.
2.7 Time of trial
Time is an important device for companies, it can be used to persuade the consumer into making the decision to buy or not (Faugère et al, 2006). Time also has a significant influence on the optimal profit of a company, which makes it an important strategic decision for the company (Cheng et al., 2012), Faugére et al. (2006) & Heiman et al. (2001) show that effective prolonged learning makes the service more valuable for the consumer, hence the consumer gets more information out of the trial with extra time. Increasing time has also its negative effect, an increase in the trial period results in an increase in the number of free riders (Cheng et al., 2012). Dey et al. (2013) question that consumers might not have sufficient time to realize that they might lose value when the trial period is finished. Taking the theory into account, increasing the time is expected to have a positive influence on the perceived value of the consumer in this study.
2.8 Ad Avoidance
Ad avoidance can be explained as, companies allowing consumers to pay for the removal of
14 advertisements from an advertisement-based product (Tåg, 2009). So, Ad avoidance is the skipping of commercial content by using ad avoidance technologies (Anderson & Gans, 2010).
Free advertisement based products have a negative influence on sales of the commercial version without advertisement. Therefore, companies make a deliberation on the revenues generated from consumers who are paying for advertisement free content against the revenues of connecting consumers to advertisements (Tåg, 2009). When introducing a commercial version without advertisements, companies should increase the advertising quantity in the free version when consumers perceive a high disutility from advertisements. This will reduce the perceived quality of the free version, because it is perceived as more annoying when consumers have the ability to pay for removing the advertisements (Tåg, 2009). Anderson et al.
(2010), find that the consumers who continue using the free version are less aversive to advertising. Subsequently companies could increase the advertisement quantity in the free version.
Based on the discussion about the theory of ad avoidance, the advertising quantity is expected to negatively influence the utility of consumers of a trial version. Because advertising content is unwanted content, it is expected that consumers prefer to skip advertisements.
2.9 Channel Choice
Service channels are concerned with when, where and how the product is delivered to the consumer (Lovelock, 2000). When offering multiple service channels, the provider wants to offer a channel mix that satisfies the consumer the best way possible (Montoya-Weiss, Voss &
Grewal, 2003). Consequently, understanding how consumers evaluate these service channels is important. One of the most important dimension of service evaluation is usefulness (Patricio, Fisk & Cunha, 2003; Montoya-Weiss et al., 2003). Usefulness can be described as, the degree to which a person expects that using a channel will enhance performance (Davis, 1989;
Loiacono,2000). Patricio et al. (2003) described that the expectations about the performance, thus satisfying ability, of each channel determines channel choice and use.
Consumers generally use more than one channel, as not one channel satisfies all their needs (Patricio et al., 2003; Montoya-Weiss et al., 2003). If consumers have the ability to choose between different channels, they will use multiple channels in a complementary way. Therefore it is important to understand the contribution of each channel to customer satisfaction within the complete service offering and improve the complete service offer instead of each channel separately (Patricio et al., 2003).
Services can be used via different channels, hence smartphone, tablet, computer and
15 even smart TV’s have applications to use the service. All channels have their have their own satisfying ability. For this study it is necessary to understand which channels are the most valuable to the consumer of a trial version.
2.10 The Utility of a trial
As mentioned earlier, offering trials has two main effects increasing sales and cannibalizing sales simultaneous (Faugère & Tayi, 2006). This indicates that the relation between offering a trial version and sales is not unequivocally positive. The theory about quality or utility of a trial is insinuating that there is a relation between the quality of a trial and the intention to start using the paid version.
First of all, offering a trial will add extra value to product. Kempf and Laczniak (2001) found that companies who provide samples have higher expected value for the product, which increases the purchase intention for the paid version. However the quality of the trial should be carefully considered, taking into account that consumers base their expectations on the quality of the trial and subsequently adapt their willingness to pay for the commercial version to these expectations (Heiman, McWilliams, Shen & Zilberman, 2001).
Goering (1985) investigated how the knowledge gained from trials affects the expectations about the commercial version. She found that if the average quality of the trial is relatively low to the expectations of the consumer, the consumer will lower their expectations for the commercial product (Goering, 1985). For this study, it is expected that when the utility perceived from the trial version is low, the expectations about the paid version will be low.
Consequently the expectations decrease the probability that a consumer of a trial version will start using the paid version. The quality of a free trial should be at a certain level to add value for the users, otherwise there is a possibility that consumers might choose for the competition, so the effectiveness of a trial is related to the quality (Hahn, Park, Krishnamurthi & Zoltners, 1994).
Tu & Lu (2006) found that high quality of a trial is positively related to the users evaluation of the commercial version. Goering (1985) has similar findings and found that when the quality of the trial is high compared to the expectations about the trial, consumers are willing to pay a higher price for the commercial version. These findings of Goering (1985) and Tu et al.
(2006) can be questioned, Haruvy et al (2001) state that the trial version should have a high standard quality to be able to make it attractive for consumers to use the trial. Yet, there should be a significant difference between the quality of the free trial compared to the quality of a commercial version (Haruvy et al., 2001). Cheng et al. (2010) also explain why high quality can
16 lead to cannibalization. When consumers accept fewer requirements on quality and functionality of software, when comparing the trial version to the commercial version, he or she will hold on to the free trial version with its limitations (Cheng et al, 2010) and therefore cannibalizing the commercial version. Thus consumers will have no incentive to switch to the commercial version when providing a high quality trial and subsequently cannibalization takes place. Based on the discussion about high quality of the trial version, for this study it is expected that when the utility of the trial version is high, the intention to start using the paid version is low.
When using the findings of Cheng et al. (2010) and Haruvy et al. (2001), there can be concluded that companies should offer moderate quality. This can be done by offering a trial version with some limitations and thus setting or reducing the trial version to moderate quality to create distance with the commercial version. Reduction of the quality of the free trial, compared to the commercial version, is necessary to prevent consumers to always favouring the free trial version (Haruvy et al., 2001). Quality reduction implies reducing the benefits of the product, examples are reducing the number of features, offered services or reduction in time. These will result in a lower expected value and lower utility of the free trial (Haruvy et al., 2001; Faugère et al., 2006).
Based on the findings described above, it is expected that moderate utility of the trial version will make the commercial version more attractive by creating distance with the commercial version regarding quality. Furthermore, by setting the quality to a moderate level, the consumers can still learn from it and perceive value from the trial version. The difference in quality, when comparing the trial version with the commercial version, is expected to generate an incentive to start using the paid version.
Derived from the discussion above, there can be concluded that there is no linear relationship between quality and the intention start using the paid version. For this study the relation between quality and intention to start the paid version, based on the preliminary discussion, is expected to have three stages: Low utility/expectancies and low intention to start using the paid version, moderate utility/expectancies and an increased intention to start using the paid version and the final stage high utility/expectancies and low intention to start using the paid version. The expected functional form for this relationship is therefore an inverted u-shape and leads to the following hypothesis:
H1: The relationship between the utility of the trial version and the intention to start using the paid version will show an inverted u-shape.
17 2.11 Conceptual model
H1
Figure 1: Conceptual model
asd
Intention to purchase commercial version Ability to share
Price
Time
Advertising Quantity
Channels
Quality
Perceived
Utility Trial
18
3. Methodology
In this chapter the method of data collection will be explained. First the method used will be described. Then the population will be explained and the channels by which the results will be gathered will be clarified. Finally the analysis plan will be illustrated, to explain how the gathered data will be analysed. Figure 2 displays the outline of this research, blue indicating the procedure of data gathering and green indicating the procedure of data analysis.
Figure 2: Outline of the research
. 1. Qualitative
research
Determine the attribute levels for conjoint experiment2. Conjoint experiment
Get data for creating latent segments, generating the relative importance of the attributes (used in survey) and
willingness to pay.
3. Survey
Get insights about the intention to 1. purchase the paid version, 2. keep using trial version 3. to stop using trialversion, when using different attribute levels.
1. Relative importance of
attributes
Get insight in which attributes are the most important for consumers
2. Computing the model for
the trial
Determining the optimal trial and determining the value of current trial versions.
3.
Segmentation analysis
Determine the differences in preferences regarding the use of the trial version for different segments.
4. Anaysis of intentions
Get insights on what effect reductions have on the intentions.
19 3.1 Method & plan of Analysis
This study will start with qualitative research on existing music streaming services. The qualitative research will give insights on how music streaming services use attribute levels.
Understanding the different levels of the attributes is required to come up with the levels for the conjoint experiment for this study. The findings of the qualitative study can be found in table 2.
3.1.1 Conjoint Experiment
The findings derived from the qualitative research will be used to design an conjoint experiment, this will be a choice based conjoint experiment (CBC). The conjoint analysis will identify the preferences of the consumer regarding the attributes and give an impression about what consumers are willing to pay.
The levels within the CBC experiment are balanced across the attributes, each attribute has four levels. By balancing the levels across the attributes, the number-of levels effect is prevented. All attributes have four attribute levels. This design is applied to prevent giving consumers the impression that one attribute with 5 attribute levels is more important in this research than an attribute with four attribute levels. All attribute levels are designed to be mutually exclusive of other levels. The choice design uses minimal overlap, therefore the alternatives within the choice set are set to be as different as possible from one another (hence minimal overlap), so all attribute levels are shown once in each choice set. This is applied to every choice set in order to get new information about the consumers preferences. The experimental design of the conjoint analysis is displayed in table 2.
Table 2: Experimental design of attributes and attribute levels
Attribute
Channels Mobile Phone application
Tablet application TV application PC
Time 15 days unlimited streaming, then reduced to 12 hours a month
30 days unlimited streaming, then reduced to 12 hours a month
60 days unlimited streaming, then reduced to 12 hours a month
Unlimited streaming
Nr. of songs 5 million 10 Million 15 Million 20 Million Quality 80 kb per sec 160 kb per sec 240 kb per sec 320 Kb per sec Advertising
quantity
No commercials Every 2 songs Every 4 Songs Every 6 Songs
Extra’s Sharing music with friends
Offline availability of music
Creating playlists Connection with HiFi systems ( e.g. sonos)
Price € 0,- Pay € 1,99 Only
once
Pay € 2,99 Only once
Pay € 3,99 Only once
20 There are 7 attributes with each 4 levels, showing 16.384 possible trial designs. The experimental design of the study will therefore be a fractional factorial design, which indicates that only a subset of all the possible attribute levels combinations are used in this study. This study will use random allocation of the stimuli to choice sets, this will be done by using preference lab to generate the data. Preference lab will randomly select an attribute level from each of the attributes in table 2 and assign these to one of the 3 alternatives in the choice set (alongside the non-choice option). Preferencelab will then record which attribute levels are presented and will store these, together will the selection of the participant, into a database. All attribute levels are displayed 2200 time during the survey, which makes the design balanced.
The choice based conjoint (CBC) consists of 4 alternatives per choice set, hence 3 options displaying attribute levels and the none choice option. The non-choice option is added to the study to measure the price sensitivity of the participant. Consumers who are not willing to pay for products displayed in a choice set, have the ability to choose for the non-option. When participants select the non-choice option, they enable us to calculate the lower boundary of utility consumers are willing to pay for. The participant has to fill in 16 complex choice sets, this might cause the participant to be become fatigue. To prevent the fatigue effect and consumers not filling in the entire CBC, an incentive alignment will be used. An incentive alignment is used to motivate the participants by initiating the consumers that amongst the participants one participant will receive an amount of 25 euro’s.
The CBC will be distributed online and on social media. The CBC experiment can be distributed by providing the population the link directing the participants to the conjoint experiment on Preferencelab. The participants are instructed that they participate in a research for developing a new trial version according to the preferences of consumers for a music streaming service. By doing this the real purpose of the study is hidden, resulting in participants showing their real preferences. The CBC will start with questions about the demographic information of the participant, after that the sixteen choice sets will be presented to the participant.
3.1.2 Conjoint simulation
Preferencelab gives researchers the ability to simulate the CBC experiment before distributing the CBC to the participants and applying the actual experiment. A simulation is used to see if the estimates from the simulation have the same utilities as the utilities that were previously determined. By performing this simulation, Preferencelab adds weights to the attributes which can be compared to the predetermined attribute values.
21 For the simulation is assumed that 200 participants contributed in the study and 16 choice sets are used. The results of the simulation are displayed in appendix 3. When comparing the predetermined utilities with the utilities derived from the simulation, there can be concluded that the values of the simulation are of higher value. However, when comparing the ranges between the attribute levels, the results show similar distances. All attributes show significance and are expected to have explaining power. Hence, these results are based on assumptions of the attribute levels, however the results give prospects that the results of the actual study will be useful as well.
3.1.3 Model selection
Consumers base their choices on the overall utilities of alternatives they have. Products or services are combinations of attributes, where consumers attach utilities to the different attributes. Together they form the utility for the product, therefore creating the systematic utility(
sum of all utilities for a product). The systematic utility will be used in this research within the random utility model. The random utility model can give systematic utilities of different combinations of products, this allows us to understand what the effect of changing the attribute levels has on the utility. The random utility model for the expected utility for product i is explained as followed:
Û
i= V
iWhere Û= expected utility
And V= rational utility (systematic utility component)
The rational utility (systematic utility component) for product i can be explained as followed:
V
i=∑
Where k= (1,…,K) number of attributes
x= a dummy representing a specific attribute level of product I And β= utility of the consumers for attribute level of k
For this research the model for designing trial versions will be made based on the random utility model. These models will be evaluated by using information criteria. Information criteria penalizes the likelihood for the number of parameters. Information criteria is used to see how
22 well the aggregate model fits, the model fit will be evaluated by looking at different information criteria and choose the model with the minimum AIC,BIC or CAIC (these are not compared across criteria). The BIC and the CAIC have a higher penalties for complexity.
3.2 Segmentation analysis
Since consumers differ, this research will investigate the difference between participants within the conjoint experiment by performing a latent class analysis. The latent classes are based on the difference in relative importance placed on the different attributes. A latent class analysis assigns consumers within the dataset to latent classes based on the probability of belonging to that particular class.
First the amount of classes will be determined by examining the information criteria, hence BIC and/or CAIC, to see which amount of classes is optimal. When the optimal amount of classes derived from the results of the information criteria is too high, lower amounts of classes will be considered at by checking the Wald(=) statistics (indicator to check whether the preferences significantly differ from each other). When all the attributes at lower amount of classes also show significant Wald(=) statistics, it is possible to use less classes.
After the number of latent classes are determined, a latent class analysis will be performed. The consumers are assigned to the latent classes according the latent class model for categorical indicator of Vermunt & Magidson (2004). The underlying idea of latent class analysis and the assignment of the consumer is explained as following.
The basic latent class model explains that the probability of finding response pattern y (P(Y=y)), is a weighted average of the probability that that response belongs to a specific class C indicated by P(Y=y|X=x). The statistical formula of Vermunt & Magidson (2004) is explained below:
( ) ∑ ( ) ( | )
With:
Y= Entire choice set
y= Response pattern regarding the preferences of the trial of the music streaming service (y= Y1+ Y2 +.. Yk
).
Yk= Choice for attribute indicated by уℓ.
k= The number of attributes k=1,2…k.
23 уℓ = The attribute level (уℓ=1,2... Dℓ).
C= The number of latent classes (x=1…C).
x= Latent class a person belongs to.
P(X=x)= Probability a person belongs to latent class x.
P(Y=y)= Probability of obtaining a response pattern.
In Latent Gold the option to generate the posterior membership probability will be used, this is the probability of a consumer belonging to a specific latent class x. The posterior membership probability is used to assign consumers to the latent classes and will be explained by the Bayes rule (Vermunt & Magidson,2004):
( | ) ( ) ( | ) ( )
After generating the posterior membership probability, the consumers will be assigned to the latent class x with the highest P(X=x|Y=y). The posterior membership probability will be generated in Latent Gold and will be used to assign the consumers to the latent classes.
Insights about differences in preferences between the latent classes can will be made at this point and are based on the relative importance of the attributes regarding the participants of the conjoint experiment.
3.3 Survey
The utility of the separate attributes and attribute levels are derived from the CBC analysis. The questionnaire will be used to discover if, the dependent variables, 1. the intention to purchase the commercial version, 2. the intention to keep using the trial version or 3. the intention to stop using the trial version will change.
The attributes channels, advertising quantity and price will be used within the survey, since these have the ability to interpolate and extrapolate and are the most important attributes according to the conjoint analysis. For each attribute, there will be a separate questionnaire.
The participant will be participating in a survey where only one attribute will be set to particular attribute level. By performing a mixed subject design the participant is not aware of the other attributes being researched and cannot weight their decision against other attributes (manipulation). The results from these questionnaires will be more reliable and more valid when
24 using a mixed subject design. Furthermore, the survey will use a between groups design, the participant will only take place in one condition.
The first part of the questionnaire exists of questions about the demographic information of the participant. After the demographical part, the existing trial version and paid version of a music streaming service are presented to the participant. Then the participant is introduced to different trial versions where only one attribute is changed, this attribute is set to the attribute levels in table 2. The utility of the trials presented are different from the existing trial version. After each presented trial version, the participant is asked to answer what the changes of the trial version will do to their 1.the intention to purchase the commercial version, 2. the intention to keep using the trial version or 3. the intention to stop using the trial version. To investigate the relation between intentions and the reduction in utility, a Likert-scale with a seven point scale will be used in the questionnaire.
From the questionnaire the average values of the intentions can be calculated for a particular trial versions. Next to that, the utility of these trial versions are calculated to be able to plot the intentions for a particular trial version against the utility of this trial version. By using different attributes and attribute levels, there can be investigated if particular attributes are more effective in increasing the intention to start using the paid version when reducing the utility of these attributes.
3.4 Population
The participants in the conjoint experiment (N=180) and the questionnaire (M=150) are exclusively users of the trial versions of music streaming services. Participants are exclusively users of trial versions of a music streaming services. It is needed to get insights on the preferences regarding the attributes of actual trial users, to be able to understand how to move users of trial versions to the paid version. Participants for this study will be invited via social media and by personal invitations, links directing them to the CBC experiment and the questionnaire will be provided to the participants. Convenience sampling, hence snowball sampling, will be used to gather the participants for this research. The participants in this study will be mainly students, since these are, in large numbers, mainly expected to be using music streaming services. Prior to the survey, the participants will be instructed that they will take part in an online research on consumer preferences. The participants will be given an incentive alignment of that one of the participants will receive 25 euro’s. The participants are also instructed that the results only will be used for this research.
25
4. Results
The results gathered from the conjoint experiment and the survey are presented in this chapter.
The subsections will provide results of the conjoint analysis, latent class analysis and the influence of reduction on intentions.
4.1 Description of the population
In total 180 people participated in the conjoint experiment, all participants are users of music streaming services. Of these participants 55,6 % is male and 44,4% is female. The average age of the participant was 26 years. When looking at the descriptions of these participants, we see that the largest part of the population is student, hence 64,4%. The other part of the population can be defined by full time workers (22,8%) and part time workers (10,6%).
When looking at the behaviour of the population we can see that almost all participants use a smart phone (93,3%) and PC (92,2%). When looking at the other channels, we see that 53,3%
is using a tablet and 25% is using a smart TV.
Description of the population
Nr of participants 180
Gender 55,6% Male 44,4%
female
Job Student
64,4%
Full time workers 22,8%
Part time workers 10,6%
In search of employment 2,2%
Percentage of participants that in general uses a:
Smartphone 93,3%
PC 92,2%
Tablet 53,3%
Smart TV 25%
Table 3: Description of the population
4.2 Conjoint analysis
Before generating the outcomes of the conjoint analysis in Latent Gold, advertising quantity is transformed to amount of advertisement an hour, this is done to be able to change the measure to scale. For generating the new values, an average of 3,5 minutes a number is used (Spotify, 2014). Therefore the new values are, 8,57, 4,29, 2,85 and 0 advertisements an hour, these are somewhat strange numbers however by making the attribute linear it gives the opportunity to interpolate and extrapolate.
26 4.2.1 Estimation of the model
First, all results are computed in Latent Gold with all variables labelled nominal, hence this means all the attribute levels from table 2 are used. To be able test whether all attributes can be used in a model, the Wald statistic will be used to test the true values of the attributes. All Wald statistic of variables but extra option (Wald= 5,1529, p-value= 0,16) show significance, indicating that extra option cannot be used within the statistical model to estimate the utility of the consumer. However, extra option is a part of the trial version and therefore it cannot be left out of the model. Thus, extra option will be used in this research to estimate the utility with the knowledge it is not significant.
Attributes Wald p-value Mean Std.Dev.
Channel 180,5456 6,70E-39 0,3995 0 Free streaming 131,7362 2,30E-28 -0,2248 0 Number of songs 12,81 0,0051 -0,1298 0 Quality of the music 65,0233 5,00E-14 -0,295 0 Advertising quantity 113,6206 1,80E-24 -0,215 0 Extra options 5,1539 0,16 -0,0307 0 Price of the Trial 453,9296 4,60E-98 0,8108 0 None option 33,9525 5,60E-09 -0,1306 0
Table 4: output conjoint analysis with all variables nominal
The next step is to create graphs with all attributes nominal. By creating graphs, a clear view on the effect of the different variables on utility is presented, these give an indication if an attribute could have linear relationships. When looking at graphs 1, 2 and 3, there can be seen that the attributes price, advertising quantity and music quality might have linear relationships. To test if an attribute should be used as linear within the model, different models are generated in Latent Gold. The decision whether to use the attributes as linear will be made on the information criteria.
27
Graph 1: Utility price Graph 2: Utility Advertising quantity Graph 3: Utility Music Quality
To test the different models, the results need to be computed in Latent Gold again. Different models are generated with price, advertising quantity and quality of the music being linear. The results are shown below.
Table 5: Information criteria of the different models
The information criteria, the hit rate and the R2 adjusted of the different models will be used in order to determine which model fits the best. First, we will take a look at the information criteria and indicate that the model with music quality as linear has the lowest value (BIC=7143,265 and CAIC= 7163,265). When looking at the hit rate, table 5 shows that the model with music quality as linear has the highest hit rate (hit rate= 45,62%).
A high amount of parameters always improves the R2, the hit rate and Log Likelihood.
The R2 adjusted is therefore a more adequate measure, since it penalized the Log Likelihood Model 1: All
Part- Worth
Model 2:
Price linear
Model 3: Music Quality linear
Model 4: Ad Quantity linear
Model 5: Price, Quality and Quantity Linear Log-likelihood LL (B*) -3518,71 -3535,31 -3519,70 -3532,18 -3549,54
LL (0) -3992,53 -3992,53 -3992,53 -3992,53 3992,53
R² 0,1187 0,1145 0,1184 0,1153 0,110955
R² (Pseudo-R²) 0,1131 0,1092 0,1127 0,1117 0,1075
R² adjusted 0,1132 0,1095 0,1134 0,1103 0,1069
nr of parameters 22 20 20 20 16
Chi-square 947,6299 914,4293 945,6497 920,6991201 885,9781
Hit rate 45,49% 45,38% 45,62% 45,76% 44,97%
Prediction Error 0,5451 0,5462 0,5438 0,5424 0,5503
BIC 7151,6706 7174,4854 7143,265 7168,2155 7182,1647
CAIC 7173,6706 7194,4854 7163,265 7188,2155 7198,1647
-0,4 -0,2 0 0,2 0,4
Utility
Music Quality
-0,4 -0,2 0 0,2 0,4 0,6
Utility
Adv. quantity
-0,5 0 0,5 1
0 1,99 2,99 3,99
Utility
Price
28 (B*) for the number of parameters. The R2 adjusted shows that the most variance is explained in the model with music quality as linear (R2 adj.=11,34%).
Taking the information criteria into account, there can be concluded that the model with music quality as linear is the best model. However, when looking at all the models, there can be stated that the models do not differ much from each other based on the hit rate and the R² adjusted. Interesting findings can come forward when using the model with advertising quantity, music quality and price as linear. It gives the ability to calculate willingness to pay and to interpolate and extrapolate for the attributes price, advertising quantity and quality of the music.
Besides the ability to generate important findings, a little variance is lost and the predictive power (hit rate) also differs very little when comparing the models, therefore this research continues with the model 5 with price, advertising quantity and quality of the music linear.
4.2.2 The model used in this study.
To define the relative importance of the attributes, the range has to be calculated for each attribute (presented in table 6). For free streaming, number of songs and extra options the part- worth utilities are used. The linear attributes had to be calculated by using the range between the highest and lowest value of the attribute. Price for example, the highest value is 3,99 and the lowest value is 0. The non-option is excluded, because the non-option cannot be used in designing a music streaming service.
When looking at table 6, there can be concluded that the most important attributes are price (29,3%) and channel (21,5%). Price has negative effect on the utility (
β
price= -0,3283) and the most important channel is smart phone (β
smart phone= 0,3967). Streaming (15,0%) and advertisement quantity (14,3%) are also attributes that consumers find important, in particular unlimited streaming (β
= 0,4505).Table 6: Relative importance of the attributes
When looking at the parameters of the attributes (appendix 5), the following model can be derived from the results is:
Channel Streaming Nr of
songs
Quality of the music
Adv.
quantity
Extra options
Price Total
Range 0,9615 0,669 0,2189 0,5326 0,6381 0,144 1,3098 4,4739
Rel.
importance
21,5% 15,0% 4,9% 11,9% 14,3% 3,2% 29,3% 100,0%
29
Û
i= ViVi = β1TV*DTVi + β2mobile*Dmobilei + β3smarttv*Dsmarttvi + β4tablet*Dtableti + β515days*D15daysi + β630days*D30daysi + β760days*D60daysi + β8unlimited*Dunlimitedi +β95millionsongs*D5millionsongsi + β10
10millionsongs *D10millionsongsi + β1115millionsongs *D15millionsongsi + β1220millionsongs *D20millionsongsi + β13sharing*Dsharingi + β14offline*Dofflinei + β15playlists*Dplaylistsi + β16Hifisystems*DHifisystemsi + β17Musicquality* Musicqualityi - β18adQuantity* AdQuantityi - β19Price* Pricei
Index:
i= Music streaming service x
Where:
Û
= the expected utilityV= systematic utility component
The perfect trial version according to the participants of the conjoint experiment is: Mobile phone + Unlimited streaming + 20 million songs + 320 Kbps + 0 commercials + offline availability and € 0,00. This perfect model is as expected, all the attribute levels are optimal and the consumer has the most value from the music streaming service.
4.2.3 Current model
To be able to get insights about trial design, this study focusses on the existing trial version/design of Spotify (see appendix 1). For this research the current model Spotify is used since it is most commonly used in the Netherlands. All music streaming services diver slightly, so insights from the use of the model of Spotify can contribute to literature and give managerial insights about how to use the different attributes within trial design.
The current model of Spotify is based on the following attribute levels, mobile phone, unlimited streaming, 20 million songs, 160 Kbps, commercials every 4 songs, creating playlists and price €0,-. The utility of the current model is displayed below.
Û
Spotify= V
SpotifyV
Spotify= 0,3967 + 0,4505 + 0,0927 + 0,0123 + 0,0022*160 - 0,0745*4,29 - 0,3283*0 + 0,1648Û
Spotify= 1,14939530 The utility of the non-option is: -0,1648. The probability of choosing the current model over the non-choice option in table 7 are calculated as following: EXP(1,137095)/( EXP(1,149395) + EXP(-0,1648))= 0,788 78,8%%. Table 7 gives an indication on what will happen to the probability of choosing the current model over the non-choice option when increasing the price.
€0 €0,99 €1,99 €2,99 €3,99 €4,99
current model 1,149395 0,824378 0,496078 0,167778 -0,160522 -0,488822 non option -0,1648 -0,1648 -0,1648 -0,1648 -0,1648 -0,1648
Probability 78,8% 72,9% 65,9% 58,2% 50,1% 42,0%
Table 7: Probability of choosing the current model over the non-choice option
4.2.4 Willingness to pay
By equalizing the utility of the current model of the trial version to the utility of the non-option (utility level they do not want to pay for), gives an indication of what consumers are maximum willing to pay for the current state of the trail version. The calculation of the willingness to pay is shown below:
The expected utility of the current model where the price is unknown:
0,3967 + 0,4505 + 0,0927 + 0,352 +0,0123 – 0,319605 (-0,3283x)+0,1648 The utility of the non-option is: -0,1648
By equalizing both utilities we are able to calculate x, which in this chase is price. First we need to calculate all utilities but price. The equation arising when equalizing both utilities :
0,3967 + 0,4505 + 0,0927 + 0,352 +0,0123 – 0,319605 (-0,3283x)+0,1648 = -0,1648, which is the same as: 1,149395+0,1648=(0,3283x). Now we can divide the total equation, 1,149395+
0,1648=(0,3283x) by 0,3283 to come up with: 4,0030=x. 4,0030=x means that the maximum willingness to pay of an average consumer for the model of music streaming service of Spotify is €4,00.
To show what happens when increasing and decreasing the price, graph 4 is computed.
When looking at the graph we can see that price has a big impact on the utility. When increasing the price, the utility approaches the non-option line. The point where the lines cross is the cut off point for this model, this indicates that the consumer will prefer not to buy this product.
