Happily paying too much -a CONJOINT study on coffee consumption in a subscription model

(1)

Happily paying too much

-a CONJOINT study on coffee consumption in a

subscription model

Marcus Axel Kelley

(2)

Happily paying too much

-a CONJOINT study on coffee consumption in a

subscription model

Marcus Axel Kelley

Faculty of Business and Economics

Marketing Intelligence

Master thesis

Completion date: 17th_{June 2019}

Address: Lierstraat 51, 9702PB Groningen

Phone: +47 4654 4439

Email: m.a.kelley@student.rug.nl Student no. s3738353

(3)

Management summary

Consumers can subscribe to nearly everything nowadays. While your mobile and music streaming services has been around for some time, actors like Amazon is facilitating the next generation grocery shopping to be as seaming- and effortless as possible by offering you to subscribe to the groceries you are most dependent on. The flat-rate pricing model has been a popular plan to ease the consumer’s “pain” of paying. As the customer’s relationship with the brand develop, literature has shown that customers are prone to overspend on a fixed fee in a subscription for products and services they do not take the full advantage of- the flat-rate bias.

The following research question is: “Is there a stronger flat-rate bias for the focal product

when primed with a hedonic consumption goal and how do the different pricing models affect this bias?” To test this, we conducted a CONJOINT study using the attributes price, quantity,

and additional units to see if we could find evidence of any flat-rate bias for coffee in a subscription model. We primed respondents with either a hedonic or utilitarian scenario to check whether hedonizing the product would increase adoption or price sensitivity.

We find evidence of that there is a flat-rate for some of the respondents and that there is a large variation of price willingness in the category. By conducting a latent class segmentation, we identify a segment prone to adopt a product combination and we identify what attributes being most approved for this set of customers. Due to this, marketing and business managers should be particularly aware of the importance of segmentation and how to position their brand to approach the right target group in this category. We construct a multinomial logit model to better understand how consumers perceive the attributes and illustrate at what level the revenue maximizing price is for the preferred product configuration.

(4)

Acknowledgements

I would like to express my deep gratitude to my supervisor on this project, Dr. Felix Eggers. His constructive and valuable advice during the work on this master thesis have been very much appreciated. Patient guidance, useful critics and enthusiastic encouragement have been of great motivation for me during the completion of this research.

(5)

Since the early 2000s we have been witnessing an evolution within digital eCommerce that have transformed in many ways how we conduct business online. Where we subscribed to informational products like newspapers or books and phone plans from a telecom company at the beginning of the digital transformation, we can now enjoy and apply subscriptions to a wide variety of product and services. The digitalization opened opportunities for streaming services like Spotify and Netflix and more recently have started to flourish within physical goods. In a recent report McKinsey finds that 51% subscribes to at least one product or service online (2018), also stating that the eCommerce subscription market has grown by at least 100% annually the past five years. We now see some of the biggest market leaders embrace this transition, reaching consumers through both online and offline platforms. Adobe Systems went away from their licensed software system in 2011 to an annual online

subscription and cloud infrastructure. Moving their $25 stock in 2011 to $264 in March 2019 (Tzuo & Weisert, 2018). Apple conducted a similar move for their music platform and iTunes, to take competition head on against Spotify who were dominating the online music industry (Dredge, 2015). This allowed customers to subscribe to their Apple Music service for a charge of $9,99 instead of paying $1 for each new song they wanted to acquire. One of the many changes coming offline among many in the automotive industry is major brands like Mercedes-Benz offering subscription-plans as an alternative to leasing were customers pay a flat fee for the use of their vehicles including insurance, maintenance, licensing and taxes (Zhang, 2018). Several other actors as well have launched similar services but in lesser scale like the Care By Volvo, Audi Select and Access by BMW where electric automobiles is increasing in interest but not fully dominating this area yet (Delbridge, 2019).

(7)

customers are entitled to between 5-15% discounts when subscribing to this service without being charged for shipping.

Together with the newly Unilever acquired Dollar Shave Club, Amazon “Subscribe & Save” offer services among the highest long-term subscription rates among consumers on the North-American market (Chen et al., 2018). Dollar Shave Club has made great success in shipping out razor blades for a flat-fee a month, far cheaper than the dominant market leader Gillette.

While digitalizing FMCG (fast-moving-consumer-goods) and the grocery industry opens for a wide span of digital business opportunities, there are some well-known obstacles and barriers for actors who want to make their mark. Distributing, selling or dealing with groceries have often been associated with small margins and low profits. Offering a new service area within a subscription model where the consumer expects a more convenient customer-journey may create managerial difficulties in terms of pricing.

Pricing-models are heavily studied in the field of marketing and there are several different pricing plans for subscription-based services or goods in varies of industries. For

subscriptions a flat-rate is often preferred since this indicate that the customer’s payment does not vary too much and are predictable. The flat-rate pricing model charges the customer a fixed fee regardless of the usage (Uhrich, Schumann, & von Wangenheim, 2013). This is often the case for online subscriptions like streaming music from Tidal and Spotify or movie and series from Netflix or ZiggoGO. For offline providers this becomes more comprehensive since the product is not delivered immediately and often is often affected by variables like freight, extra fees and difficulties in terms of ordering extra usage through more products. This opens for further studies on how pricing models can be optimized for subscription models for offline products and tangible goods different from online services.

Pay-per-use pricing uses a predefined price per consumption unit being a variable pricing

(8)

when the consumer favors the flat-rate even though the pay-per-use option would be a cheaper option in regard to actual usage. An illustration of the mentioned pricing plans may be seen in figure 1.1.

Pay-per-use Flat-rate Limited flat-rate Cost-cap tariff

Figure 1.1- Relationship between price and units in pricing plans mentioned in literature.

While most studies have focused on services or consumption through eCommerce platforms in the flat-rate bias and self-regulation literature, there are some exceptions. Nunes (2000) and Uhrich et al. (2013) believes that since the flat rate bias implies a relative preference for fixed versus variable costs their findings might be applicable for purchased goods and not only services. Nunes (2000) did a study consisting of grocery products in bundling combinations and flat-rates from $35. Making his study not able to identify the flat-rate bias for individual products. Wertenbroch (1998) did a field experiment testing subjects self-regulation for products of a vice (“wants”) and virtues (“shoulds”) characteristics. He found that people tend to forgo savings from price promotions through quantity discounts, and that through products like cigarettes they are willing to pay a price premium for being in control of their own self. By this revealing a pay-per-use bias since the customers were willing to pay more for buying a separate package of cigarettes each time and not save money on buying a bundle of them immediately. His study was different from ours in a way that Wertenbroch did not have a subscription-based model focus as well as the consumption goals of vice/virtues differ from hedonic/utilitarian as in explained in chapter 2.2.

This research has theoretical and managerial implications for the marketing field. There is a gap in current research for hedonic and utilitarian consumption of groceries and commodities within FMCG. Uhrich et al. (2013) finds that service providers who stress the hedonic

(9)

hedonic than utilitarian products and services. Brand managers may in regard to these findings, create a hedonic dimension to their products to increase revenue streams, sales and loyalty among consumers. There has been shown evidence of flat-rate biases by Nunes (2000) and Wertenbroch (1998) in field studies where bundling of products as well as other

consumption goals have been manipulated. Though, neither have tested a limited flat-rate on groceries in a hedonic or utilitarian setting. Thus, our research question is:

Is there a stronger flat-rate bias for the focal product when primed with a hedonic consumption goal and how do the different pricing models affect this bias?

Companies offering a flat-rate can benefit from predictable revenue streams for customers. A flat-rate bias may directly increase revenues and may be more profitable than the pay-per-use revenues would have been. Managers can do wisely when knowing at to what line they can manage the pricing strategy of the company’s subscription services to hit optimal consumer preferences.

1.1 Thesis outline

Through the rest of this study we will introduce how eCommerce is embracing groceries and explain the antecedents of the flat-rate bias as well as how consumption goals affect

consumers. We will then proceed by explaining the research design and how the CONJOINT survey is composed. So, present the data and build an optimal model based on the results from participants in the survey. Analyzing how consumer differences in their preference for coffee in a subscription setting. We will then test our research question and hypothesis and present the theoretical and managerial implications of the findings from this research.

1.2 Subscriptions embracing online retailing

(10)

digitally connect directly with consumers. Online shopping of groceries is nothing new and due to the increasing popularity of subscriptions through eCommerce it is of high relevance to investigate how consumer react to different pricing models for this kind service and whether consumers are willing to pay a price premium for the products through a flat-rate bias. Subscriptions within eCommerce is nothing new and has had a tremendous growth since the early 2000s. Consumers may be willing to pay a price premium for their subscriptions that exceeds their own consumption which may be of interest for actors within FMCG that are distributing groceries to consumers. Especially finding the right pricing-plan may affect how strong this flat-rate bias is and thus why it is of relevance of managers. The consumers consumption goal is also of importance since the customers perception and purpose of the initiating a subscription or using the products of the company may differ between either hedonic or utilitarian.

2.0 Research background

2.1 Flat-rate Bias antecedents

There have been conducted several studies on the phenomenon and K. E. Train, McFadden, and Ben-Akiva (1987) was one of the first to report proof of the flat-rate bias through their study on households consumption of telephone service options. A quite extensive study by Mitchell and Vogelsang (1991) also investigating telephone usage data of 151,000 households found that up to 45% of consumers with a flat-rate paid too much. Lambrecht and Skiera (2006) has divided the flat-rate bias into four effects that classify cognitive and motivational explanations: insurance-, overestimation-, taximeter- and convenience effect.

Insurance effect. Risk averse consumers may choose a flat-rate to avoid charges for

large variation for their monthly usage. Such consumers may struggle to assess or predict their own usage and self-control.

Overestimation effect. May occur easier for consumers that have a habit of

overestimate individual usage. Lambrecht and Skiera (2006) suggest that consumers who perceive minimum and maximum usage as particularly high are more likely to choose a flat-rate. This is supported by Nunes (2000).

Taximeter effect. Refers to the consumer does not want to see “their” taximeter tick

(11)

Convenience effect. The consumer seeks to avoid having to deal with calculating the

benefits of a pay-per-use plan vs a flat-rate. Could occur if an option has several alternatives that demands a lot of cognitive work and the customer chooses the easiest option, whether that means the most promoted, cheapest or the less complexed (Uhrich et al., 2013).

Fewer researchers and studies investigates a “pay-per-use” bias- that is a preference for a pay-per-use pricing plan when a flat-rate would be a cheaper option. Wertenbroch (1998) finds that the pay-per-use bias is more imminent when involved with low-consumption products like cigarettes, where the consumer tries to rationalize own consumption to less than what it is in reality. Together with the flat-rate bias the pay-per-use bias contradicts the economic assumption that consumers pick the pricing plan leading to the lowest billing rate for the consumer’s own usage (Lambrecht & Skiera, 2006).

Lambrecht & Skiera (2006) suggests that pre-commitment to certain types of goods that consumers want to predetermine a particular amount of consumption might also affect tariff choice. Pre-committing consumers might tend to overestimate their own usage and result in an even stronger flat-rate bias than seen in services. Consumers also want to avoid variations in their billing rate and decouple themselves from having to deal with the “liability” of processing the amount payed on a regular basis. They also find that customers with a flat-rate bias are less likely to churn than customers which possess a pay-per-use bias. When deciding whether to churn or switch service offer, consumers with a flat-rate bias tended to switch to an option of the same service provider (Lambrecht & Skiera, 2006).

In relation to the taximeter effect Prelec and Loewenstein (1998) finds that consumers prefer to either pay up front before consuming products and services or getting paid by work after it is done. By doing so the pleasure is greater and can almost be enjoyed for as it was free of charge and already paid for. So, by paying a flat-fee the payment is decoupled from consumption since the cost is already accounted for (i.e. in the start of each month). Findings indicated that people preferred a flat-rate compared to a pay-per-use tariff in terms of

pleasure (Lambrecht & Skiera, 2006; Prelec & Loewenstein, 1998).

(12)

DellaVigna and Malmendier (2006) provide evidence for the flat rate bias when investigating 7,978 customers with gym memberships. They find that members with a flat-rate of $70 per month or more forgoes $600 of savings due to frequency of visits to the gym compared to a 10-day pass deal would have provided. The authors state that overconfident agents overestimates their own attendance at the gym as well as their own probability of making their training goal. The same group also delays their cancellation. They also find that churn rates are lower for the flat fee compared to the annual payment solutions implying that customer retention may benefit of a flat-rate model. This is supported by Lambrecht and Skiera (2006) who shows that a flat-rate bias has not negative impact on customer churn. This encourage the proposal of the first hypothesis as we may claim that the flat-rate bias is applicable for physical goods as well as services:

H1: There is a flat-rate bias for the focal product

Further, we will look into how products have different consumption goals and how they differ between consumers when making choices.

2.2 Hedonic and utilitarian consumption

For a marketer, being able to identify what kind of consumption goal the consumer has when subscribing to the company’s products and services is of great importance. To better

understand the consumption behavior of consumers we can divide purchasing goals into either

hedonic or utilitarian consumption. The former refers to sensations derived from experiencing

goods and the latter abilities of functions from using the products (Batra & Ahtola, 1991, p. 159; Voss, Spangenberg, & Grohmann, 2003). Typical hedonic consumption may be activities the consumer think is fun, enjoyable and find pleasuring. This can be streaming a film at Netflix or listening to Spotify playlists. Utilitarian consumption refers to functional needs to solve. This may be for instance a subscription to underwear, socks or razor blades.

(13)

benefits versus costs weights in its favor (Okada, 2005). Hedonic and utilitarian consumption will on the other hand be a trade-off between consuming the one or the other. Literature indicates that of hedonic products and goods can be linked to relative vices, though that comparison is not the case for utilitarian and relative virtues consumption (Okada, 2005).

There has been conducted several studies on how attitudes towards hedonic and utilitarian brands form and how they are consumed. Wakefield and Inman (2003) argues that the brand can be given a competitive advantage since their findings indicate that making the products more hedonic may decrease price sensitivity among the consumers. The authors also find that consumers would like to minimize their costs related to services and products that they perceive perform to functional purposes (utilitarian). It is important to distinguish that a consumption is not necessarily on the end of a continuum. For instance, subscribing and using premium tea might fulfill two different needs, both the functional thirst and nutrition as well as hedonic pleasure.

Uhrich et al. (2013) investigated how the flat-rate bias affect services and how the utilitarian and hedonic consumption goal of the customer may affect this bias. They find that service providers should try to benefit from the positive effect of hedonic consumption on the flat rate bias when tempting to increase their profits and that this effect is bigger for hedonic compared to utilitarian consumption. The same authors find taximeter effect to be significant in all of their three studies and stands out as a robust driver of the flat-rate bias. In regard to price plan decision for services Uhrich et al. (2013) finds that the more hedonic the

consumption goal of the consumer the less standard economic theory explains their

purchasing and decision behavior. This is due to that consumers feel guilt and want fulfilment of their own personal symbolic needs. By consuming products and services they perceive having hedonic associations connected to it Uhrich et al. (2013) suggests that customers might be inclined to pay a price premium. This is supported by Voss et al. (2003) who states that previous research indicates that brands and products that score high on the hedonic dimension is better suited to charge a price premium or engage in price promotions. In this sense the consumption goal of the consumer plays a vital role for what kind of money and resources the customer would like to invest in the brand or product.

(14)

present and stronger for hedonic- rather than utilitarian consumption (Uhrich et al., 2013). Customers preference for a flat-rate depends on their consumption goal even though the underlying service given is the same and the cost expected is equal in all cases (Uhrich et al., 2013). Thus:

H2: The flat-rate bias is stronger for the hedonic rather than the utilitarian product

Table 2.1- Literature overview.

Table 2.1 shows an overview over previous research covering the dimensions this study touch upon. Services have been used in studies more frequently than FMCG, and thus have received more attention from research focusing on the flat-rate bias and hedonic and utilitarian

consumption among consumers. We have seen studies indicate that there are several benefits of hedonizing a product through marketing and communication initiatives to increase the consumer’s propensity of choosing a pricing plan with a flat-rate. Due to the transformation

Author Flat rates Hedonic Utilitarian Services Products Online WTP Subscription

(Krämer & Wiewiorra, 2012) x Telecom x x (Lambrecht & Skiera, 2006) x Internet provider x x

(Uhrich et al., 2013) x x x Several x x (Mitchell & Vogelsang, 1991) x Telecom x (K. E. Train et al., 1987) x Telecom x (Nunes, 2000) x Groceries x x (Wakefield & Inman, 2003) x x (DellaVigna & Malmendier, 2006) x Gym memberships x (Wertenbroch, 1998) x Groceries (Prelec & Loewenstein, 1998) x Several x

(Van Doorn & Verhoef, 2011)

x x Groceries x

(15)

of goods and commodities appearing more frequent in subscriptions through the past years, competence when pricing these products as well as identifying its’ consumption goal may be of higher relevance and thus increases the relevance of this study. Next, we will present our conceptual model and describe how the study will be conducted.

Conceptual model and research question

Subscription-based services and products increases in relevance and understanding how consumers react to pricing plans needed. We have looked into what kind of customer goals that drives consumption and how the flat-rate is affecting subscriptions in different services. What remains to be investigated is whether there are any differences in these findings in regard to goods and products. Our research question is:

Is there a stronger flat-rate bias for the focal product when primed with a hedonic consumption goal and how do the different pricing models affect this bias?

Figure 2.1- Conceptual model.

(16)

3.0 Methodology

To test our hypothesis and the potential presence of a flat-rate bias, we performed a

CONJOINT analysis measuring respondents’ preferences through a digital survey. We chose to conduct the modelling in chapter four-Results since this is related to fitting of the data and achieve an as optimal model as possible.

3.1 Product category

As mentioned earlier consumers have different consumption goals when using products and services. In the same way products have different attributes and can elicit a variety of

different emotions and usage areas. In regard to this it is important to choose a product for this study that contains attributes towards both hedonic and utilitarian consumption. In this way it will be easier to prime the product either way for respondents in a survey, make it more realistic and strengthen the validity of the responses.

3.1.1 Coffee

Consumption of coffee is driven by several needs, entails different attributes and hold a variety of associations related to its’ usage. It is a product that is frequently consumed in Western-countries and the Netherlands as well as Norway, where this study will be

(17)

and utilitarian product can be due to the attributes associated with consuming the products such as smell and taste, which both stimulates satisfaction (Desmet & Schifferstein, 2008; Labbe et al., 2015).

3.2 Consumption goal manipulation

To check whether there are differences between hedonic and utilitarian needs towards the pricing model, we primed each sample with a story to the respective consumption goal (Appendix B and C). This will function as a moderator affecting the relationship between our independent- and dependent variables. The priming was meant to make the respondent put themselves in the shoes of Andrea as done by Uhrich et al. (2013) and subconsciously adopt her consumption goals. Respondents would only face one scenario to avoid any biases when being primed with both scenarios. First, respondents will see an introductory story (Appendix A) that is similar for both consumption goals, with a background story, cost of the product and how many cups of coffee one unit can provide. Associations to coffee consumption picked in each primed story is chosen according to Messina's (2018) categorization of utilitarian and hedonic dimensions.

3.3 CONJOINT design

To understand consumer preferences and expected utility for subscribing to coffee brand of our making, we conduct a CONJOINT analysis. This is an effective tool for deriving information from consumer’s evaluations of products, services and the relative importance they assign the attributes and utility of this type of consumption (Malhotra, 2010). The utility gained may explain consumers actual purchase intentions, how modification of attributions may predict their consumption behavior and cluster respondents to arrive at homogeneous segments of preference (Eggers, Sattler, Teichert, & Völckner, 2018).

3.3.1 Attributes and levels

The attributes of this experiment is picked based on the recommendations of Eggers et al. (2018) implying that the study uses attributes that are relevant determinants for consumer choices. Reflecting specifications of product and services that are relevant in a realistic

(18)

Attributes Levels Price flat-rates €4, €5.5, €7, €8.5, €10

Units 1, 2, 3

Price additional units in choice-set €2.5, €4 and “Not an option” in this choice-set

Table 3.1- Product attributes and levels

Price flat-rate. The price attribute is important so we can get an indication by the consumer on how willing they are to pay for the pricing models on the basis of different bundling combinations. The price is reflecting the flat-rate price and how much the consumer is willing to pay each month subscribing to the product configuration in this service. There were five pricing levels as seen in table 3.2.

Euro 4 5.5 7 8.5 10

NOK 40 55 70 85 100

Table 3.2- Flat-rate prices in study

The prices are decided on the basis of that it must be realistic to subscribe to and they will be combined with more products that can contribute to decrease the price per unit since the consumer may want to buy several units. A similar study has of the authors best knowledge yet to be conducted so there is no benchmark to consider when creating the levels in this attribute except looking on how the industry price the equivalent amount of coffee.

(19)

Quantity. The number of cartons ordered makes it possible calculate whether there may be a flat-rate bias or over consumption during the survey. Adding this attribute will also indicate in what degree consumers are willing to increase their coffee usage during the subscription period. The levels are one, two and three units. This is the same amounts as other distributors market their subscription services when consumers have to choose a subscription service (Kumstova, 2017).

Price for additional units. This attribute contains how much extra monetary value the

respondent would like to pay for one more product within the flat-rate. This is relevant for the limited flat-rate where additional products outside the current flat-rate subscription can be reached by paying an extra fee. If the respondent picks a product combination with this attribute they may be inclined to go for a limited flat-rate subscription type if they already have fulfilled their own usage through stipulating their desired quantity.

3.3.2 Design

We have accounted for minimal overlap and randomly allocated levels. The attribute levels are mutually exclusive, and we provide a fractional factorial design. This is a subset of a full factorial design of combinations of attributes and attribute levels to a product configuration. There are 12 choice-sets, which is recommended according to the findings of Eggers et al. (2018) to strengthen the reliability of the parameters. By not exceeding the threshold of 16 we try to prevent that the participants get tired and submit responses affected by fatigue

conducting a time-consuming study.

(20)

option with additional units the respondent may be inclined to pursue the former and by choosing the no-choice option the latter subscription type.

Figure 3.1- Example CONJOINT choice-set pictured as seen in survey.

3.4 Sampling approach

Through this study we wished to draw conclusions from a sample consisting of respondents in the population. We wanted to approach individuals that had a relation to, purchased or

consumed coffee. By studying a sub group of the desired population we wanted to be able to draw precise conclusions that could be generalizable for the population as a whole (Sekaran & Bougie, 2013).

We reckoned the population to be quite large since people may have a relationship to coffee either by consuming themselves or buying for others. We wanted to conduct this study on a random selection of participants to make sure that we reach a broad and varied set of

(21)

4.0 Results and model fitting

In this part we will account for the results of the analysis conducted for this thesis. Data was gathered through a survey consisting of a conjoint-based-choice section with a descriptive text part investigating behavior amongst respondents for usage and purchasing habits of coffee. We then proceed with describing the data assessment and characteristics of the respondents of this study before presenting how we fitted an optimal model through multinomial logit and latent classes. In the end of this chapter we interpret the results and present findings towards the flat-rate bias and the limited pricing model.

4.1 Data assessment and respondents’ characteristics

The survey was completed by 89 respondents and data from 70 of these where used for analysis of this thesis. One was removed due to testing and 17 dues to non-coffee drinking preferences. Non-drinkers were removed since the external validity of this survey may be questioned when respondents that do not usually find themselves in a purchase situation of coffee gets questioned how they would act in such a scenario. One respondent indicated that the preference for how much they usually spend on a unit of 200g coffee was -€0.1. Since it makes little sense to have a negative preference this was changed to €1 instead. The survey was evenly distributed, with males accounting for 58% and females 42% respectively. The largest age group were 18-29 (68%) and 50-59 accounting for the second largest group (18%). Norwegian respondents where 79% and English-speaking respondents accounted for the rest 21%. This is mainly due to the social communication channel (Facebook) this survey where distributed through where Norwegian recipients is a majority. The stimuli

(22)

4.2 Multinomial logit and latent class estimation

In this chapter we present findings from multinomial logit modelling (MNL) as well at latent classes (LC) on our survey data. Since both these approaches has different applications, we choose to present both as model options. MNL computes utilities that are averaged across respondents and is beneficial for a company that want to approach a market as a whole. This is a maximum-likelihood technique which aims at identifying partworth utilities best

representing choices observed (Eggers et al., 2018). LC on the other hand aims at gathering respondents with similar preferences for attributes in segments so that companies may easier target specific groups of people for their products and series (Eggers et al., 2018).

4.3 Model fit

To find a satisfying model assisting in solving the research question, we compared several different models and assessed their impact between each other with likelihood ratio tests, goodness of fit and accounted for multicollinearity between independent variables. The function in equation E1 indicates how the variables are set and the partworth parameter estimates are stated in table 4.1. All variables in the conjoint-study are included without any interactions. All variables in E1 is partworth combinations and 𝜀 reflects the error term.

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E1) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒 𝑓𝑙𝑎𝑡 − 𝑟𝑎𝑡𝑒) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠) + 𝜀

As seen in table 4.1 the face validity is high since price is linear and the utility decreases for every higher step price-wise. The same goes for quantity where the utility increases linearly with the units added. Reference levels are recovered by summarizing the range for all partworth utilities times -1.

Variable Beta Standard

(23)

€10 - 1.205*** 0.142 -8.475

Quantity 44%

1 unit - 1.155*** 0.099 -11.702

2 units 0.299*** 0.070 4.2646

3 units 0.855*** 0.067 12.718

Price additional units 9%

One for €2.5 0.248*** 0.064 3.899

Two for €4 - 0.095 0.068 -1.408

Not possible - 0.153* 0.067 -2.274

Non-option 0.619*** 0.089 6.945

p =.000. *** = p < 0.01, ** = p < 0.05, * = p < 0.1

Table 4.1- Estimated part-worth utilities all variables included without any interactions

The log-likelihood of E1 is -957.46 and is used to test whether this model can predict better than a NULL-model. The null model can be computed as in equation E2 as conducted in Eggers et al. (2018).

Choice-sets*n*log(1/Alt) _(E2)

Since there are 12 choice sets and four alternatives per choice sets distributed on 71

respondents LL0 is -1181.123. To test whether there is any difference between the two models we can conduct a likelihood-ratio test as seen in E3:

= 2*(loglikelihood 1 -loglikelihood 0) (E3)

We run the test and the model is significantly better than a NULL model to predict (X2_(8):

447.33, p<0.00). The goodness of fit for the model is 0.19 and is the pseudo R2_{of the model}

(McFadden) describing the percentage of variance that independent variables can describe of the dependent variable (Sekaran & Bougie, 2013). The Nagelkerke R2_{is in this case 0.25. We}

take the latter in consideration since the pseudo R2_{cannot be exactly 1 as in regular linear}

(24)

are highly correlated with each other. We check for this by conducting a variance inflation factor test. None of the variables are exceeding the threshold of >10 (Sekaran & Bougie, 2013, p. 319) and the highest value is 2.08.

4.3.1 Adding vectors

Price

In table 4.1 as in figure 4.1 we observe that price follows a linear pattern. The flat-rate prices have a slightly curvature and by replacing the price variables with a linear variable this may make our model more parsimonious, we can conserve degrees of freedom relative to

partworth models and gain more power to study interactions since there will be less coefficients in our model (Sawtooth, 2004).

Figure 4.1- Utility of flat-rate prices in model with all attributes included

We implemented price and the model can be seen in E4. Adding a linear point or vector to our model we assumed that the attribute leads to a proportional negative effect in utility (Eggers et al., 2018).

(25)

We can test whether this improved our model or not with a likelihood-ratio test. The results showed that the there was no significant improvement (X2_{(3):5.83, p<0.12) between E4 and} E3. We therefore proceeded with E4 since this was a simpler model.

Additional units

Now as we have established that adding a linear price to our model optimizes its’

functionality, we tried to make it even more parsimonious by adding the two other attributes “Quantity” and “Additional units” as vectors in our model. Since both attributes in as seen in table 4.1 had decreasing or increasing utility for all its levels it is natural to believe that we could add these as vectors in our model. We started by removing the partworth combinations for Quantity and add the linear variable instead as seen in equation E5 and do the same for

Additional units in E6:

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E5) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒 𝑓𝑙𝑎𝑡 − 𝑟𝑎𝑡𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦BCD) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠) + 𝜀

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E6) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒 𝑓𝑙𝑎𝑡 − 𝑟𝑎𝑡𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD) + 𝜀

We then run a likelihood-ratio test with E4 to see how the modifications affected the overall difference between the models. The results indicated that E4 was significantly better than E5 (X2_{(1):18.90, p<0.00). When running the same test between E4 and E6 we found that there} was no significant difference by adding additional units as a vector to our model (X2_(1):2.00,

p=0.16). Goodness of fit is identical for both of these (McFadden: 0.19, Nagelkerke: 0.24).

There were no multicollinearity issues after running a VIF test with no greater values than 2.29 (E4) and 4.2 (E6).

4.3.2 Adding interactions

To answer our research question, we needed to take into account how our moderator

(Hedonic) may affect the model. Thus, we included one interaction at the time to make sure that we have the optimal combination. We started by adding an interaction on price as seen in equation E7 and then tested with the two other attributes as well in E8 and E9:

(26)

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E8) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD)

+𝛽E(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD∗ 𝑀𝑜𝑑H,I) + 𝜀

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E9) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD) + 𝛽E(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 ∗ 𝑀𝑜𝑑H,I) + 𝜀

When testing all new models with a likelihood-ratio test on E6 we observed that it was only E7 that significantly improved and where different as seen in table 4.2.

P(>chi-square) E6 Degrees of freedom change

E7 7.975** 1

E8 0.39 1

E9 2.32 2

p =.000. *** = p < 0.01, ** = p < 0.05, * = p < 0.1

Table 4.2- Chi-square value after likelihood-ratio test between interactions on E6.

We build on E7 by adding Quantity and additional units as interactions. The equations may be seen in E10 and E11:

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E10) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD) +𝛽E(𝑃𝑟𝑖𝑐𝑒BCD∗ 𝑀𝑜𝑑H,I) + 𝛽J(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD∗ 𝑀𝑜𝑑H,I) + 𝜀 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E11) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠BCD) +𝛽E(𝑃𝑟𝑖𝑐𝑒BCD∗ 𝑀𝑜𝑑H,I) + 𝛽J(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 ∗ 𝑀𝑜𝑑H,I) + 𝜀

E10 did not perform significantly better than E7 (X2_{(1):18.901.95, p=0.16) but E11 did} (X2_{(2):8.67, p<0.01). E11 was also significantly stronger than E10 (X}2_{(1): 6.72, p<0.01).} When running a VIF test results showed that the highest value is 4.25 and no problems with multicollinearity amongst the variables. As there where only one possible attribute left that can be interacted, we tried to add additional units as another interaction in E11 without any significant difference on the models (X2_{(1):0.9956, p=.318).}

4.3.3 Comparing with other models

(27)

proof of having additional costs as partworth we kept this as a vector to make the model simpler and more parsimonious. In E12 we reversed this. E13 had the same composition as E12 with partworth utilities on all attributes and only interacted with the price levels. We included this to show that our proposed model (E11) was optimal.

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E12) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒BCD) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠)

+𝛽E(𝑃𝑟𝑖𝑐𝑒BCD∗ 𝑀𝑜𝑑H,I) + 𝛽J(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 ∗ 𝑀𝑜𝑑H,I) + 𝜀

𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 =_./,,-_- (E13) Where 𝑧 = 𝛽.(𝑃𝑟𝑖𝑐𝑒) + 𝛽:(𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦) + 𝛽>(𝑃𝑟𝑖𝑐𝑒 𝑎𝑑𝑑. 𝑢𝑛𝑖𝑡𝑠) + 𝛽E(𝑃𝑟𝑖𝑐𝑒 ∗ 𝑀𝑜𝑑H,I) + 𝜀

Attributes E11 E12 E13

Log-likelihood -953.05 -952.02 -949.86 AIC 1922.101 1922.037 1925.716 AIC3 1930.101 1931.037 1935.716 BIC 1971.172 1977.242 1997.323 Price flat-rate €4 0.589*** €5.5 0.52*** €7 -0.048 €8.5 -0.136 €10 -0.925*** Price linear -0.392*** -0.393*** Quantity 1 unit -0.921*** -0.922*** -1.171*** 2 units 0.23* 0.23* 0.302*** 3 units 0.691*** 0.693*** 0.869*** Quantity linear

Price add units

One for €2.5 0.25*** 0.249*** Two for €4 -0.097 -0.1 Not possible - 0.153 -0.149* Cost add linear -0.208***

Non-option -1.287*** -0.873*** 0.631*** Interactions

(28)

1 unit*Moderator -0.536** -0.538** 2 units*Moderator 0.17 0.173 3 units*Moderator 0.366** 0.364** €4*Moderator 0.627*** €5.5*Moderator 0.162 €7*Moderator 0.112 €8.5*Moderator -0.327 €10*Moderator -0.572. McFadden 0.19 0.19 0.20 Nagelkerke 0.25 0.25 0.25 p =.000. *** = p < 0.01, ** = p < 0.05, * = p < 0.1

Table 4.3- Comparison of different models

4.4 Interpretation of model fit

In table 4.1 we presented the findings for E1 including all variables without any interactions. Positive signs indicate that there is an increased probability of that an event will occur, negative meaning that there is a decreasing probability. The most preferred configuration of product attributes was a price flat-rate of €4, three units included and the possibility of

ordering another unit for €2.5. The least preferred was the direct opposite entailing the highest flat-rate price, least units included and no possibility of ordering more units. The total utility of the most preferred option exceeded the non-option (2.011>0.619) and where expected since consumers usually want “more for less”. Price and quantity showed good face validity and were also recognized largely more important than the last attribute with respectively 47% and 44%.

In table 4.3 we observed differences in three proposed models. Several of the models had a weaker pseudo R2_{than the threshold of 0.2-0.4 which is stated to be an acceptable range for a}

(29)

Price and cost for additional units where linear and the latter had no interactions so it could be interpreted as a general utility for participants receiving both hedonic and utilitarian stimulus. The most preferred product configuration for participants would be a price of €4 (as low as possible), three units and one additional unit. For utilitarian consumers this value would be -1.09 and for hedonic -1.51. In appendix F the difference in probability of purchase and non-purchase for both stimuli can be observed. We see that for hedonic consumers the utility of the most preferred configuration has a lower utility score than the non-option and would not on average be picked by consumers. The incremental willingness to pay for one unit is also quite low (Util: €2.35, Hed: €1.56) indicating that consumers are not very prone to buy the products in this subscription setting unless the price is very low. For utilitarian consumers there is a threshold of €4.5 with 50% probability of purchase. This price is still very low and do not indicate a good adoption probability even though the case is better than for consumers being exposed to hedonic stimuli.

4.5 Latent class estimation and interpretation

Drawbacks with solely computing MNL values is that they assume independence of irrelevant alternatives, are average preferences and do not show heterogeneity across consumers. To remedy this, we found segments that had different preferences between each other by conducting a latent class estimation. This procedure estimated segment-specific partworth utilities (Eggers et al., 2018). From the fitting of multinomial logit models, we observed that E11 and E12 had more satisfactory results that alternative models. We therefore estimated latent classes for these two models respectively.

We run both models with two and three segments. R2_{and information criteria values can be}

observed in table 4.4.

R2 _AIC _AIC3 _BIC

E11 2 seg .35 1555.349 1567.349 1628.955

E11 3 seg .39 1472.24 1491.24 1588.784

E12 2 seg .36 1544.766 1558.766 1630.64

E12 3 seg .41 1445.222 1467.222 1580.167

(30)

We observe that for all columns E12 with three segments was performing better than

alternative models on R2_{as well as all the information criteria. It was also performing better}

than a NULL model (X2_{(19): 961.02, p=.000). In table 4.5 we can observe more closely this} model and the segment-specific utility values. The sizes are based on membership

probabilities.

Attribute Segment 1 Segment 2 Segment 3

Relative segment size 21% 64% 15%

Price flat-rate Price linear - 0.403 -0.386*** - 1.333*** Quantity 1 unit -8.775 -1.211*** - 1.36*** 2 units 3.841 0.313*** 0.329* 3 units 4.934 0.898*** 1.031*** Quantity linear

Price additional units

One for €2.5 4.774 0.269*** 0.528** Two for €4 -8.586 - 0.119 -0.196 Not possible 3.812 -0.149

.

-0.332

.

Non-option 11.924 -2.998*** - 2.355*** Class 2 1.119*** - 0.37* Moderator 0.023 1.159*** p =.000. *** = p < 0.01, ** = p < 0.05, * = p < 0.1. Table 4.5- Overview latent class model E12 with three segments and attribute coefficients

(31)

adopting the product since the non-option is lowest. When exposed to hedonic stimuli the probability of belonging to this segment is 76% (ModClass 2 / 1+exp(ModClass 2 ). Since the moderator for both the second and third segment is positive participant being exposed to hedonic stimuli had higher utility in this segment in combination with product attributes compared to participants with utilitarian stimulus. When recommending a segment to target the second one was more favorable since there where hard to reach a utility higher than the option in the two other segments. For segment 1 this were reflected through the high non-option value which were very high. Members of the third segment were quite price sensitive and made it difficult to create a desirable product combination since the price would very easily become too high.

In table 4.6 we observe that for E12 with three segments there is only segment 2 that produced optimal results for price when running a revenue maximizing price optimization. The prices were generated by calculating the utility of the optimal product attribute

combination which where three units and one additional unit (price variable is linear). In appendix G the price and demand probability curve for segment two may be observed and the optimal price between purchase and non-purchase probability lies between 10 and 11 euros.

Segment 1 Segment 2 Segment 3

Optimal product 3 units and 1 additional unit

€4 €0,09 €3,73 €0,78

€5.5 €0,06 €4,87 €0,17

€7 €0,05 €5,68 €0,03

€8.5 €0,03 €6,01 €0,01

€10 €0,02 €5,75 €0,00

Table 4.6- Revenue maximizing price best product combination for E12 with three segments.

Through model fitting of the MNL and LC model we found that there were clear differences between the two models. The MNL model which gave a more overall assessment of the average utility of the sample did not provide results that where satisfactory enough and struggled creating an optimal product configuration that where higher than the non-option. The latent class on the other hand provided a segment that could be approached with a

(32)

suggest a price-level that the consumers in segment 2 may find more appealing than proposed in this survey.

4.6 Flat-rate bias Our first hypothesis stated:

H1: There is a flat-rate bias for the focal product

To investigate this, we checked for differences between the price stated by the respondent on how much they were willing to spend on 200g of coffee, their benchmark, and if they

exceeded this benchmark when they chose alternatives in the choice-sets. E.g. if a respondent stated that their individual price for one unit was €4.30 and they were willing to pay €5.5 for the same bag of coffee in the choice-set then a difference of 1.2 would be made. When choosing a choice set with less than own preference the value would be zero. This difference is visualized in figure 4.2. We observe that there are some respondents that stated that their WTP to be very high (>€10). By running an individual sample t-test we show see that the mean is significantly different from zero (MDiff.Benchmark: -1.19, p=.000) and that overall, people did not exceed their own benchmark during the choice-sets. Since this may have been affected by respondents that have a high benchmark for this product, we want to test the respondents that are exceeding their benchmark and see if their choices are significantly different from zero. To do this the flat-rate bias requirement must be realized which is that participants did not exceed their own usage in the choice-set (usage explained in 4.7) but also overspent (16% of sample size). The t-test was statistically significant and different from zero (MFR-bias:1.25, p=.000, Confidence int.[1.07, 1.43]). There was a flat-rate bias for our

(33)

Figure 4.2- Payment differing from own benchmark

When addressing our second hypothesis:

H2: The flat-rate bias is stronger for the hedonic rather than the utilitarian product

we needed to see if there were any differences between overspending on coffee purchases while not exceeding own usage after being exposed to a hedonic stimulus. We used the same benchmark as in we did to answer H1. Then we created a dummy consisting of people

overspending in the choice-set while not over consuming their own usage. We then performed a chi-square test on the relationship between the binary variable and each observation treated with hedonic stimuli. The null-hypothesis indicated that the two variables are independent and bear no association to each other. The test resulted in a barely significant relationship

(34)

4.7 Consumption bias and the limited flat-rate pricing strategy

We checked if respondents where consuming more than their own need during the survey. We therefore conducted a test to check whether there where evidence for a consumption bias among the respondents. We created a benchmark based on how much and often respondents stated their own consumption and if it was less units than they preferred in the choice-sets a dummy would be created. E.g. by drinking two coffee daily * six times a week * four weeks divided on 30 cups (total amount of cups in a unit) equals 1.6 units needed (Musage: 2.24). This is not directly possible, so we converted this to two units (E.g. 1.4 converted to one unit). If the respondent requested more than one unit this will be indicated by a dummy. We did not find any relationship between participants being exposed to hedonic stimuli (X2_(1):0.00,

p<.97) and neither with respondents that were exceeding their spending (X2_{(1):0.41, p=.52).} Meaning that we could not state with significance that participants consuming more than their own usage were also overspending.

One of our attributes in this survey accounted for if respondents where prone for a limited flat-rate pricing strategy. To test this, we calculated the WTP among respondents including the additional product if possible (E.g. two units picked for €7- and one-unit addition for €2.5 equals €3.17), spending more than their own benchmark and identified the individuals that were eligible for a limited pricing subscription. To be eligible the respondents had to exceed their own usage by the number of units in the alternative they picked (E.g. monthly usage two units, chose alternative with three- eligible). We then tested these choice-sets to our

moderator but could not claim a claim a .05 significance level (X2_{(1):3.05, p<0.08,}

correlation: -0.06). We can therefore not with certainty state that respondents framed with

(35)

4.8 Manipulation check

To assess whether the respondents where affected by the moderators we checked for hedonic and utilitarian manipulation of the consumption. Voss et al. (2003) developed a scale for hedonic and utilitarian measure of attitude that was implemented to this study. The where 10 scale questions in total and five for either consumption goal. Both samples received the same questions (Appendix D) and they were measured on a five-point Likert scale (1- Fully

disagree and 5- Fully agree) as in Uhrich et al. (2013). A student t-test between the means of the same manipulation score on cross of hedonic and utilitarian framing shows that the

hedonic scenario was significant (MUtil_H.Scale: 2.99, MHed_H.Scale: 3.43, p= .000). The utilitarian stimuli on the other hand showed not to be (MUtil_U.Scale: 3.64, MHed_U.Scale: 3.55, p= .19). That the hedonic framing was significant strengthens the internal validity of this research and the claim that the results of this study may be applicable for business purposes.

5.0 Discussion

The main purpose of this study was to determine whether there was evidence a flat-rate bias for coffee in a subscription model and whether priming this product with a hedonic moderator had any effect on adoption. To do this we applied a CONJOINT study through an electronic survey. Attributes where price, quantity and additional product. The latter reflecting a limited flat-rate pricing model. Data where analyzed and we constructed a multinomial logit and latent class model to see what the revenue maximizing price options for the focal product were as well looking into how the hedonic utility affected adoption.

5.1 Applying models

(36)

members where lower on all price levels, indicating that we might have priced the product to extensively.

5.2 Flat-rate bias and hedonic stimulus

When addressing our first hypothesis we identified that there was a flat-rate bias among respondents that where overspending on coffee. In general, did people not overspend but this may have been affected by some respondents having a high benchmark for how much they were willing to pay for one unit. We found that on average participants in this study that had a flat-rate bias were overspending €1.25 per unit of coffee.

Our second hypothesis showed as we have seen in the literature section that a hedonic framing may enhance the propensity of spending. When isolating choice-sets where the flat-rate bias was present we identified that there was a significant positive relationship with participants that had received hedonic stimulus.

5.3 Consumption and a limited pricing model

(37)

6.0 Theoretical and managerial implications

Marketing and business managers can apply findings discovered in this research in several areas. As subscriptions are increasing in popularity and firms in different industries are embracing such models for their businesses, the importance of understanding how consumers adopt physical goods online is great. In this research we identify important variables when subscribing to coffee online. Understanding how consumers perceive essential attributes in the product offering is vital so the company may design services that meet the customer requirements. I the literature section we identified that there has been little research on the flat-rate bias in FMCG and for physical products and not services. Our results show that there exists a flat-rate bias for coffee in a subscription model and that hedonizing this product may enhance the consumers’ willingness to adopt.

Managers should be aware of that segmenting and properly target customers is important to succeed for this particular product on the marketplace. Customers are also price sensitive so wrongly price this service will make it hard for customers to adopt if one cannot justify a price-premium with their offerings. Since we established that there exists a flat-rate bias for the product, marketing managers can estimate the business potential underlying this bias and generate the market potential among their customers. If in thread with firm’s value

proposition can they profit on hedonizing this product to enhance subscriber’s willingness to pay.

7.0 Limitations and future research

This study was initiated to investigate whether there existed a flat-rate bias for coffee consumption in a subscription setting through a CONJOINT analysis positioned towards respondents in the Netherlands and Norway. The research has its’ limitations as well as potential for further research.

First, since the product presented was purely hypothetical with realistic attributes, the respondents did not get a picture that visualized the product. Such touch-points as design, texture, size and brand may stimulate the interest and curiosity of the respondent and may be a driver for the individual’s adoption of coffee.

(38)

analysis where the smallest segment contains few participants makes it difficult to apply results in the population.

Lastly, to get a better understanding of the respondents WTP we should have asked for preferences for two and three units and not only one. This study accounted for units becoming cheaper when the participant chose an alternative with more coffee bags as it would be in a realistic setting. This may have decreased the overall effect of overspending. By knowing the individual WTP for more than one unit would make it easier to measure a main effect

between multiple units and willingness to adopt.

(39)

8.0 References

A. Nielsen Company, Batra, R., & Ahtola, O. T. (1991). Measuring the hedonic and utilitarian sources of consumer attitudes. Marketing Letters, 2(2), 159-170.

Bhumiratana, N., Adhikari, K., & Chambers IV, E. (2014). The development of an emotion lexicon for the coffee drinking experience. Food Research International, 61, 83-92. CeCafé. (2019). Consumption. Retrieved from

https://www.cecafe.com.br/en/about-coffee/consumption/

Chen, T., Fenyo, K., Yang, S., & Zhang, J. (2018). Thinking inside the subscription box: New research on e-commerce consumers. Retrieved from

https://www.mckinsey.com/industries/high-tech/our-insights/thinking-inside-the-subscription-box-new-research-on-ecommerce-consumers

Delbridge, E. (2019). How Do Car Subscription Services Work?- Learn More About This Alternative to Leasing. Retrieved from https://www.thebalance.com/how-do-car-subscription-services-work-4584566

DellaVigna, S., & Malmendier, U. (2006). Paying not to go to the gym. american economic

Review, 96(3), 694-719.

Desmet, P. M., & Schifferstein, H. N. (2008). Sources of positive and negative emotions in food experience. Appetite, 50(2-3), 290-301.

Dredge, S. (2015). Apple Music launches to take on Spotify – and traditional radio. The

Guardian. Retrieved from

https://www.theguardian.com/technology/2015/jun/30/apple-music-launch-spotify-radio

Eggers, F., Sattler, H., Teichert, T., & Völckner, F. (2018). Choice-Based Conjoint Analysis.

Handbook of Market Research, 1-39.

Haaijer, R., Kamakura, W. A., & Wedel, M. (2001). The'no-choice'alternative in conjoint choice experiments. International Journal of Market Research, 43.

Hoyer, W. D., & Stokburger-Sauer, N. E. (2012). The role of aesthetic taste in consumer behavior. Journal of the Academy of Marketing Science, 40(1), 167-180.

Krämer, J., & Wiewiorra, L. (2012). Beyond the flat rate bias: The flexibility effect in tariff choice. Telecommunications Policy, 36(1), 29-39.

Kumstova, K. (2017). COFFEE VIA MAIL: DISCOVER SPECIALITY COFFEE SUBSCRIPTIONS IN EUROPE. Retrieved from

(40)

Labbe, D., Ferrage, A., Rytz, A., Pace, J., & Martin, N. (2015). Pleasantness, emotions and perceptions induced by coffee beverage experience depend on the consumption motivation (hedonic or utilitarian). Food quality and preference, 44, 56-61.

Lambrecht, A., & Skiera, B. (2006). Paying too much and being happy about it: Existence, causes, and consequences of tariff-choice biases. Journal of Marketing Research,

43(2), 212-223.

Malhotra, N. K. (2010). Marketing research : an applied orientation (6th ed. ed.). Boston: Pearson.

Messina, R. (2018). Hedonic and utilitarian aspects of coffee consumption in China. Università Ca'Foscari Venezia.

Mitchell, B. M., & Vogelsang, I. (1991). Telecommunications pricing: Theory and practice: Cambridge University Press.

Nunes, J. C. (2000). A cognitive model of people’s usage estimations. Journal of Marketing

Research, 37(4), 397-409.

Okada, E. M. (2005). Justification effects on consumer choice of hedonic and utilitarian goods. Journal of Marketing Research, 42(1), 43-53.

Prelec, D., & Loewenstein, G. (1998). The red and the black: Mental accounting of savings and debt. Marketing science, 17(1), 4-28.

Sawtooth. (2004). The CBC Latent Class Technical Paper. Sawtooth Software Technical

Paper Series(Version 3).

Schulz, P., Eggers, F., & Clement, M. (2017). Competing with Free Content - Pricing

Strategies for Cross-platform Bundling Services in the News Industry. Proceedings of

the 46th Annual Conference of the European Marketing Academy (EMAC.

Sekaran, U., & Bougie, R. (2013). Elements of research design: Chichester: John Wiley & Sons.

Train, K. (1991). Optimal regulation: the economic theory of natural monopoly: MIT Press Cambridge, MA.

Train, K. E., McFadden, D. L., & Ben-Akiva, M. (1987). The demand for local telephone service: A fully discrete model of residential calling patterns and service choices. The

RAND Journal of Economics, 109-123.

Tzuo, T., & Weisert, G. (2018). Subscribed : why the subscription model will be your

(41)

Uhrich, F., Schumann, J. H., & von Wangenheim, F. (2013). The impact of consumption goals on flat-rate choice: can “hedonizing” a service increase customers’ propensity to choose a flat rate? Journal of Service Research, 16(2), 216-230.

Van Doorn, J., & Verhoef, P. C. (2011). Willingness to pay for organic products: Differences between virtue and vice foods. International Journal of research in Marketing, 28(3), 167-180.

Voss, K. E., Spangenberg, E. R., & Grohmann, B. (2003). Measuring the hedonic and utilitarian dimensions of consumer attitude. Journal of Marketing Research, 40(3), 310-320.

Wakefield, K. L., & Inman, J. J. (2003). Situational price sensitivity: the role of consumption occasion, social context and income. Journal of Retailing, 79(4), 199-212.

Wertenbroch, K. (1998). Consumption self-control by rationing purchase quantities of virtue and vice. Marketing science, 17(4), 317-337.

Zhang, B. (2018). You can now drive as many Mercedes-Benz cars as you want for $1,095 a month. Retrieved from

(42)

9.0 Appendixes

A: General introduction for all participants

Andrea has recently seen an ad for the Brazilian coffee brand “Brazza”. The company is offering a subscription service where consumers may buy a subscription and get their coffee beans shipped home to her home without any delivery fee. She likes the idea of not having to go to the super market to get her coffee and she is curious about trying out a new brand that has a good reputation. Andrea is not sure about though; how much she is willing to pay for this product and how long she is willing to commit.

One bag of 200g of coffee powder/beans is equivalent to 30 cups of coffee. There is no delivery fee to take into consideration.

The service can be paused or cancelled at any time after individual need.

B: Utilitarian story

Andrea is an average consuming coffee drinker. She likes to make a cup of coffee in the morning so she can wake up and gain energy before heading off to work. During the day she might make herself another cup between breaks so she can stay energetic keep going without having unnecessary breaks. For Andrea it is the functional value of consuming coffee that matters since it is important for her to perform on all aspects of life at all times.

C: Hedonic story

Andrea is an average consuming coffee drinker. She likes to enjoy a cup in the morning since it helps on her mood and it benefits her peace of mind. During the day she prefers her cup of coffee in breaks at work, but preferably with co-workers that make her enjoy the experience as much as possible. For Andrea it is important that the raw materials are of good quality, so the taste is as satisfying as possible and the experience is at a maximum each time.

D: Manipulation check for hedonic and utilitarian consumption goals

Hedonic:

Ordering this product is fun Ordering this product is exciting Ordering this product is delightful Ordering this product is thrilling Ordering this product is enjoyable

Utilitarian:

(43)

(44)

(45)

//

(46)

(47)

(48)

Happily paying too much -a CONJOINT study on coffee consumption in a subscription model

Happily paying too much

-a CONJOINT study on coffee consumption in a

subscription model

Happily paying too much

-a CONJOINT study on coffee consumption in a

subscription model

Management summary

Table of Contents

3.0 Methodology

.

.