Effect of two-sided ads, when products require behavioural change, on willingness to pay

(1)

!

Effect of two-sided ads ,

when products require behavioural change, on willingness to pay

!

Bachelor's thesis University of Amsterdam

!

Olaf Bakker

!

Student number: 10114033/6375006 3/8/2014

Thesis Seminar Business studies Semester 2, blok 3

Supervisor: Bram Kuijken Academic year: 2013/2014

(2)

!

Abstract

!

Products that are either difficult to use, or require commitment from the consumer to use, are hard to launch. Add to that to the growing scepticism consumers have towards all the claims in advertisement and it becomes nearly impossible to get the product adopted by consumers. In this message dense society where ads are visible everywhere, credibility is what counts. This study aims at showing how two-sided ads can enhance the willingness-to-pay for products which require a form of commitment by the consumer to use, or are generally hard to use. The way this is researched is by selling certain products in a second-price sealed bid auction experiment. Different treatments concerning the advertisement are being compared to see which one has the highest influence on consumers willingness-to-pay. The results show that two sided ads in combination with products that require behavioural change, works.

!

(3)

Table of content

!

Abstract 2 Foreword 5 1. Introduction 6 2. Literature review 9

2.1 Scepticism towards one-sided ads 9

2.2 Two-sided ads 11

2.3 Effect of brand evaluation on WTP 12

2.4 Behavioural change 13

2.5 Defining the gap 16

3. Methodology 17

3.1 Research method 17

3.2 The product 18

3.3 Sample Frame 19

3.4 Treatment 1 - Control group 19

3.5 Treatment 2 - Benefits 20

3.6 Treatment 3 - Behavioural costs of change 20

3.7 Treatment 4 - Benefits and behavioural costs of change 21

3.8 Measurements 21

4 Results 23

4.1 Descriptive statistics 23

4.1.1 Descriptive statistics of the auction bids 23

4.2 Reliability 28

4.3 Correlations 31

4.4 Statistical tests 32

4.4.1 Data transformation! 32!

(4)

!

4.5 Regression 38 5. Discussion 40 5.1 Findings 40 5.2 Managerial implications 42

5.3 Limitations and recommendations for further research 43

6. Conclusion 44

7. References 45

Appendix A - Treatments 49

Appendix B- Histogram bid amount 50

Appendix C- Histogram bid amount top 50 51

Appendix D- Histogram ln bid amount 52

Appendix E- Histogram bid amount excluding 0 53

(5)

Foreword

!

This Thesis was written for a Bachelor degree in Business studies at the University of Amsterdam. The reason I choose this particular subject and method of research is

because of my interest in behavioural economics and psychology. There are some people who I would like to thank for making this thesis possible and their help in writing this thesis. The first person who I would like to thank is my supervisor mr. B. Kuijken. Special effort was made by Bram to make this thesis happen, also the support I got helped me in finishing this thesis. Furthermore I would like to thank Lisa Brown, marketing director at Fitbug for her willingness to trust me with 3 of her products to be auctioned by Veylinx. Finally I would like to thank my parents for their support in the last few days of writing this thesis.

!

Olaf Bakker

!

(6)

1. Introduction

!

New year resolutions are a yearly reoccurring topic. Each year people look back the year before and start looking on how to improve certain aspects in the year following. However, it seems to stop there. A study, that was done by Quirkology (2007) showed that from the the 3,000 resolutions made, 52% was confident of success but only 12% followed through. This suggests that making promises to change yourself or certain aspects of your life, is different from actually doing so. Vroom (1964) confirms this theory by stating: “The

consequences and costs associated with verbal and overt behaviours are not identical; a person may obtain satisfaction by merely saying (or thinking) something, and the positive outcomes associated with verbal (and cognitive) behaviours are often relatively easy to attain”. Meaning that costs and benefits in each situation, making a promise to change and

actually changing behaviour, differs. While the cost in promising a change is merely to express the change verbally, the cost in changing behaviour is every action that is needed to change that behaviour. So while resolutions are fairly easy to express, following through is not. This need or want, from consumers, to change certain aspects of life is not only confined to the days around new-year but is shown throughout people’s entire lives.

The need to change certain behaviour is also expressed through govern policy and medical concerns. In other words, the need for behavioural change can come from

consumer themselves or can be ‘imposed’ by the government or the world of medicine. No matter from who the reasoning comes from, the outcome often overlaps. Situations where behavioural change of the consumer is wanted or needed by the government or medical world, whether it’s for improving sustainability or living healthier, are often

beneficial for the consumer in one way or another. The most common examples are found in ‘green-product’ lines. Green products can be subsidised or can wear certain labels so that the cost-benefit relation changes. However, the situation just portrayed gives a false sense of the ability to choose whether to accept the change or not. This choice is not always present, both the government and medical world have the ability to impose behavioural change with various tools.

Following the ‘rational man’ model , where the consumer bases his choice

comparing costs and benefits, both parties have the ability to raise the cost and benefits of the consumer, to push the consumer into changing his or her behaviour (Halpern, David, 2004). Especially with trends such as sustainability and healthier lifestyles, the government has tools such as ‘Legal punishments’, ‘Price signals’ and ‘ Information’ that promote or deter changes in behaviour. Where the tools ‘ Legal punishment’ and ‘ Price signals’ are

(7)

used to increase the costs and therefore discourage certain behaviours, the tool ‘Information’ is purely used to let the user make an informed decision (Halpern, David, 2004). Price signals, often int he form of taxes or subsidiaries, are respectively used to make an alternative more attractive or to promote a certain product. Legal punishment is specifically used to deter consumer from certain actions. The tool ‘information’ is used to make consumers aware of the true costs, or consequences of using the product.

The last named tool(‘Information’) is used because people are not fully rational, there are certain cognitive limits that prevent us from being completely rational (S&O article). This means that people make decisions based on certain emotions when they, for example, hear that they can buy a car now and will not have to pay until two years later. The government then ads a general warning for these, and similar claims, such as: “Be careful, loaning money costs money”. This example shows how this tool is used to deter consumers from taking certain actions. However, the tool can also be used to promote a certain action. This is mostly the case when certain ‘green’ or eco-friendly alternatives are being introduced. While ‘Legal punishment’ and ‘Price signals’ can only be used by the government, the use of the tool ‘Information’ is not restricted to the government. Any organisation can make use of this tool. Even though every product has its flaws or

consequences, this tool has not been frequently used. However, marketeers are beginning to see the potential influence this tool has. Especially with the recent research done about the disbelieve in advertising(Obermiller, Spangenberg, MacLachlan, 2005).

With all the exaggerations and false promises that ads contain, consumers have become increasingly skeptical about the ads’ content (Obermiller, Spangenberg,

MacLachlan, 2005). This skepticism has lowered the willingness to pay for the products in the ads (Krishnamurthi, Mazumdar, Raj, 1992). If that is true, common sense would

suggest the opposite is true too. Thus, when ads are more credible, it should increase the willingness to pay. Following this logic, communicating the benefits and the costs would increase an ads credibility. Clearly, this theory does not apply to every ad-product

combination. Products that have costs which are easy to establish or ads where the claims are mild would not induce high levels of scepticism. However, looking at innovative and technological products, it is often hard to establish the true costs since these products require some form of experience to do so. A recent trend in technology-innovations is ‘Lifestyle improvement’. This trend introduces products that will ease your life or will improve your life in some way. Naturally, these products require that consumers change certain aspects of their lives to be able to improve their lives, which is hard to establish the costs of. The products are merely tools or reminders to accomplish lifestyle improvements.

(8)

In this market, activity trackers are growing in popularity and therefore also in numbers. As stated before, products like activity trackers do not change consumer habits, but can only support the change in behaviour. Each trackers lays different claims on their capabilities, however, all claim to be superior in one or more attributes. Due to all these different claims, the ads are vulnerable for certain scepticism consumers might have about these claims. This main goal of this thesis is to find out whether communicating positive and negative attributes in ads, decrease the ad scepticism and therefore increase the value the product has for consumers. The difference in previous research is that the negative

attributes are defined as costs of behavioural change, rather than ‘high price’, ‘poor design’ or ‘high in calories’ which were researched in previous articles.

!

(9)

2. Literature review

!

This section discusses relevant literature pertaining the subject. Literature is selected on relevance and how well the literature discussed the topics that need to be explained. First, the term scepticism is explained, and how that correlates with one-sided ads. The second topic discussed is the two-sided ad. Thirdly, brand evaluation and the correlation with willingness to pay is discussed, followed by the an explanation how certain product require the consumer to change his or her behaviour. Finally, the gap in research which this

experiment hopes to fill is mentioned.

!

2.1 Scepticism towards one-sided ads!

!

Ads are part of our everyday life, whether you turn on the TV, take the bus or subway home or even open your mail, ads are everywhere. Most of the time we discard these unwanted ads and get back to what we were doing. The goal of an ad is to persuade the consumer into buying that particular product or service. But in a world filled with ads, how do you attract the eye of the consumer? And even when the ads are intentionally being watched, it doesn’t mean that they reach their goal.

To acquire a (long term) competitive advantage, advertisers generally portray their brands as superior to other competing brands (Aaker and Myers 1987; Reeves 1961). The ads are focussed on communicating the superior determinant attributes of that particular brand. Advertisements that communicate only positive attributes are called One-sided ads (Pechman, 1992). However, this form of advertising is said to provide very little direct information, but are said to provide information to consumers indirectly by signalling the quality of the goods advertised (Becker, Murphy, 1993). This signalling does gives

“favorable notice “ to the goods advertised and therefore increases the value of the goods advertised (Becker, Murphy, 1993). Furthermore Sethuraman, Tellis and Briesch (2011) state that advertising is most effective in the early stages of the product lifecycle. This way, no attitude towards the product has been formed yet and the ad has a higher chance of influencing the consumers attitude towards the good. These articles all pertained one-sided ads which communicated one or more superior determinant attribute(s). This leads to the first hypothesis:

(10)

Hypothesis 1: One sided ads communicating only the beneficial attributes of the

product, will receive a higher willingness to pay compared to the ‘ bare-ad’.

!

Consumers have a predetermined scepticism when it comes to advertising(Burton, Lichtenstein, 1988). Scepticism towards advertising is defined as: “ The tendency towards disbelief of advertising claims” (Obermiller, Spangenberg, 1998). Skepticism towards ads, also called ‘ad skepticism’, is influential and probabilistic and not uniformly deterministic (Obermiller, Spangenberg, 1998). Which means that not every highly skeptical consumer

disbeliefs every ad claim, and that less skeptical consumers not necessarily believe every

ad claim(Obermiller, Spangenberg, 1998). However, as Obermiller and Spangenberd (1988) state: “The highly sceptical consumer should be more likely to disbelief and the less

skeptical consumer more likely to believe”.

However, there are factors that can mitigate responses to ad claims of even the most sceptical consumers. Known factors, that have been proven to influence persuasion, are: Claim substantiation, source characteristics, prior knowledge, message variables and product type. According to Obermiller and Spangenberg (1988) those factors play a role in determining the acceptance of claims in specific advertisement. The last named variable, product type, has been extensively researched. There are three categories of goods: search, experience, and credence goods (Nelson, 1970; Darby and Karni, 1973). Search

goods have characteristics that can be determined by information search before the

purchase or use. Experience goods have characteristics that require use experience. Credence goods have characteristics that cannot be determined by either information search or use experience and are too complex or require too much expert knowledge to be evaluated. The harder the characteristics are determined, the more skeptical consumers are about their ad claims. The reason consumers are skeptical is because they believe that advertising is often untruthful; it attempts to persuade people to buy things they do not want (Calfee and Ringold, 1994). However, they do not believe that consumers disbelief every ad claim. In fact, they suggest that consumers are aware of the untruthfulness in the ads, but are still able to extract valuable information from it. Though this theory assumes an equal level of scepticism and consumers to be rational. According to Calfee and Ringold (1994), consumers, as a group, recognise the intent and exaggerations of advertisers and discount ad claims accordingly. This theory contradicts the theory of Obermiller and Spangenberg (1988) that consumers differ along this dimension of

(11)

scepticism. Furthermore, they also suggest that there are consumers who reject

advertising completely, and therefore not gather any valuable information from these ads.

!

2.2 Two-sided ads!

!

As the above states, ads are meant to persuade consumers, but persuading factors in ads are often mitigated or even negated by consumer scepticism. Studies have shown that changing certain factors does improve the acceptability of the ad claims. The one that is important for this experiment is the two-sided ad.

While one-sided ads contain only benefits and important attributes that differ the product(s) or service(s) from competing brands, two sided ads work differently. As the name suggests, two-sided ads try to capture both the positive and the negative attributes. As stated before, One-sided ads receive high levels of scepticism. Advertisements are often associated with exaggeration, especially when the claims are more difficult to substantiate(Obermiller et all, 2005). Consumers are aware of this and therefore the ad becomes less credible. Researchers have suggested that one way to enhance the credibility of ads is to make it Two-sided (Settle and Golden 1974;Smith and Hunt 1978; Swinyard 1981). Which means that the add will cover both positive and negative attributes. Swinyard (1981) further states that they are also somewhat more effective at enhancing perceptions of the advertised brand on the primary featured attribute. This is congruent with other research, including the research of Pechman (1992). While most research studied two-sided ads with relatively uncorrelated attributes, Pechman (1992) began researching two sided chocolate bar ads showing negatively correlated(i.e. ‘richness’ and ‘higher in calories’) attributes as well as uncorrelated attributes (i.e. ‘richness’ and

‘sodium’) . The conclusion was that that the added uncorrelated attribute had a negative impact on the overall brand evaluation compared to two-sided ads with correlated

attributes. While in both cases, ads with uncorrelated and correlated attributes, brands received a better evaluation on the first attribute, overall brand evaluation was more positive when negatively correlated attributes were communicated. This suggests that choosing your secondary, negative, attribute is impacts overall brand evaluation. Pechman (1992) explains that ‘sodium’ is perceived as something that can be avoided. While ‘more calories’ is simply a byproduct of the extra ‘richness’ and can therefore not be avoided. This lead us to the second and third hypothesis:

(12)

Hypothesis 2: One-sided ads will receive a higher level of scepticism than two

sided ads. 1

Hypothesis 3: Two sided ads yield a higher willingness to pay than one sided ads.

!

2.3 Effect of brand evaluation on WTP!

!

The literature above has shown that scepticism can influence brand evaluation and brand credibility. Although it might seem logical that a negative brand evaluation is followed by a lower Willingness To Pay, support for this theory is scarce. However, research of Erdem et al (2002) offers insight into brand credibility and consumers price sensitivity. Erdem et al (2002) claims that consumer choice behaviour is affected by uncertainty about product attributes and/or benefits. This type of uncertainty is caused by imperfect and asymmetric information. They further state that uncertainty affects consumer price sensitivity. Erdem et al (2002) state 3 reasons for this theory:

1. High level of uncertainty about product attributes or claims may cause consumers to minimise expenses or losses, also known as ‘ risk aversion’ (Kahneman & Tversky, 1979). When the uncertainty is high enough to invoke risk aversion, and consumers are more likely to gain disutility from such purchases, Tellis and Gaeth (1990) state that consumers should then be more price sensitive.

2. Secondly, credibility may decrease information costs. If consumers can save on “information gathering and processing cots” then price sensitivity might decrease. Brands like “McDonalds” and “Apple” offer information about the quality of products and what sort of products they sell. Therefore, consumers could ‘save’ on information gathering.

3. Furthermore, credibility may enhance expected or perceived quality which, in turn reduces price sensitivity. While there are examples to be found of price premiums associated with higher perceived or expected quality brands or products (Aaker, 1991), this does not necessarily imply that higher perceived or expected quality lowers price sensitivity. However, Krishnamurthi, Mazumdar, & Raj, 1992 suggest that higher quality brands do have customers with a lower price sensitivity while lower (perceived) quality brands have customers with a higher price sensitivity.

The level of scepticism is based on the research of Mohr et all. (1998) where a 4 item 7-point liker

1

(13)

Furthermore, the points discussed above mat lead to brand differentiation. However, any of these mechanisms (e.g., decreased risks and information costs) could increase

consumers' willingness to pay higher prices because brand credibility increases even if the weight attached to a given price remains constant (Erdem et al, 2002).

Other research, done by Pelsmacker, Driesen and Rayp (2005), indicates that only consumers which share the values and believes of their brand (i.e. fair-trade) are actually willing to pay a premium price, and therefore have a higher WTP. However, this does not necessarily mean that positive brand evaluation is only acquired by means of having brand-‘lovers’ but rather indicates that simply having a faro-trade logo might not be convincing enough. Castaldo et al (2009) state that, in order to be viewed as credible in case of CSR activities, the products must: (1) comply with ethical and social requirements, and (2) the company must have an acknowledged commitment to protect consumer rights and interest. Since the article of Pelsmacker, Driesen and Rayp (2005), does not offer any insight into the the actual credibility of the ad claims, ad scepticism could possibly be the determinant factor in their non-conclusive results.

!

Hypothesis 4: The level of ad-scepticism is negatively correlated to the willingness

to pay.

!

2.4 Behavioural change!

!

The definition of the actual costs of product is a widely discussed topic. There are different views on what a product truly costs. While some economists state that products merely have a transactional value, where the intrinsic value is only considered, others argue that costs are revealed over time and thus have a utilization value(Meijkamp,Rens 1998). They argue that products with an utilization value are merely instruments that produce utilities or results (Meijkamp,Rens 1998). This view on products costs is further supported by the ‘means-end’ chain. The means- end chain is defined as the connection between product attributes, consumer consequences and personal values (Gutman, 1982). Van Raaij and Verhallen (1990) explain this further by stating that products or services have certain attributes, these attributes turn into functional and socio-psychological

consequences through the use of the product. These consequences are more determinant in consumers’ values than the mere product attributes. These consequences can be negative or positive. Behavioural change, in this case, can be both. The desired outcome

(14)

consumers hope to achieve with using this product is an example of a positive consequence while the actions needed to use the product can be seen as negative consequences.

As stated earlier, intentions to change behaviour is not the same as actually changing behaviour. Therefore, the short term benefits we receive from intending to

change and imagining the change on long term, might not outweigh the costs of the actual transformation. The ‘rational choice model’ states that consumers base their decisions on 2

calculating the costs and benefits of a particular action (Jackson, 2005). While doing this, the consumer will maximise the net utility. While the ‘Prospect theory’ has an interesting addition to this theory, namely that the benefits of a certain action have to outweigh the costs by a ratio of 2:1 in order for it to be accepted (Kahneman, Tversky 1979). However, this theory is based on experiments with monetary costs and benefits and it is unclear if this is can be generalised to take the ‘costs of behavioural change’ into account. Even though the ‘rational choice model’ is extensively criticised, the essence of this model still exist in newer theories. Theories such as the ‘ expectancy value’ theory , ‘ theory of 3

reasoned action’ and the ‘ Theory of Planned Behaviour’ are all extensions of the 4 5

‘rational choice model’ (Jackson, 2005). These models help us understand the structure of intentional behaviour. However, Jackson (2005) further states that they do leave out

important aspects, such as normative(moral), affective(emotional) and cognitive (e.g. habitual) dimensions of consumer behaviour. More specifically, the Expectancy value models assumes that behaviour is the result of deliberative, cognitive processes while it does not account for the habitual dimensions of consumer behaviour (Jackson, 2005). But in fact, our everyday behaviour consists mostly of decisions which do not require a

significant amount of conscious deliberation. Habits, routines and automaticity, which do not require serious thoughts, make us function more effectively especially in a

Rational decision making is a cognitive process. However, consumers are limited in thinking

2

rationally because we use several mental ‘short-cuts’ such as habits, routines, cues and heuristics (Jackson, 2005). These ‘short-cuts’ reduce the amount of cognitive processing which is needed to decide rationally.

Jackson (2005) writes: “Rational choice theory is a form of ‘expectancy value’ theory. In this kind

3

of theory, choices are supposed to be made on the basis of the expected ! outcomes from a choice and the value attached to those outcomes. A range of ! ‘adjusted’ social psychological models of consumer behaviour seek to use this basic ! idea to go beyond assumptions of rational choice and unravel the psychological ! antecedents of consumer preferences.”

Attempts to account for the influence of other people’s attitudes on individual behaviour.

4

extends the same model to incorporate the influence of people’s perceptions about their own

5

(15)

dense society (Jackson, 2005). Thus, even if we do have the intention to change certain behaviour, habits, routines and automaticity prevents us from following through with these changes. This leads us to the fourth hypothesis:

!

Hypothesis 5: One sided ads communicating only the costs of behavioural change of the product, will receive a lower willingness to pay compared to the ‘bare-ad’.

In order to change our behaviour, certain habits have to be changed. Anderson (1982) has defined three stages in the formation of a cognitive skill, otherwise known as a new habit. Stage one, the declarative stage, consists of the consumer receiving

instructions and information about a new skill. This stage challenges attitudes and affective responses to a certain action, but are deliberately bypassed by the consumer. Stage two is the transition between the declarative stage and a later stage. This transition is known as

knowledge compilation, where the consumer starts using the product without the

intercession of other interpretive procedures. In this stage the action is still deliberate. The third stage is called the procedural stage. This autonomous stage involves further learning the procedures required for product use. After this stage, the use of the product becomes a habit and bypasses most rational deliberation (Anderson, 1982). The action is then

accepted and considered part of a certain routine.

Behavioural change is hampered by already existing routines or habits. In order to accept a change, a person has to weigh the possible benefits against the costs of

behavioural change. These costs lie in the transition of undertaking a certain action once, and fitting them into a (daily) routine or habit. Only if the benefits outweigh the costs, and the action survives the three stages of forming a new cognitive skill, the product will be accepted.

The above literature suggests that one of the potential ‘costs’ of a product, besides the price, is the need for the consumer to change his or her behaviour in any way.

Literature about behavioural change shows that changing the behaviour of a consumer is difficult to realise (Jackson, 2005). Jackson (2005) further states that even when social norms change in such a way that behavioural change is wanted, certain aspects of human behaviour obstruct the acceptance of these changes. Recent (failed) innovations support this theory. This, in combination with the theory of two sided ads, leads us to the 6th hypothesis:

(16)

Hypothesis 6a: Negatively correlated attributes involving behavioural change act

the same way as negatively correlated attributes which are unrelated to behavioural change.

Hypothesis 6b: The negative correlation between costs of behavioural change and

several beneficial attributes of the product, positively influences WTP

!

2.5 Defining the gap!

!

While there is a growing need for products that change our way of living, ad scepticism and the extra cost of behavioural change pose an obstacle for advertisers to persuade consumers into buying these products. Even though consumers can tell that some

products require some form of behavioural change, they often cannot see the full extend of these changes. A way to counteract the scepticism that consumers feel when they aren’t being told the whole story, is the use of a two-sided ad. However, research has been done where positive and negative attributes of the product were shared, but not when positive attributes and extra costs (i.e. behavioural change) were shared. In other words, honesty about what it will truly cost the customer. The correlation between the costs of behavioural change and the positive attributes should be perceived as a negative correlation. In order to change your behaviour you have to change habits and routines, therefore when the product is believed to truly change your behaviour, it should also be clear to the consumer that there will be extra ‘costs’ with it(i.e. changing habits and routines). However, these types of costs of behavioural change are not seen as negative attributes, which makes this research fill a gap.

!

(17)

3. Methodology

!

Separately, ad scepticism and (costs of) behavioural change are extensively researched. However, as noted in paragraph 2.5, they have not been researched in relation to each other. With the help of an auction tool, ‘Veylinx’ an attempt is made to research different explanatory variables on the outcome variable; ‘Willingness To Pay’. And thereby fill the research gap of the effects of two sided ads, when products require behavioural change, on the WTP.

!

3.1 Research method!

!

This experiment will use a tool called ‘Veylinx’ (www.veylinx.nl). Veylinx is a second price sealed-bid Vickrey auction (Vickrey, 1961). The highest bidder wins the auction and pays the second highest bid. These rules ensure that an interested buyer reveals his true willingness to pay. To see this, suppose the ’ th bidder has a reservation value of the product indicated by and considers placing a bid , which is lower than his reservation value . If the largest of all the other bids, , exceeds , another buyer wins the bid so that buyer i’s gain remains zero. If buyer wins the bid and gains . However, if player i’s bid is lower than all other bids, which is also lower than ’s reservation value ( ), player receives a zero gain again. If player had bid his reservation price, his gain would be (Riley, Samuelson, 1981). This proves it is not beneficial to bid lower than your reservation value. The opposite is true as well, there is no advantage in bidding higher than the reservation value since that would yield a loss. Therefore the optimal strategy is for each bidder to submit his reservation value.

Veylinx currently has over 3200 subscribers. Each of these subscribers receives an email when an auction is about to start. Once the user enters the auction, by clicking on the link in the email, the user has 6 minutes to place one bid. All bidders are notified by an email the next day containing their bid, the highest bid and if they have won the auction. The panel roughly consist of 50% males and 50% females. Each participant is randomly assigned to a particular treatment or to the control group.

The following paragraphs goes through each treatment and explains its purpose. There will be three treatments groups and one control group. The first treatment will be communicating the benefits of the product only, the second treatment will be

communicating the behavioural costs of change, and the third treatment will communicate both the benefits and the costs. The matrix below explains this in a more visual way.

(18)

!

3.2 The product!

!

The product that is auctioned is a FitBug ORB activity tracker(Appendix 1). The Fitbug ORB is mountable on the waste or, in combination with the wristband, on the wrist. The ORB tracks your steps, calories burned, distance and sleep. With one click of a button, the information is send to your smartphone or tables. It also comes with a free lifetime

membership to Fit Bug KiK, which is Fitbug’s platform where you can log your nutrition, set goals, see your progress and keep track of your health. The retail price of Fitbug is 49$.

The category of activity trackers is a known category, however, this product is failry new and therefore less known. Moreover, the product falls into the experience goods category, and therefore the theory implies that the ad claims should receive a moderate to high amount of scepticism. Reading consumer reviews about this product tells us that the skepticism is justified because the use of this product is not as easy as the ad claims(“One

push of a button”).

This product not only stimulates a healthier life by changing the behaviour of the consumers, but also requires a certain amount of behavioural change to use. The consumer first has to find a trigger to start using the product, then it has to learn how to use it, after which the product becomes part of a routine or habit. Because the product is aimed to be part of your daily life, behaviour has to be changed. Thus, it holds (hidden) behavioural costs.

Also, this product is not widely available in the Netherlands. It is currently only sold in America, although Amazon has started offering this product internationally as well. Therefore the retail price of 49$ is an estimation of the true price. The true price is unclear because it is hard to establish the values of variables such as taxes, import costs and

Benefits

No Yes

Costs

No Control group 

Treatment 1

Treament 3

(19)

delivery time. However, the price for which it is sold in America and Amazon.com is easily accessible and may therefore still influence the bidding.

!

3.3 Sample Frame!

!

The panel of Veylinx consist of more than 3200 people. Of the 416 people that places a bid, 56% is male, 44% is female. This differs from the Dutch population where 54,5% is male and 45,5% is female (Centraal Bureau Statistiek, 2013). 65% of the respondents is older than 30 years old, 45,9% is older than 40, 3,5% is younger than 20. Further more, the youngest person is 16 years old, the oldest person is 103 years old female. That a 103 year old took part in this auction is highly unlikely, luckily she bid 0 cents and therefore no action has to be taken. 273 respondents enjoyed a high education, 40% of the participants are students at a University. 157 participants own an iPhone (37.8%), 219 participants own an Android mobile (52.8%).

Not everyone answered the questions at the end of the bid and not everyone answered them completely. The scepticism scale was, completely, answered by 382 people, while 2 persons only filled in half the 4 item 5 point liklikert scale. The first control question was answered by 382 persons, while the second one got answered by 383.

!

3.4 Treatment 1 - Control group !

!

The control group serves as a baseline. An ad with only the product and its specifications is shown. Even though, theoretically, there is still a possibility for ad-scepticism it is unlikely to occur since no extra information is shown and therefore the common reasons for

scepticism, exaggeration and clear persuasion signs, are absent. This condition is defined as Treatment 1, not because a treatment is applied but to make the data outcome

consistent. The ad is designed in such a way that the there are no variables that can interfere with the results. Colours have been kept basic, text to the minimum, and

placement of text and product are consistent through each treatment. There are, however, still visuals that can influence the dependant variable (willingness to pay) but those are present in every ad (compatibility, specifications). By keeping these parameters constant, we can assure that the outcome can be attributed to the treatment.

(20)

Appendix A figure 1 shows the ad that is used for this group.

!

3.5 Treatment 2 - Benefits !

!

The first treatment contains an ad with benefits only. This is the most common form of advertising. The advertisement attempts to persuade the consumer into buying the product by communicating superior determinant attributes. In this case, a general exaggerated consequence of using this product. So in addition to the bare-ad, with only specifications and compatibility, a text (showed below) is added. Ad-scepticism should be generated here since the claim is exaggerated. There seems to be no chance that the exaggeration goes unnoticed because it makes a claim which is very hard to realise for a product like this. Even though there is a chance that ad scepticism influences the price, it is expected that the beneficial ad claim does raise the willingness to pay.

!

The benefits communicated are:

Become fitter, healthier and happier.

!

Appendix A figure 2 shows he ad that is being used for this group.

!

3.6 Treatment 3 - Behavioural costs of change!

!

In the second treatment, the advertisement will only communicate the costs of behavioural change. This is an unusual way of advertising since persuasion by negative attributes does not work. Again, the parameters are held constant so that we can relate the outcome to the treatment. There should be little to no scepticism here because there is no reason to disbelieve negative statements. Where positive claims benefit the seller, and are therefore disbelieved, negative claims do not benefit the seller and are therefore not disbelieved. The negative claim is similar to the claims used by the government in ads silently claim that paying in terms is costs free.

!

Costs of behavioural change:

PAY ATTENTION - Daily use is required to book results.

!

(21)

!

3.7 Treatment 4 - Benefits and behavioural costs of change!

!

In the third treatment, both the benefits and the costs of behavioural change are

communicated. Where communicating the benefits only raises the level of ad scepticism but also raises the willingness to pay, communicating negative claims lowers the

scepticism and lowers the willingness to pay. Communicating both, as explained in the literature, this should make the ad more credible by mitigating ad scepticism. Furthermore, due to a more credible perceived ad and brand, the WTP should be higher than those of Treatment 1, 2 and 3 (Hypotheses 6). However, due to the number of visuals and texts in this ad, participants might not read it all. This will be checked by the control questions.

!

The benefits communicated are:

Become fitter, healthier and happier.

!

Costs of behavioural change:

PAY ATTENTION - Daily use is required to book results.

!

Appendix A figure 4 shows the ad that is being used for this group.

!

3.8 Measurements!

!

After the auction is completed and the respondent has placed his or her bid, 4 statements concerning scepticism and 2 control questions are posed. Because there is no time to develop a scepticism scale for this category (i.e. ads concerning health related products), a scale based on scepticism towards “green” products is used (Mohr et all, 1998). This model consist of a 4 item 7-point likert scale. However, due to the existing profiles in Veylinx, this is scaled down to a 4 item 5 point likert scale.

!

SKEP variable:

1. I perceive health-claims in ads to be true. 1-5 2. Health claims in ads are intended to mislead rather than to inform 1-5 3. I don’t believe most health-claims in advertisements 1-5 4. Health-claims in most ads are exaggerated 1-5

(22)

!

After the level of scepticism is established, control questions are asked as manipulation checks. The first check is to see if the treatments worked.

!

5. Do you think that using this product makes you fitter, healthier and happier?

y/n

If the treatments worked, treatment 1 and 2 should receive a low number of “yes” answers, whereas treatment 3 and 4 should receive a high number of " yes” answers.

Next manipulation check is to see if the respondents understood that this product requires a behavioural change. Especially the difference between no treatment, and the treatments where the behavioural change was mentioned.

!

6. Do you think using this product requires a change in behaviour? y/n

!

The 4 item 5 point likert scale will be converted into a new variable ‘SKEP’. This variable is needed to test some hypotheses. The control questions will be used to select some parts of the data and hopefully explain some relations or variances. Also the answers to the control questions can serve as a check, to see if the treatments had their desired effect.

!

(23)

4 Results

!

This section will provide the results and an interpretation of the results. First, some relevant descriptive statistics are given. Secondly, some correlations are examined and discussed. The third part consists of statistical tests on the sample data, followed by statistical tests on a selected portion of the sample data.

!

4.1 Descriptive statistics!

!

This section will provide all relevant descriptive statistics that are gathered from the auctions, the data already acquired on previous auctions and the questions after the auction. The aim is to provide insight into how the data is distributed. In order to run statistical tests some assumptions have to be made. The first assumption is to consider the test normally distributed, the second one is the assumption of homogeneity of variance (Field, 2009). Normally, when the sample size is larger than N=30, the sample can be assumed to be normally distributed (Field, 2009). All treatments received more than N=30 participants, however, due to the large amount of 0 cent bids (N=160) it is important to asses what part of the data is to be perceived relevant for further use.

!

4.1.1 Descriptive statistics of the auction bids!

!

A total number of 1562 Veylinx members got an email asking for their participation of the Fitbug Auction. Of those 1562, 813 were men, 749 were women. Of those xxx that were asked to participate, 416 subscribers placed a bid. Of those bidders, 231 were men, 185 were women and 273 were older than 30 years.. The lowest bid was 0 cent and the highest bid was 55,00 euro. However, due to the nature of this auction, were people are ‘forced’ into bidding on a product that they don’t want or have no need for, the frequency of 0 cent bids is large. In this case 160 persons bid 0 cent.

!

Figure 1 shows a means plot of the auctions in eurocents for the four different treatments.

!

(24)

!

The mean auction bid for the total sample was 783.01 eurocents. The maximum bid was 5500 cents (55,00 euro) and the minimum 0 cents. The mean auction bid for

treatment 1 is 764,73 cents (n=106, SD=1072,801), 904,65 cents for treatment 2 (n=110, SD=1130,120), 517,08 cents for treatment 3 (n=87, SD=837,019) and 886,50 cents for treatment 4 (n=113, SD=1172,433). As stated before, the 0 cents bids are made by those who are not interested in the product, and therefore their willingness to pay is 0 or close to zero. That means that the distribution is probably highly positive. This is further supported by the skewness and kurtosis descriptives shown in Table 4.1.1.

!

Table 4.1.1Descriptive statistics for normality

bid amount Treatment 1 Treatment 2 Treatment 3 Treatment 4

Skewness 1.956 1.740 2.397 1.519

Std Error Skewness 0.235 0.230 0.258 0.227

Kurtosis 4.094 3.320 6.715 2.107

Std. Error Kurtosis 0.465 0.457 0.511 0.451

(25)

Computing the z-scores of the skewness for treatment 1,2, 3 and 4 yields 8.32 for treatment 1, 7.56 for treatment 2, 9.29 for treatment 3, 6.69 for treatment 4 (p<.001), indicating a significant positive skewness (Field, 2009). The standardised z-scores of the Kurtosis are, respectively, 8.8, 7.26, 13.14, and 4,67. The standardised z-scores for Kurtosis are significant (p<.001) for each treatment which indicates that the distribution is heavily tailed and peaked (Field, 2009). Even though we can assume the data to be normally distributed (N>30), the previous statistics suggest otherwise. An explanation for this could be the amount of 0 cent bids. As stated before, 0 cent bids are placed because participants are not interested in the product and therefore their willingness to pay is 0 cent. Appendix B shows the corresponding histograms.

In order to see what the effect of the treatments was on people that were interested in the product, the data of the bid amounts are split at the 50th percentile. This way only the respondents interested in the product are considered. Figure 2 shows the means plot of the top 50th percentile bidding.

!

The lines follow the same pattern, however, treatment 4 has a higher mean bid amount than it had when the data was not split. The mean of treatment 1 is now is 1417,84 cents (n=56, SD=1127,853), 1589,93 cents for treatment 2 (n=61, SD=1115,091), 1022,41 cents

(26)

for treatment 3 (n=44, SD=934.095) and 1741.67 cents for treatment 4 (n=57, SD=1106,934).

!

The z-scores of skewness for treatment 1, 2, 3 and 4 are now 4.78, 5.15, 5.15, and 3.8 respectively and are therefore found significant (p<.001). he standardised z-scores of the Kurtosis are, respectively, 3.65, 4.22, 2.62, and 2.26. The standardised z-scores for

Kurtosis are significant (p<.001) for the first 3 treatments and significant (p<.05) for the 4th treatment, which indicates that the distribution is a little less heavily tailed and peaked than the previous data manipulations(Field, 2009). Furthermore, even though the bid amounts in the top 50th percentile are still significantly positively skewed, they are less skewed. It has been pointed out that in large data samples, it is more likely that the skewness is significant even though the data seems normally distributed (Field, 2009). This statement is further supported by the histograms in appendix C showing the bid amounts per treatment in the top 50th percentile.

Another correction, for the positively skewed data set, that can be applied is taking the natural logarithm of the bid amounts.

!

Doing so yields the values for skewness and Kurtosis presented in table 4.1.3. Computing the z-scores of skewness for treatment 1, 2, 3 and 4 yields -1.59, -2.94, 0.5, and -0.77

Table 4.1.2 Descriptive statistics for normality Bid Amount top

50 percentile

Treatment 1 Treatment 2 Treatment 3 Treatment 4

Skewness 1.526 1.577 1.839 1.224

Kurtosis 2.290 2.551 1.839 1.411

Table 4.1.3 Descriptive statistics for normality

ln bid amount Treatment 1 Treatment 2 Treatment 3 Treatment 4

Skewness -0.373 -0.677 0.129 -0.174

Kurtosis -1.680 -1.277 -1.881 -1.864

(27)

respectively shows that the bid amounts in each treatment are not significantly positively skewed (p<.001). Computing the z-scores for kurtosis yields -3.61 for treatment 1, -2.79 for treatment 2, -3.68 for treatment 3 and -4.13 for treatment 4. The standardised z-scores for Kurtosis are significant (p<.001) for treatment 1, 3 and 4, and significant (p<.01) for

treatment 2. However, the values are lower than with the previous alterations of the data, which indicates that the distribution is less heavily tailed and peaked (Field, 2009). The histograms in appendix D furthermore show that the moderate negative skewness can be explained by the high number of 0 cent bids. Taking the natural logarithm of a positively skewed distribution does make the data more normally distributed (Field, 2009). However, this applies to data which is not 0, since trying to take the natural logarithm of 0 would result in an error, which means the 0 is not changed. So all this transformation does is equalising the data by lowering the higher bid amounts. This is further illustrated by the graph below, the treatment with the highest mean, is lowered the most.

!

Another viable option is to subtract the 0 cent bids from the data set. By doing so, the largest cause of the positive skewness is removed. This leaves a total sample size of N=256 with a mean of 1272.39 (SD=1126.99).

(28)

The Skewness and Kurtosis values are given in table 4.1.4 Computing the z-scores of the skewness for each treatment 1, 2, 3 and 4 yields 5.4, 5.59, 5.15 and 3.71 respectively shows that the bid amounts in each treatment are significantly positively skewed (p<.001). Computing the z-scores for the Kurtosis yields 4.45 for treatment 1, 4,7 for treatment 2, 5.4 for treatment 3 and 2.00 for treatment 4. The standardised z-scores for Kurtosis are

significant (p<.001) for each of the first 3 treatments, and significant (p<.05) for treatment 4. which indicates that the distribution is heavily tailed and peaked (Field, 2009). Below, figure 5, a visualisation of the mean bid excluding 0 cent bids per treatment.

Appendix E shows the corresponding histograms.

!

! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !

Table 4.1.4 Descriptive statistics for normality Bid Amount

Excluding 0

Treatment 1 Treatment 2 Treatment 3 Treatment 4

Skewness 1.580 1.520 1.840 1.094

Kurtosis 2.555 2.538 3.765 1.165

(29)

! ! ! ! ! ! ! ! ! ! ! ! ! 4.2 Reliability!

!

After the respondent placed a bid, 6 questions were asked. The first four questions were posed as items with a 5-point likert scale, aiming to establish a level of scepticism for each participant. Each answer yielded a certain number on a 5-point likert scale as data.

Choosing 1 would mean the participant strongly agreed with the statement, choosing 5 would mean the participant strongly disagreed with the statement. The last 2 questions were “yes or no” questions, serving as a manipulation check.

To establish a level of scepticism, the 4 item 5-point likert scale is transformed in one variable ‘SKEP’. This variable quantifies a persons scepticism towards ads. A 4 item 5-point likert scale yields a summated value of 5 to 20. This scale is an adjusted version of an existing and previously used scale (Mohr et all.,1998). It is adjusted to fit with the health product group, rather than the green product group by replacing the phrase “environmental

claims” with “health claims” . Because the scale adapted, it is necessary to asses the

scale reliability by computing the Cronbach’s Alpha. Generally, a scale with a Cronbach’s Alpha of r>.70 is seen internally consistent and reliable (Field, 2009). Computing the Cronbach’s Alpha from the existing questions yields r=.311. This means that, in the current state of the items, the scale is highly inconsistent and unreliable. This low Cronbach’s Alpha is caused by the first question, which is positively stated while the other three are negatively stated. The 5-point likert items range from 1 (“strongly agree”) to 5 (“strongly

disagree”). Agreeing with the first statement suggests that the person believes health

claims in ads, and is therefore not skeptical. Agreeing with the second, third, and fourth statement suggest the person does not believe most health claims and is therefore sceptical. This difference in weight on the scale causes an unreliable scale. The

researchers whom developed this scale (Mohr et all.,1998) therefore suggest to reverse the scale of the first item so that choosing 1 (“stronly agree”) transform into 5 (“strongly

disagree”). This means that choosing to agree with the first statement, which corresponds

with a person not being skeptical, has the same value as disagreeing on the last 3 items, which also corresponds with a person not being skeptical. The Cronbach’s Alpha of the transformed 4 item scale is now r=.664. Which is still not reliable. Further testing shows that the Cronbach’s Alpha can be improved by deleting the 1st item, which is the item that was reversed in the first place.

(30)

!

Even though Item 1 is item 3 stated positively, item 3 seems to be a better fit. Since the corrected item- Total Correlation of item 1 is the lowest (r=.207) and the fact that the Cronbach’s Alpha statistic improves to r>,70 when the item is deleted the decision has been made to delete item 1 from the scale. Doing so resulted in each participant receiving a scepticism scale (“SKEP”) which ranges from 3 (very sceptical) to 15 (believes most claims). The sample mean of SKEP 7.84(SD=2.11) with a minimum of 3, indicating that people are very skeptical about ads concerning health claims, and a maximum of 13 which means that not one participant completely believes every claim made. The mean SKEP per treatment is shown in figure 5.

!

Table 4.2.1 Reliability scale

Cronbach's Alpha Cronbach's Alpha Based on Standardized Items

n

0.664 0.639 4

4.2.2 Item scale improvements

Scale Item Cronbach’s Alpha if item deleted

Item 1 0.716

Item 2 0.558

Item 3 0.389

Item 4 0.606

(31)

!

4.3 Correlations!

!

This section provides further insight in the data set gathered from the auction. Correlations between various variables are presented in appendix F. Variables selected are those that can be expected to have some correlation with the amount bid on the Fitbug. Only a few of these variables have a significant correlation. Surprisingly, some other variables are not significantly correlated.

The first variable that has a significant correlation with the bid amount, is the birth year (r=.118, p<.05). Which suggests that the higher the birth year, the higher the amount that was bid on the Fitbug, In other words, the younger the participant is, the higher the

willingness to pay is.

The second and third variable that are significantly correlated with the bid amount are the persons that visit the gym at least once per month (r=.186, p<.01) and the number of fruit pieces consumed per day (r=.140, p<.01). Both of these variables can be seen as an indication of a healthy lifestyle. The significant positive correlation between the auction bid amount and the persons that visit the gym at least once a month means that people that visit the gym at least once a month, have a higher willingness to pay. Same principle applies for the amount of fruit pieces consumed, people whom consume more fruit a day have a higher willingness to pay. On the one hand, this is surprising because people who seem to live a relatively healthy lifestyle would not need such a product. On the other hand, it is reasonable to assume that when people are living a healthy lifestyle they would buy products to enhance that healthy lifestyle even more.

The last significant correlation was suggests that the bid duration is positively correlated with the auction bid amount (r=.232, p<.01). Which tells us that the longer it takes for a person to bid at the auction, the higher the bid is. However, correlation analysis can not show the actual causal connection. While this last statistic might raise the idea that the longer you think about a product the higher your willingness to pay is, it can not be proven by this statistic alone. In fact, it could also mean that the opposite is true, higher bids require longer contemplation time.

Interesting to see is that the duration before placing a bid, is not significantly

correlated with the different treatments (r=.032, p<.520). This means that no matter which treatment the user was assigned to, it did not impact the bid duration.

The last significant correlation was found between the last control question (“do you

(32)

(r=.320, p<.01). This suggests that respondents who truly believed this product would make you fitter and healthier, also had a higher willingness to pay.

!

4.4 Statistical tests!

!

Statistical tests are used in order to see if there are differences in the willingness to pay for different treatments. Based on the information given in paragraph 1 of this chapter, the distribution is not normally distributes(Saunders et al., 2007; Field, 2009). Furthermore, the Levene’s statistic of the entire data set is 4.051 (p=.007) which is significant, therefore no equal variances can be assumed. A one-way independent Anova, taking the entire sample size, yields no significant mean bid amount differences between the treatments (p=.051) (Saunders et al., 2007; Field, 2009). However, when each group is being examined separately by a Post Hoc test (LSD), there is one treatment that has a significant different mean compared to the other three treatments. As shown in the mean plot in paragraph 1 of this chapter, treatment 3 shows the lowest mean bid amount(U= 517.08). The one-way Anova shows a significant difference in means between treatment 3 and 2(p=.012), and 3 and 4(p=.016). However, the data is positively skewed due to a high variance caused by a high amount of 0 cent bids and therefore not normally distributed. As stated before, several actions, which require the removal or adjustment of some data, can be taken to come to a normally distributed dataset.

4.4.1 Data transformation

!

Paragraph 1 of this chapter showed 3 possible ways adjust the data to a normal distribution. The first action is to ignore all the bids under the 50 percentile. As stated before, this made the distribution less positively skewed. Examining this further, by running a test of homogeneity of variances, yields a Levene’s Statistic of 1.234 (p=.298) which is not significant. This suggests that we can assume equal variances. However, Kolmogorov-Smirnov statistics (D(56)=.234, p<.001; D(61)=.193, p<.001; D(44)=.236 p<.001; D(57)=. 163 p=.001) show that the distributions are still non-normally distributes, for all 4

treatments.

!

(33)

!

The second option is to look at the natural logarithm of the bid amounts. As stated in Paragraph 1, both the Skewness and Kurtosis further declined to which neither was significant. The Levene’s statistic however, has increased to 5.730 (p=0.001) and is now significant, which means equal variances can not be assumed. And again, the

Kolmogorov-Smirnov statistics (D(106)=.256, p<.001; D(110)=.234, p<.001; D(87)=.329 p<.001; D(113)=.287 p=<.001) show that the distributions are still non-normally distributes, for all 4 treatments.

!

The third option is to take the entire data sample, and only subtract the 0 cent bids. Even though the Skewness and Kurtosis suggest that the distribution is significantly positively skewed, the Levene’s Statistic of 1.483 (p=.220) is not significant which mean equal variances can be assumed. However, also in this transformation, the Kolmogorov-Smirnov statistics (D(68)=.229, p<.001; D(78)=.197, p<.001; D(44)=.236 p<.001; D(66)=. 141 p<.001) shows that the distributions are non-normally distributed for all 4 treatments

!

Table 4.3.2 Tests of Normality and Homogeneity of Variances for the natural logarithm of the bid amounts (N=416) Kolmogorov-Smirnov Statistic df Shapiro-Wilk Statistic df Levene Statistic df1 df2 Treatment 1 0.234 56 0.831 56 Treatment 2 0.193 61 0.820 61 Treatment 3 0.236 44 0.804 44 treatment 4 0.136 57 0.881 57 Auction bid 1.234 3 214

Table 4.3.2 Tests of Normality and Homogeneity of Variances for the natural logarithm of the bid amounts (N=416) Kolmogorov-Smirnov Statistic df Shapiro-Wilk Statistic df Levene Statistic df1 df2 Treatment 1 0.256 106 0.777 106 Treatment 2 0.234 110 0.777 110 Treatment 3 0.329 87 0.744 87 treatment 4 0.287 113 0.751 113 Auction bid 5.730 3 412

(34)

4.4.2 Hypothesis

!

Even though the above statistics suggest that the data is not normally distributed, ANOVA tests are known to work on non-normally distributed data (Field, 2009). Field (2009) states: “ANOVA assumes that variances in each treatment are fairly similar, within-condition

distributions are normally distributed, observations are independent and that the

dependent variables are measured on at least an interval scale”. Therefore, the hypothesis

proposed in chapter 3 will be tested with one-way Anova tests. The next step is to asses what transformation of the data will be used. Overall, without deleting the 0 cent bids, the top 50 percentile data transformation holds the best source of data on which the

hypothesis can be tested. The data transformation which resulted in the most normally distributed data is the dataset in which the top 50 percentile is separated. Therefore, this data transformation will be used to test the different hypothesises.

Hypothesis 1 states that one sided ads, which lay certain claims, receive a higher willingness to pay compared to the ‘bare-ad’. To test this a one-way Anova, followed by a Post Hoc test, is ran to see if there’s a significant difference in the mean auction bid for treatment 1 and 2 in the top 50 percentile.

!

Table 4.3.3 Tests of Normality and Homogeneity of Variances for t bid amounts > 0 (N=256)

Kolmogorov-Smirnov Statistic df Shapiro-Wilk Statistic df Levene Statistic df1 df2 Treatment 1 0.229 68 0.835 68 Treatment 2 0.197 78 0.841 78 Treatment 3 0.236 44 0.804 44 treatment 4 0.141 66 0.905 66 Auction bid 1.483 3 252

Table 4.4.2.1 descriptive values bid amount top 50% 95% Confidence Interval Mean Bid amount top 50 percentile N Mean Std. Deviation!

!

Std. Error Lowe bound

!

Upper bound

!

Min

!

Max Between comp. variance 1 56

!

1417.84 1127.853 150.716 1115.80 1719.88 300 5500 2 61 1589.93 1115.091 142.773 1304.35 1875.52 500 5000 3 44 1022.41 934.095 140.820 738.42 1306.40 10 4050

Table 4.4.2.1 descriptive values bid amount top 50%

Bid amount top 50 percentile

(35)

!

The mean bid in treatment 1 is 1417.84 cents (SE=150.716) and in treatment 2 it is 1589.93 cents (SE=142.773), a difference of 172.095 cents. This difference is not

significant (p=.392). Therefore, hypothesis 1 is not supported. A manipulation check is in order to see if more participants in treatment 2 answered “yes(=1)” to the question whether or not you can get fitter and healthier from this product. Treatment 1 had N=101

participants answering that question (Mean=.15 SE=.036), Treatment 2 had N=100

participants answering that question (Mean=.15 SE=.039). The difference is not significant (p=.552) and therefore communication of the positive attributes, did not make a difference in how the participant perceived the product.

Hypothesis 2 states that one-sided ads receive higher level of scepticism than two-sided ads. Even though the items of the scale of scepticism were posed as ‘general’ statements, not concerning the Fitbug claims alone, there seems to be some influence of the Fitbug ad, on the level of scepticism. The plot below shows the mean of the variable SKEP, per treatment (lower SKEP is more sceptical).

!

4 57 1470.85 1116.053 147.825 1445.54 2037.80 400 5000 Total 218 1106.934 74.971 1323.09 1618.62 10 5500 Model -Fixed!

!

-Random 1084.813 73.473 1326.03 1615.68

!

147.571 1001.22 1940.49 64639.36

Table 4.4.2.1 descriptive values bid amount top 50% 95% Confidence Interval Mean N Mean Std. Deviation!

!

Std. Error Lowe bound

!

Upper bound

!

Min

!

Max Between comp. variance Table 4.4.2.1 descriptive values bid amount top 50%

Bid amount top 50 percentile

(36)

!

What this graph (figure 6) shows is that persons in treatment 1 (‘Bare-ad’) are least sceptical about ad claims, and persons in treatment 4 (‘Two-sided ad’) are most sceptical about ad claims. Since the hypothesis states that treatment 2 should receive higher

skepticism than treatment 4, no statistical test is needed as this graph shows the opposite. However, after running a one-way Anova test on these variables, we can state that

persons in treatment 4 are significantly more sceptical than persons in treatment 1 (p<. 0.5). To see if there is a difference between women and men in skepticism per treatment, two new graphs are computed.

The graphs show that men and women respond very different to the ad in terms of

scepticism. However, both men and women have a higher mean SKEP in treatment 2 than

(37)

in treatment 4. Therefore either ad number 2 is considered to be more believable, or men and women in treatment 2 were already less sceptical.

Hypothesis 3 states that two sided ads (Treatment 4) receives a higher average willingness to pay than one sided ads (Treatment 2). The same one-way Anova is ran as with Hypothesis 1. The mean bid amount for treatment 2 is 1589 cents, while that of treatment 4 is 1741 cents. This difference of 151.73 cents is not significant (p=.449). Therefore Hypothesis 3 is not supported. However, when testing men and women separately, the next 2 graphs are computed.

This suggests that the mean bid amount of men, does differ significantly in treatment 4 compared to treatment 2. Further anova testing supports this with a mean difference of 572.08 cents which is significant(p=.039). Therefore, for the men, this hypothesis is supported.

Hypothesis 4 states that the level of scepticism is negatively correlated to the willingness to pay. This hypothesis is tested by mean of running a correlation analyses. The variable SKEP has a value between 3 and 13.5, the lower the number the more sceptical a person is. Would the level of skepticism influence the willingness to pay, then the Pearson Correlation would be r<0. However, r=0.010 which means that there is almost no interaction between these two variables. Not surprisingly, the correlation is not

significant (p=.444). Therefore Hypothesis 4 is not supported.

Hypotheses 5 states that Treatment 3 should receive a lower willingness to pay compared to Treatment 1 (No treatment). The mean bid amount of treatment 1 is 1417.84 cents while that of treatment 3 is 1022.41 cents. The difference of 395.43 cents, according to the Post Hoc test, is considered to be significant (p=.036). Therefore Hypotheses 5 is supported. Manipulation check with the control question whether the participant thought if this product required him or her to change his or her behaviour was executed. More than