Dynamic Pricing: The Impact on Price Fairness, Trust and
Repurchase Intentions
by NICK ALBERS
2 MASTER THESIS
Dynamic Pricing: The Impact on Price Fairness, Trust and
Repurchase Intentions
by NICK ALBERS
University of Groningen Faculty of Economics and Business
3
Abstract
Dynamic pricing occurs when people pay a different price for the same product from the exact same seller, and is a phenomena that is increasingly becoming more common
.
This is mainly due to the fact that e-commerce is becoming more prevalent. When implemented properly, dynamic pricing can improve firms’ revenue and profits by up to 8% to 25%. However, do people accept this type of price discrimination? The possible financial benefits for companies on one hand, versus possible the negative customer feelings and reactions on the other hand, leaves an important question for the use of dynamic pricing for a certain company. To what extent do consumers accept being disadvantaged by dynamic pricing? The main findings in this research were (1) personalized dynamic pricing has negative effects on perceived price fairness, trust in the company, and repurchase intentions; (2) discount framing mitigates the negative effects of personalized dynamic pricing; (3) wealthiness and customer segments do not moderate the main effect between personalized dynamic pricing and perceived price fairness, trust in the company and repurchase intentions.4
Preface
About two years ago, after completing my bachelor at the Hanze University, I decided that I wanted to go for the next step; obtaining my master’s degree. Now, after two years of hard work at the University of Groningen, I can proudly deliver my final piece of work.
I imagined my last months at University a lot different than what actually happened. Something happened that no one could ever predict; a deadly virus broke out, causing a pandemic. No more physical lectures, no more physical meetings, no more physical contact with people that over the years at University became friends, and also, no more physical meetings with the thesis group and supervisor Dr. A.E. Vomberg, which definitely did not make things easier.
Everything switched to online. In the beginning, I thought that a period like this wouldn’t harm me too much, however I was wrong. I noticed that, especially mentally, such a period asks a lot of us humans. Therefore, I really want to thank every single person that helped me during finishing my thesis. I would like to start by thanking Dr. A.E. Vomberg for his advice and support while writing this thesis. I would also like to thank my mother, for the support I got from her, even when times haven’t been easy for her. Finally, I want to thank everyone that took their time to respond to my survey. Without everyone mentioned, I wouldn’t be able to deliver this piece of work, and therefore, I am truly grateful.
I hope you will enjoy reading this thesis.
5
Table of contents
1. Introduction………...………6
2. Literature review………...………..9
2.1 Personalized dynamic pricing………..…………..9
2.2 Acceptance……….……….11
2.2.1 Perceived price fairness……….………...……….….11
2.2.2 Trust & repurchase intentions………..………..……….12
2.3 Price framing as a moderating role…………..………...……….……….14
2.4 Wealthiness as a moderating role…………..…………...………16
2.5 Customer segments as a moderating role……….………...……….17
2.6 Conceptual model……….………..……….19
3. Methodology……….……….………….……….20
3.1 Study design………..………….…….20
3.1.1 Online experiment………..……….20
3.1.2 Personalized dynamic pricing & price framing…………...……….…….20
3.1.3 Pre-test……….…………..….21 3.2 Measurements………...……..22 3.2.1 Dependent variables………22 3.2.2 Wealthiness……….22 3.2.3 Customer segments……….……….23 3.3 Analyses………..25 3.3.1 Hypothesis testing………25 3.3.2 Sample description………..26 4. Results………..………27 4.1 Customer segments………...………...27 4.2 Regression analyses………...………..31
4.2.1 Effect of personalized dynamic pricing on price fairness……….31
4.2.2 Effect of personalized dynamic pricing on trust in the company…………..32
4.2.3 Effect of personalized dynamic pricing on repurchase intentions……..….33
4.3 Overview of the results………..……..……34
4.4 Regression assumptions………..……35 4.4.1 Linearity………..………35 4.4.2 Normality………..……..35 4.4.3 Homoscedasticity………..……..36 4.4.4 Autocorrelation………..……….36 4.4.5 Multicollinearity………..………37
5. Conclusion and discussion…………...………..……….38
5.1 Conclusion……….….38
5.2 Implications………..……...39
5.3 Limitations and directions for further research………...…….39
References………..…………..41
6
1. Introduction
Imagine yourself browsing on the internet, where you are looking for a new phone to buy. You click on a certain website, see a nice Samsung phone, look at the price, and you think the price is reasonable. However, before making such a big decision, you call a friend which knows a lot about smartphones, to ask for his opinion about the phone. Since he does not know all the phone specifications of every single phone by heart, he goes to the same website to take a look at the phone. After some talking, your friend advises you to definitely buy the phone, because a price of €650,00.- is very cheap for a phone with such specifications. However, the price stated at your screen is €700,00.-, even after refreshing the page a couple of times. The price of the exact same phone, on the exact same website, is simply different. Sounds unreal, but this phenomena is becoming more common every day (Haws and Bearden, 2006).
This is a form of personalized dynamic pricing. Dynamic pricing occurs when people pay a different price for the same product from the exact same seller (Haws and Bearden, 2006). More and more retailers are starting to adopt dynamic pricing. To give an example, a children’s clothing store admitted in 2012 that they changed their online prices every 15 minutes. This form of dynamic pricing is called time-based dynamic pricing, however both forms of dynamic pricing serve the same goal; improving revenues. When implemented properly, dynamic pricing can improve firms’ revenue and profits by up to 8% to 25% (Sahay, 2007).
This offers opportunities for companies, however dynamic pricing can come with a price. Since other people paid less for the exact same product, from the exact same retailer, people could feel discriminated and experience feelings of unfairness, that eventually could lead to a loss of trust in the company, and thus weaker repurchase intentions (Garbarino and Maxwell, 2010; Grewal et al., 2004; Xia and Monroe, 2010). On top of this, price unfairness can have a decisive influence on the reaction of a consumer to a certain price, such that the consumers are, most of the time, are not willing to pay a price that they perceive as unfair (Campbell 1999; Kahneman, Knetsch, and Thaler 1986a, b; Martins and Monroe 1994; Urbany, Madden, and Dickson, 1989).
7 The next week, Amazon dropped the price to $22.74 without mentioning this to anyone; the
baseline price just dropped (Feinberg, Krishna and Zhang, 2002). Another way of lowering the
price for a certain product, is by putting a discount on a product. A discount of $1.75 on the DVD would lead to the same result; a price of $22.74. There is only one difference, which is the way of framing. Since price framing can influence peoples’ purchase intentions heavily, it could therefore lead to different feelings of acceptance (Weisstein et al., 2013). The answer to the question if framing moderates the (negative) effect of dynamic pricing, can have very important implications for retailers, since it is very important for them to keep a good relationship with their customers (Jain and Singh 2002; Shin and Sudhir 2010; Venkatesan and Kumar, 2004).
Next to this, previous research did not account for the wealthiness of a person and different customer segments which could possibly influence feelings of acceptance. The more wealthy one is, the less worried one is about money matters. Feelings of worry are a form of anxiety, which is shown to have a great influence on perceptions (Yang, Saini and Freling, 2015). It therefore could have a influence on how people perceive fairness, trust and repurchase intentions. Customer segments, on the other hand, could also have a big influence on these topics. Since different type of customers could react different to various stimuli (in this case dynamic pricing), it therefore is possible that different customer segments will have different feelings of acceptance.
The aim of this study is finding out to what extent people accept the fact that they are disadvantaged by dynamic pricing, and if the effect gets moderated by different price framing methods, wealthiness and customer segments. Of course, one can also be advantaged by dynamic pricing, however it is obvious that people would accept the fact that they are being advantaged, hence there is chosen for being disadvantaged by dynamic pricing as the central aspect in this study. By combining the variables mentioned, the research gap is filled, since this hasn’t been studied before. Thus, the following research question has been conducted:
To what extent do people accept being disadvantaged by dynamic pricing, and does this effect differ for different price framing tactics, for how wealthy one is and for different customer segments?
9
2. Literature review
To become familiar with the concepts in this thesis, definitions of the independent variable ‘Personalized dynamic pricing’, the dependent variable ‘Acceptance’ and the moderators ‘Price
framing’, ‘Wealthiness’ and ‘Customer segments’ will be presented. Next to the explanations
of all the variables, relevant literature is discussed in order to form a relationship and to create hypotheses.
2.1 Personalized dynamic pricing
Dynamic pricing, often mentioned as the economic term ‘individual-level price discrimination’ occurs when people pay a different price for the same product from the exact same seller, and is a phenomena that is increasingly becoming more common (Haws and Bearden, 2006). This is due to the fact that people buying products via an online store (e-commerce) is getting more popular every year (ecommercenews, 2019). In most supermarkets, prices are fixed; employees cannot change the price for a bottle of cola within a split second. For online stores, this is different; prices of products in online stores can change within minutes.
10 Author IV DV Outcome Garbarino and Maxwell, 2010 Dynamic (posted) pricing Perceived price fairness
Dynamic pricing leads to lower perceived fairness
Benevolence trust
Dynamic pricing leads to lower benevolence trust towards the firm
Purchase intentions
Dynamic pricing leads to lower intention to purchase from this retailer Haws and Bearden, 2006 Consumer-based dynamic pricing Perceived price fairness Consumer-based dynamic pricing leads to different perceptions of price fairness, based having a good deal or not
Purchase satisfaction
Consumer-based dynamic pricing leads to different perceptions of purchase
satisfaction, based having a good deal or not
11 Dynamic pricing comes in different types, however the literature mainly divides dynamic pricing in two forms: time-based dynamic pricing and consumer-based dynamic pricing. Time-based dynamic pricing implies that prices differ (for the same product from the same seller) within short periods of time, while consumer-based dynamic pricing implies that prices differ between consumers at the same time (Vomberg, 2019). The type of dynamic pricing that is relevant to this research is consumer-based pricing, since personalized dynamic pricing is a form of consumer-based dynamic pricing.
Scholars and industry experts agree that personalized dynamic pricing can benefit a company, because a company can target incremental consumers with a lower price for product X, without giving a lower price for product X to the consumers who would purchase from the company anyway without any lowering in price (Shaffer and Zhang 2002). As personalized dynamic pricing becomes more and more common, it is very important that companies understand how consumers react to personalized dynamic pricing (Haws and Bearden, 2006).
The possible financial benefits for companies on one hand, versus possible the negative customer feelings and reactions on the other hand, leaves an important question for the use of dynamic pricing for a certain company. To what extent do consumers accept being disadvantaged by dynamic pricing?
2.2 Acceptance
Acceptance is a very broad topic. In this research, acceptance is categorized in three different aspects, as these three aspects are strongly related to consumers’ perceptions of dynamic pricing (Weisstein et al., 2013). The aspects that together decide whether people accept being disadvantaged by personalized dynamic pricing are perceived price fairness, trust and
repurchase intentions.
2.2.1 Perceived price fairness
12 1999). Social norms and fairness are closely linked, in fact, social norms are ‘rules of fairness’. When the rules are broken, the action will be perceived as unfair (Kahneman, Knetsch and Thaler, 1986).
A very important social norm is that everyone should be treated the same. Based on equity theory, “equity occurs only when all parties receive the same ratio of outcomes to inputs. Inequity occurs when the ratio of an individual’s outcomes to inputs differs from those of others” (Adams, 1965).
When linked to personalized dynamic pricing, this concept of equity suggests that every consumer should pay the same price, for the same product, at the same time, from the same seller (Darke and Dahl, 2003). A violation of this social norm could very likely lead to situations where consumers judge the outcome as unfair (Maxwell, 2002), which can have a negative influence on the relationship between the customer and seller. Consumers do not like buying a product, to later find out that a neighbour or a friend paid a lower price for the same product (Turow et al., 2005). In addition, research has shown that perceived price unfairness can have a decisive influence on the reaction of a consumer to a certain price, such that the consumers are, most of the time, not willing to pay a price that they perceive as unfair (Campbell 1999; Kahneman, Knetsch, and Thaler 1986a, b; Martins and Monroe 1994; Urbany, Madden, and Dickson, 1989).
Consequently, personalized dynamic pricing could lead to feelings of price unfairness. Research shows that consumers could perceive feelings of unfairness when paying a higher price, for the same product, from the exact same seller. Therefore, the effect of personalized dynamic pricing on perceived price fairness is as follows:
H1a: Personalized dynamic pricing will have a negative effect on consumers’ perceived price
fairness.
2.2.2 Trust & repurchase intentions
13 violation of a social norm), would not per se break both dimensions of trust. When linked to dynamic pricing, the violation of a pricing norm (everyone paying the same price, for the same product, from the same seller) is likely to have an effect on the benevolence trust rather than it would have an effect of the competence/credibility trust.
This statement is supported by the findings of Garbarino and Maxwell. Customers that have paid a higher price for a product bought from the same seller than other customers, perceive lower price fairness and have a lower benevolence trust in the company, this is due to the fact that dynamic pricing runs against the concept of equity, that a seller should charge the same price for the same product for all customers (Garbarino and Maxwell, 2010). Hereby, the statement of Colquitt is confirmed as well, that says that perceived trust and perceived price fairness are highly interrelated (Colquitt, 2001).
Not only can personalized dynamic pricing have an influence on the feelings of the customer (perceived price fairness and trust), but these feelings can lead to increasing punishment behaviours, such as negative repurchase intentions. Especially customers that feel disadvantaged by dynamic pricing, perceive it as unfair. Because they feel like they’re treated unfair, they utilize the company as less trust-worthy, and express lower repurchase intentions (Garbarino and Maxwell, 2010; Grewal et al., 2004; Xia and Monroe, 2010). In conclusion, there seems to be a positive relationship between perceived price fairness, trust and repurchase intentions. Therefore, the hypotheses are as follows:
H1b: Personalized dynamic pricing will have a negative effect on consumers’ trust in the
company.
H1c: Personalized dynamic pricing will have a negative effect on consumers’ repurchase
14
2.3 Price framing as a moderating role
Perceptions of price fairness are generated not only from the perceived similarity between comparative participants, but more importantly from the perceived similarity between comparative transactions. In easier words, this means that the similarity of the transaction is a very important factor that influences perceptions of price fairness when differences in prices paid occur. If consumers pay a higher price for the same product, they may perceive the price as unfair (Xia, Monroe and Cox, 2004). To make these differences in price more acceptable to customers, the perceived similarity between the two transactions can be decreased by strategically offering the products differently; price framing (Weisstein, Monroe & Kukar-Kinnet, 2013).
Price framing, or presenting prices of products in different ways, are ways for a company to try to influence customers’ purchase decisions (Weisstein et al., 2013). Price framing can be done in various ways, this research divides price framing in two dimensions: presenting the lower price for the same product, from the same seller as a discount/price promotion, and presenting the lower price for the same product, from the same seller as just having a lower baseline price. Price framing in the form of a discount implies that someone else paid a lower price for the same product from the exact same seller, because he or she got a discount on her product. On the other hand, price framing in the form of a lower baseline price means that someone else paid a lower price for the same product from the exact same seller, but just for a lower price, without any reason. The big difference in these two forms of price framing, is the reasoning. With discount framing, the company explains the customer that the price has changed, and that he or she is paying a lower price than the normal baseline price. On the other hand, with just a lower price shown (lower baseline price) to the customer, the customer is not aware that the price is being changed, and that he or she might be paying more/less for the same product than someone else does.
15 An important use for effects of framing involves opportunity costs versus out-of-pocket costs, where people tend to underweight opportunity costs relative to out-of-pocket costs (Thaler, 1985). This tendency could have implications for perceived price fairness, and therefore for trust in the company and repurchase intentions, since these variables seem to positively relate. Most firms refer to a decrease in price as a ‘discount’ or ‘sale’, rather than just having a reduction in the selling price. This is due to the fact that cancellation of a discount or the end of a sale is perceived as more fair and acceptable than just an out of the blue price increase to the regular price again (Haugtvedt, Herr and Kardes, 2018). An important part of this phenomena, is the fact that the effect on the price is exactly the same; it just goes back to normal. However, the way of telling the people that the price is going back to normal, is differently framed. It can be assumed that this theory holds when it goes the other way around; the price going from a normal price, to a lower price (discount or a lower baseline price). Since the cancellation of a discount is perceived as more fair and acceptable than an increase in price to the regular price, it is expected that consumers’ negative perceptions of dynamic pricing could be mitigated when retailers use discount framing to consumers, as opposed to someone paying just a lower baseline price. Therefore communicating a lower price as a discount should lead to higher perceptions of price fairness, trust and repurchase intentions than it would do for someone just paying a lower baseline price (Garbarino and Maxwell 2010).
In conclusion, it is shown that discount framing is perceived as more fair and acceptable than just changing the baseline price. When linked to the expected main effects; a negative effect of personalized dynamic pricing on consumers’ perceived price fairness, trust in the company and repurchase intentions, it therefore is expected that discount framing could mitigate this effect. Because of this, the hypotheses are as follows:
H2a: The negative effect of personalized dynamic pricing on price fairness will be stronger
when prices are framed as a lower baseline price, as opposed to discount framing.
H2b: The negative effect of personalized dynamic pricing on trust in the company will be
stronger when prices are framed as a lower baseline price, as opposed to discount framing.
H2c: The negative effect of personalized dynamic pricing on repurchase intentions will be
16
2.4 Wealthiness as a moderating role
“Money is a fundamental source of power, status, and a crucial tool to facilitate achieving goals and desired outcomes” (Albalooshi, 2019). According to the World Bank (2018), 10% of the world is living in absolute poverty, which means earning less than $1.90 a day. In the U.S. alone, 40% of the Americans throughout the income distribution are concerned, every day, about having enough money to cover all their daily costs. These numbers are comparable to the numbers in Europe (Albalooshi, 2019).
Lacking money is threatening and aversive in nature, it reduces a sense of control and shifts the attention towards the present and now (Albalooshi, 2019). Since lacking money is threatening and aversive in nature, it is very likely that these type of people want to do anything in their capabilities to keep costs as low as possible, in order to not go bankrupt.
On the other hand, wealthy people do not have to fear about having enough money to pay ones bills. Also, these wealthy people seem generous, looking at donations. For example, Azim Premji (tech billionaire from India), Warren Buffet (Wall street magnate) and Bill Gates (Microsoft) together donated more than 100 billion dollars to charities (FHM, 2019). Donations of multibillionaires are not a measure of how much wealthy people care about their money, however it shows that wealthy people are way less anxious about money matters than people who are not wealthy.
This is supported by the findings of Kahneman and Deaton (2010), who found that additional income does not enhance happiness, however it decreases negative feelings, distrust, sadness and anxiety, especially in relation to ones finances. For example, Americans with a lower net worth, report a stronger need to get the most for the money they spent, and were anxious about not having enough money compared to the people with greater assets (Hayhoe, Cho, DeVaney, Worthy, Kim and Gorham, 2012). Next to this, Lim and Sng (2006) found that family income is negatively related to money anxiety.
17 expected that these main effects are stronger for less wealthy people. This is due to the fact that wealthy people are less anxious about money matters, while less wealthy people are (very) anxious about money matters. Therefore, the hypotheses are as follows:
H3a: The negative effect of personalized dynamic pricing on price fairness will be stronger for
less wealthy people than for wealthy people.
H3b: The negative effect of personalized dynamic pricing on trust in the company will be
stronger for less wealthy people than for wealthy people.
H3c: The negative effect of personalized dynamic pricing on repurchase intentions will be
stronger for less wealthy people than for wealthy people.
2.5 Customer segments as a moderating role
Customer segments are self-created segments, based on online/offline shopping behaviour (purchase channel, information channel, shopping affinity and risk), and need for cognition. Nowadays, many retailers made it possible for consumers to shop both online and offline. You could say that online & offline are two different segments; people who like online shopping, and people who like offline shopping. There are many reasons to think of why one could prefer one type of shopping more than the other type of shopping. One could simply perceive a type of shopping as more fun or less risky, and could therefore choose between online/offline shopping, for example. However, research claims that online shopping can be divided in two segments as well; those who value their time and convenience & those who compare prices between multiple retailers and afterwards make their purchase. Looking at these two segments, there is one big difference; price-sensitivity, which is how one is worried about the price/is sensitive for the price of a product. The ones valuing their time and convenience may not be very price sensitive (however, they could be), while the ones that browsing on the internet looking for the lowest price definitely are price sensitive (Kannan and Kopalle, 2001). Knowing that there are differences in different segments in online shopping, it is hard to make a clear price-sensitivity difference between online and offline shoppers.
18 there are could also be less-price sensitive online shoppers out there (the ones valuing their time and convenience). Since people that shop online to save time and for their convenience can at the same time also be price sensitive, there can be concluded that, most of the time, offline shoppers are less price sensitive than online shoppers, which can have an impact on acceptance of dynamic pricing.
Next to online/offline behaviour, need for cognition might also influence the effect of dynamic pricing on acceptance. “Need for cognition is a personality trait that represents the varying level of cognitive resources and styles that individuals employ when processing information. It examines the extent to which individuals enjoy and perform an effortful cognitive process of information” (Cacioppo & Petty, 1982). Research shows that individuals with a high need for cognition, appreciate more complex tasks and are less likely (than people with a low need for cognition) to reduce their efforts in situations where cognitive efforts typically reduce (Cacioppo & Petty, 1982, 1984; Petty, Cacioppo, & Kasmer, 1985). Related to personalized dynamic pricing, people with a high need for cognition, are the people who like to think/think a lot about personalized dynamic pricing. It is expected that people with a high need for cognition, will think more about the individual differences that people pay for the same product, and therefore will more likely to think of it as unfair. On the contrary, people with low need for cognition are expected to not think about the individual differences too much, and therefore are less likely to think of it as unfair.
In conclusion, people differ, and different type of people are likely to respond different to situations. By clustering people, groups of similar people are made, in order to see whether the effects of personalized dynamic pricing differ for different customer segments. Because of the argumentation given above, the hypothesis is as follows:
H4: The negative effect of personalized dynamic pricing on acceptance differs for different
19
2.6 Conceptual model
Based on the literature, relationships between the chosen variables have been constructed. The independent variable ‘personalized dynamic pricing’ is expected to have a negative effect on the dependent variable ‘acceptance’, which consists of three variables; perceived price fairness, trust in the company and repurchase intentions. Since these three variables are expected to have a positive relationship, the moderators are expected to have the same effect on each individual variable of the dependent variable ‘Acceptance’. The moderators in this study are price framing, wealthiness and customer segments. Customer segments will mainly be based on offline/online shopping behaviour and need for cognition. The conceptual model is shown in figure 2.6.
20
3. Methodology
This chapter explains the study design, the measurements, and the analyses that have been used in order to answer the research question.
3.1 Study design
The primary data used in this study is gathered through an online experiment, using the platform Qualtrics. Internet is selected as the best method to gather data, since it is a relatively cheap and fast method. Besides this, the social desirability is low, which could lead to more honest answers than when such an experiment would have been performed in an offline setting. The respondents received a Qualtrics link which led them to the online survey via Whatsapp, Facebook, Instagram, Twitter and E-mail. Since it is an online experiment, the respondents needed to have access and knowledge of the internet. The language chosen for the online experiment is English, since some of the respondents cannot read Dutch. English on the contrary, is a universal language that almost everyone can read. The respondents have been reached through the researchers network, and contained people of all genders and ages, in order to get the most reliable answers.
3.1.1 Online experiment
The primary data for the study is gathered through an online experiment. In this experiment, participants were asked to read a story, where they had to imagine that they bought a brand new Samsung television for €700,00 at the website of the electronics shop that they visit occasionally. When the television got delivered the next day, the respondents found out that their neighbour bought the exact same TV, at the same website and moment. However, the neighbour did not pay €700,00, but €600,00, for the exact same tv, at the same website and moment.
3.1.2 Personalized dynamic pricing & price framing
What the participants did not know, is that they were randomly assigned to one of the three groups, see table 3.1.
Groups Form of dynamic pricing
Control group No presence of dynamic pricing
Experimental group 1 Lower baseline price (Low_price)
Experimental group 2 Discount
21 Randomly assigning people to groups ensures that the results of the experiment are not being affected by pre-existing differences between respondents, but are only caused by the manipulation of the variables. By making a distinction between these three groups, the effect of personalized dynamic pricing can be measured by comparing the control group to both experimental groups.
Each group got presented a different story. In the control group, there was no form of personalized dynamic pricing. This means that the neighbour also paid €700,00 for the same television; the respondent and the respondent’s neighbour paid the same price. In both experimental groups, this was not the case. In the experimental groups, participants found out that their neighbour paid only €600,00.- for the same tv, from the same seller. By making a dummy variable with a value of ‘1’ if one is in one of the two experimental groups, versus a value of ‘0’ if one is in the control group, the independent variable personalized dynamic
pricing is created. However, there is a difference between the two experimental groups. In
experimental group 1, the neighbour paid €600,00.- for the television, because the price at the website was simply stated at €600,00.-. In experimental group 2, the neighbour paid €600,00.- for the television, because he got a discount. The difference between these two groups is they way the prices are framed. By making a dummy variable with a value of ‘1’ if the respondent was in the lower baseline group, versus a value of ‘0’ if one was not, the moderating variable
lower baseline price has been created, which is one way of price framing.
3.1.3 Pre-test
22
3.2 Measurement
This paragraph contains information about how the dependent variables and moderators have been measured and constructed.
3.2.1 Dependent variables
After the manipulation, a total amount of nine questions based on perceived price fairness, trust
in the company, and repurchase intentions were asked, to measure acceptance. For these three
variables, three questions per variable have been asked, see table 3.2.3 on page 23. Respondents answered these questions based on a measurement scale. To increase validity, measurement scales from previous researches (see table 3.2.3) have been used for the measurement of the variables related to the constructs. These nine questions have all been measured on a 6-point Likert scale (1 = “strongly disagree” and 6 = “strongly agree”). The researcher consciously chose for a 6-point Likert scale. By choosing a 6-point Likert scale, there is no ‘middle point’, which means that the respondent is being forced to really think about the question, and cannot simply go for the middle option without thinking about the question too much (Thompson, 2018). In order to test for scale reliability, the Cronbach’s alpha is calculated. By calculating this, the internal consistency has been measured. All Cronbach’s alpha scores have been put in table 3.2.1. Looking at the Cronbach’s alpha scores, we see that all scores are above .8, which implies that the internal consistency is good (Kreulen, 2010). Because the internal consistency is positively tested, the questions for each DV can be combined, which leads to averaged scores per DV. With this manipulation, the data for the DV’s are in a form that can readily and accurately be analysed (Friedland, 2020).
Variable Cronbach’s alpha (a) Number of items
Price fairness .833 3
Trust in the company .870 3
Repurchase intentions .920 3
Table 3.2.1: Cronbach's alpha scores
3.2.2 Wealthiness
Wealthiness will get measured by asking two questions concerning financial stress as opposed
23 negative thing), it means that the higher one scores on this variable (the more stress one has), the less ‘wealthy’ one is. In order to test for scale reliability, the Cronbach’s alpha is calculated. The reliability analysis on the questions regarding ‘wealthiness’ showed that the questions have
a = .619, which means that the internal consistency is good (Kreulen, 2020). Because the
internal consistency is positively tested, the questions for wealthiness can be combined. After this, the wealthiness scores got mean centered, in order to improve interpretation in the linear regressions.
3.2.3 Customer segments
Furthermore, customer segments have been identified based on measured criteria. The criteria segments are based on the active variables purchase channel, information channel, shopping
affinity, risk and need for cognition. Respondents answered three questions per (active) variable
(see table 3.2.3). For the purchase channel variable, respondents answered three questions based on a 5-point Likert scale (1 = “always offline” and 5 = “always online”). The reason a 5-point scale Likert scale has been used instead of a 6-point Likert scale, is because people could equally purchase a product offline/online. For the other four active variables mentioned above, a 6-point Likert scale has been used (1 = “strongly disagree” and 6 = “strongly agree”). Again, the Cronbach’s alpha scores have been calculated, which are shown in table 3.2.2. Looking at the Cronbach’s alpha scores, we see that all scores are above .79, which implies that the internal consistency is good (Kreulen, 2010). Because the internal consistency is positively tested, the questions for each active variable can be combined, and can be used in hierarchical clustering.
Variable Cronbach’s alpha (a) Number of items
Purchase Channel .947 3
Information channel .798 3
Shopping affinity .835 3
Risk .795 3
Need for cognition .840 3
Table 3.2.2: Cronbach's Alpha scores
24
Construct CA AVG SD
Acceptance
Price fairness (own development based on Campbell, and Margaret, 2007) .83 3.50 1.17 €700,00 is a good price for the TV
Comparing the price you paid for the TV to the price your neighbour paid for the TV, the price is fair
The price paid for the TV is fair
Trust in the company (own development based on Gunia, Brett, and
Nandkeolyar, 2011)
.87 3.49 1.13 The company keeps promises and commitment to all their customers
The company that I bought the TV of is honest I trust the company that I bought the TV of
Repurchase intentions (own development based on Zhang, Agarwal, and Henry,
2011)
.92 3.26 1.09 I will consider this company as the first choice to buy electronic products in the
future.
I will buy more electronic products from this company in the future I will come back to this company to buy a similar product in the future
Wealthiness (own development based on Doss, Rhoades, and Stanley, 2009) .62 2.64 1.13 I have difficulties paying my bills and having money remaining at the end of each
month.
Financial concerns cause me stress
Customer segments
Purchase channel (own development) .95 3.16 1.11
My most recent electronic purchases were bought ... b
In the near future, I am most likely to buy my electronic products ... b
Generally, I purchase my electronic products ... b
Information channel (own development based on Homburg, Lauer, and Vomberg,
2019)
.80 4.37 1.12 Generally, I inform myself at an online shop about products I intend to purchase
Generally, I inform myself at an online shop about electronical products I intend to purchase
I prefer the internet over a physical store when it comes to informing myself about a product I intend to purchase
Shopping affinity (own development based on Homburg, Lauer, and Vomberg,
2019)
.84 3.28 1.20 I prefer online shopping over offline shopping
I like online shopping more than offline shopping
Going online shopping is one of the enjoyable activities in life
Risk (own development based on Malhotra, Sung, and Agarwal, 2004) .80 3.44 .94 In general, it would be risky to give my information to online companies
Providing online companies with my information would involve many unexpected problems
I do not feel safe giving my information to online companies
Need for cognition (own development based on Tam, and Ying Ho, 2005 ) .84 3.87 1.11 I like to have to do a lot of thinking
I prefer complex problems to simple problems
I prefer to do something that challenges my thinking abilities rather than something that requires little thought
Table 3.2.3: Overview of all constructs
Notes: Items measured on a Six-point Likert scale (1 = “strongly disagree” and 6 = “strongly agree”), unless
otherwise stated
aFormative measurement
bAnchors: 1 = “always offline” and 5 = “always online”
25
3.3 Analyses
The tests conducted in this experiment all serve one goal; finding an answer to the research question: ‘To what extent do people accept being disadvantaged by dynamic pricing, and does
this effect differ for different price framing tactics, for how wealthy one is and for different customer segments?’. In order to get an answer to this question, several analyses have been
conducted.
3.3.1 Hypothesis testing
In order to test the hypotheses, regression analyses have been conducted for every dependent variable. This means that there have been conducted separate regression analyses for price fairness, trust and repurchase intentions. The independent variables in these analyses are the same in every regression analysis. Below you can find the regression equations, where 𝛽0is the intercept, and 𝛽1 is the independent variable personalized dynamic pricing.
𝑃𝑟𝑖𝑐𝑒𝐹𝑎𝑖𝑟𝑛𝑒𝑠𝑠 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 + 𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀 𝑇𝑟𝑢𝑠𝑡 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 + 𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀 𝑅𝑒𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝐼𝑛𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑠 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 +𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀
26
3.3.2 Sample description
Before going into the analyses, the dataset needs to be checked for missing values and outliers. In total, 94 people started the survey, however, 12 people did not finish the survey. These responses got removed, because not filling in the questionnaire for 100% can imply that the respondent was not responding to the survey in a serious way, which means that the sample size is reduced to n = 82. These 82 people are divided over three different groups; the control group, experimental group 1 (Low_price) and experimental group 2 (Discount), see table 3.3 for demographics (per group).
Distribution Control Group Experimental Group (Low_price) Experimental Group (Discount) Total
Age (in years) 36.4 32.4 34.8 34.5
Gender Male Female 28.6% 71.4% 50.0% 50.0% 38.5% 61.5% 39.0% 61.0% Household Income < €20.000 €20.000 - €40.000 €40.000 - €60.000 > €60.000 39.2% 32.1% 7.1% 21.4% 25.0% 32.1% 25.0% 17.9% 42.3% 19.2% 30.8% 7.7% 35.4% 28.0% 20.7% 15.9% Respondents 28 28 26 82
27
4. Results
This chapter describes all the different analyses that were used in order to investigate whether the drawn hypotheses in the theoretical part were correct. Prior to effect-testing, a hierarchical cluster analysis has been conducted in order to create customer segments, since ‘customer segments’ is one of the moderating variables that gets tested. After preparing the data for analysis, the effect of personalized dynamic pricing gets tested on perceived price fairness, trust in the company and repurchase intentions. Next to this, the effect of the moderators ‘price framing’, ‘wealthiness’ and ‘customer segments’ are taken into account. Finally, all results will get visualized in order to see what new insights this study found.
4.1 Customer segments
After selecting the active variables that will get used in the cluster analysis (see paragraph 3.2.3), a hierarchical cluster analysis has been conducted using Ward’s method. This method allows one to create segments which are minimizing the variation, which makes the segments internally resemble each other most (Gijsenberg, 2019). Next to this, the Squared Euclidean distance is chosen, and the variables are standardized to Z scores in order to not let the scales impact the results of the analysis (Gijsenberg, 2019).
After performing the hierarchical cluster analysis, the number of clusters have to be chosen. The coefficients of the agglomeration schedule have been plotted in order to visually see what number of clusters suits the data the most. Looking at the plot (figure 4.1.1), it is clear to see that the ankle in the line is at two clusters, which implies two clusters is the most suitable number of clusters for this dataset.
28 Another way of looking how many clusters is best, is by looking at the dendrogram (figure 4.1.2). When looking at the dendrogram, it is clear that two clusters indeed is the best number of clusters, since the distance from three clusters to two clusters is relatively small, while the distance from two clusters to one cluster is very big, which implies that the costs would be high and a lot of information would be lost. Therefore, there is decided to proceed with two clusters.
Figure 4.1.2: Dendrogram
In order to test clusters on significance of differences on the active variables, a One-Way ANOVA test has been conducted. To test if the differences between the mean are significant, the p-value is compared to the significance level (=.05). The test shows significance, since all the p-values are <.05, see table 4.1.1. Since the cluster analysis contains less than three groups, post-hoc analysis are not needed, because if a significant difference would be found, it would be between the two groups (cluster one and cluster two) measured (Allen, 2017).
Variable P-value
Purchase Channel .000
Information channel .000
Shopping affinity .000
Risk .003
Need for cognition .001
29 Now that two clusters have been created, they should be interpreted. For this interpretation, the passive variables – gender, age and household income – come in. For the age variable, there is decided to recategorize them into age groups, because the average age of a cluster does not tell a lot, however what kind of age groups belong to what cluster, could be important. Looking at the distribution of the age groups, there can be concluded that the data is quite skewed, since 54.9% of the respondents are between 16 and 25 years old. Because of this, having an equal amount of people in equal ranged grouped is not possible, therefore the age groups look as follows, see table 4.1.2.
Age group Number of people
16-25 years old 45 (54.9%) 26-35 years old 5 (6.1%) 36-45 years old 6 (7.3%) 46-55 years old 12 (14.6%) 55+ years old 14 (17.1%)
Table 4.1.2: Age group distribution
The clustering is done, and all variables are ready for interpretation. It has to be mentioned that the scores of the active variables are calculated by using the output of the comparisons of the means (see appendix 4b), and on the other hand, the scores of the passive variables are calculated by using the output of the crosstabs. Table 4.1.3 summarises the interpretations of the two clusters by means of the active and passive variables.
Variables Digital innovators Conservatives
Total people 51 31
Gender 37% (M) / 63% (F) 42% (M) / 58% (F)
Relative dominant Age Group
16-25 years old 55+ years old
Annual household income Many low earners; <20.000 (39.2%)
Many high earners; >60.000 (19.6%)
Income between €20.000 and €60.000 (61.2%)
Purchase channel Mostly online
(AVG = 3.76, SD = .78)
30 Information channel Mostly online
(AVG = 4.86, SD = .69)
Slightly more online than offline
(AVG = 3.57, SD = 1.23) Shopping affinity More affinity with online
shopping (AVG = 3.83, SD = .90)
More affinity with offline shopping (AVG = 2.37, SD = 1.08)
Risk Medium risk
(AVG = 3.20, SD = .85)
Above medium risk (AVG = 3.83, SD = .96) Need for cognition Higher than average need for
cognition (AVG = 4.18, SD = 1.04)
Average need for cognition (AVG = 3.37, SD = 1.05)
Table 4.1.3: Cluster interpretation
Digital innovators: This group contains relatively young people, 71.1% of the people between
16 & 25 belong to this group. What is interesting to see, is that this group contains relatively much low income (69% of the people that have an household income of < €20.000 belong to this group), but also high income people (76.9% of the people that have an household income of > €60.000 belong to this group). The thing that differentiates this group the most, is that they seem to prefer the online channel to the offline channel. Next to this, these people are not scared to give companies their private information, and have an higher than average need for cognition.
Conservatives: This group contains relatively older people, 78.6% of the people that are 55+
31
4.2 Regression analyses
In this paragraph, the (main) effects of personalized dynamic pricing on perceived price fairness, trust in the company and repurchase intentions get tested. Next to this, (moderating) variables that could possibly moderate these main effects get tested. By doing this, the formulated hypotheses in the literature review all get tested. Based on the outcome of the tests, the hypotheses will either get accepted or rejected.
4.2.1 Effect of personalized dynamic pricing on price fairness
In order to test the (main) effect of personalized dynamic pricing on perceived price fairness, a linear regression analysis has been conducted.
𝑃𝑟𝑖𝑐𝑒𝐹𝑎𝑖𝑟𝑛𝑒𝑠𝑠 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 + 𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀
When looking at the linear regression output (see appendix 5a), we see that the output reveals a marginally significant main effect of personalized dynamic pricing, since p = .10, which is slightly above the threshold of .05. The coefficient shows β = -1.072, which implies that personalized dynamic pricing has a (marginally significant) negative effect on consumers’ perceived price fairness. Therefore, H1a is partially accepted.
Next to this, it is tested whether the moderating variables could possibly moderate the effect between personalized dynamic pricing and perceived price fairness. When we look at the interaction effect of the LowerBaselinePrice, we see that the output reveals a non-significant moderating effect of price framing on the relationship of personalized dynamic pricing on perceived price fairness, since p = .32. The coefficient shows a negative effect of a Lower Baseline Price of -.247, which implies that the price framed as just a lower baseline price has a stronger negative effect on perceived price fairness as opposed to discount framing, which was as expected. However, we cannot interpret this coefficient since it is not significant, and therefore, H2a can be rejected.
32 Finally, we look at the interaction effect of customer segments. Yet again, the moderating effect is insignificant, since p = .51. The coefficient shows β = -.286, which implies that effect of personalized dynamic pricing on perceived price fairness differs per customer segment, which is as expected. However, the outcome is insignificant, and therefore we cannot state that the main effect differs for different customer segments.
4.2.2 Effect of personalized dynamic pricing on trust in the company
In order to test the (main) effect of personalized dynamic pricing on trust in the company, a linear regression test has been conducted.
𝑇𝑟𝑢𝑠𝑡 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 + 𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀
When looking at the linear regression output (see appendix 5b), we see that the output reveals a marginally significant main effect of personalized dynamic pricing, since p = .10. The coefficient shows β = -1.067, which implies that personalized dynamic pricing has a (marginally significant) negative effect on consumers’ trust in the company. Therefore, H1b is partially accepted.
Again, we look at whether the moderating variables could possibly moderate the effect between personalized dynamic pricing and perceived price fairness. When we look at the interaction effect of LowerBaselinePrice, we see that the output reveals a significant moderating effect of
price framing on the relationship of personalized dynamic pricing on trust in the company, since p = .05. The coefficient shows β = -.504, when compared to the main effect, this difference in
price framing is very strong. This implies that the price framed as just a lower baseline price has a stronger negative effect on trust in the company as opposed to discount framing. Therefore, H2b can be accepted.
33
4.2.3 Effect of personalized dynamic pricing on repurchase intentions
In order to test the (main) effect of personalized dynamic pricing on repurchase intentions, a linear regression test has been conducted.
𝑅𝑒𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝐼𝑛𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑠 = 𝛽0+ 𝛽1𝐷𝑃 + 𝛽2 𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3 𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝛽4𝐷𝑃∗𝐿𝑜𝑤_𝑃𝑟𝑖𝑐𝑒 +𝛽5𝐷𝑃∗𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽6𝐷𝑃∗𝐶𝑢𝑠𝑡𝑆𝑒𝑔𝑚𝑒𝑛𝑡𝑠 + 𝜀
When looking at the linear regression output (see appendix 5c), we see that the output reveals a significant main effect of personalized dynamic pricing, since p = .02. The coefficient shows β = -1.439, which implies that personalized dynamic pricing has a (significant) negative effect on consumers’ repurchase intentions. When comparing this effect to the other main effects, it is clear to see that the negative effect is stronger for repurchase intentions than it is for price fairness and trust. This implies that not everyone may perceive personalized dynamic pricing as (very) unfair, and not everyone will immediately distrust the company, however,
unconsciously, it affects their repurchasing behaviour, which is an interesting finding. We can
conclude that H1c is accepted.
Looking at the interaction effects of the moderating variables, all the signs of the betas are as expected. However, all moderating effects show not to be significant, since every p value is above the threshold. This implies that we cannot interpret these findings, which implies that
H2c and H3c are rejected, and that the main effect does not differ for different customer
34
4.3 Overview of the results
This paragraph functions as a visual overview of all the results gathered in this study. By looking at table 4.3, it is clearly visible what kind of relationships and effects have been found. For every hypothesis, Personalized dynamic pricing is the independent variable, therefore it is not in the table. Although the fact that every expected sign is – (minus), this is incorporated in the table, so it is easily viewable if the coefficient matches the expected sign. Appendix 1 shows a table with all hypotheses in one picture, so that it is easily visible what hypothesis number represents which hypothesis.
Hypothesis Dependent variable
Moderator Expected sign
Coefficient P-value Accepted or Rejected H1a Price fairness N/A - -1.072 .095* Accepted H2a Price fairness Price framing - -.247 .321 Rejected H3a Price fairness Wealthiness - -.245 .193 Rejected H1b Trust in the company N/A - -1.067 .101* Accepted H2b Trust in the company Price framing - -.504 .048*** Accepted H3b Trust in the company Wealthiness - -.131 .491 Rejected H1c Repurchase intentions N/A - -1.439 .023*** Accepted H2c Repurchase intentions Price framing - -.301 .216 Rejected H3c Repurchase intentions Wealthiness - -.018 .923 Rejected H4 N/A Customer segments
N/A N/A >.05 Rejected
35
4.4 Regression assumptions
The results of this study have been found by performing several linear regression analyses. In order to validate these results, regression assumptions have been tested. Violation of these assumptions can result in outcomes that are biased and misleading (Flatt, Turner, Baker and Passmore, 2019), which means that every regression assumption should hold. The assumptions that are tested are linearity, normality, homoscedasticity, and absence of autocorrelation and multicollinearity.
4.4.1 Linearity
The linearity assumption requires the relationship of the independent variables and the mean of the dependent variable to be linear (Nimon, 2012). Nonlinear data fitted into a linear model causes incorrect estimations/predictions, therefore a violation of this assumption is considered as extremely serious (Nau, 2018). In this research, the independent variable is a binary/dummy variable, where 0/1 is just the absence/presence of personalized dynamic pricing. Dummy variables meet the assumption of linearity by definition, this is due to the fact that a dummy variable creates two points, and two points define a straight line. In other words, there is no such thing is a non-linear relationship for a dummy variable, and therefore it does not have to be tested for. We can conclude that the linearity assumption holds.
4.4.2 Normality
The normality assumption requires the error terms to be normally distributed (Osborne and Waters, 2002). A violation of this assumption causes difficulties for the calculation of p values, in order to test for significance. To test the normality assumption, the standardized residuals have been plotted for each dependent variable, see figure 4.4.1, 4.4.2 and 4.4.3.
Figure 4.4.1: Normality plot price fairness
36 In the y axis, you see the expected cumulative probability, where you see the observed probability in the x axis. If the error terms are perfectly normal, all dots would be on the line. In this case, there is a little deviation. However the deviation is not drastically, which means we can assume normality. This is also in line with the theory of Field (2009), which says that the error term can be assumed to be normal if the sample is greater than 30, which is the case. Therefore, we can conclude that the normality assumption holds.
4.4.3 Homoscedasticity
The assumption of homoscedasticity requires the error terms to have a constant variance. The opposite of homoscedasticity is heteroscedasticity, when this is present, the standard errors are biased, which leads to biased results. To test for homoscedasticity, the standardized residuals have been plotted, see figure 4.4.4, 4.4.5 and 4.4.6.
When looking at the graphs, we see that there is slight heteroscedasticity. However, Osborne and Walters (2002) state that only slight violations of the assumptions are generally acceptable. Therefore we can conclude that the homoscedasticity assumption holds.
4.4.4 Autocorrelation
The assumption of autocorrelation requires observations to be independent from one another (Nimon, 2012). Autocorrelation occurs when residuals are dependent on each other. A violation of this assumption can cause problems in regression analyses. To test for autocorrelation, Durbin-Watson tests have been conducted for each regression analysis. This test gives a value between ‘0’ to ‘4’, where the value of ‘2’ would assume no autocorrelation, therefore one would want the outcome to be close to a value of ‘2’. Field (2009) suggests that a value in the range of 1.5 and 2.5 are relatively normal, and that values beneath ‘1’ or above ‘3’ cause for concern. When applying the Durbin-Watson test on the regression models used in this research, the
37 values are all close to a value of 2 (see table 4.4.1). Therefore we can conclude that the autocorrelation assumption holds.
Dependent variable Durbin-Watson score
Price Fairness 1.960 Trust in the company 1.886 Repurchase intentions 1.918
Table 4.4.1: Durbin-Watson scores
4.4.5 Multicollinearity
The assumption of multicollinearity requires the predictor variables not to be correlated, because otherwise the individual effects of predictor variables cannot be separated. To test if predictor variables are multicollinear, VIF scores have been calculated. According to Field et al. (2012), VIF scores below 10 assume no multicollinearity. Looking at the VIF scores (table 4.4.2), we see that only one variable included in the regression analysis exceeds a score of 10. However, this is an interaction effect, which is likely to be multicollinear, since it is an interaction with another variable included in the regression, therefore it is not considered as a problem. We can conclude that the multicollinearity assumption holds.
Independent variable VIF
PersonalizedDP 9.573 PD*LowerBaselinePrice 1.457 PD*Wealthiness 2.874 PD*Customer segments 10.825 Wealthiness 2.813 Customer segments 2.854
Table 4.4.2: VIF scores
38
5. Conclusion & discussion
The aim of this study was finding answers to the research question: To what extent do people
accept being disadvantaged by dynamic pricing, and does this effect differ for different price framing tactics, for how wealthy one is and for different customer segments? After conducting
several analyses, answers have been found. Next to this, managerial implications, limitations and directions for future research are discussed.
5.1 Conclusion
First, there can be concluded that personalized dynamic pricing has a negative effect on acceptance, since this research shows a negative effect for perceived price fairness, trust in the company and repurchase intentions. These findings confirm the findings of Garbarino and Maxwell, 2010. As expected, there is a positive relationship between these three dependent variables. An important new finding on top of prior research, is that the negative effect of personalized dynamic pricing is stronger for repurchase intentions than it is for perceived price fairness and trust in the company. This implies that not everyone may perceive personalized dynamic pricing as (very) unfair, and not everyone will immediately distrust the company, however, unconsciously, it affects their repurchasing behaviour, which could really harm a company.
Second, there can be concluded that price framing affects the relationship between personalized dynamic pricing and acceptance. Although the fact that people did not consciously perceive discount framing as more fair than lower baseline price framing (p = .321), it had negative implications for peoples’ trust in the company, which unconsciously could lead to lower for repurchase intentions, just like the effect that personalized dynamic pricing has on repurchase intentions. This study showed that framing a lower price as a lower baseline price as opposed to a discount, results in less trust in the company, which are in line with the findings of Garbarino and Maxwell, 2010. Therefore, framing the price as a lower baseline price as opposed to a discount, has stronger negative implications for companies.
39 it is very hard to see a clear distinction between different customer segments and wealthy/less wealthy people, which could be the reason that the effects of these moderators were insignificant, but in the expected direction. In conclusion, since the main results (effect on price fairness, trust in the company and repurchase intentions) are almost the same for everyone in the experimental groups, it is very likely that there is a less clear distinction in main results between different type of people, which resulted in non-significant findings for the moderators that are linked to people; wealthiness and customer segments.
5.2 Implications
These findings have important implications for retailers. Based on the findings in this research, retailers are facing a very important question; are the possible financial benefits on the short-term more important than the possible financial benefits on the long-short-term. I’d suggest, based on the literature and findings in this research: no. In the long run, it is very important to have a good relationship with your customers (Jain and Singh 2002; Shin and Sudhir 2010; Venkatesan and Kumar, 2004), because these customers are the most important source of income for most companies. Since it is shown that when people find out that a certain company makes use of personalized dynamic pricing it leads to lower perceived fairness, less trust in the company and lower repurchase intentions, it could hurt a company badly when people indeed find out that it makes use of personalized dynamic pricing. Therefore, my advice would be not to implement personalized dynamic pricing as a company, since it can hurt the company on the long term. For now, not many people know that personalized dynamic pricing occurs in the market, so it does not really hurt the companies yet. However, based on this research, when people in the future find out that personalized dynamic pricing does happen in reality, I expect negative implications for the companies that make use of personalized dynamic pricing. As explained, this could harm the company badly. However, if a company decides to take the risk, and make use of personalized dynamic pricing, it should frame the lower price as a discount. Framing a lower price as a discount mitigates the effect of people losing trust in the company, in case people find out that it makes use of personalized dynamic pricing.
5.3 Limitations and directions for future research
40 between different type of people, which resulted in non-significant findings for the moderators related to people; wealthiness and customer segments.
The second limitation of this study is that it is conducted in the form of an online survey instead of a real situation. Participants of the research were asked to imagine that the story presented to them actually happened, however, it did not in real life, since they were just sitting behind their laptop/mobile phone. This could have an effect on the decisions made (Sheeran, 2005), especially because the story told is money related, which is a tricky topic in real life. Future research could test the impact of personalized dynamic pricing in a real life setting, where people actually buy a product, and someone else buying the same product for a lower price because of a discount/lower baseline price, and look at the effects of this situation.
The third limitation of this study is the fact that this study is conducted during times of a COVID-19 pandemic. A lot of people cope with stress during times of this pandemic (LHV, 2020), which could possibly have influenced results in this research. Next to this, there were questions about online/offline behaviour. Since the COVID-19 virus forces people to do as much as possible online as opposed to offline, it is very likely that this has pushed the results of the questions regarding online/offline behaviour more to the ‘online’ side (which could be a reason why the ‘digital innovators’ group consisted of 51 people, and the ‘conservatives’ consisted of 31 people). Also, questions have been asked about need for cognition. It could very well be that in times of a pandemic - where people are stressed -, people are not in the mood to think about things a lot, which could also have pushed the results towards a lower average score for need for cognition. In conclusion, without the COVID-19 virus, the results of this study could have been different. For future research, I suggest to do a similar research when the pandemic is over, to see if the results differ.
