Effect of consumer characteristics and contextual variables on the dynamic pricing implementation

(1)

0

Effect of consumer characteristics and

contextual variables on the dynamic pricing

implementation

Master Thesis (Resit)

By

Yauheniya Vauchok S2713810

y.vauchok@student.rug.nl

Rijksuniversiteit Groningen - Faculty of Economic and Business

MSc. Marketing Intelligence

August 02, 2020

(2)

1

practice of varying the price for a product or service to reflect changing market conditions, in particular the charging of a higher price at a time of greater demand.” Harrison, Keskin, and Zeevi (2012) mention that dynamic pricing is an attempt of the retailer to set the optimal (possibly high) prices in alignment with the highest market’s response, McAfee and Te Velde (2006) mention that DP is “a set of pricing strategies aimed at increasing profits”(p.2). This phenomenon continues to gain more and more attention and promises even wider use in the near future.

DP may have unpredicted outcomes. It might help build a great company with the whole new

market, such as Amazon, Coca-Cola, and Airbnb (Shartsis, 2019; Weiss & Mehrotra, 2001) or

face lawsuits for unfair treatment and discrimination, such as Uber or Amazon did (Beck, 2016). Some may view it as a topic of ethical concern or even price discrimination (Weiss & Mehrotra, 2001; Faruqui, 2012; van der Rest, Wang & Miao, 2020). Others may see it as a great opportunity to earn the highest possible revenue (Garbarino & Lee, 2003; Hinz, Hann, & Spann, 2011). The importance of the right pricing with the great certainty about the predictions of outcomes cannot

be overestimated. For that the understanding of the consumer is the key (Bonney, Zhang, Head,

Tien & Barson, 1999).

Successful dynamic pricing requires estimating the best pricing solutions for the customers (McAfee & Te Velde, 2006). However, the estimation of the optimal price for undifferentiated and expectedly homogeneous population of customers is myopic and primitive. The ability to differentiate between consumers and their varying preferences, and tailoring these differences with the use of DP models in e-commerce has a direct impact on their profits (Dasgupta & Hashimoto, 2004; Hinz, Hann, & Spann, 2011).

(6)

5

segmentation (Smith, 1956; De Keyser, Schepers & Konuş, 2015; Konuş, Verhoef & Neslin, 2008). The ability to target the right segment, understand the needs of the segment and purchase probability of the consumers in that segment is critically important for the success of any company. Luckily, Big data and technologies allow us to collect and analyse such an amount of information that it becomes possible to study in-depth any segment. The right price setting allows the company to maximise their profits and benefit the customers (Garbarino & Lee, 2003; Hinz, Hann, & Spann, 2011; McAfee, Te Velde, 2006). Such a strategy is not new. As an example of this, Hinz, Hann and Spann, (2011) mention that a firm can increase their profit up to 20% without losing their customers. Same Faruqui and Palmer (2011) mentioned their theory about the trade off for each consumer and optimal price in dynamic pricing.

Nowadays, the legal restrictions and privacy policies do not allow the companies to collect individual data as marketeers would wish to. Due to that the self-segmentation becomes widely used by online retailers. The so-called fences (rules that the customer accepts) are often met in the market, which allow customers to self-segment by the basic criteria (age, gender; and by which may include him/herself to the specific group). But the extent of effectiveness of such segmentation is doubted (Wirtz & Kimes, 2007). The ability of the company to determine and use the most important characteristics of consumers available for the company may determine the ability to assign consumers to the right segment, understand the segment, and thus set the right pricing strategy.

(7)

6

implementation. As such, the research helps to find what is the effect of dynamic pricing for the specific types of consumers.

3 LITERATURE REVIEW

The dynamic pricing is “the practice of varying the price [...] in particular the charging of a higher price at a time of greater demand.” (Oxford Dictionaries; 2020). Harrison, Keskin, and Zeevi (2012) mention that dynamic pricing is an attempt of the retailer to set the optimal (possibly high) prices in alignment with the highest market’s response, thus “the seller offers prices sequentially to individual customers, observing either success or failure in each sales attempt. There is an underlying demand model, also called a demand function or demand curve, that gives the probability of success as a function of the price that is offered.” (p.570). McAfee and Te Velde (2006) mention that DP is “a set of pricing strategies aimed at increasing profits”(p.2). This phenomenon continues to gain more and more attention and promises even wider use in the near future. Some may view it as a topic of ethical concern or even price discrimination (Weiss & Mehrotra, 2001; Faruqui, 2012; van der Rest, Wang & Miao, 2020). Others may see it as a great opportunity to earn the highest possible revenue (Garbarino & Lee, 2003; Hinz, Hann, & Spann, 2011). Anyway, the algorithms of the DP take an increasingly important position.

Pricing is one of the most important determinants of the business strategy and a company’s success. Currently, the most popular approach is to use the demand curve and probability of purchase. The models used in practice contain mainly fixed and initially uncertain parameters with fixed variation ranges, while the demand curves may be in a state of constant change. (Harrison, Keskin, & Zeevi, 2012). However, it could possibly lead to huge errors and disruptions which are often failed to be explored or understood. However, it is easier for many organisations to use the optimal model that is already known and exploit it for months or even years, rather than experimenting and learning. At the same time, the exploration of underlying parameters and `discovering of tendencies in the market and variations among consumers may result in maximisation of the profit.

(8)

7

unrealistic. Thus, the strategy of segmentation - splitting market into homogeneous groups- is often used as a solution.

Segmentation might be one of the most important determinants of pricing (Smith, 1956). Weinstein (2004) gave a nice-looking and clear definition of segmentation, “the process of partitioning markets into groups of potential customers with similar needs and/or characteristics who are likely to exhibit similar purchase behavior.” (p.4). Various authors emphasise the importance of market segmentation (Smith, 1956; De Keyser, Schepers & Konuş, 2015; Konuş, Verhoef & Neslin, 2008;). The idea that segmentation criteria could have a clear effect on the understanding of consumer behavior comes from Smith’s (1956) description of its importance as the strategy. Pride and Ferrel (1983) continued to develop the idea of dividing consumers into different groups and pursuing them. Continuing this idea, Lin (2002) and Konuş et al (2008) both emphasized how different criteria could be used to predict the behavior of consumer groups.

Privacy policies and legal restrictions limit the number of information available to marketeers. Thus, the self-segmentation becomes widely used by online retailers. The ability of the company to perform a good segmentation with the information available may determine the ability to assign consumers to the right segment, understand the segment, and thus set the right pricing strategy.

The privacy policies allow little flexibility in data collection. However, following research aims to clarify which characteristics may be used for the successful segmentation, which method is optimal to use for that, and which moderating factors may be present.

(9)

8

in further depth below. In all cases the demographic and psychographic variables are available, measurable, and considered to be very important in segmentation and pricing. Such aspects may be used as active segmentation criteria.

Active segmentation criteria are usually direct and easy to measure. However, research must also include a number of aspects which might not be measured by retailers in advance. The aspects may include the characteristics of the consumer as well as characteristics of their context or environment. For example, it is difficult to measure a consumer’s perception of fairness, presence of information, and willingness to compare alternatives. Despite the difficulty to measure such aspects, it might still be important to know them in order to understand their tendencies regarding the consumer choice. In other words, such aspects may influence consumer decisions, but there are less opportunities to gather this data about the consumer in advance. For example, it is difficult to measure the willingness to compare and availability of comparison websites. However, in case age would correlate with willingness to compare, we could expect that some age groups would rather spend more time for comparison than complying with the price.

(10)

9 Table 1: Groups of aspects selected for segmentation

3.1 ASPECTS THAT PLAY A ROLE AS ACTIVE SEGMENTATION CRITERIA

3.1.1 Frequency of purchase online

(11)

10

as ever before. Many consumers had no choice but to use the Internet as a channel for shopping. However, online shopping was apparently not the same as in store.

Online shopping assumes a number of obstacles, especially for retailers. Consumers might be more sensitive to the prices in online shopping, show more willingness to search for other options, and feel a higher risk and insecurity with online shopping. However, according to Wirtz and Kimes

(2007) and Heo and Lee (2011) the frequency of purchase may change the situation and perception

of consumers. Literature provides an idea that the more frequently online purchases are made, the

more their prices are perceived as fair (Heo and Lee, 2011). So, positive correlation of the online

shopping experience and perceived fairness of price might be expected.

3.1.2 Company loyalty

Purchase history based pricing is often used by companies for the dynamic pricing determination (van der Rest, Wang & Miao, 2020). As more frequent purchases may lead to the higher

willingness to comply with dynamic pricing (Heo and Lee, 2011), the consumer’s loyalty may also

keep them devoted to the company despite their use of dynamic pricing. The aspect of the company loyalty, however, is more complex than it may seem at first glance.

Consumers would not stay loyal to a company that treats its buyers unfairly. As mentioned by Klaus and Nguyen (2013), the perception of retail fairness plays a comparably important role in context of dynamic pricing. They define three determinants of it: Product, Service, and Company image. The company’s dimension describes the perceived fairness and transparency of the company by the consumer. If the consumer sees the company as honest, transparent and fair in it’s communication, s/he would rather be loyal to it.

(12)

11

believe that the company aims to establish the relations fairly profitable for both parties: consumer and the company. Findings of Campbell (1999a) provide evidence that consumers who perceive retailers as fair and honest are more likely to accept the price changes of a firm and remain loyal to them.

Two other aspects mentioned by Klaus and Nguyen (2013) are product and service satisfaction. The customer’s satisfaction was mentioned by numerous authors as the key to future customer loyalty (Campbell, 1999a; Lo, Lynch & Staelin, 2007; Hinz, Hann, & Spann, 2011). In the context of dynamic pricing, however, specific criteria may affect the satisfaction. As mentioned by

Dickson and Kalapurakal (1994) and Jung, Cho, and Lee (2014) the price changes and adaptations

may be perceived as fair in case that any consumer may get the same price (even if it assumes the costs of search, comparison, transaction, and acquisition). This thought compliments the product and service dimensions discussed by Klaus and Nguyen (2013). A consumer might not only refer to the internal reference or what is the satisfactory combination of the quality/price, but also compare whether s/he can acquire it with the same cost and conditions as the other consumers. In cases where the price may differ between customers, dynamic pricing is deemed as unfair and decreases customer loyalty significantly.

However, consumer loyalty always has its limits. As previously mentioned, the aspects of cost difference and availability of alternatives, loyalty and purchase history, and satisfaction of customers may influence a consumer’s perception. Although, it is also assumed that the extent to which the price changes, the type of product and whether the consumer has access to other price levels may change depending on the previously mentioned aspects. The changes which prices may undergo and the perception of these changes may affect or may be affected by company loyalty. Therefore, a consumer’s loyalty to a firm and their attitude to the dynamic pricing may be connected.

(13)

12

(1994) - whether the individual can access the same conditions as the other consumers, which is measured as comparable fairness. These aspects may be considered as both active and as descriptive segmentation variables. However, in the rare cases in which they are measurable in advance, they can also be used as a way to analyse the experience of the consumer. It is believed that they might also play the active role as well. It will be used in both types of analysis.

3.1.3 Demographics

Demographic criteria are often used for the segmentation. It is considered by many companies to be the easiest, cheapest, and most convenient way to see which of the potential customers are “most likely prospects for their products” (Weinstein, 2004, p.7). Demographic segmentation criteria are easy to scale, understand, and apply. Age and Gender are the very first which come to mind, and often appear to be first among widely used for the differentiation among consumer groups and their needs (Weinstein, 2004; Greenberg & Frank, 1983). However, as mentioned by Wells et al (2010) such variables may not have a great predictive power alone. Thus, the variables of household income and education are also taken into account. Wells et al (2010) also takes into account variables closely related to age. These may include social status and informative power, which are, however, not included in the following research. (see more in limitations).

3.1.4 Household income and education

Beldona and Namasivayam (2006) mentioned that a higher household income leads to the perception of a fairer price. In other words, individuals with lower household income tend to perceive prices as more unfair. Similarly, Sirvanci (1993) mentioned that lower price income leads to higher price sensitivity.

(14)

13

independent variable.However, in the literature reviewed it is assumed that such a variable might also serve as the base for the segmentation.

3.1.5 Gender and age

Zwick and Chen (1999) mentioned that gender should not play a role in the perception of fairness, as it brings about certain ethical concerns.Without the purpose to make the following research offensive, it is believed that gender could be an important psychographic variable. Despite recent movements of emancipation, gender equality, the aspect of gender might play a role in perception of the price. Numerous studies in the field of marketing and psychology include gender as an important factor. Beldona and Namasivayam (2006) claim that women are more likely to perceive prices as unfair rather than men. Oppositely, Eckel and Grossman (1996) and Sirvanci (1993) mentioned that women show more price elasticity than men. In any case, it is important for the following research to study gender as one of possible segmentation criteria.

Another debatable segmentation criteria is age. Age may be considered an important segmentation variable not only due to different types of products and interests of the consumer. Some authors also have the conflicting thought about the way the age may affect price sensitivity. Tellis (1988),

Sirvanci (1993), Raab, Mayer, Kim and Shoemaker (2009), Beldona and Namasivayam (2006),

and Scriven and Ehrenberg (1999) all mentioned in their research articles that age affects the price sensitivity of the consumer, but all in varying ways. They do not agree on whether price sensitivity increases or decreases with the age. For instance, Tellis (1988) and Sirvanci (1993) find the decrease of sensitivity to price to be in correlation with aging, while Scriven and Ehrenberg (1999) agree with the opposite. In the context of dynamic pricing - with the core idea of the perfect price setting - the price sensitivity and the factors affecting it should be of high importance. If age as a factor affects price sensitivity, it should be taken into account.

(15)

14

not be underestimated or neglected. Thus, the related factors should be taken into account. Therefore, a consumer’s culture may be named one of the most important factors.

3.1.6 Culture

Numerous authors in sociology and marketing present the idea that the specific age categories of consumers share some characteristics. For example, the generation of so-called baby boomers is widely perceived as the wealthiest generation, while Gen X is perceived as the most rational (Williams & Page, 2011). This type of ideas sound almost anecdotal to the marketeers in Eastern Europe, where the image of real life proves to be the complete opposite. Consequently, the idea that a specific group or segment of consumers could share the same characteristics across the globe may sound fantastic.

Why does it matter for the following research? Specific variables may be more, less or not useful at all for segmentation in different countries and their markets. Furthermore, the clear differences in cultures directly influence the strategies built by marketologists across the globe. Therefore, the way in which culture influences segmentation and marketing strategies may not be neglected.

The most convenient and easy to understand way to analyse a culture and its influence is to evaluate and describe it in simple and measurable aspects. Such aspects were clearly described by Geert Hofstede in 1967-1973 and become world-widely known and used. According to Hofstede (2001) the dimensions include:

● Power distance index, which explains how the individuals in the lowest level of the hierarchy perceive the organization and heights of the hierarchy itself. Low index describes more equal power distribution. The feeling may affect the purchase intention and willingness to pay, same as price elasticity;

(16)

15

● Masculinity and femininity, which explains the aim for heroism and material rewards or caring for weak or feelings. This dimension does not refer to gender or equality, rather to the setting of one’s life goals from an emotional aspect.

● Uncertainty avoidance, which describes the importance the individuals give to the laws, rules, guidances, procedures, plans, availability of alternatives, options, and literal avoidance of uncertainties.

● Long- / Short- term orientation, which explains the extent to which the individual or cultural group prefer long-term planning and actions over circumstantial adaptation. ● Indulgence and resistance, which explains how acceptable it is for individuals in a society

to follow their desires and fulfill their human needs and wishes.

All the dimensions mentioned above may be used for clear and concise narration of the impact of culture on an individual's perception and behavior. Culture may be a very powerful aspect with ramifications of wrong marketing and dynamic pricing strategy. It can also be connected with the passive (or descriptive) aspects of the individual or their environment.

3.2 THE CONTEXTUAL (PASSIVE, DESCRIPTIVE) CHARACTERISTICS

3.2.1 Price consciousness

One of the first aspects to discuss is the price consciousness, specifically, why it cannot be used for active segmentation. First, price consciousness cannot be measured in advance. Secondly, the scale to measure such an aspect may be difficult to determine and agree upon.

(17)

16

s/he considers as standard, and decides whether it is high, low, or acceptable (Alford & Biswas, 2002).

It should be mentioned that the aspect of price consciousness itself does not consist solely of the three aspects mentioned previously, but is rather connected with them. The level of price consciousness depends on the contextual information and internalised price reference, but also leads to attempts of additional search and verification of the “attractiveness of the deal” (Alford & Biswas, 2002, p.781). So, price consciousness is the reason and the result at the same time. It can lead to the willingness to compare and information search, same as can be caused by the available acquired information.

While deciding whether to buy the product or not, the consumer often refers to the comparison

with the reference (Heo & Lee, 2011; Jung, Cho, & Lee, 2014). According to Heo and Lee (2011)

the “customer often accesses internalised reference prices, such as the last price paid, and/or externalized reference prices, such as the price most frequently paid” (p.244). Without going too deep into the details of external and internal references, it can be assumed that the presence of such references directly influence the consumer’s perception of the price. According to the adaptation level theory, the standard the individual uses for comparison “is the adaptation level price that can be considered the mean of previously observed market prices.'' (Alford & Biswas, 2002, p.776). It is important to know the reference price and the price elasticity of the consumer in order to determine the acceptable dynamic price range for the specific consumer. Numerous authors have their own ideas on which factors may influence the price sensitivity (Sirvanci, 1993; Wells et al, 2010). However, the extent of price elasticity is connected to the specific price that the consumer

would find acceptable and fair (Heo & Lee, 2011). Thus, the reference price is the aspect taken

into account.

(18)

17

However, the lack of ways to measure or determine it in advance makes it difficult to segment the consumers accordingly. Thus, price consciousness will not be used as a criteria for segmentation, but will be tested for the correlation with any segmentation criteria or tendencies in consumer behavior.

3.2.2 Valuation and willingness to compare

The consumer might have plenty or no alternatives among the products and/or retailers. The higher speed of the product use leads to the lower intentions to compare, while the higher importance or

the price lead to the higher attention to the transaction and acquisition value (Jung, Cho & Lee,

2014). However, the presence and awareness of such alternatives does not necessarily make consumers wish to compare it or decide that it is worth it (Marmorstein, Grewal & Fishe, 1992). The willingness to compare prices and search for price alternatives may be influenced by aspects such as loyalty and company image (Kocas, 2002; 2005; Jung, Cho, & Lee, 2014). Also, in perception of consumers the perceived costs of the search and comparison may overcome the costs

of price difference (Hinz, Hann, & Spann, 2011).

Nowadays, numerous websites allow consumers to compare not only competitor prices, but also the price changes for a specific retailer. The price changes and awareness of such may significantly affect the attitude towards the price and how fair it seems to the consumer ( Bolton et al, 2003; Jung, Cho, & Lee, 2014). Even if the price seems fair at the first glance to the consumer, the awareness of the recent variations may change this perception. The changes of the price set by the same retailer may affect the source of purchase or the time of purchase. Besbes and Lobel (2015) mentioned that the product valuation and willingness to wait for a better price are negatively connected: the lower valuation allows longer waiting for a better price.

3.2.3 Price fairness

(19)

18

Some authors assume the difference in the price the consumer is aware of and cannot manage leads to dissatisfaction, as it motivates the feeling of being treated unfairly. (Dickson & Kalapurakal, 1994; Useem, 2017). At the same time, a lack of information may lead to a shift in attitude as well (Choi & Mattila, 2005; Levin, McGill & Nediak, 2009; Hogan, 2010). Thus, the following paper focuses on aspects related to the subjective feelings of the consumer and their attitude toward the offer rather than the price of the offer itself.

The awareness of the price difference may play an important role in the probability of purchase. Some authors mention that consumers mainly perceive noticeable price differences as unfair and unreasonable. The consumers who do not receive enough information tend to perceive the price as unfair (Choi & Mattila, 2005; Hogan, 2010). It is suspected that awareness of the use of dynamic pricing may be related to the probability the consumer makes the purchase. The question is whether the awareness of the use of dynamic pricing algorithms would affect the consumer’s perception of fairness and probability of purchase, or whether the consumer would agree with the dynamic price change and perceive it as fair? (Choi & Mattila, 2005).

3.3 MODERATORS

Expected that the awareness of DP implementation may have the effect on the perception of price and thus on the purchase probability. Such an effect of the third variable (awareness) on the relations of independent and dependent variables (consumer characteristics and purchase) is called

a moderating effect (Cohen,J., Cohen,P., West & Aiken, 2013). Such a variable is characterised

by the interaction effect and may be an important determinant of the outcome. For the following research two variables are surmised to have moderating power: awareness of the DP use and type of the product that experience DP use. Each one has to be discussed.

3.3.1 Awareness of the dynamic pricing

(20)

19

consumers who do not receive enough information tend to perceive the price as unfair (Choi and Mattila, 2005; Hogan, 2010). So, the price changes result in the feeling of unfair treatment. However, the awareness that the price differs for the consumers may have an even greater effect.

There are no two different individuals who may perceive or react to a price change in the same

way. (Dickman, 1985; Ritchie, 1974). The perception of different people among the consumer

group varies so significantly that it cannot be explained away by an error term (Ritchie, 1974). One of the most important aspects is the asymmetry in the evaluations among people (Bloom & Price, 1975; Peeters & Czapinski, 1990; Taylor, 1991; Campbell, 1999a). There might be plenty of reasons behind it, such as previous experience, consequences, and the consumer’s mood. In context of the following research, it may be that the consumer who is aware of the DP may react in a completely different way depending on an even minor difference of culture and background. That is why the awareness may not be implemented as the variable, as it is expected to have the different (sometimes even opposite) effect on different groups of consumers.

Levin, McGill and Nediak (2009) suggest that the awareness of the price adaptation has a negative effect on perceived fairness of consumers and thus the sales. They assume that limitation of the information given to consumers would be more beneficial. At the same time, McGuire and Kimes (2006) and Wirtz and Kimes (2007) have the opposite view and mention that the open communication regarding DP may result in the consumer’s compliance with price changes. Thus the question of whether the awareness of the DP use is beneficial or not remains open.

(21)

20 3.3.2 Product type

By the simple logic and collection of the reasons mentioned with other variables, is assumed that any of the aspects would vary depending on the product type (ex. the price sensitivity and wingness to compare).Thus, it was decided to use the differentiation used by Jung et al. (2014): separate products into two categories: non-look-and-feel (ex. computers) and look-and-feel (ex. clothes). It is expected that the availability of comparison is rather important for the non-look-and-feel category and rather leads to higher attention to the price for products within that category. (Jung et al., 2014). And once more, the aspect discussed may be closely related with the variation of other independent variables for segmentation. Thus, the aspects should be tested.

3.4 CONCLUSION (LITERATURE REVIEW)

The numerous characteristics of the consumer are assumed to be potentially influential for the success of the dynamic pricing implementation by online retailers. In addition, the effect may vary based on the type of product and whether the consumer is aware of the use of dynamic pricing. The previously mentioned theories might seem to be randomly collected from the multiple sources, however, every single variable is included either because it is often used as a segmentation variable, or because it is believed to have an impact. The little number of studies concerning customer segmentation in a context of dynamic pricing serves as a limitation yet at the same time opens the field for testing numerous variables and possible combinations of these variables. Thus, it is believed that the combination of all the characteristics mentioned above might give interesting insights on the effect of some differences among consumers on the success of dynamic pricing. In order to test all the theories mentioned above in application to the dynamic pricing and possible interaction effects, the experiment for the empirical data collection must be created.

4 RESEARCH PURPOSE

(22)

21

studied in a more precise manner. Because of this exact reason the ability to understand the consumer and predict the reaction on the specific prices precisely and clearly is needed. The success of implementation of the dynamic pricing may fully rely on the expediency and credibility of differentiation among consumers, and the robustness and implementability of the model used. It is believed that many, or at least some, aspects which were previously mentioned, may have the moderating effect on how price changes are perceived. The perception of the price changes may have a direct impact on the probability of purchase, which is why it appears to have great importance for any company implementing dynamic pricing. The individual differences may not be neglected. However, the moderating role of the certain variables should be taken into account and studied as well. The final purpose of the research consists of three points:

1. Find an optimal clustering technique that may be used for segmentation

2. Based on the information collected draw the clear segments and find out which segment have the highest purchase probability in the context of dynamic pricing

3. Find out whether the moderators have the effect on the perception of the dynamic pricing and purchase probability.

The following research does not contain the hypothesis because of the number of conflicting views on same aspects in the researched literature, limited possible level of precision, and final goals of the research. However, the second and third goals mentioned above may be visualised with the conceptual model for the clear understanding.

(23)

22 5 METHOD AND METHODOLOGY

5.1 DESIGN OF EXPERIMENT

The respondents are asked to evaluate the chances that they are going to buy the product presented in the hypothetical situations, their mood in this moment and their feelings towards the offer. In the story the information regarding the historical aspects is given, while the aspects of demographic, psychographic and ethical parts are asked afterwards. The control group post-test-only design of experiment was believed to be the most efficient for the following research. The main goal was to compare the values of the segmentation aspects among the consumers and their effect on the comparable attitude to the dynamic price between test and control groups. The control groups received the same story as the test group, however, they were told neutral conditions of the historical aspects.

5.2 METHODS OF DATA COLLECTION –PRIMARY AND SECONDARY DATA

The data was first collected through the literature search. This research approach allowed to establish assumed theories which could be tested with the quantitative survey later on and analysed. Despite the fact that the research does not contain clearly determined hypotheses that can be tested (as the list of such would be impossibly long due to the number of variables included), the specific relations are assumed by theory and some effects are expected to be found. Thus, the research is assumed to be deductive in its nature (Saunders, Lewis & Thornhill, 2009).

(24)

23

groceries). Therefore, it is believed that the respondents for the survey would represent a good heterogeneous sample of the potential target group of online shoppers.

Randomisation is an important element of the experiment. The respondents are randomly and equally allocated to 4 groups. Two groups with durable goods, and two groups with non-durable goods; two aware of DP us, two - non-aware. So, Group A reads the case with the microwave for which the price was changed by a 5% increase, the delivery information, and service information of the store. Grouproup B receives the same information, but they are also informed that the website uses the algorithm of dynamic pricing, which may determine the price based on the characteristics of consumer, environment, and other variables. Group C receives the information about the T-shirt, the delivery, service, and recent possible price change (the respondent receives a hint that s/he recently saw the 15% lower price). However, group D also knows that the website uses algorithms of dynamic pricing. As the survey allows the respondent to participate once only, the “consumers” would not be aware of the other cases.

It should also be mentioned that despite the substantial number of variables to be measured, the issue of the size was faced. According to Saunders et al., (2009, p. 151) the longer survey had a lower likelihood of being completely answered towards the end. Thus, the approximate number of questions was equal to the number of variables. At the same time, the research does not declare the preciseness of the measurement of variables. However, it has the purpose to explore the basic ideas and possible relations, and the questionnaire fits this purpose well. Any part of the outcomes of the following research may serve as the idea for the future more precise investigations.

5.3 MEASUREMENTS AND SCALING TECHNIQUES

(25)

24

the statement, “The boss should make all the decisions,” would show whether the respondent belongs to high or low power distance culture. The scale also excluded the neutral answer, so the respondent was supposed to show polarity to at least some extent (“somewhat dis-/agree”).

The neutral option was excluded with the purpose so that the respondent would show to which extent s/he belongs to the cultural segment. The cultural segments are mentioned in accordance with Hofstede cultural dimensions and stand in opposite to each other on the scale. So answering the question “Do you agree with the statement… Boss should take all the decisions?” the respondent should determine whether s/he supports the high or low hierarchy of power, and to which extent.

The variables as company loyalty can be measured by the company in advance. For this purpose, the data of purchase history can be used. Assuming that the company loyalty might affect the probability of purchase. However, due to the limitations this assumption cannot be tested in this research and must find place in further investigations.

At the same time, the effect may be very extensive: the use of dynamic pricing may have the same effect on company loyalty as on the perceived fairness. In other words, the awareness of the use of dynamic pricing may affect the loyalty of the customer and their perception of the price. For this purpose, the questionnaire includes the question regarding the perceived fairness which is asked after the hypothetical situation is presented. The situation may or may not contain the information about the use of dynamic pricing by the website. For the comparison the average evaluation of fairness and opinion about the company would be compared for the different test groups.

(26)

25

Also, because this question is not the most important for the following research and does not require great certainty, it was presented in a simple form.

In order to measure perceived quality, the information about the delivery was mentioned - the delivery had the usual conditions as much of the online stores. However, due to the general feeling of un-/fairness, the perceived quality of service might be affected. So despite the perceived quality of service and customer loyalty in general, there is something that the company should measure in advance. The questionnaire includes the Likert scale for the evaluation of the perception of fairness.

In order to measure the impact of the price consciousness, the attempt of additional verification of the deal (Alford & Biswas, 2002) will be measured. As was mentioned in the literature review, Alford and Biswas (2002) mentioned that the consumer with higher price consciousness would try to additionally verify the price. In other words, the consumer would attempt to compare the prices and think about it more if his/er price consciousness is higher (Alford & Biswas).

In order to measure such a hypothesis, the following conditions were implemented during the experiment: the consumer was already given the information of the price changes and the contextual information (aligning with adaptation level theory of Helson (1964)). Then, the connection of the price consciousness with the willingness to additionally verify the price assumed by Alford and Biswas (2002) is used. The willingness to attempt additional price verification was measured by the direct question. The respondent was asked whether s/he would rather use the comparison website for the following case or in normal life. Thus, the conditions to at least obliquely measure the price consciousness are set.

5.4 PROCESSING AND DATA ANALYSIS

(27)

26

Second, the data was tested for the possible patterns and correlations in the variables. As the optimal segmentation model cannot consist of all the variables measured, the most relevant variables got to be selected. This step consisted of two sub-steps. To begin, the variables were analysed in terms of correlation and logic. For example, the variables of age and income were highly correlated, which may be explained by the simple logic (the lowest age group “below 18” have lower income then groups “above 35”). For that three tests were used. The first was Pearson correlation used to find which categorical variables are highly correlated and which variables may be merged and/or used for the segmentation and clustering of the respondents.

Following correlation testing, two tests for independence of variables were made. The test which best suited the type of data was the Chi-square test. However, as the number of respondents is limited, the risk of bias became too high (Potthast, 1993; Raymond & Rousset, 1995) and the additional test for the small sample size was a need. Fisher’s exact test became a solution (Raymond & Rousset, 1995). Two tests for independence showed mainly very similar or identical results. Thus, it can be assumed that the results should be considered trustworthy (results of both tests are used in the text, ex. the results as “p-value <0.01” means that both tests are significant at this level).

Then, following the procedure proposed by Raftery and Dean (2006) in simplified form, each of

the variables was analysed with the information criteria and importance for the clustering (for each cluster analysis). As such, the most relevant variables were selected with the information criteria of each. Also, later the set of variables was tested by adding or removing one of the variables and observing the changes in the quality of clustering (Eshghi et al., 2011).

Third step of the analysis included the clustering procedure with the use of various models. This step aimed to select the optimal model for the segmentation. The criteria for the model comparison were drawn from the literature and analysed. Such include hierarchical clustering, k-means clustering, two-step clustering, and latent class analysis. The detailed analysis of each model follows below in the “Results - CLustering” section.

(28)

27

the moderating effect of the variables was studied. The moderating effect was studied with the binary logic regression model.

5.5 ETHICAL CONSIDERATIONS

The ethical considerations of the following research created numerous limitations. As such, the case in the experiment might only be hypothetical in nature, as the questions can only be generalised and personal information could only be asked to a certain extent. Ignoring such aspects would clearly harm the willingness of the respondents to answer, and consequently affect the sample size. However, due to complying with the limitations, the research suffers from the loss in persizeness, as well as validity and reliability. Such issues with ethical consideration might be overcome with specific means in case of further research (see the paragraphs “Limitations”).

6 RESULTS AND DISCUSSION

6.1 DESCRIPTIVE

The data received consists of responses collected in between April-May of 2020. The respondents come from different cultures and countries and include all the age groups, which makes the sample well-mixed in terms of variables (such as culture and demographics). However, the questionnaire was mainly (65%) distributed among East-European users of the network.

(29)

28

As previously mentioned, because the questionnaire was spread online, it was expected that a significant percentage of respondents would also use an online channel for shopping. As expected, up to 25% of respondents make each fourth purchase online, and 43% mentioned that they use online channels for almost half of their purchases (with the exception of groceries).

The price comparison is an important variable for the following research. Up to 18% of respondents do not use price comparison websites, while 45% acknowledge using them quite often. At the same time, for the cases given in the questionnaire for the durable goods, 85% of the respondents showed the will to use the price comparison websites, in comparison to 50% for non-durable goods. The tests of independence showed that the willingness to compare is not independent from type of goods. (Chi-square p-value = .002; Fisher p-value=.002)

6.2 CORRELATIONS

Numerous variables showed the high correlations among them (Appendix 2). Such include cultural dimensions. As such the variable of power distance highly correlates with the variables of masculinity, uncertainty avoidance, and individualism. As the respondents consist mainly of East-Europe residents, the results align with the simple description of the culture and expected

correlations in dimensions (Woldan, 2009; Wackowski & Blyznyuk, 2017).

Indulgence and age correlate in Fisher with the high significance (p-value = .01). The testing also showed that the higher the age, the lower the indulgence (each additional 10 years decreased the average indulgence by 23% (p-value < 0.01)). So, the higher age, the lower the probability that the respondent would agree with the high indulgence probability choice).

Furthermore, the dimensions of power distance and age showed the highest informative importance for the further clustering. Thus, the variables were selected as the most representative.

(30)

29

The variable of education level highly correlates with the variable of education. That may be resulting from the same problem with age, same from the fact that level of income may depend on the education level. It is interesting to note that the willingness to use price comparison websites has a negative correlation with the education level, indicating that the people with higher education levels are using less price comparison websites. However, the correlation among comparison websites and income was not significant.

The variable of gender did not show any significant correlation, only a barely significant result for the independence test (p-value=.052) with the willingness to compare the goods in case of durable goods. However, the variable showed high importance as the clustering variable.

Another important variable was found as the use of price comparison websites. The variable showed the high importance and the significant positive effect on the models. Use of websites showed very high correlation with the willingness to compare in cases given, which is very logical. Thus, it was decided to use the variables as well.

Based on the analysis, part of which could be seen above, and the reasons mentioned, the variables for the clustering were selected. Such include: Use of price comparison websites, Cultural dimension: power distance, Gender, and Age.

6.3 MODELS COMPARISON

(31)

30

following research includes the determination of the criteria for clustering (variables), the agglomerative clustering is preferred (Madhulatha, 2012).

6.3.1 Hierarchical clustering

The first step was selected to be hierarchical clustering. As the sample size may be named relatively small, the Hierarchical clustering may be used as the good way to determine the optimal number of clusters.

Ward method designs the clusters based of the measurement of the variance of the clusters by measuring the changes in sums of squares after merging the clusters. The Ward’s method works by “bottom-up” principle by merging similar clusters until the optimal number is reached. Ward’s method uses the F-values to maximise the differences between clusters. Unlike the other methods with similar approach (such as Manhattan or Euclidean), Ward’s method uses analyses the variances instead of distances (Madhulatha, 2012)

As mentioned by Eshghi et al. (2011), the merging should be stopped when observed homogeneity of cases within each cluster experience the significant decline or drop in it’s value. Agglomerations schedule showed that the optimal stopping point for merging might be 4 and 6 clusters solutions with the significant dtops of the coefficients after these points. The dendrogram (Appendix 1) however, showed that the 4 cluster solution is more logical and clear.

Clustering by only cultural of only demographic variables was too poor in quality ( Silhouette measure of cohesion and separation < 0.3)

6.3.2 K-means clustering

The K-means clustering is often used for clustering bing datasets? However, normally it cannot be

used for categorical data. Following the steps discussed by Huang (1998) the algorithm can be

(32)

31

The non-hierarchical methods of the clustering such as k-modes may be used when the optimal number of clusters is already determined. As such is 4, the other methods for clustering may be

used. Spiegler (2007) by the comparison of the techniques on the real data concluded that the

non-hierarchical clustering is often performing better than non-hierarchical. Similarly Eshghi et al. (2011) mentioned that the different techniques might give other results, which should be kept in mind by the researcher. Thus, the next step for the following research is to try k-modes clustering and Two-step clustering.

Despite K-mean clustering is rather suitable for the large datasets and ratio type of data, following the idea of Eshghi et al. (2011) following analysis was used for the comparison and alternative view on the clusters in our case. While using the k-modes clustering the 3 iteration for the 4 clusters gave the .000 change after already 3 iterations, which is the sign of a strong and stable cluster solution showing the strength of the center points positions (Garbade, 2018).

Ought to be mentioned that the test usually used for testing clustering performance of k-means clustering such as ANOVA, t-test and Homogeneity tests as Levene’s may not be used in the case of k-modes as the type of data would not fit. However, to measure the performance of the model other criteria should be used (see below in “Choice of method and metrics”) .

Table 2: K-means (k-modes) clustering results.

6.3.3 Two-step clustering

Two step clustering is based on the one of the methods described above: hierarchical clustering. The two-step clustering is applied when multivariate methods of analysis come in combination with traditional methods of clustering (Gensler, 2016). The algorithm used does the clustering in two steps: “pre cluster” by grouping the observations into small clusters, and hierarchical

clustering that groups these small clusters together (Horn & Huang, 2009). The method has the

(33)

32

For the evaluation of how well the clustering was performed, following the validation method

proposed by Norusis (2008) the number of tests were used. Such include Silhouette measure, to

see the consistency of within cluster data or how clear the cases may be segmented(Rousseeuw,

1987); the cluster ratio to see how equally the cases are distributed. Then, the Pearson chi-square test was used as it fits the type of data. The information criteria AIC was added to the results table

as the important determinant of model performance (Center IBMK, 2012).

In such a way the Two-step clustering was performed as one of the methods selected. The test showed that for the same variables as used before the optimal number of segments is 3. However, the cluster quality (Silhouette measure) for the 4 segments was much higher, same as ratio of qluster sizes was much smaller, which means a more equal distribution of the cases (Table 3) .

Table 3: Two-step clustering results

6.3.4 Latent class analysis

(34)

33

The procedure of Latent class analyses was runned several times for the data. First, the LCA algorithm determined the optimal number of segments, which was 3. Then the process was repeated, however with a pre-specified number of segments (2,4,5) to see whether the fit of the model decreased or increased. The information criteria such as AIC and BIC in the cases of pre-specified number of segments decreased (table 4), however the entropy increased. The entropy of the values could not serve as the criteria for choosing the model, but may indicate clear separation

of the classes if approaching 1 (Asparouhov & Muthen,2018; Chen & Liu,2005). Furthermore, the

statistical base of the LCA model allows to use of likelihood to evaluate the distribution of cases based on their posterior probability of segment membership (Eshghi et al., 2011).

Table 4: LCA results

The best model of Two-step cluster analysis is considered the model of 4 clusters. It has one of the highest log-likelihood values, however it does not experience such a significant drop in AIC and BIC values as the next model (for 5 clusters). Furthermore, the entropy test shows comparably good value (.79).

6.4 CHOICE OF METHOD AND METRICS

The success of the specific method of clustering depends on the measures used evaluating this

success (Nguyen & Rayward-Smith, 2008). To compare the clustering methods some measures

applicable to all and each method should be implemented. Rand (1971) proposed the list of criteria

(35)

34

allow to compare techniques by how clearly the segments were drawn. Among the clusters of each analysis the best performing number of clusters was chosen and then compared with the cases of the other methods.

The first criteria mentioned by Rand (1971) include the easiness of use. In case of following

research the data represent a list of categorical and ordinal descriptive variables. Due to that, the technique of k-means clustering had to be modified to k-modes technique (Huang,1998). This makes the process of clustering much longer and difficult. According to Rand (1971) the ease of application of clustering technique may be an important reason for model choice. For the following case, the use of other techniques as Two-step analysis and LCA require much less effort. Thus, from the point of first criteria of Rand the k-means (k-modes) clustering may be rejected as the least convenient method for the researcher in the following case.

Following the criteria of Rand (1971), next the performance of the model should be compared. As proposed by Eshghi et al., (2011, p.276), first the homogeneity of the observations within the segment (the measure of similarity of the cases in one cluster) and heterogeneity of clusters as differences between clusters. In the case of a two-step model there are two ways to measure distance: Euclidean distance and Log-Likelihood. As Euclidean distance can only be used for continuous data, the Log-likelihood measure was used as it could be used for categorical data as well. This method also allows us to evaluate the latent class models.

Another useful criteria to evaluate the model compared with other models is information criterion. In such a case AIC and BIC may be used. However, AIC is preferred as it can be applied for the models with categorical variables (Fonseca, (2013). Table 5: Models comparison

The results appear as following:

(36)

35

2. K-means clustering requires too much effort for application to the case with categorical data which makes it least attractive for researchers.

3. Two-step clustering and LCA are similar in the values of Log-Likelihood, however 4 clusters from two-step clustering procedure show more attractive AIC value.

Based on the reasons mentioned above, the 4 clusters given from the two-step clustering analysis are considered to be preferred ways to segment the cases of the following research.

7 RESULTS AND DISCUSSION

The two-step model showed comparably best results with 4 clusters. These 4 clusters should be described and analysed in terms which of them show which reaction in terms of purchase in context of dynamic pricing. First the clusters should be described with characteristics; second, the cluster with highest purchase probability determined, third, which effect the moderators (awareness and type of good) have on purchase probability of each segment has to be determined.

7.1 SEGMENTS AND MODERATORS

The segments vary in the characteristics.

First segment consists of women, with the average level of indulgence, young (19-35). Probability of purchase in general is the lowest for this segment (14%), which slightly increases in case of awareness of dynamic pricing use (9% increase; p.=.009) and decreases for durable goods by 50%. The findings for this segment support the ideas of McGuire and Kimes (2006) and Wirtz and Kimes (2007) that claim that the honesty about the price changes may lead to higher compliance. Also, it might be assumed that such a type of consumer as the young female might feel less confident with the online purchase of durable goods than the second segment.

(37)

36

may directly impact the probability of purchase. In the same time, in contrast with segment 1, the opposite theory of McGill and Nediak (2009) was supported with the findings: awareness of the DP use leads to lower purchase probability. The difference between segments might be explained by the habit to use price-comparison websites. However, there is not enough evidence to claim something like this in the following research. Should be mentioned, the variable of product type showed not significant result.

Third segment is characterised by non-use of the price-comparison websites and highest indulgence. The segment has the probability of purchase of 38% which might be slightly affected by the awareness of the dynamic price use (4% increase in probability), however, the type of good did not show significance. So, again the awareness of the DP use have a slight positive impact on the purchase probability, which supports the idea of positive effect of honesty (McGuire & Kimes, 2006; Wirtz & Kimes, 2007). However, the results for the independence tests between the perceived honesty of company, transparency, and awareness were not significant for that specific segment.

Fourth segment is characterised by the older age and comparably low indulgence with a little use of price comparison websites (the least significant variable). The segment has the 25% probability of purchase, which is the most significantly affected by the awareness of DP use (2 times decrease in probability of purchase). That might relate to the little use of the price comparison websites. It could be that the consumer may experience less need in comparison and rather use own internal reference prices (Heo & Lee, 2011; Jung et al., 2014) and perceive the price changes as already forwardly unfair. It also could appear that the older consumer groups rather have lower valuation for the specific products and experience less willingness to compare the prices which might been changed (Alford & Biswas, 2002).

(38)

37

7.2 OTHER VARIABLES

Few words should be said about the other variables. The valuation of the good has a relationship with willingness to compare. For the microwave example it was found that 93% of respondents would spend up to a few hours attempting to find a cheaper option. The test also showed that the willingness to compare for the case given (the question, “Would you use price comparison in the following situation?”) is much higher for durable goods when the consumer is aware of the DP use (p-value =.007). Thus, the theory of Besbes and Lobel (2015) of the longer waiting time for more valuable goods (described by Jung et al., 2014) finds the place. However, as the effect was not significant for any specific segment, it is possible to talk about the effect on the sample in general only.

The effect was also noticed for the perceived fairness of the service and company regardless of the type of product. The awareness of the DP negatively correlates with the perception of service fairness (p-value < .001) and comparable fairness (p-value < .001). The awareness of the DP use showed that the consumer is slightly more likely to consider the prices of the company as unfair (p-value <.001) and more likely to see the website as unfair compared to the websites which do not declare to use DP (p-value <0.01). However, the effect is significant only for the sample in general. While testing for separate segments, the significance is only approaching the critical value for segment 4.

(39)

38 8 CONCLUSIONS

The conclusions of the following report can be summarised in the number of points:

- The K-means clustering requires moderation for such kind of research which makes it less attractive as the method for clustering

- Latent Class analysis shows good results as the segmentation method, however is outperformed by the Two-step clustering

- Two-step cluster analysis is the optimal method for the simple segmentation of the categorical variables with the small sample

- The segments drawn and analysed show that the highest purchase probability in the context of dynamic pricing belongs to the segment 2. The segment consist mainly of man of the average age which used to use price-comparison websites.

- Type of good as the moderator showed almost no effect, which may be due to sample size or ineffectiveness in the context of DP. Further research may be required.

- Awareness as the moderator shows the different effect on purchase probability (significantly negative or slight positive) depending on the segment.

The research above was done in the context of Master thesis and only superficially describes the topic of segmentation in the context of dynamic pricing. It may serve as the source of ideas for further more narrow and deep investigation in the field.

9 LIMITATIONS

(40)

39

Second was the timing of the research. Unfortunately, the research and specifically data collection coincided in time with the world pandemic. It is believed that the data collected may be significantly affected by the moods and consequences of the pandemic, which leads to lower reliability of the data and results.

(41)

40

10 REFERENCES

Alford, B. L., & Biswas, A. (2002). The effects of discount level, price consciousness and sale proneness on consumers' price perception and behavioral intention. Journal of Business research, 55(9), 775-783. Asparouhov, T., & Muthén, B. (2018). Variable-specific entropy contribution. Los Angeles, Muthén & Muthén. Beldona, S., & Namasivayam, K. (2006). Gender and demand-based pricing: Differences in perceived (un) fairness

and repatronage intentions. Journal of Hospitality & Leisure Marketing, 14(4), 89-107

Benassi, M., Garofalo, S., Ambrosini, F., Sant’Angelo, R. P., Raggini, R., De Paoli, G., ... & Piraccini, G. (2020). Using Two-Step Cluster Analysis and Latent Class Cluster Analysis to Classify the Cognitive Heterogeneity of Cross-Diagnostic Psychiatric Inpatients. Frontiers in Psychology, 11, 1085.

Besbes, O., & Lobel, I. (2015). Intertemporal price discrimination: Structure and computation of optimal policies.

Management Science, 61(1), 92-110.

Bloom, H. S., & Price, H. D. (1975). Voter response to short-run economic conditions: The asymmetric effect of prosperity and recession. American Political Science Review, 69(4), 1240-1254.

Bolton, L. E., Warlop, L., & Alba, J. W. (2003). Consumer perceptions of price (un) fairness. Journal of consumer

research, 29(4), 474-491.

Bonney, M. C., Zhang, Z., Head, M. A., Tien, C. C., & Barson, R. J. (1999). Are push and pull systems really so different?. International journal of production economics, 59(1-3), 53-64.

Campbell, M. C. (1999a). Perceptions of price unfairness: antecedents and consequences. Journal of marketing

research, 36(2), 187-199.

Campbell, M. C. (1999b). Pricing strategy & practice “Why did you do that?” The important role of inferred motive in perceptions of price fairness. Journal of Product & Brand Management.

Chiao, L. T. B. Y. C., & Bei, L. T. (2006). The determinants of customer loyalty: an analysis of intangible factors in three service industries. International Journal of Commerce and Management, 16(3/4), 162-177.

Chen, K., & Liu, L. (2005). The" Best K" for entropy-based categorical data clustering.

Choi, S., & Mattila, A. S. (2005). Impact of information on customer fairness perceptions of hotel revenue management. Cornell Hotel and Restaurant Administration Quarterly, 46(4), 444-451.

Center, I. K. (2012). TwoStep cluster analysis. Retrieved July, 25, 2016.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2013). Applied multiple regression/correlation analysis for the

behavioral sciences. Routledge.

Dasgupta, P. R., & Hashimoto, Y. (2004, July). Multi-attribute dynamic pricing for online markets using intelligent agents. In Proceedings of the Third International Joint Conference on Autonomous Agents and Multiagent

Systems-Volume 1 (pp. 277-284). IEEE Computer Society.

(42)

41

Dickman, S. (1985). Impulsivity and perception: individual differences in the processing of the local and global dimensions of stimuli. Journal of Personality and Social Psychology, 48(1), 133.

Dickson, P. R., & Kalapurakal, R. (1994). The use and perceived fairness of price-setting rules in the bulk electricity market. Journal of economic psychology, 15(3), 427-448.

Eckel, C. C., & Grossman, P. J. (1996). The relative price of fairness: Gender differences in a punishment game.

Journal of Economic Behavior & Organization, 30(2), 143-158.

Ehrenberg, A. S., & Scriven, J. (1999). Brand loyalty. The Elgar companion to consumer research and economic

psychology, 53-63.

Eshghi, A., Haughton, D., Legrand, P., Skaletsky, M., & Woolford, S. (2011). Identifying groups: A comparison of methodologies. Journal of Data Science, 9(2), 271-291.

Faruqui, A., & Palmer, J. (2011). Dynamic pricing and its discontents. Regulation, 34, 16. Faruqui, A. (2012). The ethics of dynamic pricing. In Smart Grid (pp. 61-83). Academic Press.

Fonseca, J. R. (2013). Clustering in the field of social sciences: that is your choice. International Journal of Social

Research Methodology, 16(5), 403-428.

Garbade, M. J. (Sep 13, 2018). Understanding K-means Clustering in Machine Learning. Retrieved 20.07.2020 from: https://towardsdatascience.com/

Garbarino, E., & Lee, O. F. (2003). Dynamic pricing in internet retail: effects on consumer trust. Psychology &

Marketing, 20(6), 495-513.

Gensler, S. (2016). Finite Mixture Models. Handbook of Market Research. Doi: 10.1007/978-3-319-05542-8_12-1. Greenberg, M. G., & Frank, R. E. (1983). Leisure lifestyles: Segmentation by interests, needs, demographics, and

television viewing. American Behavioral Scientist, 26(4), 439-458.

Harrison, J. M., Keskin, N. B., & Zeevi, A. (2012). Bayesian dynamic pricing policies: Learning and earning under a binary prior distribution. Management Science, 58(3), 570-586.

Helson, H. (1964). Adaptation-level theory: an experimental and systematic approach to behavior.

Heo, C. Y., & Lee, S. (2011). Influences of consumer characteristics on fairness perceptions of revenue management pricing in the hotel industry. International Journal of Hospitality Management, 30(2), 243-251.

Hinz, O., Hann, I. H., & Spann, M. (2011). Price discrimination in e-commerce? An examination of dynamic pricing in name-your-own price markets. Mis quarterly, 81-98.

Hofstede, G. (2001). Culture's consequences: Comparing values, behaviors, institutions and organizations across

nations. Sage publications.

Hogan, W. W. (2010). Fairness and dynamic pricing: comments. The Electricity Journal, 23(6), 28-35. Horn, B., & Huang, W. (2009). Comparison of segmentation approaches. Decision Analyst: Arlington, TX. Huang, Z. (1998). Extensions to the k-means algorithm for clustering large data sets with categorical values. Data

mining and knowledge discovery, 2(3), 283-304.

Jung, K., Cho, Y. C., & Lee, S. (2014). Online shoppers' response to price comparison sites. Journal of Business

Research, 67(10), 2079-2087.

Effect of consumer characteristics and contextual variables on the dynamic pricing implementation