Is odd-number pricing strategy working?

(1)

(2)

(3)

Management Summary

The strategy of setting a product’s price one cent or a few cents lower, e.g. €0.99 instead of €1.00, is odd-number pricing strategy. Odd-number pricing strategy is widely used in FMCG product categories and markets. A lot of marketers, manufactures, and retailers simply adopt this strategy because they see other players in the market are doing so. But does it really work compared to the rounded-number pricing strategy? And if it works, is there any difference of the effectiveness when products in the product category can be either PL (private label) or NB (national brand)? How can such differences combined in a product influence consumers’ shopping behaviors, and will they prefer to choose the €0.99 product or the €1.00 one?

To test how this potential relationship actually looks like, two more moderating factors were included, which are consumers’ shopping characteristics, i.e. their shopping frequency and consumption volume, and perceived product quality towards both PLs and NBs. The research was conducted in the fluid milk product market in the Netherlands from February till July 2018, and it was with a preliminary data collection and analysis. In the data collection, a sample size of 195 people, who were at the moment living in the Netherlands, participated in the online survey powered by Qualtrics RuG in May 2018, and their data were collected and put in R programming for a multinomial logistics model analysis.

Key findings suggest that most of the participants chose Friesland Campina and Albert Heijn PL as their current favorite brands. Such phenomenon indicates that both NBs and PLs are well developed and widely accepted in the Netherlands. And odd-number pricing strategy generally works on consumers’ choice preference and willingness-to-pay (WTP) in the FMCG categories, in a market that both NBs and PLs are mature. When an odd-number price is offered for a product, a consumer will show less choice preference towards the product. When the price of the product would change from, for instance €0.99 to €1.09, a consumer would show much less choice preference towards the product. A product being either a PL or a NB, influences consumers’ choice preference, but this research could not yet prove its moderating effect on the relationship between the odd-number pricing strategy and consumers’ choice preference in the current research setting.

(4)

Preface

(5)

(6)

(7)

(8)

1. Introduction

1.1 Background

Even now, some retailers still do general practice and procedure when it comes to pricing, and they ignore how the market reacts towards the different pricing strategies. Some retailers adopt the odd-number pricing strategy without having the right pricing knowledge or the knowledge of how the markets and consumers react to different prices. They simply do mark-up-pricing, which is neglecting the predominant practice and how consumers really react to different pricing strategies in a market (Monroe and Cox, 2001). One of the general pricing strategies is “intuitive” pricing; for instance, some retailers simply take rounded-number pricing strategy, as it is well known as a popular and widely adopted measure (Holdershaw and Gendall, 1997); and some others simply adopt the odd-number pricing strategy, while neglecting the markets’ and consumers’ real reactions towards the strategy. These retailers conduct the pricing strategies based on their own knowledge or they simply follow what other competitors in the market do. At the same time, markets and consumers in those markets are dynamic, they keep changing, the knowledge that retailers hold/have are not up-to-date, and the quality and efficiency are highly doubtable. Knowledge about consumers’ willingness-to-pay (WTP) or choice preference for a certain product plays a crucial role in many areas, like pricing decisions for both manufactures and retailers (Breidert, Hahsler and Reutterer, 2006). Supported by Balderjahn (2003), consumers’ choice preference is an essential element to look into for developing and deciding an optimal pricing strategy. To ensure easy readability and comprehension, the research includes definition of key terms. Odd-number pricing is the practice of setting prices just below the nearest round figure, and it is often referred to as psychological pricing (Wedel and Leeflang, 1998) e.g., labeling a product as €1.99 rather than €2.00 (rounded-number pricing strategy).

A private label (PL) is a brand owned not by a manufacturer or producer, but by a retailer or supplier who gets its goods made by a contract manufacturer or by itself nowadays under its own label (Businessdictionary.com). Those PL products only sold in the brands own retailing chains, e.g. Albert Heijn (AH)’s own brand of milk is a PL product, and it is only sold at AH supermarkets. Comparably, a national brand (NB) is marketed throughout a national or international market. NBs are owned and promoted usually by large manufacturers and companies (Businessdictionary.com). Those products are sold across different retailers and stores that generally do not specify a chain brand’s exclusion. For example, Campina milk is a NB in the Netherlands and Belgium, and it is widely sold at AH, Jumbo, COOP, and other supermarkets and retailers.

(9)

than €2.00 (Kinard, Capella and Bonner, 2013). After such an overwhelming moment, the consumer would try to calm down and re-consider the impulse of purchasing some brand of the same product that is only €0.01 cheaper. On the other hand, the consumer could be attracted by the €0.01 cheaper odd-number pricing strategy and choose a different label category instead of the regular choice (e.g. choosing a PL instead of an NB that he or she consumes regularly). So, odd-number pricing strategy seemingly works on many consumers. What is the trade-off in the consumers’ mind here, and how about all the other consumers in general? And what can be found out from this research and tell retailers? Should all retailers follow odd-number pricing strategy for all, or should they look at the strategy efficiency on PLs and on NBs differently?

Odd-number pricing strategy is well researched, but not well studied with regards to testing how effective such a pricing strategy is when a product could be either a PL or a NB on consumers’ choice preference. In real life, some consumers personally do not care if the product is from a PL or a NB, such as in the Netherlands; both of them have comparably good quality (Kadirov, 2015). So, are there any differences between them, a PL and a NB, that the odd-number pricing strategy can actually influence consumers’ choice preference differently? If such a combination can be researched successfully, it could provide fresh eyes on how odd-number pricing strategy is really sufficiently influencing consumers’ choice preference under those mentioned moderating conditions. Then it could help and direct retailers to re-consider their pricing strategies of FMCGs with future better sales and profits, and it could lead to a new direction in the research field of odd-number pricing strategy.

1.2 Research Questions

To clear out the doubts, this research looks deeper and closer at pricing strategies, and breaks it down to how odd-number pricing strategy compared to rounded-number pricing strategy for both PLs and NBs, influences consumers’ purchase choice preference in the FMCG category.

Main research question How does odd-number pricing strategy influence consumers’ choice

preference in the FMCG category for both PLs and NBs in the FMCG category?

To answer the main research question, more sub research questions are defined and raised, and they also need to be answered to provide a full view on odd-number pricing strategy’s efficiency based on the difference of PLs and NBs. What could be the other moderating factors interfering the relationship between odd-number pricing strategy and consumers’ choice preference with interactions?

(10)

separately. It is interesting to explore and see how the product category from either PL or NB can moderate consumers’ choice preference out of odd-number pricing strategy. So the sub research questions are summarized and presented below.

Sub research question 1 How does odd-number pricing strategy influence consumers’ choice

preference differently when there is a difference of PL and NB?

Second, consider consumers’ perceived quality opinions towards products and brands: consumers could have pre-knowledge or judgment about the products and the brands they would like to purchase, it could be a PL or a NB, in different product categories. Here, it is about the perceived quality difference how consumers think of certain brands themselves. Some of them may not think that there is much of a quality difference between those brands, but some others may think the opposite. The derived sub research question is summarized and presented below.

Sub research question 2 Does perceived product quality influence odd-number pricing strategy’s

influence on consumers’ choice preference?

(11)

At the beginning of the introduction, it talks about the fact that some retailers adopt odd-number pricing strategy without having the right pricing knowledge or the knowledge of how the markets and consumers actually react to different prices. In practice, the outcome of this research can provide manufacturers and retailers an up-to-date view how odd-number pricing strategy works towards consumers’ choice preference in the real world, especially when a store wants to conduct such a strategy on both its popular PLs and NBs from same or similar product categories at the same time, and how consumers think of the brands’ quality and how they do shopping could influence their shopping experience and choices in general.

(12)

2. Theoretical Framework

Mentioned in Chapter 1 Introduction, when it comes to the odd-number pricing strategy’s influence on consumers’ choice preference; certain differences of product traits can play important roles. Here, Chapter 2 continues and looks further into the research questions with corresponding literature support, hypothesis sets and an overall conceptual model at the end.

(13)

2006; Thomas, Simon & Kadiyali, 2007), with an increase in sales. For example, it is often presented along with advertisement for promotion campaigns in practice; it develops a habit of consumers that a product with such an odd-number price tag is considerably cheaper, and consumers’ choice preference could therefore be influenced to be different.

All the literatures above support that odd-number pricing strategy has certain significant influence on consumers’ choice preference. To analyze further on how exactly odd-number pricing strategy influences consumers’ consumers, compared to rounded-number pricing strategy in the FMCG environment, the following hypothesis is proposed.

Hypothesis 1 (H1) Odd-number pricing strategy has a significant positive influence on consumers’

choice preference, compared to rounded-number pricing strategy. When the odd-number pricing is adopted for a certain product, consumers will shorten their consideration process, think the product is significantly cheaper, and the odd-number pricing strategy will result in a more preferable purchase choice.

2.2 PL vs. NB

(14)

Balan, 2012), that odd-number prices could signal lower product quality, and retailers prefer to use it for PLs rather than for NBs. Conclusively, the studies talking about odd-number pricing strategy for both PLs and NBs are rare. So, is there really a difference between a PL and a NB, and can the difference moderate the effect of odd-number pricing strategy on consumers’ choice preference? To answer this question, the following hypothesis is proposed for further research.

Hypothesis 2 (H2) The difference of a product being a PL or a NB significantly moderating the

relationship between odd-number pricing strategy and consumers’ choice preference. When it is a PL product, there is a stronger positive effect compared to a NB product on consumers’ choice preference.

2.3 Perceived Product Quality

Second, does perceived or expected product quality influence odd-number pricing strategy’s influence on consumers’ choice preference? And if so, how does it work? In sub question 1, the research already discusses the valuable points on the relationship between odd-number pricing strategy and consumers’ choice preference. Here, it continues with the focus on the trade-off between perceived product quality and consumers’ choice preference.

The trade-off between quality and choice preference has been widely researched. The term “perceived product quality” is defined as the relative- instead of the absolute value that consumers have in their mind; it is the knowledge consumers have before purchasing. It is the value that consumers hope to receive in exchange for the price paid for the product or service (Valle et al., 2017). Researched by Qi, Chu and Chen (2016), consumers look at a product’s price and compare it with the perceived quality and value this product could bring. They consider quality is in line with the distinctions between brands, and quality perceptions play a major role in determining whether a consumer will buy a PL or not (Sethuraman and Cole, 1999). Quality differentiation usually exhibits itself through the perception that PLs are of lower quality than the corresponding NBs (Choi and Coughlan, 2006).

(15)

Supported by Nielsen Company Report (2014), the perceived quality difference of two products in the same product category can drive consumers to show more favorable choice preference for a product with “better” quality (i.e. NB), and pay a price premium for that.

In the research of Yanhong, et al. (2011), it talks about hedonic pricing, and suggests that the price of a product can be decomposed to account for the value provided by its attributes, and some consumers prefer branded/premium products (i.e. NB) over PLs, because they are more valuable as those branded products are associated with better quality and prestige images. Sometimes, consumers believe the products, which are more expensive, are with higher quality (Erickson and Johansson, 1985). They use the perceived quality perspective to decide how much they would like to pay, or which product or brand to choose. In other words, when consumers believe that a certain product has better quality, they would like to pay more for that, which shows a more favorable or preferable purchase choice (Wolinksky, 1983). And in Bertini, Wathieu and Iyengar’s research (2012), a greater sensitivity to quality results in higher choice preference for high-quality options than for the low-quality options.

Conclusively, all the researches mentioned above support that there is a relationship between perceived product quality and consumers’ choice preference. And the following hypothesis is proposed to find out how such a relationship could be moderating the relationship between odd-number pricing strategy (as the explanatory variable) and consumer’s choice preference.

Hypothesis 3 (H3) Perceived product quality moderates the relationship between odd-number pricing

strategy and consumers’ choice preference. When a consumer has good perceived product quality knowledge towards a specific brand or product, he or she shows a more favorable purchase choice towards the product with odd-number price, instead of the product with rounded-number price.

2.4 Consumers’ Shopping Characteristics

Last, recall the last sub research question in Chapter 1, which is about “how do consumers’ shopping characteristics influence odd-number pricing strategy’s influence on consumers’ choice preference?” So, how can shopping characteristics moderate the relationship between odd-number pricing strategy and consumers’ choice preference? By definition, consumers’ shopping characteristics are about segmenting consumers based on typical demographic-based groups, age or gender and similar buying behaviors and lifestyle habits (Holmes, 2015), and what specific shopping habits such consumers share in a group, for instance, shopping frequency and consumption volume. In the research of Kinard, Capella and Bonner (2013), consumers’ specific shopping characteristics show a positive effect on odd-number pricing strategy’s efficiency.

(16)

It has a wide scope, whether they are heavy users, regular users, or lighter users from the volume segmentation perspective; whether they have brand involvement or attachment and so on, could influence their purchase choice preferences in certain ways.

Furthermore, Harris and Bray (2007), and Baumgartner and Steiner (2007) both explain in their researches conducted in Europe that some buyer segments (e.g. women) are more sensitive to odd-number pricing strategy (i.e. nine-ending prices).

In line with previous sub research questions, in countries where PLs are already at the maturity life cycle stage, consumers are more educated, mature and rationally thinking. Consumers perceive equal quality for both PLs and NBs, but paying a price premium for NBs still happen. And this could come from that consumer’s loyalty or attachment towards the brand (Krishnamurthi and Raj, 1991). And once a consumer becomes more or less a PL user, he or she will be more price-insensitive to a brand of PLs, and once a consumer is more or less a NB user, he or she would also be more price-insensitive to a product of NBs (Goldsmith et al., 2010). Thus, consumers’ shopping characteristics do play an important role. For retailers, this is an interesting part to work on, and it can provide useful information for further pricing and promotion strategies. All the literatures presented above support that there is a relationship between consumers’ shopping characteristics and their product choice preference, but not all the characteristics are related to this specific research. In this research, the consumers’ shopping characteristics are fragmented down to shopping frequency and consumption volume (Solomon, Russell-Bennett, and Previte, 2012). The reason of doing so is that consumers’ shopping characteristics represent a combination of personal traits whilst shopping, and other traits cannot be easily designed in a CBC environment for collecting data. Therefore, those two specific traits are listed and they could be used for segmentation in the survey. And the following hypotheses are listed to find out how such a relationship could be further interacting with odd-number pricing strategy involved as the explanatory variable. Hypothesis 4a (H4) Consumers’ shopping characteristic, specifically shopping frequency, significantly

moderates the influence odd-number pricing strategy has on consumers’ choice preference. Consumers who shop more frequently show stronger choice preference for products with odd-number pricing.

Hypothesis 4b (H4) Consumers’ shopping characteristic, specifically consumption volume, significantly

(17)

2.5 Conceptual Framework

(18)

3. Research Design

In this chapter, the research explores further into the corresponding target group, research method, data collection, and plan of testing analysis in details.

3.1 Target Group and Market

Based on the conceptual model and the fact of lacking original secondary data support, this research requires primary data collection and analysis. It focuses on the relationship between odd-number pricing strategy and consumers’ choice preference, along with three designed moderating factors, PL vs. NB, perceived product quality, and consumers’ shopping characteristics. To achieve a realistic and successful research in a short period of time (from February till June 2018) for five months, simply studying all FMCG categories would not work, and it was also important to choose a target market to research. Combined with the research of Breidert, Hahsler and Reutterer (2006), it was decided as an indirect, in the format of a within-group CBC design with full profile method, conducted as a computer-assisted online survey experiment (Miller et al., 2011). To narrow down and specify the research design, the dairy product category in the Netherlands was chosen to be the target market. Again, the dairy product category was narrowed down to the specific 1L fluid milk carton packaging, as the fluid, original-flavored, in three levels of richness, with full fat, semi-skimmed, and skimmed (“volle melk”, “halfvolle melk” and “magere melk” on the packaging in Dutch, which are sold the most in the Netherlands) product sold at all different supermarket chains/retailers, such as AH or Jumbo in the Netherlands.

Why was the dairy product category chosen to be the target product category? Dairy product category is widely adopted. Reported by Nielsen Company (2014), dairy products, especially fluid milk products, firstly have minimal differentiation and low brand equity. There is a rather low level of product difference among them, and there are many suppliers and brands in the market. Second, those products are inexpensive, and consumers show high price sensitivity, high purchase frequency, and low involvement. Consumers who are especially in developed countries and markets, they are highly price-sensitive, less brand-loyal, and they are always looking for the best deal in the market. Last, milk products are with low innovation rate, which provides a stable product status in the commodity category (Nielsen Company, 2014) for a realistic research purpose.

(19)

question 3 mentioned in Chapter 1, consumers’ shopping characteristics. On the other hand, this research has a focus on whether odd-number pricing is working differently under the condition of the variability of PL vs. NB. A market such as the Netherlands shows strong PL sales, and PL brands/products here are well developed on a national level. PLs in the FMCG categories are already at the maturity life cycle stage (Steenkamp, Van Heerde and Geyskens, 2010), with a wide extension of product categories (Nielsen Company, 2014) in the Netherlands. For example, the PL milk price is between €0.89 to €1.49 on average, and FrieslandCampina Company’s NB product Campina milk price is between €1.02 to €1.69 on average (Ah.nl, 2018).

To achieve a successful research in a short time period, it was decided to use all real market brands for the research online survey. As it also included PLs, the AH supermarket chain was chosen. The reason of choosing AH is that it is the most popular supermarket chain brand in the Netherlands, with 954 stores compared to 585 of Jumbo (Statista.com, 2018) and other supermarket brands by the end of 2017. AH also has a rather complete product portfolio with lots of product categories and lots of product brands under each product category. Further in the online survey, AH Supermarket’s own PL milk brand AH brand was picked as the PL product, and Friesland Campina Company’s own milk brand Campina was picked as the NB product for comparison. Therefore, it was decided to conduct the research on the milk product category in the Dutch market.

3.2 Research Method

As a quantitative research study in marketing intelligence, conducting an online survey is the optimal option. Online survey experiment is digital, fast, real-time, comprehensive, and scalable (Eggers et al., 2017). It is with achievable measures that University of Groningen supports students with a professional version of Qualtrics online survey service, and the survey could be widely spread to different kinds of milk consumers in the Netherlands via the paid Facebook Advertisement function.

The research has one IV and one DV, along with other three moderating factors. The IV is odd-number pricing strategy vs. rounded-odd-number pricing strategy; the DV is consumers’ choice preference, and followed by moderators PL vs. NB, perceived quality difference, and milk consumers’ shopping characteristics/habits. To conduct such a quantitative research on the milk consumers in the Netherlands, a CBC analysis method was picked.

(20)

moderators into independent attributes and levels to measure the moderating influences on choice preference.

Conducting this research in the format of an online survey requires specification of which attributes to include, and their individual attribute levels. What are the representative attributes of milk carton product here? In the research of Pilelienė and Liesionis (2014), milk products have six general attributes for conjoint analysis studies, which are country-of-origin, naturalness, package size, package type, fat richness, and price. Here, along with the main effect, there are only three moderators included. So the attributes should be odd-number pricing strategy vs. rounded-number pricing strategy, and PL vs. NB brand. Consumers’ shopping characteristics questions and perceived product quality question are asked at the beginning of the survey, as they cannot fit in the CBC condition, same with the simple demographic questions about gender and age, which would be asked in the end of the survey. Therefore, the demographic questions and consumers’ shopping characteristics questions would be presented in the single multiple-choice question setting, and the perceived product quality question is in the scale condition. To provide a clear over view of the design layout, the following table is presented.

Research Design Layout in Survey

1. Multiple Choice Part

Variable Consumers’ shopping characteristics Shopping frequency Consumption volume 2. Scale Part Variable Perceived product quality Very poor Poor Fine Good Very good 3. CBC Part

Attributes Variables Levels

Attribute 1 Odd-number pricing vs. rounded-number pricing (4 _levels)

0.99 1.00 1.09 1.10

Attribute 2 PL vs. NB (2 levels) _NB: CampinaPL: AH

(21)

number-of-level effect. It means that the numbers of levels are not distributed equally across different attributes, and this could lead to higher importance of some attributes with more levels than others (Eggers et al., 2017), and further biased analysis outcome. For instance, the first attribute odd-number pricing strategy vs. rounded-number pricing strategy is the main effect, and it has four attribute levels (€0.99, €1.00, €1.09, and €1.10). Also, the second attribute PL vs. NB brand has only two attribute levels, which is a binary variable. All above means that attributes’ levels are to be adjusted to deliver a reliable attribute design, and the multicollinearity check and outcome are presented in 4.9 Multicollinearity on page 33. In this way, only pricing strategy and PL vs. NB could be adopted as attributes in the CBC design part, and perceived product quality and consumers’ shopping habits would be scales and in the multiple-choice setting. Only two attributes in a CBC condition with rather few attribute levels cannot deliver good research quality, and to control that, a third control attribute was added, and it was the fat richness of milk products in three levels, full fat, semi-skimmed, and skimmed. Adding this control variable in the moderating model helps that a fluid milk product has six general attributes for conjoint analysis studies, which are country-of-origin, naturalness, package size, package type, fat richness, and price (Pilelienė and Liesionis, 2014). In such a way, the moderating model has three CBC variables, which are odd-number pricing strategy vs. rounded-number pricing strategy, PL vs. NB, and fat richness. All other variables are presented in scales in the single multiple-choice questions. And how should such a research look like in an online survey format? The minimum sufficient sample size is set to be around 200 (achieved 195 in the end), to provide generalizability of representing the Dutch market with a population size of 17 million by the end of 2017 (Worldpopulationreview.com, 2018). All the participants would be reached out from the Internet (supported by RuG Qualtrics), of various age groups, genders, education, and household backgrounds during the time period 22nd April to 10th May 2018. The Facebook Advertisement function served the need to promote a certain post or link to a wider range of audience, and the scope of audience number and region could be chosen as wanted. Therefore, using it to spread the survey and to reach the target group was an effective option.

3.3 Data Collection

To ensure the quality of the data collected, a pre-test was firstly conducted between 19th to 21st April 2018. A group of five participants was chosen to fill in the survey first, to check how the data collection procedure would work out and for further survey revision. During the pre-testing period, the paid Facebook Advertisement function was not yet active, as the pre-test surveys were delivered and spread to the social circle of the researcher. The pre-testing environment was face-to-face, and all advice and revision were therefore finished immediately and sufficiently.

(22)

moment. This question was created specifically for the paid Facebook Advertisement function to control the participant scope (which country they were staying in at the moment), as only the people who were in the Netherlands were selected in the target group. Third, it asked the question if the participant had an allergy or tolerance problem with milk products, the participants who were reached out but with allergy/tolerance problems, were directed to the end of the survey right away. Fourth, each participant had to answer shopping habit questions (about shopping frequency and consumption volume), those questions define the consumers into possible sub groups, and to answer the consumers’ shopping characteristics part. Next, an introduction of the product, with a question of “What is your current favorite milk brand in the market (with also an open choice)?” was provided. After that, the survey introduced the focus of this research with what a PL was, what a NB was, and the AH and Campina brands were presented, for testing the perceived product quality moderator. After all the previous steps, the survey continued with the major part of the survey. After this part would be finished, the participant would then be exposed to all attribute choice set combinations. As a random factorial design, all attribute levels were combined and a part of them (eight combinations) was randomly chosen and exposed to each participant; because this was mostly efficient when there were not that many attributes or levels.

As a 4*2*3 (four pricing levels * two PL/NB brand levels * three fat richness levels) random full-factorial design, every combination of attribute levels were included, and a full-full-factorial design included all potential interaction effects of the IV and moderators. Therefore, there would be 24 choice combinations in total. And it is a within-group design, with only the difference in the survey that attribute choice alternatives and sets were randomly chosen and exposed to every participant by Qualtrics’ randomization function (Randomization question list in Appendix 5 on page 62), in the cue card & image presentation. Here, the necessity of using the randomization function is that this research focuses on the interaction effect between moderator NB vs. PL, and the IV odd-number pricing vs. rounded number pricing and how consumers react differently when both factors are included in one scenario. Each participant was exposed to eight different questions out of 50 questions (choice sets) were created by Preferencelab.com in an automatic randomization way. In each question, there were three choice alternatives, along with a no-choice option (consumers could show price sensitivity, so presenting a no-choice option is optimal). The inclusion of a no-choice option ensures the realism of CBC and that a threshold could be identified and indicated the utility that is actually needed to make the participant switch from one choice option to another (Eggers et al., 2017).

(23)

(24)

included a control variable fat-richness, which could have shown certain taken-over power compared with the IV and moderators, e.g. a consumer only chose milk product options with the same level of fat-richness, despite the brand or price differences, using R programming could provide better and more complete analysis on it. The data analysis took care of both main effect and the other three moderating effects. Besides that, all the interaction effects between the IV and moderating variables were taken into account. By the nature of the online survey asking for consumers’ choice preference, this research was presented as a part-worth utility model below. Null Part-worth utility model Y= β1*X1 + β2*X2 + β3*X3 + β4*X4 + β5*X5 + β6*(Xi*XJ) Y: model dependent variable, the overall utility-consumers’ choice preference; β: utility size of the variable’s effect; X1: main effect of variable odd-number pricing strategy vs. rounded-number pricing strategy; X2: moderating effect of binary variable PL vs. NB; X3: moderating effect of control variable milk product fat richness; X4: moderating effect of variable consumers’ shopping characteristics; X5: moderating effect of variable perceived product quality; Xi*Xj: potential interaction effect between variable i and variable j; It was important to first test if the model was overall presenting significantly influenced outcome. After that, the analysis broke down to test the main effect, moderating effects, and the inclusions of interaction effects, which all were done by using Multinomial Logit Model Analysis (MNL), as a MNL is designed for the nominal or categorical type of explanatory variables, in this research, the IV and moderators are nominal. MNL is a rather easy model designed for choice-based-conjoint analysis for choosing out of multiple options (Dehmamy, 2018), and it is easy to be conducted in the R environment. For potential multicollinearity, it could be detected by conducting regression analysis, or check either the VIF- or Tolerance score. After all, all the data analysis details are presented in Chapter 4. Chapter 4 focuses on the procedure after data collection, and presents the data analysis and the relevant details.

(25)

(26)

Furthermore, the question about the current favorite milk brand was included (Figure 5). 28% of the participants did not have a specific favorite brand, indicated as “NA” in the dataset, followed by two most popular brands Friesland Campina (21%) and Albert Heijn (18%). Interestingly, both AH and Campina were mentioned as PL and NB examples explained in the survey, but the question of current favorite milk brand was presented at the beginning, which indicates that there was no interfering effect of the PL and NB example presentation in the survey. Such phenomena of Campina and AH being the current most popular brands confirmed the good brand examples in the survey, and showed that PL brands owned a widely adopted market in the Netherlands. What’s more, as an open question, participants filled in their favorite brands, PLs such as Jumbo, AH Zaanse Hoeve, Lidl Milboa, AH Biological (presented as AH Organic in the data), and Jumbo Biological were often mentioned. Figure 5 Current Favorite Brand

4.2 Data Recoding

After separating variables “age”, “gender”, and “current favorite brand”, the dataset was not yet ready. Qualtrics has a disadvantage of having purely the original data, recoding the dataset were necessary (Eggers, 2018). To conduct the recoding procedure successfully, all questions and their answers in the survey were given different numbers as choice levels, and the details are presented in Appendix 9 on page 97.

(27)

4.3 Part-worth Utility Model

After a successful recoding, it went to the procedure of running the MNL Analysis in R, and testing out the overall significance of the model, and the potential autocorrelations among explanatory variables (IV and moderators).

First of all, Table 2 below presents how the model was tested and adjusted step-by-step. Attribute & effect Null model Model 1 Model 2 Model 3 Model 4 Optimal model

Price.1 _✓ _✓ _✓ _✓ Price.2 _✓ _✓ _✓ _✓ _✓ Price.3 ✓ ✓ ✓ ✓ ✓ ✓ Price.4 _✓ _✓ _✓ _✓ _✓ Brand.1 _✓ _✓ _✓ _✓ _✓ Brand.2 ✓ ✓ ✓ ✓ ✓ Fat.1 _✓ _✓ _✓ _✓ Fat.2 _✓ _✓ _✓ _✓ Fat.3 _✓ _✓ _✓ _✓ None_option _✓ _✓ _✓ _✓ Brand.2 * Price.2 ✓ ✓ Brand.2 * Price.3 _✓ _✓ Brand.2 * Price.4 ✓ ✓ Consumption_Volume * Price.2 ✓ ✓ Consumption_Volume * Price.3 _✓ _✓ Consumption_Volume * Price.4 _✓ _✓ Shopping_Frequency * Price.2 _✓ _✓ Shopping_Frequency * Price.3 _✓ _✓ Shopping_Frequency * Price.4 ✓ ✓ AH_Quality * Price.2 _✓ _✓ AH_Quality * Price.3 ✓ ✓ AH_Quality * Price.4 ✓ ✓ C_Quality * Price.2 _✓ _✓ C_Quality * Price.3 ✓ ✓ C_Quality * Price.4 _✓ _✓ Table 2 Model Adjustment Overview

To test points mentioned above, the MNL model analysis was done twice with the difference of control variable “fat-richness” being excluded, demonstrated in “part-worth utility MNL model 1”; and control variable “fat-richness” being included, demonstrated in “part-worth utility MNL model 2”. And the MNL analysis output is in line with each model, respectively. To double check which model would be better, Log-Likelihood, and AIC methods were conducted for comparison.

Model 1: part-worth utility MNL model Seletion_Dummy = Price.2 + Price.3 + Price.4 + Brand.2 +

None_option

(28)

Attribute Estimate P-value

Price.2 -0.043 0.383 Price.3 -0.139 0.004* Price.4 0.061 0.185 Brand.2 0.056 0.046* None_option -1.721 < 2.2e-16* Df: 5 Log-Likelihood: -1934 AIC: 3877.927 *are the significant ones Table 3 Part-worth Model 1 MNL Analysis Output

Model 2: part-worth utility MNL model Seletion_Dummy = Price.2 + Price.3 + Price.4 + Brand.2 +

Fat.2 + Fat.3 + None_option (better model)

Attribute Estimate P-value

Price.2 -0.021 0.681 Price.3 -0.154 0.002* Price.4 0.071 0.129 Brand.2 0.055 0.051* (marginally significant) Fat.2 -0.049 0.224 Fat.3 -0.007 0.851 None_option -1.722 < 2.2e-16* Df: 7 Log-Likelihood: -1932.8 AIC: 3879.604 *are the significant ones Table 4 Part-worth Model 2 MNL Analysis Output Demonstrated in Table 3 and Table 4, though AIC score increased from 3877.927 to 3879.604 from model 1 to model 2, model 2 included the control variable “fat-richness”, which were presented as “Fat.2” and “Fat.3” in R, and the model’s significance did not change very much. Attribute levels “Price.2”, “Brand.2” and “None_option” were still significant. Having “fat-richness” in the model included more variables, and maximized the Log-Likelihood (from -1934 to -1932.8) for better predictive validity. Therefore, model 2 was chosen to be the part-worth utility model.

(29)

Attribute Odd ratio P-value Price.2 0.979 0.681 Price.3 0.857 0.002* Price.4 1.073 0.129 Brand.2 1.056 _{(marginally significant)}0.051* Fat. 2 0.952 0.224 Fat.3 0.993 0.851 None_option 0.179 < 2.2e-16* *are the significant ones Table 5 Part-worth Model 2 Odd Ratio Output

Model 2 was chosen to be the part-worth utility model; its likelihood was then calculated and provided in Table 5. The output indicates that the odd ratio of “Price.3” is smaller than 1, and it means that it is less likely for a consumer to choose a product priced with odd number pricing strategy, which is €1.09, than the other options. The same goes for the “None_option”, that it is less likely for a consumer to choose this option than the others. Oppositely, “Brand.2” shows an odd ratio larger than 1, which indicates that a consumer is more likely to choose “Brand.2”, the Friesland Campina, a NB, other than the other option, “Brand.1”, AH, a PL product.

After adjusting the model, likelihood increased from -2162.619 to -1932.802, and the Pseudo R2

score decreased from 0.123 to 0.103, which was not very much a significant difference (only -0.020). Overall, it showed overall a good model fit. Part-worth utility model 2 LL0 -2162.619 LL2 -1932.802 Pseudo R2 0.123 Adj. Pseudo R2 _0.103 Table 6 Model 2 MNL Analysis Output

4.4 Part-worth or Linear

As the part-worth utility model 2 was chosen, there was still some doubt about whether the model should be part-worth or linear. To be noticed, the coefficients of a MNL model cannot be interpreted or used directly, to conduct the linearity test, the marginal effect table output is presented first. From output in Table 7, the results suggest that all of those significant variable levels showed negative marginal effect on participant’s choice preference.

(30)

Attribute Coefficient Odd Ratio Marginal Effect Price.2 -0.021 0.979 -0.0004 Price.3 -0.154 0.857 -0.022* Price.4 0.071 1.073 -0.005 Brand.2 0.055 1.056 -0.003* Fat.2 -0.049 0.952 -0.002 Fat.3 -0.007 0.993 -0.00005 None_option -1.722 0.179 -1.414* *are the significant ones Table 7 Model 2 Marginal Effects

After running the MNL analysis in R, it provided the visual linearity assessment of using attribute “Price”, which is the IV “odd-number pricing strategy vs. rounded-number pricing strategy” based on the current part-worth model. And the corresponding linearity test is presented below in Figure 6. Figure 6 Price Attribute’s Linearity Test The effect of the IV “odd-number pricing strategy vs. rounded-number pricing strategy” did not seem to be linear but in an S shape. Therefore, it was decided to stick to the part-worth utility model instead of a linear model.

4.5 Attribute Effect Estimation

(31)

Attribute Estimates P-value Price.1 0.104 0.038* Price.2 -0.021 0.681 Price.3 -0.154 0.002* Price.4 0.071 0.129 Brand.1 -0.055 0.255 Brand.2 0.055 0.051 (marginally significant)* Fat.1 0.056 0.150 Fat.2 -0.049 0.224 Fat.3 -0.007 0.851 None_option -1.722 < 2.2e-16* *are the significant ones Table 8 Model Attributes’ Estimates

From what can be found in Table 8, after recovering those levels in the model, “Price.1” shows significant effect. Also, all levels of attribute “Fat-richness” were surprisingly not significant, which could also be certain supportive fact that “Fat-richness” only worked as a control variable in the research survey design.

As coefficients of a MNL model cannot be interpreted and used for the size of effect, the odd ratio and marginal effect of “Price.1”, “Brand.1” and “Fat.1” are calculated and presented below.

Attribute Coefficient Odd Ratio Marginal Effect

(32)

choice option (None_option) showed a strong negative marginal effect on participant’s milk product choice preference.

4.6 Attribute Importance

In this CBC research design, it was vital to look at the comparison of attributes’ importance, and check out which one could seemingly influence consumers more strongly on choice preference than others. Attribute/variable Importance Price 54.60% Brand: PL vs. NB 23.17% Fat-richness 22.23% Table 10 Model Attribute Importance

Presented in Table 10, the IV “Price”, which represents odd-number pricing vs. rounded number pricing strategy, showed relatively strong importance to participants, which is over 50% (54.60%) on the overall importance. Such a result indicates that participants made product choices heavily based on the influence price instead of others. And moderators “Brand”, which represents the variability of a product being a PL or a NB, and “Fat-richness” showed equal importance to consumers, but “Fat-richness” was overall not significant as a control variable, and it did not influence the participants on their choice preferences.

4.7 WTP

Consumers’ WTP is measured based on choice preferences, and this research followed the same research design, and tried to look into consumers’ choice preference and WTP (Eggers et al., 2017). A choice preference equals to the utility per attribute level, which is “utility/level”, and as not all levels of each attribute in this research were significantly influencing the participants’ choice preferences, the part-worth utility model was then adjusted by including only the significant attribute levels.

Model 3: adjusted part-worth utility model Seletion_Dummy = -0.011*Price.1 - 0.022*Price.3 -

(33)

Attribute level P-value WTP Price.1 0.038* 0.060 Price.2 0.681 -0.012 Price.3 0.002* -0.089 Price.4 0.129 0.041 Brand.1 0.255 -0.032 Brand.2 0.051 (marginally significant)* 0.034 Fat.1 0.150 0.033 Fat.2 0.224 0.028 Fat.3 0.851 0.004 None_option < 2.2e-16* -1.000 *are the significant ones Table 11 WTP of Attributes Table 11 lists the incremental WTP of significant attributes, and it indicates that a €0.99 milk product increased the participant’s WTP by €0.060, and a €1.09 milk product decreased the participant’s WTP by €0.089, so the €0.99 product was more favorable. And a NB product increased participant’s WTP by €0.034.

4.8 Moderating Effects

To achieve a good research survey design, moderators “consumers’ shopping characteristics” and “perceived product quality” were fragmented down to “Shopping_Frequency” (consumers’ shopping characteristics, specifically shopping frequency), and “Consumption_Volume” (consumers’ shopping characteristics, specifically consumption volume), and “AH_Quality” (perceived product quality of AH PL) and “C_Quality” (perceived product quality of Campina NB), respectively. Those two moderators were asked as general multiple-choice questions at the beginning and before the CBC choice set questions. All previous points talked about the IV and moderating effects of attributes “Brand” and “Fat”, and the moderating effects of “Shopping_Frequency”, “Consumption_Volume”, “AH_Quality” and “C_Quality” are presented further in this section.

Model 4: added moderators’ part-worth utility model Seletion_Dummy = -0.011*Price.1 -

0.022*Price.3 - 0.003*Brand.2 - 1.414*None_option + Brand.2 * Price.2 +

Brand.2 * Price.3 + Brand.2

* Price.4 + Consumption_Volume * Price.2 + Consumption_Volume * Price.3 + Consumption_Volume * Price.4 + Shopping_Frequency * Price.2 + Shopping_Frequency * Price.3 + Shopping_Frequency * Price.4 + AH_Quality * Price.2 + AH_Quality * Price.3 + AH_Quality * Price.4 + C_Quality * Price.2 + C_Quality * Price.3 + C_Quality * Price.4

First, to test the moderating effect of attributes “Shopping_Frequency”, “Consumption_Volume”, “AH_Quality”, “C_Quality”, and “Brand” on “Price (odd-number pricing strategy vs. rounded-number pricing strategy)” were put in the MNL model in R programming for analysis.

(34)

Mlogit model Log-Likelihood Df

Without moderators -1932.8 7 With moderators -1926.1 22

Table 12 Model Fit of Adding All Moderators

In Table 12, including the moderators in the MNL model analysis increased the model fit and predictive validity, as Log-Likelihood increased from -1932.8 to -1926.1, with 15 more DF (degree of freedom) included. The moderating effects, which were tested as interactions, surprisingly did not indicate any significance of all those moderators on the relationship between odd-number pricing strategy and consumers’ choice preference. To retest for certainty, variables “Brand”, “Consumption_Volume”, “Shopping_Frequency”, “AH_Quality”, and “C_Quality” were separately added to the MNL model for further moderating effect test. Below, Table 13 (*are the siginificant effects) presents how the significance and coefficients of each variable level changed by adding more in the model (output in Appendix 6 on page 67).

Variable and

level Estimate null

(35)

Overall, the significances did not change in the models between variables, and it could not provide significant moderating effect of the interactions. When the moderations of “Shopping_Frequency”, “AH_Quality”, and “C_Quality” added into the model, “Price.3” became insignificant. Therefore, those three variables showed multicollinearity in the interactions, and they should be left out for the optimal model.

4.9 Multicollinearity

After adjusting the model for multiple times, it was still important to check multicollinearity between variables, and to finalize the model (output in Appendix 10 on page 98).

Variable “fat-richness” showed strong multicollinearity with most of other variables, which is a marginal issue due to the choice design. As it was a control variable, and it was left out from the model formula after the finding of insignificant P-value in the MNL model analysis, its multicollinearity is not much of importance. In the future, new research in the same area of odd-number pricing strategy vs. rounded-number pricing strategy with PL and NB could look further into including “fat-richness” in the original research question and hypothesis, instead of only including in a research design as a control variable, or choose a more generalized and easy-to-control FMCG category for survey or experiment, which is also mentioned in 5.2 Limitations. Other than variable “fat-richness”, some other variable and their different levels showed minimal multicollinearity, and this could be caused by the choice randomization design in the survey, and it should not be much of a concern. Overall, the research design and the model showed a good level of statistical power with rather good coefficient estimate and marginal effect stability. Therefore, the optimal utility model stays the same as what model 3 represents. Therefore, the optimal CBC MNL part-worth utility model is presented below. Optimal part-worth utility model Y= -0.011*X11 - 0.022*X12 Y: model dependent variable, the overall utility-consumers’ choice preference utility; β: utility size of the variable’s effect; X11: main effect of variable odd-number pricing strategy vs. rounded-number pricing strategy, when the product is priced as €0.99; X12: main effect of variable odd-number pricing strategy vs. rounded-number pricing strategy when the product is priced as €1.09;

Next, Chapter 5 will continue the flow, and summarize the data analysis findings with further recommendations and limitations of the current research.

(36)

5. Conclusions and Recommendations

Chapter 5 is the final part of the research paper, and the data analysis outcomes are summarized and presented below. Also, the recommendations and current research limitations are included.

5.1 Conclusions and Recommendations

(37)

Third, Hypothesis 2 proposed that a product of being a PL or a NB would moderate the relationship between the odd-number pricing strategy vs. rounded-number pricing strategy and consumers’ choice preference. The difference of a product being a PL or NB itself has a significant negative influence on consumers’ choice preference. When a NB milk product was provided, the consumer showed a lightly lower possibility and preference to choose it by a factor of 0.003. But it did not show significant moderating effect on the relationship between odd-number pricing strategy vs. rounded-number pricing strategy and consumers’ choice preference, as a moderator.

Fourth, Hypothesis 3 proposed that perceived product quality could moderate the relationship between odd-number pricing strategy and consumers’ choice preference. But the result could not prove so, and perceived product quality did not show significant moderating effect on the relationship between odd-number pricing strategy vs. rounded-number pricing strategy and consumers’ choice preference.

Next, Hypothesis 4 proposed that consumers’ shopping characteristics could significantly moderate the influence odd-number pricing strategy has on consumers’ choice preference. But consumers’ shopping characteristics neither showed significant moderating effect on the relationship between odd-number pricing strategy vs. rounded-number pricing strategy and consumers’ choice preference. Last, the no choice option, “None_option” in the survey CBC design, showed significant influence on consumers’ choice preference.

Overall, the CBC design and MNL model analysis showed the negative main effect of odd-number pricing strategy on consumers’ choice preference, and the negative main effect of a product being a NB on consumers’ choice preference, but not the moderators’. It has good validity and generalizability, and the research findings can be used and adopted for other FMCGs, which are low involvement products in a market areas such as Netherlands, the countries that share similar economical situation and consumer features and habits. It starts the research field of odd-number pricing strategy combined with branding strategy PL vs. NB in the academic world, and it sheds light on how retailers should price their FMCG products. For future academic research, more moderators (e.g. product shelf presentation position) or even mediators can be included into the relationship between odd-number pricing strategy vs. rounded-number pricing strategy and consumers’ choice preference, and new product categories and markets can be tested for better validity, generalizability, and new research directions.

(38)

could negatively influence consumer’s purchase choices in a market like the Netherlands. From the outcome of this research, the moderators of a product being either PL or NB does not influence the relationship between the pricing strategy and choice preference, and the same goes for perceived product quality’s and consumers’ shopping characteristics. Therefore, marketers and retailers should do better market and customer research before conduct such pricing strategies on FMCG goods to their customers when the market is mature and the product is widely adopted, and they could use their resources and focus on other potential factors, which could influence the pricing strategies’ efficiency, and combine in-store data, e.g. membership system and scanning data, and conduct better target store customer research for better pricing strategies and sales.

5.2 Limitations

This research also has limitations. Mentioned that it could suffer from number-of-level effect, as the research design had a 4*2*3 full factorial model, none of the three attributes in the model shared the same number of attribute levels. Researched by Eggers and Sattler (2009), the optimal solution to tackle this issue is to conduct the hybrid individualized two-level (HIT) CBC analysis in a full and random fraction environment. HIT CBC includes only the best and worst levels to generate an adaptive efficient choice design, and it includes the possibility to using individualized consumers’ choice preference measures such as price levels, which is perfect to continue to conduct further research on this specific topic area.

The research did not include other factors. For example, it did not include the influence of shelf sets in supermarkets on consumers’ choice preference. Sales from previous time period could have potential effect on the relation between odd-number pricing strategy and consumers’ price elasticity and choice preference, due to lack of secondary data support. Instead, asking consumers what their current favorite brands are as an explanatory variable could have comparably provided the information needed from lagged sales, and helped with finding certain benchmarking scope of sales and consumers’ current status of preferences.

Other marketing instruments, such as advertising and promotion can also have potential effects. Advertising influences how consumers perceive certain brands, and it could be combined with sales promotion strategy. Would it be more appealing to consumers when a premium NB has odd-number pricing sales with advertising promotion, or would it be the case of a PL? Including both advertising and promotion instruments can strongly cause multicollinearity in a model for sales, it can make the research also rather complicated with too many attributes and their levels. In this research, advertising and promotion are not the focus and therefore were not included.

(39)

as more expensive compared to PLs are. But it is decided to control the price factor and design the survey with products from both PL and NB in the same price range.

(40)

(41)

(42)

(43)

(44)

(45)

(46)

Appendix 2 Preference Lab Conjoint Choice Design

(47)

(48)

(49)

(50)

(51)

(52)

(53)

(54)

(55)

TheChoice=dfResponses[row , ColName ] if (!is.na(TheChoice)){ if (TheChoice==a) { BigData[BigDataRow2,14]=1 }else{ BigData[BigDataRow2,14]=0 } } #fill out Question and Alternative_ID BigData[BigDataRow2,2] <- q BigData[BigDataRow2,3] <- a } } } #convert Matrix to DataFrame (BigData to cbc) for easier column naming etc cbc <- data.frame(BigData) colnames(cbc) <- c("RespondentID", "Question","Alternative_id",

(56)

(57)

LL0<-195*8*log(1/4) LLml1<-ml1$logLik[1] LLml2<-ml2$logLik[1] PseuoR2<-1-LLml1/LL0 PseudoR2_adj<-1-(LLml1-5)/LL0 PseuoR2<-1-LLml2/LL0 PseudoR2_adj<-1-(LLml2-7)/LL0 install.packages("mlogit") library(mlogit) #Price linear? plot price utility function plot(c(1, 2, 3, 4), c(coef(ml1)[1:3], -sum(coef(ml1)[1:3])), xlab="Price", ylab="Utility") lines(c(1, 2, 3, 4), c(coef(ml1)[1:3], -sum(coef(ml1)[1:3]))) #estimate linear model for price ml3<- mlogit(Selection_Dummy ~ Price + Shopping_Frequency + Consumption_Volume + AH_Quality + C_Quality + Brand + Fat+ None_option| 0, cbc2) summary(ml3) ml3<- mlogit(Selection_Dummy ~ Price + Brand.2 + Fat.2 + Fat.3+ None_option| 0, cbc2) summary(ml3) ml3<- mlogit(Selection_Dummy ~ Price + Brand + Fat+ None_option| 0, cbc2) summary(ml3) ml3<- mlogit(Selection_Dummy ~ Price + Brand.2 + None_option| 0, cbc2) summary(ml3)

(58)

# recover reference levels for overview of output Price <- -sum(coef(ml2)[1:3]) Price.2<-sum(coef(ml2)[1]) Price.3<-sum(coef(ml2)[2]) Price.4<-sum(coef(ml2)[3]) Brand <- -sum(coef(ml2)[4]) Brand.2 <- -sum(coef(ml2)[4]) Fat <- -sum(coef(ml2)[5:6]) Fat.2 <- -sum(coef(ml2)[5]) Fat.3 <- -sum(coef(ml2)[6]) None_option <- -sum(coef(ml2)[7]) ml2<- mlogit(Selection_Dummy~Price.2 + Price.3 + Price.4 + Brand.2 + Fat.2 + Fat.3 + None_option| 0, cbc2) summary(ml2) #instll covMatrix package for standard errors install.packages("covMatrix") #standard errors covMatrix <- vcov(ml2) # this is the variance-covariance matrix sqrt(diag(covMatrix)) # these are the standard errors you find in the summary output (the square-root of the diagonal elements of the matrix)

(59)

(60)

(61)

(62)

(63)

(64)

Appendix 5 Preferencelab.com Randomization Question List

Question Alternative Brand Fat_richness Price

(65)

(66)

(67)

(68)

(69)

(70)

(71)

(72)

(73)

(74)

(75)

(76)

(77)

(78)

(79)

Appendix 8 Research Qualtrics Survey Overview

2018/5/29 Qualtrics Survey Software

(80)

(81)

https://rug.eu.qualtrics.com/ControlPanel/Ajax.php?action=GetSurveyPrintPreview 3/20 specify a chain brand exclusion. Here, the product is from Friesland Campina. How would you rate the quality of AH private label milk? How would you rate the quality of Friesland Campina national brand milk? From here, you go to the main part of the survey, you will be provided three milk product choice options. They are all from different brands, with different price tags, and different fat richness. Simply choose the one you prefer, or you can also choose the "None" nochoice option. Block 2 Which of the following milk products would you prefer to purchase?

Very poor Poor Fine Good Very good

Friesland Campina

AH Friesland Campina

(82)

(83)

(84)

(85)

(86)

(87)

(88)

(89)

(90)

(91)

(92)

(93)

(94)

(95)

(96)

(97)

(98)

(99)

(100)

(101)

(102)

(103)

(104)

(105)

(106)

(107)

(108)

(109)