How price and promotions influence product
returns of utilitarian and hedonic products in
the online retail industry
3
How price and promotions influence product returns
of utilitarian and hedonic products in
the online retail industry
Herre Zonderland S3534715
Master Thesis
First supervisor: Prof. dr. T.H.A. Bijmolt Second supervisor: C.F. Hirche, Ph.D. student
Examiner: Dr. F. Eggers
University of Groningen Faculty of Economics and Business
Department of Marketing PO Box 800 9700 AV Groningen
5
ABSTRACT
Since consumers of online retailers cannot physically interact with products prior to purchase, ordered products are not always as expected and are often returned to the retailer. The involved operational costs of product returns cut down margins, and as such, they are a big cost driver for online retailers. Due to its big impact on the profitability of a firm, a richer picture of what drives product returns is of high value for an online retailer. Although the body of literature on product returns has grown over the years, still little is known about how returns may differ between types of product categories. This paper investigates just that and looks at how price and promotions affect the product returns of utilitarian (i.e., task-related and useful) and hedonic (i.e., emotional and feeling-based) product categories. Binary logistic regression modeling is used to shed light on how these variables influence return probability.
The study uses a dataset from a large European online retailer that contains purchase-information six product categories, ranging from January 1, 2017 – December 31, 2017. The selected product categories are chosen based on how well they represent either utilitarian or hedonic product categories.
The results show that the effect of price is positive for vacuum cleaners, shavers, and hanging lamps (utilitarian) and audio & hifi while being negative for handbags (both hedonic). The effect of promotion is negative for vacuum cleaners, meaning that when this product category is on promotion, it is less likely to be returned compared to hedonic products. Conversely, tablets and audio & hifi showed the opposite result. The effect of discount depth is the same for all significant categories; as it increases, the products are less likely to be returned. In general, hedonic products are more likely to be returned than utilitarian products.
As more expensive products are more likely to be returned, online retailers could try to reduce consumers’ fit uncertainty for these products by providing additional product information. Furthermore, retailers should be careful when promoting hedonic products since price promotions of such products may increase the sales, but also its returns. Lastly and related to promotions, online retailers could display utilitarian products more so than hedonic products when many products are discounted. In sum, the results show that price and promotions differences do, in fact, exist in how they impact the product returns of different categories. This paper provides firms and academia with good first insights to get a clearer picture of how product returns of different product categories are driven by prices and promotions.
6
PREFACE
About three years ago I decided to pursue something I had wanted for a long time: to try to obtain a master’s degree by enrolling myself at a university. Now, after two and a half years of hard work and dedication at the University of Groningen, I can finally say that I have made that happen.
I began this academic journey with the pre-master marketing, which I started four years after I received my bachelor’s degree at a university of applied sciences. Although I had to get used to studying again and was unsure if I was capable of managing, I got more and more excited and confident along the way. The broad marketing domain has always intrigued me, and as I approached the end of my pre-master, I decided to enroll for both tracks of the master marketing: Marketing Management and Marketing Intelligence. I have thoroughly enjoyed each course, ranging from Market Models and Customer Models to Consumer Psychology and Marketing Communication. This thesis is the final phase of my time as a student.
First, I would like to thank prof. dr. Bijmolt for his advice, good talks and support while writing this thesis. I would also like to thank Christian Hirche for his helpful input and my thesis group members for the useful discussions about our thesis topics. Further, a big thank you to the staff and students at the University of Hamburg for welcoming us and for organizing meetings with staff members. The support I have received from my family over these last two years has helped me to succeed, for which I am grateful. Lastly, I would like to thank Guus van der Veen, Tjalling de Jong, Mendel Koornstra, Hidde Smit, and Ilyas Aaqaoui for our laughs and new friendships.
I hope you enjoy reading this thesis.
7
TABLE OF CONTENT
1. INTRODUCTION 9
1.1 Research questions 10
1.2 Relevance of the study 11
1.3 Research structure 11
2. THEORETICAL FRAMEWORK 13
2.1 Conceptual model 13
2.2 Literature review 14
2.2.1 A definition of product returns 14
2.2.2 Effects of price on product returns 14
2.2.3 Effects of promotions and discounts 15
2.2.4 The difference between hedonic and utilitarian motivations and products 16 2.2.5 Effect of price for hedonic and utilitarian products 17 2.2.6 Effects of promotion and discounts for hedonic and utilitarian products 17
2.2.5 Control variables 18
3. DATA COLLECTION 19
3.1 European online retail environment 19
3.2 Data preparation 19 3.2.1 Data selection 19 3.2.2 Data cleaning 20 3.3 Variable description 21 3.4 Data exploration 23 3.4.1 Missing data 23 3.4.2 Outliers 24 3.4.3 Descriptive statistics 25 4. METHODOLOGY 29 4.1 Model estimation 29 4.2 Model assumptions 30 4.3 Model specification 31
4.3.1 Model 1 (only direct effects) 32
4.3.2 Model 2 (including dummy variable) 32
4.3.3 Model 3 (including interactions) 32
5. RESULTS 34
5.1 Results model 1 34
5.1.1 Effect of price 35
8
5.1.3 Effects of discount depth 37
5.2 Results model 2 38
5.2.1 Main effect of hedonic and utilitarian products 38
5.3 Results model 3 39
6. DISCUSSION 41
6.1 Effect of price 41
6.2 Effect of promotions 42
6.3 Effect of discounts 42
6.4 Hedonic versus utilitarian products 43
6.5 Interaction effects 43
6.6 Conclusions and managerial implications 44
6.7 Limitations and future research 45
8. REFERENCES 47
9. APPENDIX I 53
9
1. INTRODUCTION
Digitalization has changed the retail environment forever. The emergence of the Internet has offered the traditional brick-and-mortar retailers new ways and channels to sell their offerings while also giving rise to online retailers without physical stores. Online retail sales have grown steadily since 2000 and continue to do so at an increasing rate. Whereas online sales made up 10,2% of the total retail sales in 2017, it is expected to reach 17,5% by 2021 (eMarketer, 2018). The recent ‘Singles Day’ record ($38 billion) of the Alibaba Group is a clear example of the magnitude of e-commerce nowadays (Singh, 2019).
Despite its similarities, online and offline (i.e., physical) shopping environments are also quite different. The absence of a physical store environment, and thus, no possibility to physically touch, try on or use products before purchasing it, is perhaps the biggest difference between the two (JDA, 2015). In the physical stores of retailers, customers can inspect the product of interest before purchase. The absence of it in online retail, however, may lead to fit uncertainty, a state in which a customer does not yet know if the ordered product will be satisfactory or if it will meet his or her expectations (Abdulla, Ketzenberg, & Abbey, 2019). As a result, product returns are substantially higher for online retailers compared to brick-and-mortar retailers (Dennis, 2018). An estimated 30% of the products online are being returned, which is almost three times higher than the estimated returns for brick-and-mortar stores (Saleh, 2016).
10 Most online retailers know how marketing-related decisions, such as pricing and promotions, influence the sales of their products. As an example, discounts generally have a positive effect on sales (Guerreiro, dos Santos, da Silveira Gisbrecht, & Ong, 2004; Zuo & Iida, 2017). However, these same decisions (e.g., pricing and promotions) may also affect product returns, which in turn heavily influence the profits of an online retailer. Online retailers usually sell highly diverse products, and it is likely that the effects of pricing and discounts will differ for different types of products. One way to classify products is by shopping value. According to Babin, Darden and Griffin (1994), products can have either utilitarian or hedonic shopping values. Although these ‘values’ are not mutually exclusive (Batra & Athola, 1991), some products can be classified as mostly having mainly utilitarian values or mainly hedonic values. Hedonic products tend to be more luxurious and fun (Kivetz & Zheng, 2017), whereas utilitarian products have a more necessary purpose (Hirschman & Holbrook, 1982) and serve a more practical goal. Insights in how certain variables influence product returns and how this varies between hedonic and utilitarian products are thus of great value for an online retailer. With such insights, online retailers have a richer picture of how their marketing-mix decisions may influence the profitability of their firm.
As such, the aim of this thesis is to discover how price and promotions impact product returns for online retailers and how this may differ between utilitarian and hedonic product categories.
1.1 Research questions
The main research question of this study is: How do the effects of price and promotions on product returns differ across utilitarian and hedonic products at an online retailer? Knowing which variables impact the product returns of different product categories (and in what way) is useful for an online retailer when making future marketing-related decisions.
11 To help answer the main research question, several sub-questions have been formulated. These sub-questions are:
a) What are product returns?
b) What is the effect of price on product returns?
c) How does whether there is a promotion or not affect product returns? 1
d) How does discount depth affect product returns? e) What are utilitarian and hedonic products?
f) How do price, promotions, and discount effects vary between utilitarian and hedonic products?
1.2 Relevance of the study
As mentioned in section 1.1, the vast body of literature on product returns is growing and is relevant from multiple domains, from logistics to marketing. However, little has been written about the effects of price or the different effects for utilitarian and hedonic products. This paper aims to be a first step towards research that investigates product returns from a product-perspective.
As described in the introduction of this chapter, product returns have a big impact on the profitability of online retailers. Online retailers should, therefore, try not only to increase sales but also try to decrease the number of product returns. This thesis focuses on how price and promotions influence the returns of different types of product categories. By doing so, managers become more aware of the consequences of their marketing-mix decisions by relating these not only to sales but also to product returns. The managerial relevance of this thesis lies in the clearer picture and awareness of what drives product returns.
1.3 Research structure
The outline of this thesis is as follows. First, a theoretical framework provides an overview of relevant theories for the current study, followed by proposed relationships between variables. Next, the data collection procedure and methodology are described. The fifth chapter
1 It should be noted that sub-question c looks at whether a product was on promotion or not. Sub-question d
13
2. THEORETICAL FRAMEWORK
This chapter provides an overview of relevant scientific theories regarding pricing, promotions, and product returns. It highlights the rationale and theory behind the included variables and includes any definitions and expressions that may need clarification. First, the conceptual model, including all the variables and expected effects, are shown. Then, the rationale and details will be explained based on relevant theories. Each paragraph will end with a hypothesis.
2.1 Conceptual model
The conceptual model below gives a graphical representation of the hypotheses discussed later in this chapter. It proposes the relations between the independent variables and the dependent variable, and how this may be moderated by whether products are utilitarian versus hedonic. A set of control variables that also influences product returns is included in the model but will not be tested for.
14 2.2 Literature review
This section describes the relevant literature on product returns for the current study. The scope of this literature review is limited to the effects that are of interest, which are mentioned in the introduction part of this thesis.
2.2.1 A definition of product returns
In this thesis, product returns are studied in the context of online retailers (i.e., retailers without physical stores). This study specifically looks at how certain independent variables impact the probability that a product will be returned and how this may vary between different product categories. According to Guide, Souze, van Wassenhove and Blackburn (2006), a product return occurs “when a consumer sends a product back to the retailer, which can be for any reason”. It should be mentioned that returning a product in this context is different from a traditional brick-and-mortar store, where consumers are able to return a product themselves (e.g., by bringing the products back to the physical store).
According to Kumar et al. (2010), the decision to return a product is one of the ways customers can bring value to the firm. As such, product returns are stated to have a direct financial impact on a firm (Minnema, Bijmolt, Petersen, & Shulman, 2018). Furthermore, according to Minnema et al. (2018), customer product return behavior should thus, always be included in customer value management. Next to lost sales, product returns lead to several extra costs, which are most often paid for by the retailer (Guide et al., 2006).
2.2.2 Effects of price on product returns
Price is the amount of money charged for a product and service (Kotler, 2017). It has a big, if not the biggest, effect on a consumer’s purchase decision and, as such, is a very important marketing-mix tool for any retailer (Kotler, 2017).
15 1991; Zeithaml, Parasuraman, & Berry, 1990). It is argued that as the price of a product increases, expectations go up, and the consumer becomes less accepting of a product not living up to the expectations (Hess & Mayhew, 1997). According to Anderson et al. (2009), lower prices lead to a consumer surplus (i.e., increasing the perceived value), which in turn leads to reduced return probability. Hess & Mayhew (1997) found a positive relationship between price and the return rate. As the price went up, return rates increased (Hess & Mayhew, 1997). Adversely, a cheaper product may decrease the expectations that the consumer has. If the product disappoints upon arrival, a consumer is less quickly inclined to return when the product is lower-priced. In this case, the utility of not returning may be higher than that of returning. Several studies have shown that higher-priced products, compared to lower-priced products, result in a higher number of product returns (Anderson et al., 2009; Hess & Mayhew, 1997). Similarly, a study by Anderson et al. (2009) showed that, when all else was equal, lower-priced products had lower return rates compared to higher-priced products.
H1: Price has a positive effect on the product return probability.
2.2.3 Effects of promotions and discounts
A retailer must make many decisions within a marketing strategy. An important element within such strategies is price promotions (Mulhern & Padgett, 1995). Price promotions are temporary price changes, meaning that a product is sold at a lower price for a limited amount of time (Pauwels, Hanssens, & Siddarth, 2002). These temporary price-drops are known to be an important driver of demand for products in many markets, including consumer goods. Products at the focal firm are sometimes offered at a discounted price instead of the original price, in which case both prices (i.e., the original and the new price), as well as the discount percentage (e.g., 30%) are shown on the web page.
16 returning a product at a regular price may actually lead to a subjective gain. A study by Asdecker, Karl and Sucky (2017) shows a negative relationship between the relative value of a coupon and the return rate. In their study, as discounts went up, the return rates went down. It should be noted that in their research, customers used a coupon to get a discount instead of the discount being given through a web page. Their findings, however, still indicate the possible effect of the size of discounts on return rates. Research by Petersen and Kumar (2009) shows that price promotions affect the likeliness of customers returning a product. They state that customers become less critical (due to lower expectations) if a product is less expensive, leading to a lower return rate for products on sale.
H2: Products purchased on promotion are less likely to be returned than products purchased
at the regular price.
H3: There is a negative relationship between product discount depth and product return
probability.
2.2.4 The difference between hedonic and utilitarian motivations and products
17 Batra and Athola (1991) state that these two types of shopping motivations can be inclusive, meaning that one product can serve both utilitarian and hedonic purposes (e.g., toothpaste prevents cavities while at the same time it may provide enjoyment from its taste). Similarly, Dhar and Wertenbroch (2000) reason that a new car may be chosen for its utilitarian features (e.g., engine specifications), but also its hedonic features (e.g., car brand).
While describing its mutual exclusiveness, Batra and Athola (1991) also state that some product categories can be perceived to be predominantly either one or the other. Although many products comprise both ‘values’, customers generally characterize some products as mainly hedonic or mainly utilitarian (Dahr & Wertenbroch, 2000). Hedonic products are typically described as involving the senses or being luxurious (Kivetz & Zheng, 2017), such as tablets and perfume. Utilitarian products are typically described as more practical or necessary products that consumers are more certain of (Hirschman & Holbrook, 1982), such as socks, shavers, and vacuum cleaners. The aforementioned theory has led to the following hypothesis:
H4: Hedonic products are more likely to be returned compared to utilitarian products.
2.2.5 Effect of price for hedonic and utilitarian products
The effect of price on product returns may be different for utilitarian and hedonic products. With utilitarian products being more task-related (Batra & Athola, 1991), it could be argued that customers, in general, have a clearer purpose for that type of product (e.g., they know exactly why they want it and how they will use it). Thus, a price difference may not have as big of an impact on the return probability, compared to hedonic purchases, since the chances of customers keeping it are higher. The aforementioned findings and theory have resulted in the following hypothesis:
H5: The effect of price on product returns is smaller for utilitarian products compared to
hedonic products.
2.2.6 Effects of promotion and discounts for hedonic and utilitarian products
18 According to Kivetz and Zheng (2017), promotions then help to justify such hedonic purchases. Although their reasoning relates to sales, it could also translate to product returns. This would mean that people might be less likely to return hedonic products because of a promotion, compared to the effect of utilitarian products not changing much. The aforementioned findings and theory have resulted in the following hypotheses:
H6a: The effect of promotions on product returns is stronger for hedonic products compared
to utilitarian products.
H6b: The effect of discount depth on product returns is stronger for hedonic products compared
to utilitarian products.
2.2.5 Control variables
In this section, several control variables that may influence the proposed hypotheses are discussed. These control variables will not be interpreted as independent variables but may still have an effect on the results and thus will be controlled for and included in the model.
The control variables are gender, age_group, rel_length_days, quarter, and verkoop_aantal2.
Firstly, gender differences exist in online shopping behavior (Liu, Lin, Lee, & Deng, 2013). Women tend to return more products than men. Second, online shopping behavior may change depending on the age of consumers (Hervé & Mullet, 2009). Third, the customer relationship in years may influence product returns. Furthermore, consumer spending may depend on the time of year. An example of this is that spending, and thus returning, tends to increase towards the end of the year (Deloitte, 2017). Also, Petersen and Kumar (2009) found that products purchased during the end of the year are more likely to be returned than purchases during the rest of the year. Lastly, the number of purchased products (verkoop_aantal) may also influence return behavior due to a sense of certainty. When a customer buys, for example, five items of the same product, it is assumed that this customer is more certain about the purchase due to the higher volume.
19
3. DATA COLLECTION
This chapter explains the data collection procedure for this study. First, it addresses the research and the setting in which it takes place. Then, the data preparation is described while giving a description of the dependent, independent, and control variables of this study. When needed, the variable description includes an explanation of how variables have been operationalized for the analytical purpose of this study. Lastly, the data is explored, which includes the description of missing values, outliers, and descriptive statistics.
3.1 European online retail environment
The current research focuses on product return behavior in the online retail environment. The market in which the focal European online retailer operates consists of firms that sell physical goods, through a digital channel, to a private end-user. This is generally referred to as the business-to-consumer market (De Best, 2019).
The population of this study consists of consumers that shop at online retailers. Online shopping is highly popular in Europe. In 2017, 60% of Europeans stated they used e-commerce for shopping for goods and services at least once a month (Ecommerce News, 2019). This figure may grow in the future, since the European online retail market has grown over the last decade, with a total of over 600 billion euros turnover in 2019 (Ecommerce News, 2019).
Regarding product returns, the European law states that consumers have a ‘right of withdrawal’ of 14 days, meaning they can return a product within 14 days after they have received it (Konsument Europa, 2016). However, the focal firm allows customers to return a product for free within 30 days, which is in line with its competitors.
3.2 Data preparation 3.2.1 Data selection
20 data consists of the order information from 165,482 customers, ranging from January 1, 2017 to December 31, 2017.
As the aim of this thesis is to not only look at the direct effects of independent variables but also at how this may be different for utilitarian and hedonic product categories, the dataset contains a selection of six product categories that is gathered from a larger dataset. The six product categories are chosen based on how well they represent either utilitarian or hedonic products (Batra & Athola, 1991).
The selected ‘utilitarian’ categories are the vacuum cleaners, shavers and hanging lamps product categories. The selected ‘hedonic’ categories are the handbags, tablets, and audio & hifi product categories.
The dataset has not been supplemented with external data because the main aim of this study is to look at the effects of decisions made by the focal firm. However, that does not mean that no other factors also play a role in product return decisions. That is outside of the scope of this study.
3.2.2 Data cleaning
First, the original dataset is cleaned by deleting all irrelevant variables and variables that contain redundant information. This includes variables with only NA values, only a single value (i.e., “1”) and descriptive codes that are of no use for the analytical purpose of this study. Examples of such variables are those that indicate where products are located in the stockroom of the retailer or variables that show the number of days since a date in the past. This last variable is redundant since other variables in the dataset contain the same information by providing an actual date.
21 have been deleted from the dataset. These observations were labeled as “BEST”, “ANEZ”, “DIST” and “AFSC”.
Close to 5000 observations in the original dataset showed near-duplicate rows for the same purchase, where only the type of discount was named different. In these cases, a second row (with all the same values, except for the type of discount) meant that not only the product of interest was on discount, but that an additional discount was given (e.g., a discount related to the overall basket). Given that the dataset contains a subset of observations for selected product categories, it is impossible to tell why an additional discount was given. Thus, the ‘duplicate’ rows have been deleted while summing the values of both types of discount, leaving a clear picture of the actual discount a customer received for each specific order.
3.3 Variable description
This section provides an overview of the variables that remain in the dataset and that are used for analytical purposes. This section describes how the variables in the conceptual model (see Figure 1) are operationalized and have been prepared for analytical use.
Product return. The cleaned dataset includes a dummy variable ‘return’ to indicate whether a product has been returned or not. This binary decision is denoted with a 1 when a product has been returned, and with a 0 if a product has not been returned (i.e., the customer has kept the product). A product was considered as returned when a return date was present for the observation.
Price. The numeric variable ‘klant_verkoopprijs_bedrag’3 indicates the price of a product in
euros. The amount includes a possible discount and is thus considered to be the actual price paid for a product.
Promotion. The variable dum_promotion is a dummy variable that indicates whether a product was on promotion at the moment of purchase. A value of 1 refers to the product being on promotion, whereas a 0 indicates that the product was not on promotion at the time of purchase. Promotion refers to a temporary price-drop and does not distinguish between the type of
22 promotion, but only if a product was on promotion. The threshold for a promotion is set at 10% (0.10 in the dataset).
Discount depth. The variable discount_depth is a numeric variable that indicates the product discount at the time of purchase. The value reflects a percentage discount (i.e., 20 = 20% discount).
Gender. The variable gender indicates the gender of the customer. In this case, a 1 refers to the customer being female and a 0 refers to the customer being male.
Age. The categorical variable age group4 indicates to which age group a customer belongs.
This variable consists of seven categories: up to 25, 26-35, 36-45, 46-55, 56-65 years, over 65, and unknown.
Relationship length. The numeric variable relation_length reflects the relationship between the customer and the retailer in the number of days. A relationship starts at the date that the first purchase is made by a customer. This variable is created by calculating the difference in time between the first purchase ever and the date of an order.
Seasonality. As mentioned in paragraph 2.2.5, seasonality may have an effect on product returns. Therefore, to control for this, the dummy variable quarter has been created to indicate in which quarter a purchase was made. The beginning and end dates for each quarter are as follows: Q1 ranges from January 1st – March 31st; Q2 ranges from April 1st – June 30th; Q3
ranges from July 1st – September 30th; and Q4 ranges from October 1st – December 31st.
Basket size. The numeric variable verkoop_aantal shows the sum of the number of the same products (i.e., similar article ID) in one order. For example: if order i contains three identical lamps, the value is 3. This variable does not technically represent basket size since the actual basket size cannot be inferred from the provided dataset. However, the variable may still explain some of the variance and is thus included in the dataset and models.
23 Utilitarian / Hedonic. The dummy variable dum_hedon_util indicates whether a product is either utilitarian (i.e., the product categories vacuum cleaners, shavers, and hanging lamps are labeled as 0) or hedonic (i.e., the product categories handbags, tablets and audio & hifi are labeled as 1).
Table 1. Variable description
Variable Type Description Type
Return DV Whether a product is returned in time t
Binary (1 = Yes; 0 = No) Price IV The price of a product in euros Numeric
Promotion IV Whether a product is on promotion Binary (1 = Yes; 0 = No) Discount_depth IV Given discount in percentages Numeric
Gender C The gender of the customer Binary (1 = Female; 0 = Male)
Age_group C The age group to which the customer belongs
Categorical (0-25; 26-35; 36-45; 46-55; 56-65; 65+; Unknown)
Relation_length C The length of the relationship in days since the first purchase
Numeric Quarter C The quarter in which an order is
placed
Categorical (1, 2, 3, 4) Verkoop_aantal C The sum of the number of identical
products in one order
Numeric Dum_hedon_util Whether a product is hedonic or
utilitarian
Binary (1 = Yes; 0 = No)
3.4 Data exploration
This section looks into the data that is used for this study. It describes how things such as missing data and outliers are treated and ends with some descriptive statistics of the included variables.
3.4.1 Missing data
24 The variable rel_length_days showed two rows with missing values. Upon closer examination, the rows had a missing value for the variable eerste_bestelling_datum (i.e., date of the first-ever purchase made by a customer). Hence, rel_length_days was unable to calculate the difference between the first purchase date and the order date. These two observations are removed from the dataset.
Furthermore, two other observations in the variable relation_length showed negative values of -18 and -40 days. As described in section 3.3, the variable relation_length calculates the difference between the first purchase ever made at the retailer and an order date. A value of 0 is likely since it means that the order is simply a customers’ first ever purchase at the retailer. However, a negative difference between the two variables is impossible, considering how the variables are set up. Therefore, the two observations have also been removed from the dataset.
The variable korting_bedrag showed missing values for 58,548 observations. A missing value indicates that no discount was given on the product. Furthermore, 425 observations have a value of zero, indicating that a discount was not given. The NA’s have been set equal to 0, resulting in a total of 58,973 observations were no discount was given. The variable discount_depth is calculated based on korting_bedrag and thus showed the same missing values. After setting the NA’s of korting_bedrag to 0, discount_depth does not show any missing values anymore.
3.4.2 Outliers
Outliers are unusual observations that deviate quite heavily from the rest of the observations (Wickham & Grolemund, 2017). Just like missing data, outliers can influence the results of analyses. Therefore, it is important to identify them and, if deemed necessary, handle them appropriately. An outlier is defined as an observation that is 1.5 times outside the interquartile range above the upper quartile and below the lower quartile (Leeflang et al., 2017). Outliers have been detected for the following variables:
25 the overall tablet category. Similarly, the shaving product category includes products such as accessories and nose trimmers, which are relatively cheap compared to a shaver. Due to the plausible nature of the outliers, they will remain present in the dataset.
Discount depth. The variable discount_depth shows several outliers per product category, except for shaving and audio (no outliers found). Although these observations are technically considered to be outliers, most of them are close to the third whisker of the boxplot. Table 2 shows the percentage of outliers found per product category and, if deleted, how much this would influence the overall mean of the product category. Based on the (lack of) difference between columns 3 and 4, the outliers are not deleted and will remain present in the dataset.
Table 2. Outliers per product category5
Product category % of outliers Mean before deleting Mean after deleting
Tablets 0.008 0.092 0.092
Handbags 0.026 0.138 0.138
Audio & hifi Not applicable Not applicable Not applicable
Vacuum cleaners 0.100 0.257 0.256
Hanging lamps 0.926 0.151 0.146
Shavers Not applicable Not applicable Not applicable
3.4.3 Descriptive statistics
After an exploration of the dataset, the current paragraph gives a descriptive overview, which includes summaries and measures of the relevant variables in the dataset. It provides a basis for inferential statistics in the next chapters.
The dataset contains orders from January 1, 2017 - December 31, 2017. Given this timeframe and the selected product categories, a total of 165,482 unique customers have ordered 3,106 unique products.
Table 3. Return rate and promotion rate in percentages per product category
Product categories
Handbags Tablets Audio & hifi Vacuum cleaners Shavers Hanging lamps Return rate 46,50 7,83 13,73 6,55 5,29 74,93 Promotion rate 39,00 38,65 63,64 94,80 53,33 52,11
26 The first row of Table 3 shows the return rates of each product category. Shavers are with 5,29% the least returned product category, and hanging lamps are returned the most with a return rate of 74,93%. As they are part of a larger interior or decoration, hanging lamps seem like a product that can quite easily disappoint or not be what a customer expected (depends on aesthetics). The return rate of handbags is 46,50%, audio & hifi is 13,73%, tablets is 7,83% and lastly, vacuum cleaners is 6,55%. Overall, the return rates for each product category seem plausible.
Table 4. Descriptive statistics of price and discount depth per product category
Product category Min Max Mean Median SD
Handbags Price 7,50 311,95 49,8 35,99 36,87 Discount depth 0 83,33 13,83 0 19,02 Tablets Price 39,99 3199,00 287,30 219,00 218,75 Discount depth 0 41,11 8,68 6,58 8,44
Audio & hifi
Price 9,03 1499,00 206,90 157,00 183,56 Discount depth 0 64,27 18,08 17,36 14,84 Vacuum cleaners Price 12,99 999,00 149,23 119,00 103,80 Discount depth 0 62,94 25,65 25,09 10,53 Shavers Price 4,99 429,99 50,54 39,99 40,29 Discount depth 0 59,00 13,59 11,11 13,45 Hanging lamps Price 5,56 599,00 64,87 49,95 51,01 Discount depth 0 75,13 14,89 16,05 16,45
Table 4 shows descriptive statistics of the variable price and discount_depth for each product category. As indicated by the rows price, the overall prices range from €4,99 (shavers) up till €3199 (tablets). The price ranges per product category are quite wide, although it does make sense considering the fact that some product categories include products that start at around 5 euros (e.g., a tote bag or a very basic hanging lamp). The mean price is lowest for handbags (€49,80) and highest for tablets (€287,30).
27 The average relationship length is 2199 days (or six years), with a minimum of 0 days and a maximum of 7784 days (or twenty-three years).
Table 5 (see next page) summarizes some descriptive statistics for the relevant categorical variables in the dataset. It should be noted that the frequencies in Table 4 are grouped by unique customers, and the products they have ordered.
As indicated in the first row of Table 5, a large majority of the ordered products (85,35%) is not returned. Most customers are female (63,92%) and between the ages of 36 and 45 years old. The ordered products of the six product categories seem to be equally spread throughout the year, with a slight decrease in the fourth quarter of 2017 (29,1% of all ordered products). Differences can be observed in the number of ordered products per product category. Handbags (14,98%), tablets (16,40%) and shavers (19,46%) are quite similar, whereas audio & hifi (7,70%) and hanging lamps (8,59%) are ordered less often. The largest product category is vacuum cleaners, which takes up 32,87% of the total dataset.
Table 5. Frequencies of the categorical variables over the total dataset
Variable N (counts) % Return Return 29.807 14,65 No return 173.715 85,35 Gender Female 130.101 63,92 Male 73.421 36,08 Age group 25 and younger 10.780 5,30 26 – 35 years 46.320 22,76 36 – 45 years 52.179 25,54 46 – 55 years 48.969 24,06 56 - 65 27.905 13,71 65 and older 17.211 8,45 Unknown6 158 0,08 Quarter 1st quarter 47.480 23,33 2nd quarter 49.298 24,22 3rd quarter 47.521 23,35 4th quarter 59.223 29,10 Product category 1: Handbags 30.494 14,98 2: Tablets 33.378 16,40
3: Audio & hifi 15.670 7,70 4: Vacuum cleaners 66.903 32,87
29
4. METHODOLOGY
This chapter discusses the methodological part of the study. First, an appropriate model is chosen, after which its assumptions are discussed. Furthermore, this chapter addresses model specification, and shows the proposed equations of each model.
4.1 Model estimation
To be able to analyze the effects of the aforementioned independent variables on the dependent variable, the right model needs to be chosen. The aim of this thesis is to describe and understand the effect on product returns, which is a binary decision (i.e., a product is either returned or not). A commonly used approach for modeling a binary decision within a marketing setting is a binary logistic regression model (Fok, 2017.). Given the context of this study, a binary logistic regression is deemed appropriate.
An advantage of logistic regression models is the mathematical convenience. Such a model calculates probabilities, which are easy to calculate. Furthermore, the interpretation of the parameters is ‘easier’. Interpretation can be done by looking at the coefficients, the odd ratios, or at the marginal effects.
The assumption of a binary logistic regression model is that an unobserved, latent variable drives the product return decision of an individual (Leeflang et al., 2016). Only the outcome of the decision that a customer made is observable, not the probability. The decision outcome looks as follows:
30 In the context of this study, a product will be returned when the utility is higher than the utility of not returning. Thus, the utility variable and the decision are linked as follows:
The error term is independently distributed according to the cumulative distribution function (CDF).
4.2 Model assumptions
Logistic regression does not require many of the assumptions other types of models have, such as normality and homoscedasticity (Leeflang et al., 2016). However, some other assumptions still apply to (binary) logistic regressions. This section addresses these assumptions and whether they hold.
First, in a binary logistic regression, the dependent variable should be of a binary nature. The dependent variable return takes only two values (0 and 1) and, thus, this assumption holds. Second, observations should be independent from each other. This means there should be no repeated measures (i.e., multiple measures from the same variable), which also holds.
31 kept in. Lastly, model 3 showed signs of multicollinearity with VIF scores > 10, which was caused by the interaction between price and discount depth. Instead of estimating one model 3 with all interactions, three separate models are estimated. The equations of the original and adjusted versions of model 3 are presented in the next section.
4.3 Model specification
In order to answer the sub-questions and the proposed hypotheses, three binary logistic regression models are used. This section explains the different models that are used for analysis. It starts by showing a full model of all variables. It is followed by three smaller models, each of which will be used to answer a share of the proposed hypotheses.
The (fully) conceptualized model described in the second chapter is translated into the following equation: 𝑈" = 𝛼 + 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛" + 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ" + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒"+ 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟" + 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" + 𝛽J(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙"∗ 𝑃𝑟𝑖𝑐𝑒") + 𝛽)N(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" ∗ 𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛") + 𝛽))(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ") + 𝜀" With:
𝑈" = The utility that product i is returned in period t
α = The intercept
𝛽)−𝛽)) = Coefficients of the explanatory variable 𝑃𝑟𝑖𝑐𝑒" = The price of product i in euros
𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛" = Dummy for whether product i was on promotion 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ" = The percentage discount of product i
𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" = The number of days between the first and current purchase of customer i
𝐺𝑒𝑛𝑑𝑒𝑟" = Dummy variable for whether customer i is female
𝐴𝑔𝑒" = Age group to which customer i belongs
𝑄𝑢𝑎𝑟𝑡𝑒𝑟" = The yearly quarter in which product i is ordered 𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" = Sum of product i in order i
𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" = Dummy variable for whether product i is hedonic
32 4.3.1 Model 1 (only direct effects)
As mentioned in section 3.2.1, the dataset includes six product categories. Model 1 looks at the direct effects only per product category. This model is estimated six times, one for each product category. Differences in these estimations may be indicative that there are indeed differences in the effects of variables between utilitarian and hedonic product categories, which will be tested in model 2 and 3. The proposed equation of model 1 looks as follows:
𝑈" = 𝛼 + 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛" + 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ" + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒"+ 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟" + 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" + 𝜀"
4.3.2 Model 2 (including dummy variable)
After looking at the direct effects of the independent variables in model 1, the aim of model 2 is to look at whether products being either hedonic or utilitarian in itself influences product returns. Instead of estimating a model for each product category separately, as is done in model 1, model 2 uses the data of all six product categories at the same time but now includes dum_hedon_util. Thus, the model is estimated only once. Although the equation of model 2 does include all independent and control variables, only the dummy variable dum_hedon_util is of interest. The signs of price, promotion, and discount depth may change compared to the separate estimates for model 1, because of differences between the product categories (e.g., products are not all and always similar in regards of price range and other variables). However, these estimates are not of interest in this model. The proposed equation of model 2 looks as follows:
𝑈" = 𝛼 + 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛" + 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡𝑑𝑒𝑝𝑡ℎ" + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟"+ 𝛽C𝐴𝑔𝑒" + 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟" + 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙"
+ 𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" + 𝜀"
4.3.3 Model 3 (including interactions)
33 The original equation of model 3 looks as follows:
𝑈" = 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛" + 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒"+ 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟" + 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙"
+ 𝛽J(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" ∗ 𝑃𝑟𝑖𝑐𝑒") + 𝛽)N(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙"
∗ 𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛") + 𝛽))(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙" ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ") + 𝜀" A check for multicollinearity is performed by means of checking VIF scores. The output shows there is a slight multicollinearity issue between dum_hedon_util and its interaction with discount_depth. Therefore, instead of estimating one model with all interactions, three versions of model 3 are estimated, each containing one interaction - first, a model with the price interaction. Second, a model with the promotion interaction. Third, a model with the discount depth interaction. The adjusted model 3 equations look as follows:
𝑈" = 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛"+ 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒" + 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟"+ 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" + 𝛽J(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙"∗ 𝑃𝑟𝑖𝑐𝑒") + 𝜀" 𝑈" = 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛"+ 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒" + 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟"+ 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" + 𝛽J(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙"∗ 𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛") + 𝜀" 𝑈" = 𝛽)𝑃𝑟𝑖𝑐𝑒" + 𝛽/𝐷𝑢𝑚_𝑝𝑟𝑜𝑚𝑜𝑡𝑖𝑜𝑛"+ 𝛽8𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ + 𝛽<𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛_𝑙𝑒𝑛𝑔𝑡ℎ" + 𝛽A𝐺𝑒𝑛𝑑𝑒𝑟" + 𝛽C𝐴𝑔𝑒" + 𝛽E𝑄𝑢𝑎𝑟𝑡𝑒𝑟"+ 𝛽G𝑉𝑒𝑟𝑘𝑜𝑜𝑝_𝑎𝑎𝑛𝑡𝑎𝑙" + 𝛽J(𝐷𝑢𝑚_ℎ𝑒𝑑𝑜𝑛_𝑢𝑡𝑖𝑙"∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡_𝑑𝑒𝑝𝑡ℎ") + 𝜀"
34
5. RESULTS
This chapter describes the relevant outcomes of the performed analyses. It identifies the relationships between the aforementioned independent variables and the dependent variable product returns in the online retail industry. The estimation outcomes of the models are presented and discussed by means of tables and text.
5.1 Results model 1
This section describes the results of model 1. It discusses the effects of the independent variables on product returns for each of the six product categories. The impact of the independent variables is assessed using odds ratios (as mentioned in section 4.1). Significant results will be interpreted, while insignificant results will be commented on, as to underpin possible category differences. To check whether the models for each product category outperform a null-model, Likelihood-Ratio tests are performed (see Appendix, Figures a - f). This test shows that all six models significantly (p < 0.001) outperform the null model without any explanatory variables. The interpretation of the estimates for each product category can be found in section 5.1.
Table 5. Results of model 1, estimated per product category
35 Age (46-55) 0.241*** (0.058) -0.126 (0.137) 0.051 (0.132) -0.138 (0.109) -0.078 (0.132) 0.102 (0.130) Age (56-65) 0.098 (0.067) -0.190 (0.144) 0.169 (0.137) -0.146 (0.113) -0.090 (0.138) 0.118 (0.139) Age (65+) 0.116 (0.089) 0.218 (0.143) 0.299** (0.143) 0.067 (0.116) -0.037 (0.144) 0.194 (0.167) Age (Unknown) -0.127 (0.427) -11.123 (111.146) 0.359 (1.105) -10.946 (102.978) -0.337 (1.027) 2.262*** (0.832) 2nd quarter 0.133*** (0.040) -0.077 (0.070) -0.166** (0.078) -0.122** (0.052) -0.025 (0.078) -0.120* (0.062) 3rd quarter 0.239*** (0.038) 0.034 (0.070) -0.073 (0.080) -0.051 (0.052) 0.085 (0.081) 0.098 (0.060) 4th quarter 0.050 (0.038) 0.075 (0.066) 0.052 (0.072) 0.044 (0.050) 0.069 (0.081) 0.056 (0.056) Verkoop_aantal 0.146 (0.155) -0.241*** (0.074) -0.471*** (0.121) 0.191* (0.104) 0.195 (0.148) -0.036 (0.037) *0.05; **0.01; ***0.001 5.1.1 Effect of price
As can be seen in Table 5, four out of the six product categories show significant results for price, while one product category is moderately significant, and one is not significant at all. For the price variable, a one-unit increase is 0,01 euros. For the purpose of interpretability, the results in this section are multiplied by 100, so that a one-unit increase is a 1 euro increase.
Handbags. The negative coefficient (p < 0.001) shows that an increase in price decreases the probability that a return is observed, although the effect is small. The odds ratio equals 0.9966, which indicates there is a slight negative relationship between price and return probability. It can be interpreted as that a one-unit increase in price, decreases the odds of the return probability by 3,4%.
Tablets. The coefficient is not significant and is therefore not interpreted.
36 Vacuum cleaners. The positive coefficient is moderately significant (p < 0.05) and shows that an increase in price slightly increases the return probability. The odds ratio equals 1.0003, which means that a one-unit increase in price increases the odds of the return probability by 0,3%.
Shavers. The positive coefficient (p < 0.001) shows that an increase in price increases the return probability. The odds ratio equals 1.0041, which means that a one-unit increase in price, increase the odds of the return probability by 4,1%.
Hanging lamps. The positive coefficient (p < 0.001) means that an increase in price slightly increases the return probability. The odds ratio equals 1.0019, which means that a one-unit increase in price, increases the odds of the return probability by 1,9%.
5.1.2 Effect of promotions
As can be seen in Table 5, the effect of whether a product is on promotion (dum_promotion) has significant effects for three of the six product categories. Promotions do not have a significant effect on the return probability of handbags, shavers, and hanging lamps.
Tablets. The positive coefficient (p < 0.01) shows that if tablets products are on promotion compared to not being on promotion, the return probability increases. The odds ratio equals 1.2263, which can be interpreted as that when a tablet product is on promotion, the odds of the return probability increase by 22,63%.
Audio. The positive coefficient (p < 0.01) shows that if audio products are on promotion compared to when they are not on promotion, the return probability increases. The odds ratio equals 1.2905, which can be interpreted as that when an audio product is on promotion, the odds of the return probability increase by 29,05%.
37 Interestingly, differences can be observed in the direction of estimates between two product categories (see Table 5). Whereas audio & hifi and tablets show a significant and positive relationship with return probability (p < .01), the estimate for vacuum cleaners is almost exactly the opposite (p < .01).
5.1.3 Effects of discount depth
The fourth row of Table 5 shows the estimates for the effect of discount depth on return probability per product category. The variable shows significant and negative estimates for four out of six product categories. For each product category, discount depth seems to have a negative relation with return probability. For this variable, one unit equals 1%. However, for interpretability purposes, the results in this section are multiplied by 10 so that the variable is interpreted in steps of 10%.
Handbags. The negative coefficient (p < 0.001) shows that an increase in discount decreases the return probability. The odds ratio equals 0.9920, which can be interpreted as that when the discount for a product in the handbag category increases by ten units, the odds of the return probability decrease with 8%.
Tablets. The negative coefficient (p < 0.001) shows that for tablets products, an increase in discount decreases the return probability. The odds ratio equals 0.9792, which means that when the discount for tablets increases with ten units, the odds of the return probability decrease with 21%.
Audio & hifi. The negative coefficient (p < 0.001) shows that also for audio products, an increase in discount decreases the return probability. The odds ratio equals 0.9881, which means that when the discount for audio increases with ten units, the odds of the return probability decrease with 12%.
38 5.2 Results model 2
As mentioned in section 4.3.2, the aim of model 2 is to check whether there is a difference in return probability between hedonic and utilitarian product categories when the data are taken together, instead of a separate estimation for each product category. Since the results for each separate product category are discussed in the previous section, only the dummy variable that labels whether a product category is hedonic or utilitarian (dum_hedon_util) is interpreted.
Table 6. Results of model 2
Coefficient Std. Error Odds ratio
Intercept -2.379*** 0.0476 0.0926 Price -0.0033*** 0.0001 0.9968 Dum_promotion -0.4688*** 0.0289 0.6258 Discount_depth 0.0049*** 0.0009 1.0049 Dum_hedon_util 1.225*** 0.0163 3.4042 Relation_length 0.0001*** 0.0001 1.0001 Gender 0.5517*** 0.0182 1.7362 Age (26-35) 0.1508*** 0.0356 1.1628 Age (36-45) 0.05355 0.0364 1.0550 Age (46-55) -0.0003 0.0375 0.9997 Age (56-65) -0.1304** 0.0409 0.8777 Age (65+) -0.109* 0.0449 0.8967 Age (Unknown) 0.01469 0.2703 1.0148 2nd quarter -0.0344 0.0222 0.9662 3rd quarter 0.0655** 0.0219 1.0677 4th quarter 0.0622** 0.0212 1.0642 Verkoop_aantal -0.0108 0.0249 1.0109 *0.05; **0.01; ***0.001
A Likelihood-Ratio test is performed to check whether model 2 outperforms a null-model (see Appendix 1, Figure g). The test is significant, which means that the model is significantly better than the null-model (p < .001) and can be interpreted.
5.2.1 Main effect of hedonic and utilitarian products
39 observed products and has a return rate of almost 47% (see Table 3, section 3.4.3). Furthermore, the two utilitarian product categories, vacuum cleaners and shavers together, represent 32,87% and 19,46% of the observed products while both having very low return rates (6,55% and 5,29%, respectively). As mentioned in section 3.4.3, the return rate for hanging lamps is the largest, however, it represents only a small part of the total observations (8,59%). To check if the abovementioned results of model 2 are influenced by the inclusion of hanging lamps, a second version of the model is run without this product category (see Appendix 1, Figure h). Overall, the coefficients and significance levels do not change much. The estimate of dum_hedon_util, however, does become stronger, underlining the significant main effect for the return probability between hedonic and utilitarian products.
5.3 Results model 3
As described in section 4.3.3, model 3 is the fullest version of the models as its aim is to discover whether interaction effects are present between hedonic versus utilitarian products and the independent variables price, promotion and discount depth. The focus of this section is mainly on whether the effect of the interactions become stronger or weaker compared to the main effects. Its focus is less on the direction of the estimates, as these are tested in model 1 in section 5.1 for each product category individually.
For the output of these models, see Appendix 1 (Figures i-k). A Likelihood-Ratio test is performed for each model 3 version to check whether they outperform a null-model (see Appendix 1, figures l-n). The tests are significant, which means that the models significantly outperform a null-model (p < .001) and can be interpreted.
40 Promotion. The interaction between dum_hedon_util and dum_promotion shows an insignificant coefficient. It seems that whether a product is hedonic, or utilitarian does not influence the relationship between promotion and product returns. Hence, this result cannot be further interpreted.
41
6. DISCUSSION
This chapter provides conclusions that are based on the results of the previous chapter. The conclusions are related to the literature review in chapter 2. It provides answers to sub-questions b - f and the main research question of this study. This chapter starts with a hypothesis table, indicating whether the proposed hypotheses are accepted or rejected. Next, each hypothesis is addressed and discussed.
Table 4. Hypothesis table
Hypothesis Conclusion
H1a: Price has a positive effect on the product return probability. Partially accepted H2: Products purchased on promotion are less likely to be returned
than products purchased at the regular price.
Partially accepted H3: There is a negative relationship between product discount
depth and product return probability.
Accepted H4: Hedonic products are more likely to be returned compared to
utilitarian products
Accepted H5: The effect of price on product returns is smaller for utilitarian
products compared to hedonic products.
Accepted H6a: The effect of promotions on product returns is stronger for
hedonic products compared to utilitarian products.
Not applicable H6b: The effect of discount depth on product returns is stronger for
hedonic products compared to utilitarian products.
Accepted
6.1 Effect of price
As stated in section 2.2.2, the aim of sub-question b is to investigate how the price of a product affects its product return probability. Based on utility theory (Thaler, 1983) and customer expectations (Teboul, 1991; Zeithaml et al., 1990), it is argued that as the price of a product becomes higher, the return likelihood increases.
42 An opposite (negative) effect is found for the handbag product category. For this category, as the price increases, its return probability becomes smaller. Hence, lower price then increases the return probability. It could be so that for such a hedonic product category, handbags in a lower price category are more tempting to initially buy (“That looks like a fun bag”), while the purchase is followed by a state of “I did not actually need this”, and thus, a return follows.
The aforementioned results show that a majority (4 out of 6) of the relationships between price and product returns are in line with what is proposed based on the literature review. Therefore, H1 is partially accepted. The answer to sub-question b is that the effect of price on return probability is positive for the majority of selected product categories.
6.2 Effect of promotions
As section 2.2.3 described, the aim of sub-question c is to look at how promotions (i.e., whether something is on promotion or not) influence product returns. Based on prospect theory (Kahneman & Tversky, 1979) and potential values of losses and gains, it is hypothesized that products on promotion are less likely to be returned than products that were not on promotion.
The results in section 5.1.2 show that this holds for vacuum cleaners, but not for tablets and audio & hifi product categories. Out of all product categories in this study, the vacuum cleaner category may be considered the most utilitarian category. Since utilitarian products are more task-related, the expectation is that consumers have a clear goal in mind with the product. When such a product is on promotion, this goal may become even clearer and, as such, increase the likelihood of a fit and decrease the return probability.
Based on these findings, it seems that when the two hedonic product categories are on promotion, the return likelihood actually increases compared to when they are not on promotion. Interestingly, again, a slight difference in effect is noticeable between hedonic and utilitarian product categories. The difference in the signs of the coefficients leads to the partial acceptance of H2.
6.3 Effect of discounts
43 As the results in section 5.1.3 indicate, this negative relation is true for all significant product categories, namely handbags, tablets, audio & hifi, and shavers. Two observations can be made from the estimates. The first observation is that the coefficients are slightly more negative for the three hedonic product categories. Discount depth thus seems to have a bigger impact on product return for these products. Second, there are no significant effects for the vacuum cleaners and hanging lamps product categories (which are both labeled utilitarian products). Although the result of vacuum cleaners seems contradicting with the previous section at first, it may be explained by the high degree of promotions for that category, while the given discounts are not that high. This may explain the significant effect of promotions and the insignificant effect for discount depth.
Similar to the previous paragraph (section 6.2), differences in the effects of variables between hedonic and utilitarian products seem to appear. The signs and significance of the coefficients lead to the acceptance of H3.
6.4 Hedonic versus utilitarian products
As described in section 5.2.1, the results of model 2 show a significant main effect between the type of product category and product returns. The model shows that when a product is hedonic (versus utilitarian), the odds of the likelihood of returning are 3,4 larger. This outcome may be best explained by the task-related purchase goal (Batra & Athola, 1991) when people acquire utilitarian goods. The usage situation and purpose of utilitarian products may be clearer, whereas hedonic purchases may be more impulsive for which such a goal has not (yet) been formed.
The significant and positive relation between hedonic product categories and return probability lead to the acceptance of H4.
6.5 Interaction effects
The results in section 5.3 describe which interactions exist between the aforementioned variables price, promotion, discount depth, and utilitarian versus hedonic product categories. This section discusses the results related to sub-question f.
