Examining the Influence of Basket Compositions and
Price Promotions on Product Returns
Erik Webers
S3854744
Master Thesis
University of Groningen
Faculty of Economics and Business
MSc Marketing Intelligence
22-06-2020
First supervisor: prof. dr. T.H.A. Bijmolt
Abstract
The phenomena of product returns for both offline as online stores has been debated a lot in the past years, which has led to many studies about these product returns and their impact to a firm. The statistics of product return show that this phenomenon is extremely costly for these companies and therefore, it is important for retailers around the globe to get insights in the drivers of product returns to reduce their costs in the end. This study focusses on the effect of basket compositions and price promotions on product returns.
Multiple factors affect the product return behaviours of consumer for retailers worldwide. First, the study of Andrew Petersen and Kumar (2009) showed that product returns cost the firm in terms of both profits from sales and reverse logistics. Second, many other different aspects act as drivers of product returns. For example, product return behaviour is also influenced by subjective norms (customer attitudes), product compatibility, perceived risks, costs and complexity, social group influences, desires for uniqueness, variety seeking, materialism and patronage intentions.
Another important topic in the literature of marketing research is basket compositions and basket analyses, which is often referred to as shopping cart compositions and analyses. Kamakura´s (2012) study examined the market basket analysis (MBA), also known as the product affinity analysis, which idea is one in which the MBA find pairs of sets of products that are jointly observed in large sample baskets. The assumption made here is that purchasing one product would lead to purchasing another product. Hence, in the context of this study, it is quite interesting to examine whether this joint occurrence of pairs of products would also lead to increased return probabilities in general, or that the purchase of certain product pairs leads to increased return probabilities in specific product categories.
To examine the effects of the three topics in marketing research described above together, data from an online Dutch retailer has been used. The dataset available for this study contains circa 120.00 observations, with a total of 62 variables. For the product categories, there are 17 level one product categories, divided into 87 subcategories (level two) and even more specific, the data contains 605 specific product variations (level three). The data is based on a day-level of 365 consecutive days, starting at 1st of July 2016 until 30th of June 2017. The prices per product and per order are within the data set, as well as the price promotions, shipping fees and the detailed order information.
For examining the product return probabilities, a binary logistic regression was used. The results have shown that whenever a basket contains multiple product categories, the return probabilities are influenced significantly, and the return probabilities decrease slightly by 0.02%. Also, whenever a basket contains only items from one product category, the return probabilities are influenced significantly, too. However, the study has in the end not examined the actual cross-category and within-category effects due to time and resource limitations and the complexity of the study. Furthermore, contrary to what was expected, price promotions decrease the amount of product purchases within one basket. This could be due to the possibility that price promotions result in a decrease of total products per basket but could imply that there would be purchases more baskets instead of one large basket. However, this effect should be investigated by future research. To continue, price promotion does lead to multiple product categories within one basket. Lastly, results have shown that a price promotion at the time a customer purchased a product would lead to a decrease of 56,83% in product return probabilities. Hence, the psychological process in which customers are less likely to return products which had a price promotion going on due to the lower prices is supported by data.
Preface
I started my academical journey with the pre-master’s in marketing, after I have completed my bachelor’s degree in Commerce at the university of Emmen. After finishing my bachelor’s degree, I had one year of trying to figure out what I really wanted to do in my life. I could have started working right away, but I felt there was more to it and that my academical capabilities were not challenged yet. Hence, I decided to apply for the pre-master’s in marketing, which I knew would be a challenge for me to continue my personal development. In the beginning, it was quite difficult to get used to studying again after being focussed on alternative matters in life, but I was confident I could make it and along the way, I became more confident in my academical capabilities. The broad domain of Marketing always interested me during my bachelors’ study and in my everyday life. At the end of my pre-master, I decided to enrol for the master’s track Marketing Intelligence, due to my analytical capabilities and my diverse interests. I have to admit that during the pre-master, I really enjoyed all the courses offered and learned that my general interests in the world expanded to other fields of research such as Consumer Psychology, Market Models and Customer Models. This thesis is the final phase of my academical journey, before I go out into the real world and apply my theoretical knowledge and capabilities.
First, I would like to thank prof. dr. Bijmolt for his intriguing advice, good and honest talks and support while writing this thesis, besides having his own very busy schedule. I would also like to thank Christian Hirche for providing me with the dataset, which was used for writing this thesis, and would like to thank other fellow students of my master thesis group for useful discussions about our thesis topics. Lastly, I would thank my family and friends for supporting me during this intense period of challenging myself and to help me succeed, for which I am truly grateful.
I hope you will enjoy reading this thesis.
Kind regards
Table of Contents
Abstract ... 2
Preface ... 4
Chapter 1. Introduction ... 7
1.1 Product Returns ... 7
1.2 Basket Compositions and Analyses ... 8
1.3 Cross Category Influences ... 9
1.4 Price Promotions ... 10
1.5 Study Purpose ... 10
1.6 Literature Contribution ... 11
1.7 Study Consecution ... 11
Chapter 2. Theoretical Background ... 12
2.1 Research Framework ... 12
2.2 Product Returns ... 12
2.3 Basket Compositions and Analyses ... 14
2.4 Cross Category Influence ... 15
2.5 Price Promotions ... 17
Chapter 3. Data Collection ... 19
3.1 Dutch Online Retail Environment ... 19
3.2 Research Design ... 19 3.3. Data ... 19 3.4. Variables ... 21 3.4.1 Key Variables ... 23 3.4.2 Control Variables ... 24 3.5 Data Preparation ... 24
3.5.1 Missings and Recoding ... 24
3.5.2 Correlations and Multicollinearity ... 25
3.6 Model ... 26
3.6.1 Model Specification ... 26
Chapter 4. Results ... 27
4.1 Model Fit ... 28
4.2 Model results ... 28
4.2.1 Results Main Variables ... 30
4.2.2 Results Control Variables ... 32
5.1 Theoretical Implications ... 36
5.2 Managerial Implications ... 37
5.3 Limitations and Future Research ... 37
Chapter 6. References ... 39
Chapter 7. Appendices ... 42
Appendix A: VIF Scores ... 42
Appendix B: Output model 1 ... 42
Appendix C: Output model 2 ... 43
Chapter 1. Introduction 1.1 Product Returns
Product returns is a widely known phenomena in the real-world context and is a real challenge to many retailers within many countries like the Netherlands. In 2019, a total of 13% of all online ordered products has been returned in the Netherlands. This percentage is significantly higher than in other countries such as Germany and Poland and is 3% higher than the average product return percentage within Europe, which is 10%. (Hoeijmans, 2020). Especially within the fashion environment, product returns are a huge problem for online retailers. In 2019, approximately 51% of all fashion products was being returned. This percentage is up until today reduced to 44%, which is still a quite large percentage (Hoeijmans, 2020). Dutch consumers consider online shopping as “the ease of ordering products at home” for 49% and explain that ordering product online allows them to consider purchases from a “bigger selection” (Terpstra, 2019). Dutch people are ordering a lot of online product within Europe, as they have bought for a total of 1.9 billion euros at European web shops, which is 19% more compared to 2018. On the other hand, Dutch web shops generated 17% more online sales compared to 2018. (CBS, 2020).
The statistics above show that product returns are a real challenge for retailers and hence, a topic of interest for many scientific practitioners. The phenomena of product returns for both offline as online stores has been debated a lot in the past years, which has led to many studies about these product returns and their impact to a firm. For example, the study of Petersen and Kumar (2009) showed that product returns are an important and necessary part of the exchange process between companies and customers. This exchange process is considered as a firm-customer exchange process and consists of three key parts: (1) firm-initiated marketing communications, (2) customer buying behaviour, and (3) customer product return behaviour. The last one, will be examined more extensively in this research. In some cases, certain product lines have had return rates of greater than 25% which brings a lot of costs with them. For example, product returns in the U.S. costs the manufacturers and retailers approximately $550 billion by 2020, which is 75,2% more than prior years (Shopify, 2019). Hence, getting insights into the product return behaviours of consumers is an important driver for firms.
firm in terms of both profits from sales and reverse logistics. However, the results of the study showed that product returns empirically increase future customer purchase behaviour. Not only the statistics and the process of product returns is an interesting part of the topic, but also other aspects are interesting to take into consideration. To continue, the study of Powers & Jacks (2015) found that not a lot of product are returned to the retailers because they are defect. Their findings found that customer return products because 1) it does not meet their expectations and 2) they found a better price or product elsewhere. Although many previous studies examined the topic of product returns, yet a lot of topics within this phenomenon are unexplored.
1.2 Basket Compositions and Analyses
Figure 1. Influence Basket Composition on Individual Product Return Probabilities
For example, when a customer would purchase a product from the Electronics category and a product from the Garden category at the same time, what are the mutual influences regarding the returns of these products? Does the purchase of the product from the Electronics category stimulate the return of the product of the Garden category and vice versa, or is there no underlying relationship at all? These questions could possibly be answered within this study, as it will provide deeper insights in the phenomena of product returns regarding the influence of basket compositions, which will lead to interesting insights for managers of retailers
1.3 Cross Category Influences
the firm to offer products which meets his or her needs. Second, a higher ratio of product returns could be a result of the manifestation of possible negative return behaviour by the customers, whereas they would misuse/abuse the return system. To conclude, cross-category influences do have an impact on product return probabilities. However, the question remains which product categories, solely and jointly, affect the product return ratio’s mostly.
1.4 Price Promotions
Nowadays, online retailers have a good understanding of how to use their marketing mix and what influence their marketing-related decisions have on the sales of their products. Some elements of these decisions are pricing and price promotions, which influences their total sales in positive and negative ways. For example, discounts generally have a positive effect on the total sales (Zuo & Lida, 2017, Drechsler et al. 2017, Guerriro et al. 2017). However, these price promotions and discounts influence multiple matters within the environment of, for example, an online retailer. Price promotions could also affect the product returns of an online retailer. Normally, an online retailer has a long-range product portfolio with many diversified products and applies different kind of price promotions per product category. Hence, the effects of these price promotions could also have different effects for the product returns ratio per product category. Therefore, price promotions are taken into consideration for this study due to their possible effects on both product returns and basket compositions.
1.5 Study Purpose
The aim of this study is to fill up the gap in the literature described above, as the results of the study would contain valuable information for managers whom try to comprehend basket compositions and the influence of these compositions on the probabilities of product returns of their customers. This study will examine these basket compositions and their influences on product returns for a large online retailer. The main research question that will be answered within this study is the following:
“How does basket composition and price promotions influence the probabilities of consumers returning products across categories?
2. Which items are frequently bought together? 3. Which items are frequently returned together?
4. Does more products purchased leads to more products returned? 5. Do price promotions lead to more or fewer product returns? 6. Do price promotions influence basket compositions?
7. Which control variables should be taken into consideration to examine the cross-category product return influences?
1.6 Literature Contribution
This research contributes to the existing literature and scientific marketing studies in several ways. First, it shows that the probability of product returns in specific product categories varies across categories. Second, the study will examine the mutual influence between product returns in specific product categories and the likelihood of returning products in other product categories. Basket composition plays an important role within this study, as it will be used to examine whether certain basket compositions influence the likelihood of product returns in general and in specific categories.
1.7 Study Consecution
Chapter 2. Theoretical Background 2.1 Research Framework
The conceptual model below is an illustration for visualizing the hypotheses and theoretical procedure of this study. The model shows the relationship between the independent and dependent variables and the influence of another main effect, which influences both the independent as the dependent variable. Also, a set of control variable below the dependent variable are included but will not be tested and analysed.
Figure 2. Conceptual Model. 2.2 Product Returns
induce more purchases but also more product returns. Hence, the psychological effect of searching for information by reading reviews, influences customers their product return behaviour in a negative way, in which overly positive review results in more product returns.
Other studies, like the one of Lepthien and Clement (2019), examined the effect of shipping fee schedules and return behaviour. Their study analysed the behaviour of customers based on a randomized field experiment where customers were assigned to one of the seven different shipping fee structures. Their study found that minimum order values leads slightly to more returns and that threshold-based free shipping leads to across all customer towards a higher number of strategic returns because the customers tend to return more. Hence, in the context of this study, the minimum order values an online retailer induces towards their customers shows that their shipping and returning fees could influence their product returns in an obstructive manner, leading to more product returns. This could be of use for this study, as the online retailer for which this study will be executed started to handle returning fees for their customers and induced shipping fees, too.
To continue, the study of Walsh et al. (2016) examined the relationship between an online retailer’s reputation and product returns. Their findings indicate that an online retailer’s reputation is a powerful means of reducing product returns, which is influenced by the return motivation. To illustrate this, another big and well-known online retailer, Zalando, had in 2019 a return ratio of more than 50%, meaning that for every two products purchased, one will be certainly be returned (Zalando, 2019)
Hence, there are many different aspects that act as drivers of product returns. To conclude, not only customer reviews, shipping fees and online reputation act as drivers of product returns, but other aspects act like drivers too. For example, product return behaviour is also influenced by subjective norms (customer attitudes), product compatibility, perceived risks, costs and complexity, social group influences, desires for uniqueness, variety seeking, materialism and patronage intentions. Hence, product returns are influenced by a wide variety of drivers and this illustrates the importance of the phenomena of product returns.
Operator yields a predictive model achieving the best accuracy for future product return volume. Hence, the possibility to develop a model for predicting future product returns is possible and shows the importance of product returns nowadays, as it is a great deal for many retailers and manufacturers.
2.3 Basket Compositions and Analyses
Another important topic related to this study is basket compositions and analyses. Basket composition is considered as the variety of specific products within one order, containing different product categories prices and return probabilities. Basket composition is often referred to as Market Basket Analyses (MBA). According to the study of Kamakura (2012), Market Basket Analyses (MBA), also known as product affinity analysis, are already widely known and utilized by traditional and Internet retailers. The basic idea behind the MBA is to find pairs of sets of products that are jointly observed in large samples basket. This assumes that the purchase of one or more products would lead to the purchase of other, remaining products. The underlying assumption in the basket analysis is that the joint occurrence of two or more products in the same basket imply that these products are complementary to each other. Hence, the purchase of one product will lead to the purchase of another product. This part is crucial to this study, as the study aims to examine whether certain purchased products together would lead to an increased probability of returning one-to-many products within and across product categories.
2.4 Cross Category Influence
Another important aspect of this study is cross-category influence. This is due to the interest and reasoning of the researcher in which previous studies do not really examine the effects of basket compositions, product returns, price promotions and cross-category influences all at once. In other words, the four major topics of this study are examined individually often, but there is still a big gap in the literature about these four topics together in one study. This interest lays especially in the part in which a certain basket composition (e.g. purchase of products of multiple product categories) has underlying relationships regarding product returns. The information that possibly could be obtained from this study could induce significant contributions to managers in the eCommerce environment, as it could show that certain product categories together into one basket, could lead to more product returns and how one specific product could alter the return ratio’s for the whole basket. However, there must be noted that for cross-category influences there has been recently a growing interest (e.g. study of Manchada et al. 1999 and Niraj et al. 2008 regarding cross-category models). Most studies about this topic examine the responses to marketing mix activities such as prices and promotions (e.g., Ainslie and Rossi 1998). Hansen et al. (2006), using a multicategory model, explore the extent to which consumers exhibit similarities in their purchases for store brands across categories. They find that household-specific factors are more prominent in explaining preferences across product categories than demographics. To continue, Singh et al. (2005), utilize a multicategory framework for analysing attribute-based preferences across categories while allowing them to be a function of both observable and unobservable household-specific traits, and they find that consumers have strong preference correlations for certain attributes such as brand names. Furthermore, Ma et al. (2012) found in their study about dependencies in brand choice outcomes across complementary categories that the magnitude of cross-price elasticities varies greatly across brands, as certain brands within a product category (e.g. Betty Croker frosting) have a much larger impact on sales and profits of another related product category (e.g. cake mix). In a real-world situation, estimating the full set of cross-elasticities across all possible pairs of products becomes very complex. Also, in this case, the risk of finding certain elasticities that just by chance are statistically significant becomes very relevant (Montgomery, 1997). Hence, examining these cross-category influences regarding product returns is a complex study, but could be useful for managers and scientific practitioners worldwide.
cross-category effects between categories that are not directly related at first sight. They suggest that future research should examine many product categories simultaneously in order to capture valuable information for cross-category effects. Cross-category effects are prevalent but many of them are still absent, which calls for an analysis or estimation procedure that succeeds in highlighting main inter-relationships within product category networks. One insight that is relevant for retailers is that they should focus on shifting their traditional category management approach towards a customer management category approach to enhance sales and profits (Musalem and Aburto, 2018). As the study of Gelper et al. (2015) stated; there are a lot of cross-category effects undiscovered and this study aims at examining at least one effect which remain unexplored. Product returns and cross-category influences are present, yet these two combined has not been studied before. As the study of Gelper et al. 2015 showed that there are a lot of unexplored effects within cross-category effects, the first point within this study will be examining whether these cross-category influences also affect product returns and hence:
H1: “Product purchases from one product category does have an influence on the possibility of returning products across categories”
For H1, the way the cross-category products influence product returns is not expected to be negative or positive yet. The first part is to examine whether these cross-category influences exists, and based on the results of this study, the way it influences product returns (e.g. positive, less product returns) could be concluded later. However, this hypothesis only examines the effects of product returns across categories as a basis for the remainder of the study. Research also has shown that fashion products as well as supplementary and complementary goods from the same product category do have several effects within this product category and therefore:
H2: “Product purchases from one product category have a negative influence on the possibility of returning product within the same category”
with each other, and if one of the intended matched products is not as he or she expected, the intended set will not be possible to wear anymore and hence, he or she will return both products.
The example above illustrates the possible return behaviour of customers with purchases form the fashion category. However, this study will not reduce the possibilities of examining influences cross-category only within one product category but will examine multiple product categories. These hypotheses will be tested to show whether product returns in, for example, the electronics product category does or does not have an influence on the product returns in, for example, the fashion product category and other product category combinations.
2.5 Price Promotions
First, expected is that price promotions have a positive influence on basket compositions (read: more product purchases, larger basket size, larger basket composition). Due to the lower prices during price promotions, it is expected that customers would buy more products and the compositions of baskets change. To continue, price promotions in specific product category could alter intended purchases and could lead to additional purchases which would not be made in the first place. Hence, price promotions increase the number of products purchased and also differentiate the basket composition into one which includes additional product categories and specific products. Scientific evidence for this matter has been examined by Raju (1992). Raju examined the effects of price promotions on the variability of product category sales and found that a higher magnitude of discounts leads to a greater variability in category sales, whereas a higher frequency leads to lower variability in category sales and therefore:
H3: “Price promotions increase the amount of product purchased”
H4: “Price promotions lead to more different product categories within one basket”
decides to keep the product, even though it is against their expectation, due to the low price and discount and hence:
H5: “Price promotions at the time that a consumer purchased a product leads to lower return rates”
On the other hand, opposing to H5, the phenomena described above could also have a reverse reasoning and reverse effect. When a customer would buy a product that has a price promotion going on, there could be the possibility that the customer bought the product too fast (due to the discount) and when he or she receives their order, the product(s) within are not as expected. The psychological process in this phenomena is one in which the customer ask him or herself why they have bought this product and due to the product not meeting expectations, they decide to return the product disregarding the discount of the product.
Chapter 3. Data Collection
This section contains the data collection, which includes a description of the Dutch online retail environment, the research design, the data, variables (key and control), data preparation, missings, recoding, the model and its specification.
3.1 Dutch Online Retail Environment
This research focusses on product return behaviour in the online retail environment within the Netherlands. The data available is collected from a large online retailer, which operates exclusively within the Netherlands. The market in which this retailer operates is a market which consists of many firms selling products to end-users through digital channels, which could be referred to as a business-to-consumer market (De Best, 2020). The population of this study are the consumers who order at online retailers for a various variety of products. The interest in online shopping has increased exponentially since the beginning and is extremely popular within the Netherlands.
3.2 Research Design
This thesis´s research design will be a conclusive one. The research will be about testing hypotheses and to examine relationships and interrelated effects across variables. Furthermore, the analysis of the data will be of quantitative nature and based on secondary data, which the researcher did not collect himself (Malhotra, 2010). Also, the research design will have a descriptive and predictive nature. Descriptive research in this context will be about describing the phenomena examined and usually involves market characteristics, functions and relationships.
3.3. Data
to a maximum of level one product categories, to make the study feasible and achievable. The dataset has been supplemented with external data from the Koninklijke Nederlandse Meteorologisch Instituut. The added data contains weather factors like temperature, which are included as control variables within this study.
In the table below, one can see the most important descriptive statistics per product category.
Product Category Sales %
Sales Returns % Returns Total Purchase value Average purchase value per order 1. Garden 1401 1.17% 76 5,42% €147.566,78 €105,33 2. Beauty 1594 1,33% 52 3,26% €28.215,92 €17,70 3. Toys 2075 1,73% 60 2,89% €68.391,54 €32,96 4. Beachwear 5661 4,73% 2792 49,31% €174.692,11 €30,86 5. Kids Fashion 10884 9,09% 3383 31,08% €263.801,42 €24,24 6. Other 285 0,24% 0 0% €157,50 €0,55 7. Mens Fashion 12222 10,21% 3446 28,20% €500.486,14 €40,95 8. Health 504 0,42% 10 1,98% €9.172,34 €18,20 9. Ladies Fashion 41498 34,67% 17704 42,66% €1.485.616,54 €35,80 10. Nightwear 1343 1,12% 418 31,12% €33.635,75 €25,05 11. Home 5621 4,70% 492 8,75% €410.740,97 €73,07 12. Sports 4551 3,80% 1584 34,81% €181.332,49 €39,84 13. Accessories 3124 2,61% 684 21,90% €88.769,90 €28,42 14. Shoes 9548 7,98% 2559 26,80% €553.989,73 €58,02 15. Electronics 7950 6,64% 421 5,30% €1.313.935,61 €165,27 16. Lingerie 5843 4,88% 1788 30,60% €160.036,53 €27,39 17. Baby 5604 4,68% 654 11,67% €160.000,27 €28,55
The mean value of purchased product within the dataset for all product categories is €46,62. As expected, the values of purchased items within the Electronics, Garden and Home category are far above the average, as the products within these categories have higher cost price per production per one item. Furthermore, the fashion and clothing categories (Kids, Ladies, Men fashion, Sports, Beachwear and Lingerie) have the highest return percentages value. This is due to the fact that within these product categories, the number of products and ratio of product return and purchased products is higher compared to other categories. Subsequently, whereas the purchase values of Electronics, Garden and Home was above the mean of all product categories, the return percentages of purchased products are lower than expected. This is due to the possibility that within these categories, product returns ratio’s and probabilities are lower compared to the fashion category.
3.4. Variables
The dataset that will be used for this study contains several variables. A complete overview of these variables can be found in the table below.
Variable Specification Description Mean SD Min/Max
Customer Information - -
Customer_id IV Identification of
the customer
- - -
Product characteristics - -
Category_level1_code IV Number of the product category a product belongs to (1-18, i.e. 1 = Garden)
- - -
Product Category IV Dummy coded;
Product_return1 DV Dummy coded; 1/0, 1 when returned
- - -
TotalBasket IV The number of
products a customer ordered
7.52 7.26 Min: 1, Max: 51
Purchase_value_eur IV The value of the products ordered in euros
46,62 81,45 Min: 0.00, Max: 3699,00 Returned_items_value_eur IV The value of the
product returned in euros 42,18 53,48 Min: 0.00, Max 3699,00 Price Promotions - -
Promotion1 IV Dummy coded;
1/0: 1 promotion ongoing at time of purchase
- - -
Promotion_value_eur IV Value of the price promotion in euros
12,44 27,63 Min: 0,10, Max: 479,00
Discount1 IV Dummy coded;
1/0: 1 = discount on products
- - -
Discount_value_eur IV Value of the discount in euros
20,55 51,10 Min: -278,00, Max: 2980,00
Voucher1 IV Dummy coded;
1/0: 1 = voucher for products
Table 2. Overview of the Variables 3.4.1 Key Variables
This study examines product returns and to measure these product returns, information regarding whether a customer has returned a product should be used (Petersen & Kumar 2009). Hence, a new variable will be added to the dataset containing dummy coding and gets a value of 1 when a product is returned. To continue, the category difference variable has been added to examine whether an order contains products from different categories (yes/no). This variable will be dummy coded with 1 being an order containing product of different categories. For modelling these variables, the total amount of items in a basket will be analysed with the TotalBasket variable. To continue, for the price promotions, a new variable will be added Promotion1, which is dummy coded with 1 being a promotion going on at the time that the customer placed an order. Also for discounts and vouchers, new variables are added Discount1, Voucher1, which indicate whether an ongoing discount and/or voucher was applied to the order (dummy coded, 1 = ongoing discount/voucher).
Voucher_value_eur IV Value of the voucher used in euros 25,86 45,04 Min: 2,84, Max: 400,00 External factors Avg_temperature CV Average temperature per day 11,18 6,64 Min: -3,80, Max: 25,40
Season CV Time of the year
(i.e. summer, autumn)
- - -
Male CV Dummy coded;
1/0: 1 when the customer is a male
Female CV Dummy coded;
3.4.2 Control Variables
The control variables which should be included into the dataset are weather variables like average temperature. These values will be extracted from the database of the KNMI, due to the influence of the weather on sales. According to Steinker et al. (2017), weather has a significant impact on online sales, especially during weekends, summers and days where the weather is quite extreme. Furthermore, seasonality will be included into the dataset as well. According to Meyer (2019), seasonality influences online sales, whereas nearly 40% of online sales are generated within the last three months of the year – October, November and December. Also, according to Kumar et al. (2004), the definition of seasonality is the underlying demand of specific product types of product category/groups as function of the time period in a year that is independent of other external factors (e.g. price, inventory and price promotions). Hence, this study must control for weather and seasonality for making it a valid and reliable study.
3.5 Data Preparation
The dataset which will be used has to be prepared for any solid analyses could be executed. Hence, data cleaning and data recoding should be applied first. The data was transferred from SPSS to R for the analyses.
3.5.1 Missings and Recoding
3.5.2 Correlations and Multicollinearity
This section contains to elaboration of the dependencies of the independent variables by testing and controlling for multicollinearity in the IV’s. This is done by the means of a correlation matrix, converted to a correlation plot which can be seen in figure 3. Moreover, the VIF scores have been calculated too, which can be found in Appendix A.
Figure 3. Correlation Plot
values exceed the threshold value of 5. Hence, the variables purchase_value_eur (VIF score: 79846,12), returned_items_value_eur (VIF score: 79850,35), voucher_value_eur (VIF score: 8.19) have been removed from the dataset in order to deal with the issue of multicollinearity. However, Voucher1 will remain in the set of variables for examining the influences of vouchers.
3.6 Model
This section contains the model and its specification. The model will be based solely on the provided data which only suggests the results of one year of the sales and product returns of the online retailers. Hence, a comprehensive and detailed report and analysis for the retailer in general cannot be made, only a snapshot of the retailers’ online operations for one year.
For analysing the probabilities for items returned, a generalized linear model is used, and more specifically a logit model is used. A logistic regression has been chosen because the dependent variable is a binary variable with a binomial distribution (items returned, yes/no) (Hosmer & Lemeshow 2000). The dataset contains information regarding orders, which could include multiple items. Due to the possibility of an order containing multiple items, a random effects logit model will be used to correct for this matter. The random effects model is used because one advantage of it is that the assumption of individual heterogeneity varies across items, according to De et al. (2013). To continue, all observations within the dataset are included in this random effect model for an increase in the representativeness of the sample.
3.6.1 Model Specification
Most of the variables described in chapter 3 will be included into the function, which leads to the discrete choice regression function. The function for the model is the following:
𝑃𝑖 = 𝐹(𝛼 + 𝐵′𝑋𝑖𝑗) = 𝑒(𝛼 + 𝐵 ′𝑋𝑖𝑗) 𝑒(1 + 𝑒𝐵′𝑋𝑖𝑗)
Complete Model for Price Promotions and Basket Compositions on Product Returns
Product_return1 ~ Α + 𝛽1(Promotion1*TotalBasket) + 𝛽2(Discount1*TotalBasket) +
𝛽3(Promotion1*CategoryDiff) + 𝛽4(Discount1*CategoryDiff) + 𝛽5category_level1_code+ 𝛽6Purchase_amount + 𝛽7Promotion1+ 𝛽8Promotion_value_eur + 𝛽9Discount1 + 𝛽10Discount_value_eur + 𝛽11Voucher1 + 𝛽12Garden + 𝛽13Beauty + 𝛽14Toys + 𝛽15Beachwear + 𝛽16KidsFashion + 𝛽17Other + 𝛽18MensFashion + 𝛽19Health + 𝛽20LadiesFashion + 𝛽21Nightwear + 𝛽22Home + 𝛽23Sports + 𝛽24Accessoires + 𝛽25Shoes + 𝛽26Electronics + 𝛽27Lingerie + 𝛽28Baby + 𝛽29Male + 𝛽 30Female + 𝛽31avg_temperature + 𝛽32Season + 𝛽33TotalBasket + ɛij.
Model for Price Promotions on Basket Compositions
Α + 𝛽1(Promotion1*TotalBasket) + 𝛽2(Discount1*TotalBasket) + 𝛽3category_level1_code+ 𝛽4Purchase_amount + 𝛽5Promotion1+ 𝛽6Promotion_value_eur + 𝛽7Discount1 + 𝛽8Discount_value_eur + 𝛽9Voucher1 + 𝛽10Garden + 𝛽11Beauty + 𝛽12Toys + 𝛽13Beachwear + 𝛽14KidsFashion + 𝛽15Other + 𝛽16MensFashion + 𝛽17Health + 𝛽18LadiesFashion + 𝛽19Nightwear + 𝛽20Home + 𝛽21Sports + 𝛽22Accessoires + 𝛽23Shoes + 𝛽24Electronics + 𝛽25Lingerie + 𝛽26Baby + 𝛽27Male + 𝛽28Female + 𝛽29avg_temperature + 𝛽30Season + 𝛽31TotalBasket + ɛij
Chapter 4. Results
and all control variables on product returns. The second model will include only the effect of basket compositions on product returns. The third and fourth model will include all variables and the different interaction effects. Hence, the third model looks at the effect of Price Promotion*Purchase Amount with all the control variables included. The fourth model includes the Promotion1 * CategoryDiff an interaction effect plus all control variables .
4.1 Model Fit
The model fit of the four models will first be discussed in this section and the results can be seen in table 3. As can be seen in table 3, all models are performing better than the null model. This is because when comparing the different models their Likelihood Ratio Tests, all the models show significance at the 1% level. When looking at the Akaike Information Criterion and Bayesian Information Criterion, all the models perform better than the null model due to the lower scores. Also, model 2 performs the best based on the AIC and BIC and thus, has a better model fit. When looking at the McFadden R2, also model 2 performs the best with the highest value (pseudo R2 = 0,0561). Hence, there can be stated that model 2 shows the best fit but will not be the only model that needs interpretation, due to the different effects measurements of the different models.
Measurement Null
Model
Model 1 Model 2 Model 3 Model 4
McFadden R2 0.0 0,04619 0,05606 0.0571 0.04655 Log Likelihood -73302.9 -69897.07 -40909 -65180 -69890 Likelihood Ratio Test - *** 6798.9 (13) ***9548.5 (104) ***16245( 31) ***6825.5 (15) AIC 146607 139824 81844 130424 139812 BIC 146617 139969 81970 130734 139967 Signif. Codes: *** p < ** p< 0.05, *p<0.1,
Table 3. Measurements of Model Fit 4.2 Model results
To examine the effects of the four models, the odds ratio results are presented below in table 4. The results will be elaborated below the table.
Variables Model 1 Model 2 Model 3 Model 4
Intercept 4.9999 5.8826 4.1721 5.1250
Discount1*TotalBasket - - 1.001 1.0006
Promotion1*CategoryDiff 1.0041 1.0403**
Discount1*CategoryDiff 1.030 1.0068**
Significance Codes: *** p < 0.01, ** p < 0.05, * p < 0.1, . p = 0.1
Table 4. Estimates per Variable in Odds Ratio 4.2.1 Results Main Variables
In all the models, several variables are always significant. As can be seen in the four models, the variables category_level1_code, TotalBasket, Promotion1, Discount1, SeasonS, SeasonH, Male and all product category variables (except the category Other) are highly significant, which means they all influence the product return probability or the amount of products purchases significantly.
Model 1
Accessoires and Shoes, the odds ratio are all above 1, indicating that whenever a product is purchases from one of the above stated product categories, the return probabilities increase. This is in line with the descriptive statistics of this research, which all indicated that the percentages of product returns of these groups are higher compared to other categories.
Model 2
For model 2, which examines the effects of all included variables on the total basket of a customer, one can see that category_level1_code, Promotion1 and Discount1 significantly influences the amount of products a customer purchases. For category_level1_code, the product category a customer buys a product from, decreases the TotalBasket by 6,56%. For the Promotion1 and Discount1 variables, the total basket decrease by 56,18 and 11,29% respectively. Hence, whenever a promotion or discount is going on, the amount of products purchases decreases, which is contrary to H3, which stated that price promotions increase the amount of products purchases. These results could be due to the possibility of customers buying more baskets but less products per basket, resulting in a decrease in amounts of products purchases, but an increase into the amount of baskets. For model 2, the effects of the values of promotions, discounts and vouchers does not influence the TotalBasket significantly, which means that the values of these three variables does not decrease the TotalBasket significantly. In model 2, the interaction effect between TotalBasket and the category_level1_code is included and show an highly significant odd ratio of 1.0042, meaning that whenever TotalBasket and the product category a product belongs to are taken into consideration together, the total basket increase by 0,42%. Hence, the product category a product belongs to increases the total basket of customer. Furthermore, the odds ratio for the CategoryDiff variable is 0.9999 and significant, indicating that whenever an order contains items of multiple product categories, the product return probability decreases by 0.01%. Also, like model 1, the following product categories increase product return probabilities: Beauty, Beachwear, KidsFashion, MensFashion, LadiesFashion, Nightwear, Sports, Accessoires and Shoes.
Model 3
Beauty, Beachwear, KidsFashion, MensFashion, LadiesFashion, Nightwear, Sports, Accessoires and Shoes. Furthermore, one can see that the interaction effect between Promotion1 and TotalBasket is significant with an odds ratio of 1.004. Also, for the interaction effect between Discount1 on TotalBasket is significant, meaning that whenever a discount is not applied to a product purchase and there is a category difference within the basket, the product return probability decreases by (0.8870*1.030 = 0.91361) 8,63%.
Model 4
Model 4 examined the complete model with all main variables and control variables included tot test their effects on the product return probability. This model shows first that the TotalBasket and category_level1_code significantly influences the return probability by +6,28% and -3,77%. Hence, the total basket increases the product return probability and the product category a product belongs to decreases the return probability. Furthermore, for model 4, the effects of Promotion1 and Discount1 are significant, with odds ratios of 0.5345 and 0.8203 respectively. Hence, when a promotion or discount is going on, the product return probabilities decrease by 46,55% and 17,97%. Also for this model, the values of the price promotions, vouchers or discounts does not significantly influence the return probabilities. Lastly, the interaction effects between Promotion1 and TotalBasket, Promotion1 and CategoryDiff and Discount1 and CategoryDiff are all significant, indicating that whenever Promotion1 or Discount1 is 0, and for example, there is no difference within the category, the return probabilites increase by 4.03% and 0.06% respectively.
4.2.2 Results Control Variables
First, when looking at the results of all models, one can see that variables like avg_temperature, SeasonW and SeasonL are never significant, meaning that their influence on product returns or total basket is not significant.
Model 1
has shown that these beachwear-related products are returned more often, compared to other product categories. To continue, interestingly, the autumn season decreases the return probabilities (odds ratio = 0.8660) by 13,40%. This is somewhat surprising, as this effect was not expected before. Hence, product return probabilities decrease during autumn. Furthermore, whenever the customer is a male, the return probabilities decrease by 21,58% (odds ratio = 0.7842), compared to whenever the customer would be a female. Hence, the probability of product returns by males decreases.
Model 2
One can see for model 2, that again, the control variables SeasonS and SeasonH significantly influence the return probabilities. For the summer season (SeasonS), the odds ratio is 1.1038, meaning that purchases during summer increases the return probability by 10,38%. For the autumn season, the odds ratio is 0.8676, meaning that purchases during autumn decreases return probabilities by 13.24%. Also, the odds ratio for male is 0.7858, meaning that whenever a male purhcases a product, the return probabilities decrease by 21,42%.
Model 3
Once more, the control variables SeasonS, SeasonH and Male are highly significant influencing the return probabilities. For the summer season, return probabilities increase by 11.71% and decrease in the autumn season by 9,45%. For gender, the results of the data show that whenever the customer is a male, return probabilities decrease by 20,99%.
Model 4
Chapter 5. Discussion
This study aimed at examining the effects of price promotions and basket compositions on product returns. For examining these effects, data from a Dutch online retailer has been used. The retailer will remain anonymous for confidentially reasons. The data provided contains data from almost 120.000 observations, containing different orders, products, product characteristics and customer characteristics. The data was solely obtained on the online sales of this Dutch retailer and did not encompass any offline or foreign observations. The timeframe of this dataset was starting at 1st of July 2016 until the 30th of June 2017. During this timeframe, the retailer offered a lot of different price promotions, discounts, vouchers and the orders included many different product categories. For an overview of the statistics per product category, see chapter 3.3. For examining the main research question: “How does basket composition and price promotions influence the probabilities of consumers returning products across categories?, several sub-questions are being answered. These sub-questions are the following:
1. What are the probabilities of product returns for the product categories? 2. Which items are frequently bought together?
3. Which items are frequently returned together?
4. Does more products purchased leads to more products returned? 5. Does price promotions lead to more or fewer product returns? 6. Does price promotions influence basket compositions?
7. Which control variables should be taken into consideration to examine the cross-category product return influences?
These different questions are answered by means of a binary logistic regression translated in four different models, which examine the different effects. Also, for answering these sub-questions, several hypotheses were formulated and the results are provided in table 5
Hypotheses Supported?
H1: Product purchases from one product category does have an influence on the possibility of returning products across categories
Partially, FR H2: Product purchases from one product category have a negative influence on the possibility of returning product within the same category
Table 5. Summary Hypotheses
For the first and second hypothesis, results have shown that there is a significant influence on the product return probabilities whenever a basket contains multiple product categories (CategoryDiff = 1). The results show that whenever a basket contains multiple items from different product categories, the return probabilities decrease slightly (0.02%), but this percentage and the differences of the odd ratios are too small to really imply that these basket compositions influence product returns across categories, despite being significant. The influence for baskets containing only one product category (CategoryDiff = 0), influence the return probabilities significantly. However, for both H1 and H2, this study has in the end not examined the cross-category and within-category effects due to time and resource limitations. Hence, these hypotheses should be studied in future research to examine the real cross-category and within-category effects for product returns.
The third hypothesis questioned whether price promotions increase the amount of product purchases. Although this was expected beforehand, results show that price promotions decrease the amount of purchased products. A possible explanation for this could be that customers would buy more baskets during price promotions, but less products per basket on average. Hence, price promotions decrease total amount of products per basket, but whether these price promotions increase the total amount of baskets could be examined by future research.
The fourth hypothesis examined whether price promotions would lead to more different product categories in one basket. The results show that the interaction effect between price promotions and total basket and price promotions and category differences is significant and hence, price
H3: Price promotions increase the amount of product purchased
No, opposite
H4: Price promotions lead to more different product categories within one basket
Partially
H5: Price promotions at the time that a consumer purchased a product leads to lower return rates
Yes
H6: Price promotions at the time that a consumer purchased a product leads to higher return rates
promotions lead to more baskets and multiple product categories within one basket. However, also here has to be noted that these results only imply whether price promotions lead to more a categorical difference within one basket, but do not imply how many more basket are being bought or how many more product categories end up in one basket, which also should be examined by future research.
The fifth hypothesis stated that price promotions at the time a customer purchased a product leads to lower return rates and this is supported by the data and its results. The results show that at the time a customer purchased a product, return probabilities decrease by 56,83%. Hence, as expected, the psychological process in which customers are less likely to return products which had a price promotion going on due to the lower prices is supported by data. In this case, customers decide to keep the product for a lower price instead of returning it. Hence, the results show that hypothesis 6 is not supported, as price promotions reduce the return probabilities instead of increasing them.
5.1 Theoretical Implications
5.2 Managerial Implications
This research has shown some managerial implications for online retailers in the Netherlands. First, whenever a basket contains multiple product categories, the return probabilities decreases significantly and therefore, managers should try to use cross-category selling efforts in order to increase product varieties within one order, leading to a decrease of overall product return probabilities. However, there has to be noted that these differences are quite small and therefore, managers should not apply these multi-category-selling efforts at once, but should rather wait for future research to acquire knowledge about which product categories specifically influence each other in a negative or positive way. Furthermore, results have shown that price promotions decrease the amount of products purchases per basket, which is contrary to what was expected in the first place. Hence, managers should take good care of their price promotion efforts per product category and should consider whether they want more products sold per order (increase in total basket) or whether they want their results to show that price promotions lead to smaller but more baskets.
5.3 Limitations and Future Research
Chapter 6. References
Ainslie A., Rossi PE. (1998) Similarities in choice behaviour across product categories. Marketing Science 17(2): 91-106.
Bandi, C., Moreno, A., Ngwe, D. and Xu, Z. (2018) Opportunistic Returns and Dynamic Pricing: Empirical Evidence from Online Retailing in Emerging Markets. Harvard Business School. Working Paper 19-030.
de Best, R. (2020), “Online Fashion in the Netherlands – Statistics & Facts” (accessed April 9, 2020) [available at https://www.statista.com/topics/4964/online-fashion-in-the-netherlands/]
CBS (2020) “ Dutch consumers spent €1.9 bn in EU webshops in 2019” (accessed April 15, 2020) [available at https://www.cbs.nl/en-gb/news/2020/17/dutch-consumers-spent-1-9-bn-in-eu-webshops-in-2019]
Cui, H., Rajgopalan, S. and Ward, A.R. (2020). “Predicting product return volume using machine learning methods”, European Journal of Operational Research, Vol 281, No. 3. pp. 612-627 Drechsler, S., Leeflang, P. S. H., Bijmolt, T. H. A. & Natter, M., (2017), “Multi-unit price promotions and their impact on purchase decisions and sales”, European Journal of Marketing. 51, 5-6, p. 1049-1074 26 p.
De, P., Hu, Y., & Rahman, M. (2013). “Product-oriented web technologies and product returns: An exploratory study.” Information Systems Research, 24(4), 998-1010.
Florian Badorf and Kai Hoberg, The impact of daily weather on retail sales: An empirical study in brick-and-mortar stores, Journal of Retailing and Consumer
Services, 10.1016/j.jretconser.2019.101921, 52, (101921), (2020).
Gelper S, Wilms I, Croux C. Identifying Demand Effects in a Large Network of Product Categories. Journal of Retailing. 2016;92(1):25-39.
Guerreiro, R., dos Santos, A., da Silveira Gisbrecht, J. A., & Ong, B. S. (2004). “Cost implications of bonus pack promotions versus price discounts”. American Business Review, 22(2), 72.
Hansen K., Singh V., Chintagunta P. (2006) Understanding store brand purchase behaviour across categories. Marketing Science 25(1) 75-90.
Hoeijmans, N. (2020) “Dutch consumers return the most online orders” (accessed April 16, 2020) [available at: https://cross-border-magazine.com/dutch-consumers-return-the-most-online-orders/] Hosmer, D. W., & Lemeshow, S. (2000). “Introduction to the logistic regression model”
Applied Logistic Regression 2, 1-30.
Kamakura, W. (2012). Sequential market basket analysis. Marketing Letters, 23(3), 505–516. Krithika M., & Rajini, G. (2018). Persuading Attributes of Online Shopping Cart Abandonment. International Journal of Engineering and Technology. 7. 739-742.
Kumar, V., George, M. and Pancras, J. (2008), “Cross-buying in retailing: Drivers and conseuquences”, Journal of Retailing, Vol. 84, No. 1, pp. 15-27
Lepthien, A., & Clement, M. (2019). Shipping fee schedules and return behavior. Marketing Letters, 30(2), 151-165.
Manchanda P., Ansari A., Gupta S. (1999) The “Shopping Basket”: A model for multicategory purchase incidence decision. Marketing Science 18(2): 95-114
Ma, Y., Seetharaman, P.B. and Narasimhan, C. (2012), “Modeling dependencies in brand choice outcomes across complementary categories”, Journal of Retailing, Vol. 88 No. 1, pp. 47-62,
Minnema A., Bijmolt T., Gensler S., & Wiesel, T. (2016). To Keep or Not to Keep: Effects of Online Customer Reviews on Product Returns. Journal of Retailing. 92(3) 253-267
Montgomery, A.L. (1997), “Creating micro-marketing pricing strategies using supermarket scanner data”, Marketing Science, Vol. 16 No. 4, pp. 315-337,
Niraj R., Padmanabhan V., Seetharaman PB. (2008) A cross-category model of households’ incidence and quantity decisions. Marketing Science, 27(2): 225-235
Orendorff, A. (2019). The Plague of Ecommerce Return Rates and How to Maintain Profitability. Shopify. Retrieved from: https://www.shopify.com/enterprise/ecommerce-returns
Petersen, J. & Kumar, V. (2009). Are Product Returns a Necessary Evil? Antecedents and Consequences. Journal of Marketing 73. 35-51.
Powers, T.L., Jack, E.P. (2015) “Understanding the causes of retail product returns”. International Journal of Retail & Distribution Management, 43(12), 1182-1202.
Raju, J.S. (1992), “The Effect of Price Promotions on Variability in Product Category Sales”, Marketing Science, Vol.11, No.3.
S. Serrano (2019). “ Top 10 Reason (and solutions) for Shopping Cart Abandonment” (accessed March 30, 2020) [available at https://www.barilliance.com/10-reasons-shopping-cart-abandonment/] Singh VP., Hansen K., Gupta S. (2005) Modeling preferences for common attributes in multicategory brand choice. Journal of Marketing Research. 42(2): 195-209
Sitetuners (2019). “4 ways to get Customers to Add More to their Shopping Carts” (accessed April 5, 2020) [available at
https://sitetuners.com/blog/4-ways-to-get-customers-to-put-more-in-their-shopping-carts/]
Song, J. (2019) A Study on Online Shopping Cart Abandonment: A Product Category Perspective. Journal of Internet Commerce, 18:4, 337-368
Sridhar K., Bezawada R. (2012) Investigating the Drivers of Consumer Cross Category Learning for New Products Using Multiple Data Sets. Marketing Science 31(4): 668-688
Terpstra, V. (2019) “The Netherlands 2019: Ecommerce Report” (accessed April 16, 2020) [available at https://www.osudio.com/en/blog/ecommerce-report-netherlands-2019]
Ting, P.-H., Pan, S., & Chou, S.-S. (2010). Finding Ideal Menu Items Assortments: An Empirical Application of Market Basket Analysis. Cornell Hospitality Quarterly, 51(4), 492–501.
Walsh, G., Albrecht, A.K., Kunz, W. & Hofacker, C.F. (2016) Relationship between Online Retailers’ Reputation and Product Returns. British Journal of Management, 27: 3-20.
Zalando (2019) “ Returns at Zalando” (accessed April 17, 2020) [available at
Chapter7. Appendices
Appendix A: VIF Scores
Appendix C: Output model 2
