WHICH FACTORS INFLUENCE THE QUALITY
PERCEPTIONS OF RETURNED PRODUCTS?
A study about the effects of the marketing mix on the quality perceptions and demand for
returned products
EVA NOORDMANS
MASTER THESIS
MSc Marketing Intelligence
WHICH FACTORS INFLUENCE THE QUALITY
PERCEPTIONS OF RETURNED PRODUCTS?
A study about the effects of the marketing mix on the quality perceptions and demand for
returned products
By: Eva Noordmans
University of Groningen
Faculty of Economics and Business Department of Marketing
Master Marketing Intelligence
June 2020
Supervisor: Felix Eggers
Second supervisor: Maarten Gijsenberg
Molukkenstraat 134
9715 NZ Groningen
0624454593
ABSTRACT
In the last couple of years, the amount of consumers product returns has increased
enormously. Besides the high cost that the number of returned products brings to a firm, the
enormous number of returned products is not beneficial for the environment either. One of the
strategies for firms to deal with returned products in an effective and sustainable way, is to
encourage customers to purchase returned products. Because of this, the purpose of this study
is to develop the sales of returned products. In this research, we used conjoint choice-based
experiment to investigate which factors, which are related to the marketing mix, influence the
quality perceptions of returned products and how the quality perceptions influence the utility
and demand for returned products. This study found that there is mediating effect between the
quality perceptions of returned products and the utility and demand for returned products.
Furthermore, the results of the study indicate that the factors related to the marketing
instruments; product and place, are most suitable to use for increasing the quality perceptions
for returned products.
PREFACE
In front of you lies my master thesis: ‘Which factors influence the quality perceptions
of returned products?’. I have been working on this thesis from February until June
2020. Writing the thesis was part of my final phase in the Master Marketing
Intelligence at the University of Groningen. Even though I am happy that this journey
now comes to an end, I am very thankful that I got the opportunity to work on this
topic. I would specially thank Maarten Gijsenberg, who firstly introduced me to this
topic, and Felix Eggers for this. While writing this thesis I did not only learn how to
apply skills and techniques that I have learned during the master, but I also learned a
lot more about possible strategies to deal with returned products in a more
environmentally friendly way. I am happy that I have acquired this knowledge and
hopefully I cannot only use this knowledge for my future career but also use it to make
the world a little bit better. I would really like to thank my supervisor Felix Eggers for
his excellent guidance during this process. I would really like to thank him for his
involvement in this project and his valuable feedback. It was a real pleasure for me to
work with him on this topic. I would also like to thank Maarten Gijsenberg in advance
for reading and evaluating my thesis.
I really hope that you enjoy reading my thesis.
Eva Noordmans
TABLE OF CONTENT
1.
INTRODUCTION ... 6
2.1
Conceptual model ... 11
2.2
Quality ... 12
2.3 Promotion: Green advertising ... 12
2.4 Product ... 14
2.4.1 State product ... 14
2.4.2 Warranty ... 14
2.5 Place: Retailer reputation ... 15
2.6 Price ... 16
3.
METHOD ... 17
3.1 Research context ... 17
3.2. Procedure ... 17
3.3 Attribute and levels ... 18
3.3.1 State of the product ... 18
3.3.2 Warranty ... 19
3.3.3 Retailer reputation ... 19
3.3.4 Price ... 20
3.3.5 Moderator: Green advertising ... 21
3.4 Choice design ... 21
3.4.1 Choice task ... 21
3.5 Data analysis ... 23
3.5.1 Multinomial logit model analyses ... 23
3.5.2 Candidate models ... 25
3.5.3 Model fit ... 26
4.
DATA ANALYSIS AND RESULTS ... 27
4.1 Descriptive analyses ... 27
4.1.1 Demographic characteristics sample ... 27
4.1.2 Familiarity with the topic of interest ... 27
4.2 Model fit ... 28
4.3 Effects on the utility and demand ... 29
4.3.1 Mediation: Quality perceptions ... 29
4.3.1 Promotion: Green advertising ... 29
4.3.2 Product: State of the product ... 30
4.3.3 Product: Warranty ... 30
4.3.4 Place: Retailer reputation ... 30
4.3.5 Price: Price ... 30
4.3.6 Moderating effect of green advertising on attributes ... 30
4.4 Effects on the quality perceptions ... 32
4.4.1 Product: State of the product ... 32
4.4.2 Product: Warranty ... 32
4.4.3 Place: Retailer reputation ... 32
4.4.4 Price: Price ... 32
4.4.5 Moderating effects of green advertising on the attributes ... 33
4.5 Relative attribute importance ... 34
5.
LATENT CLASS MODEL ... 36
5.1.2 Model fit ... 37
5.1.3 Segment sizes ... 37
5.2 Segment descriptions ... 38
6.
DISCUSSION ... 40
6.1 Findings and theoretical implications ... 40
6.2 Conclusion and managerial implications ... 43
6.3 Limitations ... 45
7.
REFERENCES ... 47
Appendix A. ... 52
Appendix B. Rcode ... 53
1. INTRODUCTION
In the last couple of years, the amount of consumer product returns has increased enormously.
In the United States alone consumer product returns are approaching over three-hundred
billion dollars (Abbey, Meloy, Blackburn & Guide, 2015). Thereby, within Europe, the
Netherlands is the top leader in terms of the number of products returned each year and the
number continuous to grow (DPD, 2020).
Consumers expect free returns and not offering free returns will slow down sales (Ecommerce
News, 2019). Retailers, however, struggle to offer this because they cannot absorb the cost
into their bottom line (Lazar, 2018). In many cases before the returned item can be resold,
some extra actions such as sorting, repacking or replacement are required (Rittman, 2015). In
the US alone, the cost of online returns come down to $41 billion (Appriss Retail, 2020).
According to AfterPay the average cost for a retailer per return shipment is €12,50 ($14,27)
(AfterPay, 2020). Besides the high cost that the number of returned products brings to a firm,
the enormous number of returned products is not beneficial for the environment either
(Taleizadeh, Haghigh & Niaki, 2019; Liu, Diallo, Chen & Zhang, 2019). This is because
transporting the package back to the seller leaves a trail of emissions. Besides that, every year
5 million pounds of returned goods end up in US landfills. Even when the returned package is
in a good condition it has a certain chance to end up in landfills (Calma, 2020).
Because of the increasing rate of environmental concerns, designing a sustainable supply
chain has attracted researchers and manufactures attention (Liu et al., 2019). Manufactures
who intend to adopt a sustainable supply chain need to deal with product returns in an
effective way (Shaharudin, Govindan, Zailani, Tan & Iranmanesh, 2017).
Another reason for firms to deal with the returned products in an effective and sustainable
way, is that it can create economic benefits. This is because the resale of returned products
can generate an additional amount of profit for a firm, even without applying a series of
reverse supply chain processes to the returned product (Abbey et al., 2015). This is because
the majority of returned products are effectively still new products, so the strategic resale of
these products offers a great opportunity to generate additional profit (Abbey et al., 2015).
Nevertheless, only a few firms have proper strategies to handle the returns and many tend to
ignore them (Jayaraman & Luo, 2007). This can be due the fact that there is still limited
research about the sale and demand of returned products.
This study addresses this research gap and will identify what drives the demand of returned
products. Specifically, it will focus on non-defective returned products, i.e., products without
or very small but not visible damages (e.g., damaged packaging; Abbey et al., 2015) and
refurbished products. A refurbished product is defined as a returned product that will be
available for sale after the product is reconditioned and damaged parts of the product are
replaced. The existing literature mainly focuses on the remanufactured category (Agrawal et
al., 2015; Subramanian & Subramanyam, 2012; Abbey & Guide, 2018). Remanufactured
products are products that were produced from the materials of returned products (Hazen et
al., 2017). As remanufacturing concerns manufacturing processes and the development of a
sustainable supply chain (Liu et al., 2019; Shaharudin et al., 2017; Xue et al., 2018), we will
exclude this category from this study that focuses primarily on the marketing perspective.
To develop the resale of refurbished and non-defective products it is important to investigate
which factors influence the demand of these products. According to the existing literature, the
quality of the product is the most important factor for customers when buying returned
2019; Subramanian & Subramanyam, 2012). Therefore, offering a warranty seems to be an
essential feature to increase the perceived quality and the demand of returned products (Liu et
al., 2019). Whereby, warranty can be defined as a contract between the manufacturer and the
customer to ensure that the manufacturer will repair, replace, or offer compensation to the
customer when the product fails in a specified time period after purchase (Shafiee &
Chukova, 2013). However, the direct relationship between warranties and quality perceptions
of returned products is less investigated and the results of other related studies are mixed
(Srivastava & Mitra, 1997; Purovit & Sivastrava, 2001). Therefore, this study is in need to
further investigate the effect of warranties of the quality perceptions of returned products.
Another factor that can possibly reduce uncertainty about the quality and increase the quality
perceptions of the returned product is the reputation of the seller. Earlier research already
found a positive effect between seller reputation and willingness to pay for returned products
(Akerlof, 1970; Sumbramanian & Subramanyam, 2012). Furthermore, the study of Purohit &
Srivastava (2001) found some evidence for the indirect relationship between seller reputation
and perceived quality perceptions of returned products. However, the direct relationship is
less investigated. Therefore, this study will investigate if seller reputation also can have a
direct and positive effect on the quality perceptions of the returned product.
demand of returned products when the price difference compared to the new product is
relatively small.
What is also quite remarkable in this research field, is the fact that a lot of studies about
returned products highlight the environmental benefits of the resale of these returned
products. However, there is very limited research about the relationship of green advertising,
which is related to ‘promotion’ of the marketing mix, and the perceptions and demand of
returned products. Whereby, green advertising can be defined as “the promotional messages
that may appeal to the needs and desires of environmentally-concerned consumers” (Zinkhan
and Carlson 1995). One of the few studies that have explored the effect of highlighting the
greenness of remanufactured products, compared to the new product, is the study of Michaud
& Llerena (2011). The results of the study suggest that consumers tend to value the
remanufactured product less than the equivalent new product, unless they are informed about
their respective environmental impacts (Michaud & Llerena (2011). If green advertising has a
positive effect on customers perceptions of returned products, this could be very useful for
(marketing) managers. Therefore, it is important to further investigate the effects of green
advertising.
The purpose of this study is to develop the sales of returned products from a marketing
perspective and therefore answer the following questions:
Research question: Which factors influences the quality perceptions of returned products?
Sub question 2: How does quality effect utility and demand for returned products?
Sub question 3: Which marketing instruments can affect the quality perceptions for returned
products?
This study contributes to the existing literature by further investing the relationship between
quality perceptions and demand for returned products. Furthermore, this study is an addition
to the existing literature by investigating the relationship between factors, related to the
2. LITERATURE REVIEW
This chapter consists of an extensive discussion of relevant literature regarding the
topic of this study. Important findings in this topic of research are addressed and
provided as background information for the conceptual model. All variables in the
models are discussed and their related hypotheses are presented.
Figure 1. Conceptual model
2.1 Conceptual model
2.2 Quality
Perceived quality is a major factor affecting purchase decision of customers when buying
returned products (Subramanian & Subramanyam, 2012; Vafadanrikjoo, Mishra, Govindan &
Chalvatzis, 2018). However, the fact that the product has been returned before does
negatively influence the perceived quality perceptions of the product and increases
uncertainty about the quality of the product (Liao, 2018; Vafadarnikjoo et al., 2018). Because
of the high uncertainty about the quality of the returned product, consumers tend to value the
returned product less than the equivalent new product (Subramanian & Subramanyam, 2012).
The lower quality perception of returned products is one of the crucial factors that slow down
the demand of returned products (Hazen, Boone, Wang & Khor, 2016; Agrawal, Atasu &
Ittersum, 2015)
Thus we hypothesize;
H1: Positive quality perceptions of returned products have a positive effect on the utility and
demand for returned products relative to the new products
We argue that quality serves as a mediator such that, when selling returned products, it is very
important to investigate which factors, directly and indirectly, influence the quality
perceptions and utility and demand for returned products, which will be discussed below.
2.3 Promotion: Green advertising
Buying returned products is considered as “green choices” (Yang et al., 2015). One of the
reasons for this, is that a lot of returned products end up in landfills (Parket et al., 2015; Xu et
al, 2005; Xue et al., 2018) More specially, buying these returned products is associated with
diverting used products from landfills and reducing harmful emissions.
product (in terms of money) less than the equivalent new product, unless they are informed
about their respective environmental impacts. Despite the environmental benefits of returned
products and customers increasing attentions for environmental issues, a lot of previous
studies failed to consider the influence how green advertising can be used to inform people
about the green benefits of the product and can change peoples’ perceptions towards returned
products. Furthermore, the moderating effect of green advertising on the relationship between
quality perceptions and the utility and demand of returned products is, to the authors
knowledge, not investigated before. One of the few studies who investigates the relationship
between green advertising and quality perceptions is the study of Newman, Gorlin & Dhar
(2014). This study found that when a company intends to make the product that is better for
the environment this can have a negative effect on the quality perceptions and demand of the
(new) product. However, when the environmental benefits are an unintended side effect this
has a positive effect on the quality perceptions and purchase intentions of the product
(Newman, Gorlin & Dhar, 2014). The result of the study indicate that green message framing
is affecting quality perceptions and demand of returned products. Besides that, the results
indicate that, independent of how the message is framed, green advertising is affecting the
relationship between quality perceptions and demand for products. Therefore, it would be
interesting to investigate if green advertising will strengthen the relationship between the
quality perceptions and demand for returned products. Because of the promising results of
previous studies (Atasu et al, 2008; Michaud & Llerena, 2011; Gorlin & Dhar, 2014) and the
increasing attention and awareness of environmental issues we expect and hypothesize that:
H2a: Using green advertising will strengthen the effect and relationship between quality
perceptions and the utility and demand for returned products relative to new products
H2b: Using green advertising will increase the utility and demand for returned products
relative to new products
2.4 Product
There are two variables included in this study which are related to ‘product’ of the marketing
mix; state of the product and warranty.
2.4.1 State product
One of the things which negatively influences the quality perceptions of returned products for
consumers, is that there is a wide range of definitions and categories that describe returned
products and consumers do not know the differences between these categories (Ovchinnikov,
2011). As mentioned, this study will focus on the non-defective returned products and
refurbished products. Despite the fact, that through strategic resale non-defective product
returns offer a great opportunity to generate additional profit, there is very limited research
about this (Abbey et al., 2015). This is the same for refurbished products. The difference
between the two categories, are the remanufacturing processes that have been applied to the
returned products before they were made available for (re)sale. A refurbished product is a
returned product that is available for sale after the product is reconditioned and damaged parts
of the product are replaced (Ovchinnikov, 2011). A non-defective product is a returned
product that because of some very small, but not so visible, damages (e.g. damaged
packaging) cannot be sold as a fully new product. However, the product is still fully
functional and will be available for sale without even applying remanufacturing processes
(Abbey et al., 2015). According to the existing literature the unobservable applied
manufacturing processes to the returned products has negative impact on the quality
perceptions of returned product (Agrawal et al., 2015). Because the non-defective returned
products are made available for sale without remanufacturing processes, we expect and
hypothesize that:
H3a: Non-defective returned products are perceived to have a higher quality than refurbished
products
2.4.2 Warranty
returned products (Liao, et al., 2015; Liu et al, 2019). However, to the authors knowledge, the
effect of offering a warranty on the perceived quality of returned products is less investigated.
The results of the related study of Srivastava and Mitra (1997) indicate that for experts a
better warranty leads to perceptions of higher quality, regardless of the reputation of the
seller, but for novices this is only true when the seller of the product has a high reputation.
However, this study did not focus on returned products. Therefore, this study is in need to fill
this research gap and investigate the relationship between warranties and quality perceptions
for returned products. Because of the high uncertainty about the quality of returned products
and the reason that warranty often signals quality, we expect and hypothesize that;
H3b: Offering a strong warranty for returned products will positively influence the quality
perceptions of returned products
2.5 Place: Retailer reputation
Furthermore, when there is a lot of uncertainty about the quality of the product, such as for
returned product, the role of the seller becomes more important (Subramanian &
Subramanyam). Besides that, the additional effort (disassembling, servicing, and testing) that
is applied to the returned product to make it “as new” is unobservable to the customers
(Agrawal et al., 2015). Therefore, the trust in the seller is very important and the seller
reputation is an important factor when deciding to buy a returned product (Subramanian &
Subramanyam, 2012; Akerlof, 1970). Akerlof (1970) suggest that when there is uncertainty
about the quality of the product or service, sellers with a good reputation should enjoy higher
prices through credible communication of quality. The result of the more recent study
are not directly sold by manufactures but often sold through retailers. Therefore, it is
important to investigate if there is direct link between retailer reputation and quality
perceptions of returned products. Because of the unobservable additional effort that is has
been applied to the returned product to make it as new, this increases the uncertainty about the
quality of the product (Argawal et al., 2015). Because the retailer provides the link between
the consumers and the manufacture, the trust in the retailer becomes very important for the
quality perceptions for returned products (
Purohit & Srivastava, 2001;
Subramanian &
Subramanyam, 2012; Akerlof, 1970). Therefore, we hypothesize and expect:
H3: There is positive relationship between the reputation of the retailer and the perceived
quality perceptions of returned products
2.6 Price
As a result of the lower perceived quality of the returned products, compared to the new
product, price seems to be also an important driver for the demand of these returned and
refurbished products (Hazen, et al., 2017). However, when offering the returned product at a
much lower price than the equivalent new product, this can also reduce the perceived quality
of the returned product as well as for the new product (Argawal et al, 2015; Liu et al, 2019).
Because, when there is uncertainty about the quality of the product consumers often use the
price of the product as a cue to evaluate the quality of the product (Akerlof, 1970; Wolinsky,
1983; Milgrom & Roberts, 1986). But this widely used quality and price relationship is put in
perspective by
Völckner and Hoffman (2007). They found that this relationship is still there
but is weakened over the years and might differ by product category. Therefore, it would be
interesting to investigate if this quality-price relationship holds for returned products and if
customers evaluate the quality of returned products more positively when the price difference
compared to the new product is relatively small. Therefore, we hypothesize:
H5: Offering a returned product with a small price differential compared to the new product
3. METHOD
3.1 Research context
Despite the fact that the online fashion industry has the highest return rate (DPD, 2020;
Charlton, 2020), earlier studies mainly focused on the demand and supply of returned
electronic products (Abbey et al., 2015; Liu et al., 2019; Subramanian & Subramanya, 2012).
One category within the fashion industry with relative high return rate (30%), is the category
of jewellery and accessories (Charlton, 2020). Despite the high return rate and the great
potential for the sale of returned items within this product category, such as refurbished
watches, there is limited to none research about this. Therefore, this study will contribute to
the existing literature by focusing on a different product category. Within the product
category of jewellery and accessories, watches have a relative high return rate of 20%
(Brandfield, 2019). Because of this high return rate and the great potential of resale of
returned watches this study will focus on the demand of returned watches. Furthermore,
because the primary goal of buying luxury products, such as watches, is to impress others,
consumers often use the price cue as evidence to access the quality of the product and to
confirm their quality perceptions (Parguel, Delécolle & Valette-Florence, 2016; Kapferer et
al., 2014
)
. It would therefore be interesting to see how the other factors of the marketing mix
will influence the quality perceptions of watches.
3.2. Procedure
As mentioned, a returned watch was used as the product stimulus in the experiment. The
experiment was randomized so that the attribute and attribute levels were presented an equal
number of times. In our experiment each participant took part in three tasks. To create some
insights in the sample and their familiarity with the product and product type of interest, at the
start of the survey the participants were asked to answer some questions related to watches
and their experience with returned, refurbished and second-hand products. The next task was
the conjoint choice task. In the follow-up task people were also asked about their
3.3 Attribute and levels
The 4 independent variables; state of the product, warranty, price and retailer reputation, are
in the choice experiment presented as the attributes of the returned product with 2 to 4
attribute levels per attribute (see table 1). The attributes of the new product, as a benchmark,
will be kept fixed.
Table 1. Attributes and attributes levels
Variable
Number of attribute levels Attribute level description
State product
2
1. Non-defective
returned watch
2. Refurbished watch
Warranty
3
1. 1 year
2. 2 years
3. 3 years
Retailer reputation
3
1. 3.0 star – average
2. 3.5 star – good
3. 4.5 star – excellent
Price
4
1. 5% discount
2. 10% discount
3. 20% discount
4. 30% discount
Green advertising
Moderator
1. Green advertising
message
2. No message
New watch
Used a benchmark
New watch
1 year warranty
Average retailer reputation
Full price
3.3.1 State of the product
The attribute; state of the product and the attribute levels, needed to measure the preferences
of consumers for the state of the product (non-defective or refurbished) and how the state of
the product influences the quality perceptions of returned products and how this influences
the demand. Therefore, the attribute was split up into two attribute levels; non-defective
returned watch and a refurbished watch. The main difference between the two levels is the
remanufactured processes that are applied to the returned watch and the two levels were
described as follows:
watch. The watch is tested, cleaned and repacked by the retailer. May have observable
cosmetic blemishes.”
2. Refurbished watch: “The watch is as new. The original package of the watch is
opened. The watch may have been worn by a customer before. Fully refurbished,
whereby damaged parts are replaced. The watch is tested and repackaged to meet the
original factory specifications. May have small observable cosmetic blemishes.”
3.3.2 Warranty
The attribute warranty and the attribute levels needed to measure the preferences of
consumers for offering a warranty and how offering a warranty influenced the quality
perceptions of returned products and how this influences the demand. Because the data is
collected in the US, warranty guidelines of (online) stores, who sell watches, and the legal
guidelines of the Federal Trade Commission were used to determine the warranty type and
length (Federal Trade Commission, 2020; Apple, 2020; Casio 2020). According to this data
the standard warranty for watches is one year and the warranty gives protection for any
manufacturing defects of the watch (Apple, 2020; Casio 2020). To measure if offering a
stronger warranty increased the quality perceptions and the demand for returned products the
attribute levels of warranty were:
1. 1-year warranty – standard warranty period
2. 2-years warranty – extended warranty period
3. 3-years warranty – very extended warranty period
3.3.3 Retailer reputation
The attribute and levels of retailer reputation, should measure the influence of retailer
reputation on consumers quality perceptions of returned products and how this influences the
demand of returned products
customers (Eggers & Sattler, 2011) the retailer reputation are also described in words. In this
study the retailer reputation is ranging from:
Level 1: 3 stars – Average reputation
Level 2: 3.5 stars – Good reputation
Level 3: 4.0 stars – Excellent reputation
3.3.4 Price
According to the results of previous studies when the price of the remanufactured product was
decreased up to approximately 40% with a mean of 27%, relative to price of the new
equivalent product, it increased the attractiveness of returned product (Abbey et al., 2015;
Subramanian & Subramanyam, 2012). As mentioned, lowering the price can result in
cannibalization of the new product (Liu et al., 2019). Furthermore, lowering the price can
reduce the quality perceptions of products (Akerlof, 1970; Wolinsky, 1983; Milgrom &
Roberts, 1986). Especially for luxury goods, such as watches, the price of the product is often
used to support consumers quality perceptions (Parguel, Delécolle & Valette-Florence, 2016).
This is because the primary goal of consumers is to impress others (Parguel, Delécolle &
Valette-Florence, 2016). Therefore, this study investigated how price influenced the quality
perceptions of returned products when the price differences compared to the new equivalent
product were relatively small and up to 30%. As earlier studies recommend, in this study we
used discounts (relative to the new product price) instead of absolute prices (Ovchinnikov,
2011). This is because customers often think in discounts rather than absolute prices
(Ovchinnikov, 2011). Therefore, the attribute price was split up in the following four levels:
Level 1: 5% discount
3.3.5 Moderator: Green advertising
To measure the effect of green advertising we used a randomized experimental design.
Meaning that half of the respondents, before they start with the choice tasks, was shown a
message where the environmental benefits of buying returned products is highlighted and the
other half of the respondents was not be shown the message (e.g. control group). The
respondents were randomly assigned to one of the conditions (green advertising vs control).
The message framing is based on the suggestions made in the study of Newman et al. (2019).
The message was as follows:
Green advertising message: “Buying returned products is associated with significantly
reducing the amount of products ending up in landfills and reducing harmful Co2 emissions
coming from the production of new products. Buying a returned watch instead of a new watch
is therefore very beneficial for the environment.”
3.4 Choice design
To measure customers’ preferences for returned products a conjoint based choice experiment
was conducted. In a conjoint based choice experiment the customers choose their most
preferred product from a set of alternatives (Eggers & Sattler, 2011). Conjoint based choice
experiment is a widely used preferences measurement (Eggers & Sattler, 2011). Preferences
measurement is used to translate specific characteristics or attributes of product into the
perceived preferences of consumers and is used to predict customers buying behaviour for
different conditions, in this case different (additional) product modifications (Eggers &
Sattler, 2011). The recommended amount of choice tasks per customer is 8 – 16 and after 12
choice sets you really need to motivate the customer to proceed (Eggers & Sattler, 2011).
Based on the recommendations and for the validity of the results each participant needed to
evaluate 12 choice sets (Eggers & Sattler, 2011). Within every choice task the respondents are
asked to choose between 3 alternatives per choice set. This is within the recommended (2-5)
number of alternatives per choice set (Eggers, Sattler, Teichert, Völckner, 2018).
3.4.1 Choice task
items presented in survey and asked participants to evaluate them on a multiple item
Likert-scale. However, as the study of DeSarbo, Rolandelli & Choi (1994) and the study of Newman
et al., (2014) suggested that a conjoint analysis can also be used to measure the perceived
quality. To measure the utility for returned product the respondents were asked to select the
alternative they prefer the most (see figure 2). Furthermore, a dual response question was used
to investigate if people will actually buy the returned watch if it was available or will go for
the new product which could also be interpreted as the none-option (see figure 3). In other
words, to investigate the demand for the returned product relative to the demand for the new
product. Using the dual response format corrects for the under selection of the no-choice
option which is beneficial for the salience of the purchase decision (Wlömert & Eggers,
2016). Another important benefit of the dual response procedure is that the preference
information is still there even if none of the alternatives would be purchased, or in our case if
they would go for the new product (Wlömert & Eggers, 2016; Brazell, Diener, Karniouchina,
Moore & Uldry, (2006). An example of the choice task is given in figure 2.
Figure 2. Hold-out choiceset of the choice task
3.5 Data analysis
This study includes 4 independent variables, 1 moderator, 1 mediator and 1 dependent
variable. Previous models and research did not allow for mediating effects in choice
experiments and models (Baron and Kenny 1986; Judd and Kenny 1981). However, the more
recent study of Burk, Eckert & Sethi (2019) developed the benefits-based choice model that
allows for prior defined mediating effect, such as perceived quality perceptions. This model
examines how changes to product features influences the perceived benefits of interest and if
these changes in product features affect consumers choices (Burk et al., 2019). In this study it
examines how changes in product features or additional features, such as warranty and price,
affect the perceived quality perceptions of returned products and how such changes affect
consumers choices. Therefore, the benefit-based choice model was a suitable method for this
study.
3.5.1 Multinomial logit model analyses
In order to investigate the effect of the marketing mix on the quality perceptions, and in turn
the effect of quality perceptions on utility of returned products, different multinomial logit
(MNL) models were conducted (see 3.5.2 candidate models). These models are also known as
the conditional logit model (McFadden, 1973). The multinomial logit model (see equation 3)
is based on the utility model and utility function (see equation 1 and 2).
Multinomial logit model relies on the random utility model:
U = Utility
V= systematic utility component, rational utility
𝜀 = stochastic utility component, error term
n = consumer
Equation 1. Random utility model
Where the systematic utility component can be obtained with the following formula:
k = (1, …, K) number of attributes
x = dummy indicating the specific attribute level of product i
β = part-worth utility of consumer n for attribute k
Equation 2. Systematic utility component
The utility function is transformed for the MNL model to estimate the probabilities of
choosing alternative i from choice set J:
prob = probability of choosing option i from a set of J alternatives.
J = choice set with alternatives {1, …, i, …, m}
c = choice sets
Equation 3. Multinomial logit model
The MNL model estimation relies on maximum likelihood procedures. The maximum
likelihood procedure relies on finding the set of part worth utilities that best represent the
observed choices in the data set and is calculated by the following formula:
n = number of consumers
c = choice sets
Equation 4. Maximum likelihood procedure
The parameters for the attributes can be found by maximizing the function subject to the path
worth utilities:
3.5.2 Candidate models
In total four MNL models are estimated to analyse the results. Where, MP-Model 1 (is ml1 in
the R code, see Appendix B) is the model where the dependent variable is the Most Preferred
(MP) alternative and with only the main effect of the independent variables (State of product,
Warranty, Price (e.g. Discount) and Reputation). MP-Model 2 (mlI in the R code, see
Appendix B)
is the most complete model including the mediating effects of the Highest
Quality alternative (included as an independent variable) and with the moderating effects of
Green Advertising. To control for other moderating effects of green advertising, interaction
effects with all the variables in the model are included. HQ-Model 1 (ml2 in the R code, see
Appendix B) is the model where the dependent variable is the Highest Quality (HQ)
alternative and with only the main effect of the independent variables. HQ – Model 2 (mlI2 in
the R code, see Appendix B) is the model with Highest Quality alternative as a dependent
variable with moderating effects of green advertising. It is important to note that effect coding
was used for the estimation of the part-worth utilities. Effect-coding is accomplished by
setting the reference level to -1. Effect-coding therefore provides a part-worth utility value for
each attribute level and it is irrelevant which level is set as the reference and the reference
level can be estimated (Eggers, Sattler, Teichert, & Völckner, 2018).
MP - Model 1:
𝑉𝑖 = 𝛽𝑠𝑝 ∗ 𝑆𝑡𝑎𝑡𝑒𝑃𝑟𝑜𝑑𝑢𝑐𝑡
!+ 𝛽𝑤 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!+ βd ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!+ 𝛽𝑟 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!+
𝛽𝑛 ∗ 𝑁𝑜𝑛𝑒
MP - Model 2:
𝑉𝑖𝑛 = 𝛽ℎ ∗ 𝐻𝑖𝑔ℎ𝑒𝑠𝑡𝑄𝑢𝑎𝑙𝑖𝑡𝑦
!"+ 𝛽𝑠 ∗ 𝑆𝑡𝑎𝑡𝑒𝑃𝑟𝑜𝑑𝑢𝑐𝑡
!+ 𝛽𝑤 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!+ βd ∗
𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!+ 𝛽𝑟 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!+ 𝛽ℎ𝑔 ∗ 𝐻𝑖𝑔ℎ𝑒𝑠𝑡𝑄𝑢𝑎𝑙𝑖𝑡𝑦
!"∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑠𝑔 ∗
𝑆𝑡𝑎𝑡𝑒𝑅𝑒𝑡𝑢𝑟𝑛𝑒𝑑
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑤𝑔 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑑𝑔 ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!∗
𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑟𝑔 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑛 ∗ 𝑁𝑜𝑛𝑒 + 𝛽𝑛𝑔 ∗ 𝑁𝑜𝑛𝑒 ∗
𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"HQ - Model 1:
𝑉𝑖 = 𝛽𝑠 ∗ 𝑆𝑡𝑎𝑡𝑒𝑃𝑟𝑜𝑑𝑢𝑐𝑡
!!+ 𝛽𝑤 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!+ βd ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!+ 𝛽𝑟 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!HQ - Model 2:
𝑉𝑖𝑛 = 𝛽𝑠 ∗ 𝑆𝑡𝑎𝑡𝑒𝑃𝑟𝑜𝑑𝑢𝑐𝑡
!+ 𝛽𝑤 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!+ βd ∗ 𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!+ 𝛽𝑟 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!+
𝛽𝑠𝑔 ∗ 𝑆𝑡𝑎𝑡𝑒𝑅𝑒𝑡𝑢𝑟𝑛𝑒𝑑
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑤𝑔 ∗ 𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑑𝑔 ∗
𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"+ 𝛽𝑟𝑔 ∗ 𝑅𝑒𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛
!∗ 𝐺𝑟𝑒𝑒𝑛𝐴𝑑𝑣𝑒𝑟𝑡𝑖𝑠𝑖𝑛𝑔
!"Vi = Systematic utility component for alternative i
3.5.3 Model fit
As mentioned, different MNL models used for analysing the results. To assess which model
fitted best to data, multiple model fit analyses were performed. First, for every model the
likelihood test was performed. This test analyses if the estimated model exceeds the
log-likelihood value of the NULL model (L0). Whereby the log-log-likelihood of the estimated model
(LL) is calculated by:
Equation 5. Log-Likelihood function estimated model
The log-likelihood function of the NULL model is calculated by:
Equation 6. Log-Likelihood null-model
The log-likelihood value of the estimated model (LL(1)) should significantly exceed the value
of the NULL model, because otherwise the estimated model is not predicting better than the
NULL model would. To test if the log-likelihood for the estimated model is significantly
better, and therefore the estimated model predicts better than the NULL model, the
log-likelihood ratio test is thus performed with the test statistic is χ
2= 2 * (LL(1)– LL(0). The test
statistic is x-squared distributed with df=difference in the number of parameters (Eggers,
Sattler, Teichert, & Völckner, 2018). Besides the log-likelihood ratio test, the Pseudo-R
2and
Adjusted- Pseudo-R
2were performed to asses and compare the goodness of fit of the different
models. Where the 𝑅
!is calculated by 1 – (LL(1) /LL(0)). When the 𝑅
!exceeds the value of
0.2 the model can generally be considered as acceptable (Eggers, et al., 2018).
In addition to the log-likelihood ratio test and pseudo 𝑅
!test the predictive validity of the
models was also tested. This is done by calculating the Mean Absolute Error (MAE) for the
holdout set (see figure 2). This is calculated by the following formula:
𝑀𝐴𝐸 =
1
𝑛
( |𝑋
!− 𝑝𝑟𝑜𝑏
!|
" !#$
With,
|x
i– probi| = the absolute errors (differences between the observed values for the holdout
4. DATA ANALYSIS AND RESULTS
4.1 Descriptive analyses
Before further analysing the data there were some descriptive analyses performed to create
insights in the sample.
4.1.1 Demographic characteristics sample
As mentioned, the total sample consisted of N=349 participants. The sample consisted of 163
males (46,7%), 179 females (51,3%) and 7 participants (2%) did not like to provide their
gender. This distribution is roughly the same as the male (48,2% and female (51,8%) as the
approximated distribution of the US population (Duffin, 2020). The minimum age class of the
sample was 18-24 years old and the oldest participant was between 75 and 84 years old. Most
participants (N=127) were between 25-34 years old, which is quite representative for the
population of interest because most people in the US are in the age class of 25-29 (Duffin,
2019). Therefore, the sample can be seen as a moderate representative sample for the
population of interest.
4.1.2 Familiarity with the topic of interest
Besides demographic questions, respondents were also asked to answer some questions
related to the research topic of this study. According to the descriptive analysis of these
questions, around 193 (55,3%) respondents were currently wearing a watch and 275 (78,8%)
respondents had ever bought a watch for themselves (see figure 4). Besides that, only 21,7%
of the respondents did not ever buy a returned, refurbished or second hand product (see figure
5.
Figure 4. Experience watch
Figure 5. Experience returned products
55% 78,8% 0% 20% 40% 60% 80% 100%
Wearing a watch Ever bought a
Amount of respondents
Experience returned, refurbished
or second hand products
4.2 Model fit
The results of the performance analyses of the models are presented in table 2. According to
the results, MP - Model 2 was slightly better performing, except for the MAE, than MP-
Model 1. Because of this, and because MP - Model 2 was more complete than MP - Model 1,
the effects of the attributes on the demand of returned products are based on the results of MP
- Model 2. The results are presented in table 4. For the same reasons, the results of HQ-
Model 2 are used to investigate the effect of the attributes on the quality perceptions of
returned products. The results are presented in table 3.
Table 2. Model fit MNL models
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
4.3 Effects on the utility and demand
To investigate how the quality perceptions of returned products affect the utility and demand
for returned products (relative to new products), and how this relationship is moderated by
using green advertising, the part-worth utilities were estimated by the MNL model analysis.
For the completeness of the study the effects of the different attributes on the utility of
returned products will be also briefly discussed. As mentioned, the results of the multinomial
logit analyses of MP-Model 2 are interpreted. These results are presented in table 3
.
4.3.1 Mediation: Quality perceptions
In order to investigate the effect of quality perceptions on the overall utility and demand for
returned product, the path-worth utility for the highest quality variable was estimated.
Because the MNL model does not allow for a mediating effect, the highest quality selection
variable was included as an independent variable (Highest_Quality) in the model. According
to the results of Model 2 (see table 3) quality has a positive and significant utility and
therefore has a positive effect on the probability that someone chooses the returned product.
This result is in line with H1; positive quality perceptions of returned products have a positive
effect on the demand of returned products and therefore H1 is supported.
4.3.1 Promotion: Green advertising
To investigate the effect of the green advertising message, which is related to “promotion” of
the marketing mix, strengthen the relationship between quality perceptions and the utility and
demand for returned products an interaction effect between quality and green advertising was
created. The green advertising message is represented by the dummy variable;
GreenAdvertising, where 1 represents the half of the respondents who saw this message and 0
represents the half of the respondents who did not saw this message. When we look at the
results (see table 3), we see that the interaction effect between green advertising and
perceived quality (Highest_Quality) is significant but negative. Meaning that using green
advertising significantly lowered the mediating effect of quality perceptions. Therefore, H2a:
Using green advertising will strengthen the effect and relationship between quality
perceptions and the utility and demand for returned products relative to new products, cannot
their most preferred option it equals to 0. According to the results; the new product shows a
negative part-worth utility so that, on average (i.e., with all attributes at their mean utility of
zero), choosing one of the returned products provides a higher utility, and is therefore more
likely than choosing the new product. To investigate the effect of green advertising on the
utility and demand for returned products relative to the utility and demand of new products,
an interaction effect between the none-option (which is representative for the demand of new
product) and green advertising was created. According to the results there is no significant
effect between green advertising and the none-options and therefore; H2b: Using green
advertising will increase the utility and demand for returned products relative to new
products is not supported
4.3.2 Product: State of the product
According to the results of MP-Model 2, the part-worth utility for attribute level;
non-defective returned is significant and positive. Because the refurbished returned product state
was the reference level is therefore negative but also significant, see table 3.
4.3.3 Product: Warranty
The attribute which is also related to “product” of the marketing mix is warranty.
When we look at the estimates for different attribute levels of the attribute warranty, we see
that strongest warranty (Warranty_3) showed the highest utility and the lowest level of
warranty (Warranty_1) showed the lowest utility see table 3. Only a 2 years warranty
(warranty_2) is not significant, which means, that it is not significantly different from zero.
4.3.4 Place: Retailer reputation
When we look at the attribute retailer reputation, which related to the “place” of the marketing
mix, we see that the higher or better the reputation of the retailer the higher the utility (see
table 7)
4.3.5 Price: Price
According to the results of Model 2, higher discounts lead to higher utilities and for the
returned product. Meaning that the utility for discount is 30% (Discount_30) is the highest
lowest discount (Discount_5) has the lowest utility (see table 7).
4.3.6 Moderating effect of green advertising on attributes
attribute levels and the green advertising message created. When we look at the results (see
table 3) we see that the interaction effect for the interaction effect between the attribute level;
non-defective returned is significant and positive. Meaning that when people saw the green
advertising message, the utility for non-defective returned products is increased (0.135 +
0.033). However, the interaction effect for refurbished returned products is not significant.
Besides that, the interaction effect for the attribute levels of price were significant (see table
3), except for the attribute level with a discount of 30% (Discount_30). When the respondents
saw the green advertising message, the utility for 5% discount (Discount_5) was less
negative. However, for the attribute level of 10% discount (Discount_10), the utility became
more negative. Also, the utility for a discount of 20% decreased. For the other attributes and
attribute levels the moderating effect was not significant.
Attribute
Estimate
St.Error
z-value
Pr(>|z|)
State_non-defective
0.135
0.030
4.490
7.138e-06 ***
State_refurbished
-0.135
0.030
4.490
9.720e-06***
Highest_Quality
0.814
0.056
14.426
< 2.2e-16 ***
Warranty_1
-0.213
0.039
-5.178
2.244e-07 ***
Warranty_2
0.011
0.027
0.277
0.781
Warranty_3
0.202
0.038
5.196
3.472e-07***
Discount_5
-1.119
0.067
-16.600
< 2.2e-16 ***
Discount_10
-0.364
0.054
-6.671
2.549e-11 ***
Discount_20
0.474
0.048
9.867
< 2.2e-16 ***
Discount_30
1.010
0.049
20.317
4.526e-61***
Reputation_35
-0.820
0.052
-15.539
< 2.2e-16 ***
Reputation_40
0.307
0.040
7.611
2.731e-14 ***
Reputation_45
0.361
0.043
8.329
1.895e-15***
I(GreenAdvertising * State_non-defective)
-0.033
0.030
-4.489
3.471612e-07***
I(GreenAdvertising * State_refurbished)
-0.033
0.043
-0.753
0.4510321
I( GreenAdvertising * Warranty_1)
-0.035
0.059
-0.607
0.5433995
I( GreenAdvertising * Warranty_2)
0.039
0.056
0.698
0.4847683
I( GreenAdvertising * Warranty_3)
-0.003
0.055
-0.062
0.950402
I( GreenAdvertising * Discount_5)
0.223
0.094
2.363
0.0180847 *
I( GreenAdvertising * Discount_10)
-0.177
0.080
-2.203
0.0275559 *
I( GreenAdvertising * Discount_20)
-0.159
0.069
-2.302
0.0213044 *
I( GreenAdvertising *Discount_30)
0.114
0.069
1.636
0.1026784
I( GreenAdvertising * Reputation_35)
0.102
0.073
1.373
0.1695688
I( GreenAdvertising * Reputation_40)
-0.088
0.057
-1.521
0.1282993
I( GreenAdvertising * Reputation_45)
-0.014
0.061
-0.224
0.8221672
I(GreenAdvertising * None)
-0.046
0.103
-0.446
0.6554367
4.4 Effects on the quality perceptions
As mentioned, the results of multinomial logit model analyses of HQ-Model 2 are used to
investigate the effect of the attributes on the quality perceptions of returned products. The
estimated part-worth utilities are presented in table 4.
4.4.1 Product: State of the product
To investigate the effect of the attribute; state of the product (e.g. non-defective returned or
refurbished) on the quality perceptions of the returned product, the part-worth utilities of the
state_non-defective (e.g. non-defective returned) and state_refurbished (e.g. refurbished
returned product) are interpreted. The part-worth utility for attribute level; non-defective
returned, is significant and positive. Because the refurbished returned product state was the
reference level is therefore negative but also significant see table 4. These results are in line
with H3a: Non-defective returned products have a higher quality perception than refurbished
products and therefore and therefore this hypothesis is supported.
4.4.2 Product: Warranty
The attribute which is also related to “product” of the marketing mix is; warranty.
When we look at the estimates for the different attribute levels of the attribute warranty, we
see that strongest warranty (Warranty_3) showed the highest utility and the lowest level of
warranty (Warranty_1) showed the lowest utility (see table 4). Only a 2 years warranty
(warranty_2) is not significant, which means, because we used effect-coding, that it is not
significantly different from zero. Therefore, H3b: Offering a strong warranty for returned
products will positively influence the quality perceptions of returned products is supported.
4.4.3 Place: Retailer reputation
When we look at the attribute retailer reputation, which related to the “place” of the marketing
mix, we see that the higher or better the reputation of the retailer the higher the utility (see
table 4). These results are in line with H4: There is positive relationship between the
reputation of the retailer and the perceived quality perceptions of returned products and
therefore H4 is supported.
4.4.4 Price: Price
with a small price differential compared to the new product will result is a higher quality
perception of the returned product cannot be supported.
4.4.5 Moderating effects of green advertising on the attributes
The results indicate that the interaction effect for the attribute levels related to the state of the
product, are significant. Furthermore, we see that when people saw the green advertising
message, the utility for non-defective returned products is increased (0.403 + 0.106) and the
utility for refurbished returned products decreased even more (-0.403 - 0.453). Therefore,
using green advertising strengthens the relationship between the state of the product and the
quality perceptions. This is also true for the attribute; warranty. However, for the other
attributes the interaction effect, and therefore moderating effect of green advertising is not
significant
Table 4. Results HQ – Model 2
Attribute
Estimate
St.Error
z-value
Pr(>|z|)
4.5 Relative attribute importance
To get more valuable managerial implications, the part-worth utilities are transformed to
relative attribute importance. This based on the range of the utilities of the different attributes
relative to the total range of the attributes. The results are shown in figure 6 and 8 and table 5.
According to the results the attribute; seller reputation is the most important attribute for the
quality perceptions of the returned product (57.69%). Besides the seller reputation, the state of
the product, with 22,7%, is also an attribute for the quality perceptions of returned products.
However, warranty seems not be that important for the quality perceptions of returned
products. For the utility and demand of returned products the most important attribute is price
(45,27%). Besides the price, the reputation of the seller, with 35,86%, is also important for
customers for selecting and buying returned products. The state of the product seems not to be
that important for the utility and demand of returned products.
4.5.1 Moderating effect of green advertising on the attribute importance
The attribute importance including the moderating effect of green advertising are also
calculated and presented in table 5 and figure 7 and figure 9. The differences are very small.
The relative attribute importance of the reputation of the seller on the quality perceptions is
slightly decreased, because the importance for the remaining attributes are slightly increased,
but it is still the most important attribute in terms of quality perceptions. Furthermore, the
attribute price is becoming even more important for participants most preferred option after
they saw the green advertising message.
Table 5. Attribute importance quality
Figure 6. Relative attribute importance
highest quality
Figure 7. Moderating effect on the relative
attribute importance highest quality
Figure 8. Relative attribute importance
5. LATENT CLASS MODEL
A drawback of the multinomial logit model, where the estimates are based on the maximum
log-likelihood, that it aggregates all choices from all respondents and therefore produces one
set of utilities that represent all consumers. In other words, it does not allow for differences
between respondents and assumes that consumers are “clones” from each other. Besides that,
this is not very realistic, from a marketing perspective it is also better to allow for differences
between respondents and target different type of segments with different marketing strategies.
A model that assumes that consumers belong to segments is the Latent Class model.
Therefore, a Latent Class Model was conducted to create segments with different
characteristics within the sample.
5.1 Statements used for segmentation
Based on the environmentally friendly aspects of buying returned products, respondents were
asked to indicate to what extent they agreed with statements regarding environmentally
friendly behaviour. One of the statements was: “I care about the environment” and
respondents were asked to respond on a 1 – 5 Likert-scale ranging from strongly disagree to
strongly agree. To create different segments, a dummy variable (care_environment) was
created for this statement. When people who respond with 4-5 were classified as people who
cared about the environment (83%), and get a dummy value of 1, and people who respond
with 1-3 were classified as people who did not care about the environment (16%) and
therefore get a dummy value of 0. Another question that was used was used for segmentation,
was the question related to their experience with returned, refurbished or second-hand
products; “Have you bought a returned or refurbished or second-hand product (not
necessarily a watch) before?”. Where respondents could select the option, which applied to
them or choose for the “none of the above” option. The returned product option is used for
segmentation. Where 0 means that they have not bought a returned product before and 1
means that have bought a returned product before. Around 40% have ever bought a returned
product before and around 60% did not ever bought a returned product before.
5.1.2 Model fit
In order to investigate how many segments could be created by using these statements, there
were 3 different models performed with classes ranging from 2 – 4. HQ - Model 1 is the MNL
model without interaction effects and where the dependent variable was highest quality
option. HQ-LC2 is the Latent Class Model with 2 classes, HQ-LC3 is the Latent Class Model
with 3 classes and HQ-LC4 is the Latent Class Model with 4 classes. According to the
information criteria (see table 6) HQ-LC4, with 4 different segments, performed the best. This
HQ-LC4 model had also the highest Pseudo R2 and Pseudo adjusted R2 (see table 7).
Therefore, the results of Model 4 are interpreted and presented in table 8.
Table 6. Model fit Latent Class Models
HQ-Model 1
HQ-LC2
HQ-LC3
HQ-LC4
AIC
6532.643
5792.613
5305.894
5250.512
AIC3
6540.643
5811.613
5337.894
5294.512
BIC
6569.338
5883.425
5446.822
5443.555
CAIC
6578.338
5905.425
5481.822
5491.555
Table 7. Performance Latent Class Models
HQ-LC2
HQ-LC3
HQ-LC4
PseudoR2
0.375
0.431
0.439
PseudoR2 adjusted
0.371
0.421
0.429
5.1.3 Segment sizes
In order to investigate the sizes of the segments the probability of each respondent belonging
the different segments is calculated. The membership probability is based on the mean of
these probabilities (see figure 9). Based on the membership probabilities the segment sizes are
calculated (see figure 10). According to the results segment 1 is the largest segment and
segment 3 is the smallest segment.
Figure 9. Mean membership probability
Figure 10. Class sizes different segments
60 80 100 120