OMNICHANNEL RETAILING
IN THE FASHION INDUSTRY
>> Differences in consumers spending behaviour Offline and Online <<
ABSTRACT
This research investigates the differences in spending behaviour on fashion of consumers offline and online. The variables age, residence type, presence of a physical store, total spending of consumers and clothes size, are used to explain the proportion of online to total
spending of consumers. The multiple regression analysis finds that factors age, presence of physical store, residence type and clothes size have a significant influence on the spending
behaviour of consumers. Furthermore, an ordered logistic regression analysis is used to investigate the differences between consumer types. The variable consumer type is categorized into three types, namely physical, online and Omnichannel consumers. The main findings of this research are: (1) Older people spend a lower proportion of their total spending
on fashion online compared to younger people, (2) The presence of a physical store in the living area of a consumer decreases the proportion of online spending and, (3) Consumers who are living in a city spend less online compared to consumers living in a village. This research contributes to the knowledge and understanding of offline, online and Omnichannel
spending behaviour of consumers in the new retail environment.
Wilhelmina Elizabeth Martina van Toor MSc. in Business Administration | EMCI Track
MSc Ieva Rozentale 10002810 23-06-2016 Words | 14 820
STATEMENT OF ORIGINALITY
This document is written by Lisette van Toor who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
TABLE OF CONTENT
1. INTRO DUCTIO N 4
2. LITERATURE REVIEW 7
2.1 THE TRADITIONAL RETAIL FORMAT 7
2.2 A CHANGING RETAIL LANDSCAPE 7
2.2.1 Technology as cause for a new retail landscape 7
2.2.2 The retail landscape in a new style 10
2.2.3 New consumer behaviour 11
2.3 CUSTOMER-CENTRIC RETAIL 15
2.4 FACTORS EXPLAINING CONSUM ERS ONLINE VERSUS OFFLINE
SPENDINGS BEH AVIOUR 16
2.4.1 Online versus offline spending behaviour 16
2.4.2 Demographical factor as age 18
2.4.3 Presence of a physical store in consumers surround area 19
2.4.4 Consumer’s residence type 20
2.4.5 Consumers total spending 21
2.4.6 Clothes size 21
3. CONCEPTUAL M ODEL 22
4. M ETH ODOLOGY 22
4.1 Research design 22
4.2 Data collection and sample description 23
4.3 Operationalization 23
5. RESULTS 25
5.1 Descriptive statistics 25
5.2 Analysis 27
5.2.1 Multiple Linear Regression Model 27
5.3.2 Ordered Logistic Regression Model 29
6. DISCUSSION 35 6.1 Hypotheses 36 6.2 Limitations 38 6.3 Managerial contributions 39 6.4 Future research 39 7. CONCLUSION 40
8. ACKNOW LEDGEM ENTS 40
REFERENCES 41
APPENDICES 46
Correlations 46
Test for multicollinearity 46
1. INTRODUCTION
71 % Of U.S. consumers say they still prefer to buy from a physical store, according to the TimeTrade survey of the U.S. Census Bureau (2015). This consumer research also finds that 85 % of the sample likes to shop in physical stores because they want to "touch and feel" items before buying them. However, consumers nowadays also seek improvements in both in-store experience and a seamless shopping process, which includes the ability to shop across channels (Accenture, 2015). The data on online spending shows a positive picture for the online retailer (CBS, 2015). On average, 53 % of the European consumers bought at least once a year in an online store. In the last ten years this percentage has doubled. The United Kingdom has the highest percentage of online shoppers in Europe, namely 81 % of the population is an e-shopper (CBS, 2015). To meet the preferences of consumers and to follow the online developments, companies are looking for Omnichannel experts and Consumer intelligence specialists. Omnichannel consumers refer to consumers who buy both on- and offline (Brynjolfsson & Rahman, 2013). Knowledge of their consumer could give them a competitive advantage and provides them on the long term the right to exist (Sato & Huang, 2015).
Retailers primarily sell products manufactured by others and create competitive advantage from exclusivity in their assortment (Sorescu et al., 2011). They always engage in direct interactions with end consumers. The retailer business model is always customer-centric and thus based on commercial value, in contrast to business models of brand-only businesses, which are mostly based on creative value and originality. The business model of a retailer is therefore only as good as the assumptions the retailer makes about what consumers value. Understanding the consumer characteristics and behaviour is thus the number one task for retailers.
Developments in technology are in this research considered as main cause of the changes in the retail landscape. These technological developments are mainly related to the introduction of the Internet and digital devices, such as tablets and mobile phones. Through these new technologies new shopping channels, such as web shops and mobile shopping, have emerged and these channels could be considered danger for the traditional shopping channels (e.g. physical stores). This can be explained by retailers who expected that consumers would choose between the different channels instead of using them next to each other. Even though retailers would be better off if they see all channels holistically together and take advantage of all touch points with the consumer, including mobile devices and social networks. Within
these channels, the physical store, as a hub linking different channels, is becoming the source of value creation (Blázquez, 2014).
The fashion industry struggled in the beginning of the digital era to translate the in-store experience to the online environment, since the physical look and feel of the product is valuable in the deciding process of customers of which fashion items to buy, which appeared to be hard to imitate online. Therefore, e-commerce is slower adopted in this industry compared to other industries (Blázquez, 2014). Luxury fashion players have been very cautious about digital and e-commerce in the past and e-commerce has even more been seen as a threat. New players from adjacent industries like Amazon, Bol.com and Tesla are setting the bar in terms of digital and Omnichannel experience, but the fashion industry is starting to become a serious player in e-commerce now as well. New information technologies have enabled consumers to evaluate fashion online, by using mannequins online who visually show the fit of clothes and possibilities to chat with sales advisors virtually. As a result, online shopping grows more and more.
The initial predictions of retailers were that online shopping would crowd out offline shopping. Previous research of Blázquez (2014) has proved that consumers do not choose between online and offline. As both channels have different strengths, the channels are considered complementary. In the offline world of shopping people usually go to a store for fun and feel the experience of the brand or retailer. You can get advice from a sales associate, try the clothes you like and have finally a higher chance of a successful purchase. On the other side in the online world of shopping, the shopping process is considered to be faster, more efficient and without distraction of possible sales associates or other people. Online, consumers have a clear overview of the products available and they can easily compare prices. Next to shopping at a computer or laptop, shopping by mobile devices also becomes popular (Wang, Malthouse & Krishnamurthi, 2015), which is referred to as M-shopping.
The differences in shopping characteristics between these two worlds show opportunities for both channels. What is still unclear in the literature is in which way consumers combine online and offline shopping. Is the division in online and offline shopping influenced by age for example or by distance to physical store? Consumer intelligence is the process of gathering and analyzing information regarding customers; their details and their activities, in order to build deeper and more effective customer relationships and improve strategic decision making (Wikipedia, 2016). Retailers gain advantage from getting to know their consumer. Therefore an increase of knowledge of three different consumers: online, offline and Omnichannel consumers could be of value for them.
The aim of this study is to gain a better understanding of different consumer types; physical, Omnichannel and online consumers. The research question is stated as follows: Which factors explain the difference between consumers’ spending behaviour offline and online in fashion retail? Especially for retailers it is valuable to know what drives consumers to shop either more online or offline.
In order to answer this question, a quantitative data analysis is performed using the data on consumer behaviour from a Dutch fashion company, which has 28 physical stores and a web shop. This research focuses on demographical, geographical factors, economic factors and consumer-specific variables, which are expected to explain different spending behaviours offline and online. More precisely, “Consumers’ age”, “Residence type”, “Presence of physical store”, “total spending” and “clothes size” are used as the independent variables to explain the variation in “proportion online to total spending”.
First of all, Parment (2013) found that different generations have different Internet habits and different shopping behaviour. Based on his findings, this research expects age to be a factor of explanation. Younger consumers are more often active online and are therefore more comfortable with shopping online compared to older consumers. The second and third factor, residence type and presence of a physical store, are geographical factors, and were found as relevant indicators for the offline and online proportion of consumers in research of Forman, Ghose and Goldfarb (2009). In their research data of Amazon.com was analysed, so that the results said something about the book market. They found that when a store opens locally, people substitute away from online purchasing, which is the base for the geographical part of this research. We expect that living near a physical store could influence the proportion of online shopping for the fashion market as well. This research also analyses the total spending of consumers in relation to their online and offline shopping proportion, wherefore the knowledge on high spenders as luxury and compulsive consumers are taken into consideration. Finally, the average clothes size is analysed, since most consumers with a different size than average feel the need to find a sales associate who can provide them from advice (Li & Yang, 2014). Therefore clothes size is expected to be a factor that influences consumer’s decision between shopping channels finally as well.
Furthermore, the academic relevance of this study is related to consumer knowledge in a new retail environment. Since these developments in retail are quite recent and even still happening, not a lot of research before has investigated the same subjects. It is expected that in the upcoming years extensive research will follow this study to broaden the knowledge of consumers, as the developments in retail are currently taking place.
Practically, this research is relevant for fashion retailers, as they are highly dependent on their consumers. A better understanding of the consumers could help them to establish a strategy that better suits their company and ultimately lead to a more profitable business.
2. LITERATURE REVIEW
The traditional retail format has been changed through the development of technology and has led to the Omnichannel retailing we know nowadays. The definition of a retailer is “a business person that sells goods to the consumer, as opposed to a wholesaler or supplier, who normally sell their goods to another business” (Business Dictionary, 2016). In this literature review firstly the traditional retail format is explained, followed by a discussion on how the technology developments change this format. Thirdly, the consequences of technology are explained, which are twofold, namely the consequences for the retail landscape and the consequences for consumer behaviour. Lastly, the new retail format of Omnichannel retailing is explained in combination with previous research about online, offline consumers and Omnichannel consumers in this landscape are explained. The gap in knowledge of these different consumers is where the analysis of this research starts.
2.1 THE TRADITIONAL RETAIL FORMAT
The traditional retailers primarily sold products manufactured by others and created competitive advantage from exclusivity in their assortment (Sorescu et al., 2011). The only way to get the products for customers was to physically go to their store. Shopping in general was more personal compared to nowadays, whereby the sales representatives knew their consumers and the consumers appreciated their advice, as this was both their only reference point as well as the only way to get a professional opinion. Similar as nowadays, all season items that weren’t sold during the season, were sold during sale-periods with a discount. Because of significant expenditures such as rents and personnel costs, the traditional retailing format entails high costs and relatively low profits (Li & Yang, 2014). In the last two decennia, new retail formats emerged, that have quite different characteristics.
2.2 A CHANGING RETAIL LANDSCAPE 2.2.1 Technology as cause for a retail change
Development in technology can be seen as main cause for the changes in the retail landscape (Brynjolfsson & Rahman, 2013). Several changes are driven by the increasing presence of the Internet in every minute of consumer’s lives. The ubiquity of Wi-Fi and 3G networks, as well
as the number of devices available to access these networks, ensure that consumers can quickly and easily access the Internet, regardless of space, time or location (Grewal, Roggeveen, Compeau & Levy, 2012). Traditional retailers had to deal with barriers such as geography to establish their positions in the market. However, technology (e.g. Internet) has removed these barriers (Brynjolfsson & Rahman, 2013). In this section the different technology developments related to retail are explained.
The first and most radical development is the web shop. This has created a shift from part of the offline sales to online. While the fashion industry was a relatively late adaptor of e-commerce, online buying in fashion currently experiences significant growth (Cormick et al., 2014). Fashion is currently the fastest-growing online category of goods bought in the United Kingdom. The growth in online shopping is enhanced by personalized e-mails with recommendations based on consumer’s prior browsing and shopping experiences in the web shop.
Another technological change came in the form of mobile developments for smartphones and tablets (Grewal et al., 2012). For example, Amazon offers a free app to their consumers to help them follow daily deals, recently released products and other functions. This app stimulates consumers to buy in the online store of Amazon. On the other side websites and apps could encourage consumers to go into physical stores as well, for example by codes consumers receive on their phone which they can validate in store. Through mobile developments companies can get in touch with customers by an e-mail notice, an alert on smartphone, an app and Facebook, all in the same day. This explains the importance for companies to fully leverage their mobile platforms, since this lead to convenient access to their consumers (Wang, Malthouse and Krishnamurthi, 2015).
Technology has also changed the shopping experience in physical stores. Nowadays, displays and tablets are used in store windows and in the stores to show behind-the-scene movies or fashion shows. Stores also make use of technology through virtual mirrors. This is an on-going development, which shows up more and more frequently, for example in stores of brands as Zara. Examples of virtual mirrors are presented in figure 1.
Figure 1: Virtual mirrors
Wang, Malthouse and Krishnamurthi (2015) refer to these developments as M-shopping. M-shopper stands for mobile-assisted shopper, which means that these consumers visit a store with their mobile device to compare prices on Internet and find the best deal available. Mobile phones and their related apps have made consumers more aware of price promotions, because they can check the price of a shopping item even while they are physically in the store. Of course, this development of comparing prices already happened online, but when customers check prices of products they see in store, it can directly influence consumer’s decision between online and offline purchasing.
Another development which influences fashion retail is Social media. Social media are online applications, platforms and media, which aim to facilitate interactions, collaborations and the sharing of content (Richter & Koch, 2007). For example networks such as Facebook and Twitter enable customers to share information and group buying, which again impacts the retail branche. The use of social networks increases exponentially (Kim & Ko, 2012) by both a growing number of individual numbers as well as business using them as communication tool. Findings suggest that social media plays an important role in communicating information to customers (Agnihotri, Dingus, Hu & Krush, 2015). Especially fashion companies primarily use the media for advertising and marketing. They are willing to pay for extra promotion of their public company pages, in order to increase their online presence, which could lead to an increase in sales. According to Hoffman and Fodor (2010) it is hard to measure the ROI of social media marketing in the fashion industry.
Finally, Internet impacts consumers shopping decisions because of the existence of product reviews, recommendations and ratings. The research of Chen, Wu and Yoon (2004) found that more recommendations positively correlate with sales. As well as the number of consumer reviews who are positively associated with sales. However, they did not find a significant relationship between consumer ratings and sales.
On the other hand, there are also some disadvantages related to technology, in the fashion industry specifically. The mobile phone is not the optimal channel to launch new products or promote products that require more consideration during the buying process (Chesbrough, 2007). The main reason for this is that pictures of the product can only be presented in a relatively small format. Another disadvantage is that it is not possible to try and touch the product. Apart from these general disadvantages, another downside of the mobile developments is that not all consumers understand the new technology; especially older generations are not familiar with shopping online. As a result of these disadvantages investments in development of apps and other might be considered less attractive.
2.2.2 The retail landscape nowadays
The research of Cao (2014) explains the process of retailers who have to switch from an old single-channel retail model to the new cross-channel model, because of the change in retail landscape. The study has developed a model that explains the change from a single-channel retailer to an Omnichannel retailer, which is called the stage-of-adoption model. This model consists of five stages, whereby the retailer becomes more and more an Omnichannel player. The first stage in this model is the (1) solo mode, in which the retailer adopted a multichannel strategy but used independent business models for different channels. The solo mode is followed by minimal integration, which means when major front-office operations of the business model were the different yet complementary across channels, with the retailers focusing on the development of online business. A third stage is the moderate integration, where the retailer integrated some operational activities across channels for optimization and invested in the back office to support the strategic shift. Full integration is the fourth stage, were the retailer aligned the offering across channels, so same prices offline and online, provided a seamless shopping experience to consumers, restructuring the organization to adapt to the full integration of activities across channels. Finally there is the new business model, where the retailer redefined its role in the whole value chain of the sector by changing its profit formula or co-creating value with other stakeholders. To facilitate the strategic shift, retailers should optimize rather than merge their activities across channels, reinforce the
strengths of the physical store and develop co-creation with stakeholders. Integration of multiple channels can provide consumers advantages as increasing confidence, convenience, control, assistance and customization as well as a sense of safety.
Brynjolfsson and Rahman (2013) argue that retailers should begin by adapting best practices form offline and online worlds in areas, including pricing, designing the shopping experience and building relationships with consumers. According to the authors, several successful strategies for Omnichannel retailing in the new competitive environment are possible. First of all, providing attractive pricing and curated content, which means avoid price wars, but focus on good prices wherefore consumers come to you and in exchange organize merchandising so that consumers will not get lost in a sea of products. Secondly, to maintain the consumers, retailers should create switching costs in the form of loyalty programs or discounts for loyal consumers. Thirdly, avoid direct price comparisons by creating distinctive features and focus on exclusivity. An offering of distinctive or customer-made versions of a product for example lead to differentiation and therefore lowers price competition. Furthermore, retailers may want to focus on product development partnerships and innovations to create exclusive products. Finally, and most important nowadays is using the power of data and analytics. This is better known as Consumer Intelligence (CI).
Consumer intelligence is information about consumers that a company collects and uses to help make future plans. It can provide insights that can change and organization’s marketing strategy (Cambridge Dictionaries, 2016). As mentioned before, retailers with the most accurate knowledge of consumers are able to adjust to the consumer preferences resulting in increase in sales. New data became available from social, mobile and local channels in particular, which makes it possible to not only understand customer transactions but also customer interactions such as visits to the store, likes on Facebook, searches on websites and check-ins at nearby establishments (Brynjolfsson & Rahman, 2013).
2.2.3 Consumer behaviour in a different retail environment
Due to the technology developments, not only retail operations have changed. Rosenblum and Rowen (2012) show that it also had an impact on consumer behaviour. They find that the digital revolution has created empowered consumers with high expectations. The retail landscape does not any longer consist of one single category of consumers, but consumers can be divided in three main categories: physical consumers, online consumers and Omnichannel consumers.
(1) First of all the traditional consumers who visit the physical store, also mentioned as offline consumer in this research. Offline consumers are mostly looking for experience. These consumers have an on going, hobby-type interest and have time available. Experimental motives play a more important role in offline shopping, since the ability to build and sustain a fantasy in offline environments is more likely then in the online environment. Consumers are more likely to select a physical store when they shop for hedonic fashion goods because strong physical environments elevate the mood through opportunities for social interaction, product evaluation and sensory stimulation (Nicholson, Clarke & Blakemore, 2002). Furthermore, offline consumers often switch to another brand when there is lack of hedonic or pleasant value, despite being highly satisfied with utilitarian needs such as quality products and fair prices (Faison, 1977; Jones and Sasser, 1995). Since shopping for fun consumers are in general more loyal to a brand or store, retailers should exploit the hedonic potential of their store (Scarpi, Pizzi & Visentin, 2013).
Furthermore, 70% of consumers use Internet for finding information about fashion products; only 16% of these consumers buy those products online (Vasiliu & Cercel, 2015). This behaviour can be explained by the lack of trust in the online retailer or because in this type of retail trying is very important to purchase decision. These consumers do not like to take a risk and feel the need of physical interaction with the store and the desired product. This explains the high percentage, 27.9% of consumers who prefer the traditional way of commerce and refuse the use of new technologies in the process of buying fashion products. The study of Benedicktus, Brady, Darke and Voorhees (2010) found that trust and purchase intention increases when a physical store exists. Furthermore, consumers make choices depending on their overall costs. Consumers take savings in transportation costs and time into consideration when deciding whether they want to shop offline or not (Wang et al., 2015).
(2) The second category of consumer is the consumer who shops online. Online consumers are characterized by their goal-oriented mind. There are two ways of shopping; goal-oriented and experimental (the last is mentioned previously for offline consumers). The goal-oriented, or utilitarian consumers are task-oriented, efficient, rational and deliberate (Wolfinbarger & Gilly, 2001). These shoppers are transaction-oriented and desire to purchase what they want quickly and without distraction. Online they have a better overview of prices available and more control over their own shopping process compared to offline. In the research of Scarpi, Pizzi and Visentin (2013) is found that online shopping could have more than utilitarian reasons only. Namely, shopping for fun can be online the case as well. They found that consumers shopping for fun enjoy deal hunting and spending more time shopping
online; whereas when consumers shop for needs they actively search for lower prices and a more efficient money allocation, especially because price comparisons are much faster and easier online than offline. Fashion consumers prefer to have a clear overview of all the brands available on the market and are therefore more likely online consumers (Kukar-Kinney et al., 2009)
Furthermore, the order rate of online consumers who use M-shopping is higher compared to offline consumers; which mean that they place higher number of orders per year. Consumers doing M-shopping shop online not only using their computer at home, but also using their smartphones or tablets when they are in the store (Wang et al., 2015). They compose, modify or place orders on these devices. Consumers utilize mobile devices because the technology provides convenient access, which leads them to incorporate M-shopping into their habitual routines.
Another specific group of online consumers is determined in the literature, named compulsive buyers. Kukar-Kinney et al. (2009) indicate that as the compulsive buying tendency increased, consumers were more strongly motivated to buy on the Internet compared to a more traditional retail store environment. This conclusion is in contrast with previous research, which indicated that the lack of social interaction on the Internet was a limitation of online shopping (Nicholson, Clarke & Blakemore, 2002). Compulsive buyers have been found to experience shame, guilt and regret because of their frequent buying episodes (Kukar-Kinney et al., 2009). Because of these feelings, compulsive buyers may not want others to see what, how frequently and how much they buy. As a consequence, these consumers may feel the need to hide their buying activities and they may fear that instant recognition by sales clerks labels them as buyers who buy too often. Internet retail environment is a solution for this group. Internet enables consumers to be alone while shopping and buying and offers a low to non-existent level of social interaction.
(3) The third category is the Omnichannel consumer. These consumers do not choose between channels, but use them all. In previous research is found that Omnichannel consumers on average spend more money and buy more frequently (Lu & Rucker, 2006). Besides that these consumers are more loyal to the brand or retailer. Kushwaha and Shankar (2013) agree on that and state that Omnichannel customers have a significantly higher monetary value than single-channel customers. The average Omnichannel customer outspends the average catalog- and web-only customer by $60.13 and $108.92 respectively (Kushwaha and Shankar, 2013). An explanation for this outcome is that customers who prefer multiple channels may become more engaged in the purchase process as they shop across channels.
Greater engagement may lead to more frequent purchases, a greater order quantity and greater spending. Therefore these consumers are really important for the retailer and should be maintained. However these consumers are also more demanding and expect more from their shopping experience. Their shopping behavior is more exploratory, seeking more variety than consumers who buy in a single channel. Furthermore, Omnichannel consumers consider their shopping experience holistically and look for an integrated and consistent experience between channels. They combine the channels and made decisions based on their mood and lifestyle demands. In fashion, the consumer’s mood is the key determinant in making the choice between channels.
Besides the consumer behavior that has changed in shopping, behavior has also changed through the development of Social Media. Consumers are more influenced by other consumers or influencing others with their opinion. Furthermore, bloggers and vloggers (video-bloggers) are playing an important role in fashion shopping (Chesbrough, 2007). In order to reach potential customers, famous bloggers are paid by brands to wear their clothes. This works in the same way for famous Instagram-accounts. Through social media consumers become a part of the marketing of brands. They can join the world around the brand by posting own pictures on Social Media and join for example photo competitions. For retailers this is a new way of doing marketing. The brand or retailer that comes up with the most creative Social Media campaigns is the big player in the fashion industry nowadays.
2.3 CUSTOMER-CENTRIC RETAIL
Traditional models of retail and fashion brands have always had a product-centric strategy, especially the luxury fashion brands competing on the best quality products. However, because of the development of the web shop and mobile shopping today’s retail environment is more competitive than ever before (Rosenblum & Rowen, 2012). Companies had to adjust to the consumer’s preferences as much as possible to get the consumers to their store. This explains the shift from product-centric retailing to consumer-centric retailing, which can be seen as a radical change. For most retail companies, one of the main key performance indicators (KPI) is nowadays customer satisfaction. They have to offer the right products, at the right time and in the right place, and more important they should understand the customer lifestyle, experience and values (Corbellini & Saviolo, 2015).
The competition in the fashion industry has been always intense because of a lot of active players and in particular nowadays because of the huge retail chains as Zara and H&M who take a large market share. Among fashion retailers an increasing focus is presented on consumer acquisition, consumer retention and consumer development. Consumer acquisition can be done by providing customers the possibility to create your own product (e.g. customer-made scarf by Louis Vuitton, customer-customer-made running shoes by Nike etc.). Another way to acquire more customers are social media campaigns (e.g. Picture contest by Adidas, Live in Levi’s Instagram campaign by Levi’s etc.). Once acquired, companies want to retain their customers. Consumer retention is most often related to loyalty programs such as a special invitation for loyal consumers, a consumer specialty card for which the consumers could get extra discounts (e.g. Zegna World Club Pass) or personal shop sessions. Furthermore, high service quality offered by the staff of physical stores leads to an increase in trust in the retailer and the quality of service has been found as an important component in establishing and retaining customers (Al-hawari & Mouakket, 2012).
Finally, retailers should try to think of the life stage of consumer, to better understand the wants and needs. This means think about where the consumer is in life. For example, will he or she get married soon? In this case the company could send an email or brochure with suggestions for dresses or jewelleries. This last tool is especially relevant when a company is part of a larger group, for example LVMH-group, which consists of different brands. At such companies, an information flow can go from brand for younger consumers to brands with an older target group when a consumer becomes older they receive for example an email from the brand with the older target group. In Figure 2 the information flow between three brands of LVMH is shown.
Figure 3: Information-flow in a fashion group
2.4 FACTORS EXPLAINING CONSUMERS ONLINE VERSUS OFFLINE SPENDINGS BEHAVIOUR
2.4.1 Online versus offline spending behaviour
The central part of this research is the proportion of consumers online spending compared to their total spending:
𝐶𝑜𝑛𝑠𝑢𝑚𝑒𝑟 𝑜𝑛𝑙𝑖𝑛𝑒 𝑠𝑝𝑒𝑛𝑑𝑖𝑛𝑔
𝐶𝑜𝑛𝑠𝑢𝑚𝑒𝑟 𝑡𝑜𝑡𝑎𝑙 𝑠𝑝𝑒𝑛𝑑𝑖𝑛𝑔 = % 𝑜𝑛𝑙𝑖𝑛𝑒 𝑠𝑝𝑒𝑛𝑑𝑖𝑛𝑔
In 2016, online shopping accounts for 9 percent of consumers total retail spending, according to the most recent figures of worldwide consumption from the U.S. Census Bureau (U.S. Census Bureau, 2016).
Figure 4: Online versus offline consumption (McKinsey, 2015)
Currently it is only one tenth of total retail spending, however this
proportion of online spending increases still every year. The graph on the right shows the online and offline spending over years with a clear growth of online spending worldwide (McKinsey, 2015). In figure 5 of “Centraal Bureau voor de Statistiek” (2015) the proportion of consumers who bought something online in 2015 is shown. On average 53 % of the European Union consumers is an E-shopper, which refers to someone who has bought at least once a year a product online. For the United Kingdom, Denmark, Luxembourg, Germany and The Netherlands this percentage is significantly higher, namely from 71 % of e-shoppers in the Dutch populations to 81 % in the United Kingdom (CBS, 2016).
Figure 5: Online shopping in Europe (CBS, 2016)
All retailers have to adapt to the digital world to meet the desires of (online) consumers. They have to deal with the three types of consumers as explained in previous paragraph: physical consumer, Omnichannel consumer and online consumer. The difference in spending behaviour between these consumer types is clear, however the question is if these consumers have different characteristics or are they all quite similar? This paragraph contains the available knowledge on five different subjects; age, residence, store presence, total spending and clothes size. Furthermore, this paragraph contains the hypotheses compiled based on the available literature.
2.4.2 Demographical factor as age
Since age groups are in a different life stage, have different preferences and different values, age could be a first factor that influences the proportion online and offline spending. Parment (2013) finds significant differences in shopping preferences between generations. The last four named generations are the generation of Babyboomers, generation X, Y and Z. The Babyboomers are born after the Second World War, so approximately between 1950 and 1960. Generation X is the generation born between 1960 and 1980 and are connected to the pop culture of the 80s and 90s they grew up in. Generation Y, also referred as the Millennials, are the ones born between 1980 and 1999. This generation grew up with world changing inventions as mass communication and the Internet. Finally generation Z is the generation of people living in Western or First World cultures that follows Generation Y. A previous study compared the Babyboomers with generation Y consumers, with respect to their shopping behaviour and purchase involvement for food, clothing and automobiles and found that for the three types of products, Babyboomers values the retail experience and in-store service higher than Generation Y (Parment, 2013). For Babyboomers, the purchase process starts with a retailer that consumer trusts and who gives advice for choosing the right product, while for Generation Y, the purchase process starts with choosing a product. Generation Y is also called the digital natives, because they are the first generation who spend their entire lives in the digital environment (Bolton et al., 2013), which leads to higher convenience in using online devices. However a research of McKinsey (2015) found that in general Babyboomers and Generation Y consumers spend approximately the same amount of time on the Internet, which is around 15 hours a week.
Since usage of technology devices becomes more important in the online shopping behaviour of consumers, the difference in usage between generations are important here as well. Figure 3 shows a comparison of the usage of mobile devices, time spending on the Internet and Social Media usage between Generation Y and Babyboomers. However the differences are not that significant, there is still a difference in technology usage between these generations. For example, the Social Media usage differs by 26 % between Generation Y and Babyboomers (McKinsey, 2015).
Figure 6: Comparing generation Y with
Babyboomers in relation to technology usage (McKinsey, 2015).
Based on the literature, this research expects that there is a difference between age groups in using Internet for shopping. Therefore the following hypothesis is created:
H1: Consumers of lower age are spending more online compared to older consumers.
2.4.3 Presence of a physical store in consumers surround area
According to several researchers the difference between online and offline spending could be influenced by the availability of physical stores in the nearby area. First of all, adding a physical store to the existing channels of a brand as catalog and Internet impacts the sales of the brand (Pauwels & Neslin, 2015). The physical store introduction increases the accessibility of the brand, which leads to an increase in purchase frequency of consumers, but on the other hand a decrease in Internet spendings. Forman, Ghose and Goldfarb (2009) studied the competition between offline and online book markets by analysing data from Amazon and found a similar effect. Namely, that when a store opens locally, people substitute away from online purchasing, which is known as cannibalization-effect.
One could argue however that the opposite relationship could also be true - the presence of a physical store could increase online sales. In this case, the retail store can be seen as an image builder. A flagship or concept store is the ultimate communication vehicle for a brand, where product quality, heritage and service combine in the manifestation of a brand’s DNA (Deloitte, 2014). The presence of a physical store can also lead to a first acquaintance of a consumer with the brand. Furthermore, a store is where the brand can provide multisensory experience to the consumer, so that the consumer can identify and aspire to various attributes (McKinsey, 2015). Afterwards this can lead to online purchases, thanks
to existence of a physical store. The web shop benefits in this way from the intensive network build by the stores.
Both theories follow a logical reasoning and could be the case; however in this research we will follow the first theory, since this has a better academic foundation. Therefore the expectation here is that the presence of a physical store will trigger consumers more to buy in that store instead of leading them to the web shop. The following hypothesis will be tested:
H2: The presence of a physical in the surrounding area of consumers’ residence decreases the proportion of online spending to total spending.
2.4.4 Consumer’s residence type
In this research two residence types are taken into account, namely village versus city. Consumers living in a village could differ in shopping behaviour from consumers living in a city for the following reasons:
Firstly, consumer’s transportation costs to come to a store is one of the factors that determine consumers choice of channels (Forman et al., 2009). The presence of transportation costs for offline shopping are often put in front of the online search costs. The transportation costs have thus a negative impact on physical consumption. Balasubramanian, Konana and Menon (2003) emphasize how Internet retailing provides a convenient substitute to local retailing when there are transportation costs. Consumers living in a village have to deal more often with transportation costs, which could result in a change to Internet shopping. Secondly, time to store can also be an obstacle for consumers to shop in a physical store. The same reasoning as for transportation costs applies for time to store as distance and time are related to eachother. Finally, as explained in previous paragraph when a store opens in a consumers living area, these consumers substitute away from online purchasing (Forman et al., 2009). This will happen more often to consumers living in a city, since the quatitiy of stores here is higher.
All three points are showing advantages of online consumption for consumers living in a village and advantages of physical consumption for consumers living in city. Therefore the following expectation is assumed:
H3: Consumers living in a city have a lower proportion of online spending compared to consumers living in a village.
2.4.5 Consumers total spending
In this research consumer’s total spending is the total quantity a consumer spends across all channels. To investigate the relationship between total spending and online versus offline consumption, the existing knowledge around high spending consumers is explained in this paragraph.
The first group of consumers with high spending are luxury consumers. Luxury good consumers spend a high amount of money and since the products they are looking for are of high value they prefer to check the products in the store instead of buying online (Deloitte, 2014). Besides that luxury products are mostly related to a broader experience that is still only present in store, although luxury brand are trying more and more to translate this experience into the digital environment (Blázquez, 2014).
The second group is the compulsive buyer. Compulsive buying does not appear to be linked to income (Dittmand, Long and Bond, 2007). Motivations as buying unobserved and avoiding social interactions during shopping are important factors for compulsive consumers (Kukar-Kinney, Ridgway and Monroe, 2009). The tendency to buy compulsively increases the preferences for Internet shopping, so therefore a compulsive consumer with high spending will buy more online compared to a non-compulsive consumer with lower spending.
However luxury consumers are important to take into account in this section, they represent a small percentage of the population. Therefore the fourth hypothesis is based on the theory behind the second group of high spenders:
H4: Consumers with higher annual spending, spend a lower proportion online compared to consumers with lower total spending.
2.4.6 Clothes size
The best selling goods on the Internet nowadays are clothes and shoes (Li & Yang, 2014). One can purchase a wide variety of styles online and it is very convenient that you do not need to stroll the streets tiredly. Nevertheless, finding the right clothes size online can be a problem and therefore an obstacle for consumers. Especially people with a size that is smaller or larger than the average sizes have a great need to find the seller who can provide them from advice or custom-made clothes. Since we expect that having a smaller size than average in the Dutch population to be less common, the hypotheses is based on having a larger size. The hypothesis about clothes size in relation to online and offline shopping is the following:
H5: Consumers with a larger size are spending more in a physical store compared to consumers with a smaller size.
3. CONCEPTUAL MODEL
Figure 7: Conceptual model of research on consumers spending behaviour
4. METHODOLOGY
This paragraph consists of the following sections: research design, data collection and the operationalization of the variables.
4.1 Research design
The purpose of this research is to find out weather the factors discussed can explain the variance in consumer online versus offline spending behaviour. Hence this research is explanatory. In order to answer the question, we have chosen a quantitative analysis with consumers as the unit of analysis. Therefore numerical data are collected and will be analyzed. The setting of this research is a fashion company with female consumers.
The dependent variable proportion of online to total spending is predicted by using a multiple linear regression model. ‘Multiple’ because this research contains more then one independent variable. A multiple linear regression is the most common form of linear regression analysis. As a predictive analysis, the multiple linear regression analysis is used to explain the relationship between one continuous dependent variable and two or more independent variables. The independent variables can be continuous or categorical. Other than correlation
analysis, which focuses on the strength of the relationship between two or more variables, regression analysis assumes a dependence or causal relationship between one or more independent and one dependent variable. Furthermore, an ordered regression analysis is executed to explain the differences between consumer types. Three consumer types are taken into consideration, which are physical, Omnichannel and online consumers. Ordered logistic regression is a class of regression where the independent variable is used to predict the dependent variable, in which the dependent variable categories are ranked.
4.2 Data collection and sample description
The data of this research is collected by means of convenience sampling, which is a type of non-probability sampling. This is because the sample is drawn from the part of the population, fashion retail companies, that are close to hand. Data is obtained from a fashion retail company, to which the researcher has access. This company consists of 28 physical stores based in the Netherlands and a web shop. The web shop comprises approximately 15% of total sales. Primary (raw) data was offered by the financial manager of the company, which consists of all the purchases that were made in all stores in 2015. In total, the dataset consists of 587 075 individually purchases. These individual purchases are computed per consumer to create the possibility to do the analyses on consumer-level. Computing the different purchases was done based on the existence of unique consumer (id) numbers in the system. Subsequently, the unit of analysis became consumers. After computing the individual purchases, the dataset consists of 75 501 consumers, which means a large dataset and is therefore useful to do the reliable analysis. No further sample is taken from this dataset, hence the whole population was analysed. The data is cross-sectional.
4.3 Operationalization
This study includes seven variables – one dependent variable and six independent variables. The dependent variable represents the percentage of online to total spending. Consumer’s total spending consists of their online and offline purchases together. This research tries to explain the proportion of online spending with five independent variables.
Proportion Online of Total Purchases is the dependent variable (Y), which is measured in percentages of online purchases divided by consumers’ total purchases. This data is on continuous ratio level. For many cases the value is zero, but this does not mean that this consumer did not buy anything. This means that this consumer bought only in an offline store in 2015. Furthermore, for offline as well as for online purchases a lot of negative values where present in the database, which are returns and possible customer compensations. Since
this research focuses on where consumers buy their clothes, information around these items do not contribute to this and therefore this data is deleted from the dataset in Excel. In this way the dataset consists only of real purchases offline and online.
Age is the first independent variable (X1). In most analyses age is a continuous
variable. However, this research also describes the generational differences, whereby age becomes a nominal variable consisting of five categories.
Presence of Physical Store is the second independent variable (X2). This variable is a
dummy variable that takes the values 0 or 1 to indicate the absence or presence of a physical store in the surrounding area of consumers’ residence.
Residence type (X3) is the third independent variable in this research and is about the
type of residence a consumer lives. The values of this variable can have two categories 0 and 1, for which 1 means city and 0 means no-city consumer. In this way this variable becomes a dummy variable.
Total spending (X4) is measured in euros, which means this is a continuous ratio variable. The total spending is the sum of online and offline purchases of a consumer in that year.
Clothes size (X5) is the last independent variable that is taken into account and is
measured as discrete ordinal data. The question is if a consumers’ size influences the decision to buy online or to go to a store. Table 2 below is created to align the different size methods in fashion. For this research the Dutch sizing with size scale from 32 to 52 is used, although most common sizes are from size 36 up to 46. For every consumer the average size the consumer bought in 2015 is calculated in Excel.
Table 1: Size method per country, focus-point the Dutch sizing method
S-M-L NL USA UK IT FR GE SP Jeans XXS 32 - 2 4 38 34 32 0 24-25 XS & XS/S 34 0 4 6 40 36 34 1 26 S 36 I 6 8 42 38 36 2 27 S/M & M 38 II 8 10 44 40 38 3 28-29 M/L & L 40 III 10 12 46 42 40 4 30 L/XL 42 IV 12 14 48 44 42 5 31 XL 44 V 14 16 50 46 44 6 32 XXL 46 VI 16 18 52 48 46 7 33 XXXL 48 VII 18 20 54 50 48 8 34 XXXXL 50 VIII 20 22 56 52 50 9 35 XXXXXL 52 IX 22 24 58 54 52 10 36
5. RESULTS
The result section is divided into four parts. First, the data are described with the use of descriptive statistics and the skewness and kurtosis are checked. Then the correlations between variables are measured and a multiple linear regression analysis is executed.
5.1 Descriptive statistics
While checking the frequencies as first step in the analysis, a range of values for the variable ‘age’ was from -918 to 113. By zooming in on the ages, more outliers were found. These negative ages, ages of 0 and one value of 3 were deleted, which means they became missing values, since there was no possibility to find out what their real age was. As a second step to check the data for age, a histogram was plotted with the ages presented in percentage. The data showed a normal distribution, but with one remarkable case, namely an unnatural high quantity (N=1967) of consumers who were 16 year. The data received were checked with the company and we discovered that in year 2000 a loyalty pas was introduced, wherefore consumers had to fill in a form with the question of their birthdate as well. People who did not fill in their birthdate got an automatic birthdate of year 2000. Therefore based on this data a lot of consumers are now in 2016, 16 years old, although this is not the case in real. This explains that the values of 16 were incorrect and that the real age of these 16 years old consumers is not available. These values were removed and became missing value in the dataset. Still more than 40 000 consumers with information about age were left in the data set. The average age of these consumers is 55 years old. In table 1 the age distribution of the Dutch female population is compared with the sample to show for which part of the population the research explains something. As shown in the table consumers between 40 and 65 years old and consumers between 65 and 80 years old are overrepresented in this research. The other three categories are under-represented. By interpreting the results this difference in distribution should be kept in mind.
Table 2: Age distribution Dutch Female population versus Sample distribution
Age Categories 1 2 3 4 5
Range < 20 year 20 - 40 year 40 - 65 year 65 - 80 year > 80 year Population Dutch Women 22,65% 24,48% 35,12% 13,46% 4,35% Sample (N=40 694) 0% 7,30% 67,00% 23,40% 2,30%
Furthermore, the generational representation is as follow: 43 % of the consumers involved in the research are part of generation X, Babyboomers represent 42 % of the sample and only 4
% of the consumers is part of generation Y.
While checking the descriptive of the other variables, unnatural values for the variable ‘Clothes_size’ were found. All average sizes below size 32 are not possible since the size method starts with size 32 (see size table 2). We assume that the sizes below 32 are a mistake made in the company system and therefore we cannot change or control this. Since the quantity of unnatural values is not high, namely below 20, these values are deleted.
After dealing first with some outliers, the skewness and kurtosis are obtained to check the distribution of the scores of the continuous variables. Skewness value provides an indication of the symmetry of the distribution. Kurtosis, on the other hand, provides information about the ‘peakedness’ of the distribution. If the distribution is perfectly normal we would obtain a skewness and kurtosis value of 0, but in social sciences this is an uncommon occurrence (Pallant, 2013). The high kurtosis of ‘Total_Spending’ is marked, which means that high values leads to a high peak in the distribution. The positive skew of ‘Total_Spending’ indicates that the tail on the right side is longer or fatter than the left side. Since normal distribution has a skewness of zero, this data are not normally distributed around the mean. The variable ‘Online_to_Total’, which is related to the values of ‘Total_Spending’, is not normally distributed as well. Since this research is not using statistical tools wherefore normal distribution is a requirement, no further changes have to be made. Finally, no further extreme values in kurtosis and skewness are found.
The mean of 7.78 % of the variable ‘Online_to_Total’ means that on average consumers purchase for 7.78 % of their clothes online. In other words on average 92.22 % of consumer’s purchases are in a physical store. Furthermore, the range of ages of consumers is broad, namely from 14 years old up to 113 years old consumers and with an average age of 55 years old. When referring to the literature, this means that the average consumer here is from Generation X. This could already indicate a relationship between age and proportion of online shopping, since this value is really low and consumers here are relatively old. Besides that the mean of 0.81 for the variable ‘Presence_Physical’ means that more consumers live in an area where a physical store is presented. Finally, the mean of ‘Clothes_size’ should be rounded to 40, since the value is closer to 40 than 38 and the size 39 does not exists.
Table 3: Mean, Standard deviation, Min and Max, Kurtosis, Skewness
Measures M SD Min Max Kurtosis Skewness
Age 54.68 12.23 14 113 .04 -.02 Presence_Physical .81 .39 0 1 .46 -1.57 Residence_type .55 .50 0 1 -1.96 -.21 Total_Spending 323.30 409.67 0 18 598.95 121.68 6.86 Clothes_size 39.30 3.65 32 48 -.439 .29 5.2 Analysis
Two types of analyses are used, namely a multiple linear regression analysis and an ordered logistic regression analysis.
5.2.1 Multiple Linear Regression Model
The multiple linear regression analysis is executed to estimate the relationships between the five independent variables, also named predictors and the continuous dependent variable ‘Online_to_Total’. The dependent variable (Y) is in this model a continuous variable; therefore a (multiple) linear regression is possible. This model is based on the following formula: Y = β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 + ε Y= Online_to_Total_Spending X1 = Age X2 = Presence_Physical X3 = Residence_Type X4 = Total_Spending X5 = Clothes_Size
In this model the R2
is equal to 0.07. This means that the addition of all independent variables into a regression model explained only 7 % of the variability of the dependent variable, proportion online to total spending. This is a small size effect according to Cohen (1988).
The statistical significance of the overall model (i.e., the model containing all independent variables) is shown in the ANOVA table. The significance value is .000, thus p < .0005, which means that the results are statistically significant. Age, physical presence, residence type, total spending and clothes size statistically significantly predicted the proportion online to total spending, F(5, 39751) = 60.031, p < .001.
Table 4: One-way ANOVA outcomes
Model Sum of Squares df Mean Square F Sig.
Regression 1794844.20 5 358968.84 60.031 .000b
Residual 23769891.80 39751 597.97 Total 25564736.00 39756
a. Dependent variable: Online/Total
b. Predictors: (Constant), Clothes_size, Presence_Physical, Total_Spending, Age, Residence_type
For interpreting the results of the multiple linear regression analysis the coefficients in table 6 are analyzed. Here the slope coefficient (B) represents the change in the dependent variable for a one unit change in the independent variable. The coefficient for age is -0.22 (P<0.05), which means an increase in age of one year is associated with a decrease in proportion online to total spending of 0.22 %. The multiple regression equation predicts that the older you are the lower your online spending while all other independent variables are held constant. The coefficient of the other continuous variable, total spending, is 0.00 (P<0.05), which means this variable does not predict the value of the dependent variable online spending. The coefficient of clothes size is -0.234 (P<0.05). This means that an increase in the average size of consumers is associated with a decrease in proportion online to total spending of 0.234 %. So in other words, consumers with a larger clothes size spend more in a physical store then consumers with a smaller size.
The values of the slope coefficient of the dummy variables have a different interpretation. The value of the slope coefficient represents the difference in the dependent variable between the two categories of the dichotomous independent variable.
The coefficient for the variable presence physical store is -15.48 (P<0.05), which means that the predicted online spending of consumers with physical store in their surrounding area is 15.48 % lower than the predicted online spending of consumers without physical store in their surrounding area. The coefficient for residence type is -0.56 (P<0.05), which means that the predicted online spending of consumers living in a city is 0.56 % lower than consumers living in a village. So, city-consumers spend less online compared to consumers living in a village.
Table 5: Multiple Linear Regression analysis outcomes
Model B SE(B) b t Sig. (p)
Age -.22 .10 -.11 -21.40 .000 Presence_Physical -15.48 .32 -.24 -48.97 .000
Residence_type -.56 .25 -.01 -2.23 .026
Total_Spending .00 .00 .01 2.75 .006
Clothes_size -.234 .04 -.03 -5.71 .000
a.Dependent variable: Online_to_total. Notes. R2=.070
5.3.2 Ordered Logistic Regression Model
This customer research investigate which factors influence the way people buy: in a physical store, online in a web shop or across both channels, i.e., as Omnichannel consumer. Ordinal logistic regression is used to predict the ordinal dependent variable given one or more independent variables. It can be considered as a generalization of the logit model, which has a dependent variable with two categories. It is especially interesting for this research, since the real difference between consumer types becomes clear by using this ordinal regression. The formula for the OLR is as follows:
y* = β0 + β1X1 + β2X2 + ε
y =
In the previous analysis the outcome variable was the proportion of online spending to consumers total spending. In this analysis we want to zoom in on the different consumer groups. Therefore the continuous variable “Online_to_Total” is changed into an ordinal variable “Consumer_Type” consisting of the following three categories:
1. Pure physical consumers are the consumers who have spent 0 % online. 2. Omnichannel consumers are in this analysis the consumers who have online
spending between 1 % and 49 %. For example a consumer who have 10 % online spending, means that this consumer have spent 90 % in a physical store. This consumer is an Omnichannel consumer in this analysis.
0 if y* ≤μ1,
1 if μ1 < y*≤ μ2,
2 if μ2 < y*≤ μ3,
.
3. Finally, online consumers have spent between 50 % and 100 % online.
Figure 8: Percentages of online spending per consumer type
So the variable “Consumer_Type” consists of three categories, 0, 1 and 2. 75395 observations (N = 75395) are measured with a mean score of .18 (SD = .55). This variable is non-normally distributed, with skewness of 2.88 (SE = .01) and kurtosis of 6.57 (SE = .02). The frequencies per consumer type are presented in table 5. Here is shown that this dataset consists of a high quantity of physical consumers (90 %). The findings for this consumer group are therefore more reliable compared to the other two groups, however every group still consists of a high quantitiy of consumers, so there is no real problem for the analysis.
Table 6: Descriptives per group of the variable Consumer_Type Frequency Percent Cumulative Percent Physical_Consumers (0) 67716 89.7 89.8 Omnichannel_Consumer (1) 1952 2.6 92.4 Online_Consumer (2) 5727 7.6 100
Missing 104 .1
TOTAL (Consumer_Type) 75499 100
A critical part of the Ordered Logistic Regression is to check all the assumptions, which have to be met before starting the analysis. The following four assumptions have to be considered and are here explained:
In this analysis the dependent variable is ranked from no online spending in the first category to higher online spending in the subsequent two categories.
2. One or more independent variables that are continuous, ordinal or categorical (including dummy variables) ✔
Ordinal independent variables must be treated as being either continuous or categorical; therefore the variable “Clothes Size” is changed from an ordinal to a dummy variable with only two categories small and large. All other variables are continuous or already a dummy variable. So this assumption is also met.
3. There should be no multicollinearity ✔
Multicollinearity occurs when you have two or more independent variables that are highly correlated with each other. This leads to problems with understanding which variable contributes to the explanation of the dependent variable and technical issues in calculating an ordinal logistic regression. Determining whether there is multicollinearity is an important step in ordinal logistic regression (Agresti, 2010) and is determined using the same method as used for multiple regression, but with the dependent variable as ordinal and not continuous. This research does not have a problem with multicollinearity as found after running the linear regression (for the complete test and outcomes, see appendix).
4. The model should have proportional odds ✔
A full likelihood ratio test compares the fit of the proportional odds model to a model with varying location parameters. Selecting the ‘Test of parallel lines’ option in the Ordinal Regression analysis generates this outcome. The assumption of proportional odds was not met, as assessed by a full likelihood ratio test comparing the fit of the proportional odds location model to a model with varying location parameters, χ2
(5) = 800.345, p = .000. This means we would expect the difference in fit between these two models to be large and statistically significant (p < .05). For this research this outcome is not unexpected, since the large samples size increases the chances of statistically significant results. Therefore, a deeper investigation of the assumption of proportional odds is undertaken by running separate binomial logistic regressions on cumulative dichotomous dependent variables. To do this test new dichotomous dependent variables, Cat1 and Cat2, are created that represent the cumulative categories of the ordinal dependent variable, Consumer_Type. Cat1 consists of the
physical consumers, which are the consumers with 0 % online spending. Cat2 represents these physical consumers plus Omnichannel consumers, which are all the consumers who spend 0 % to 49 % online. Now it is possible to run separate binomial logistic regressions on Cat1 and Cat2. For both categories the results of the parameter estimates (B) and the Odds Ratios (Exp(B)) are aggregated into one table (see table 6). To test the assumption of proportional odds it is possible to compare the parameter estimates, although it is more useful to compare directly the odds ratios in the table (Agresti, 2010). The variables Age, Total_Spending and Clothes_size have similar odds for the two categories. The other two variables, Presence_Physicalstore and Residence_Type, have a larger difference between the odds ratios. The assumption of similar odds for this variable might not hold, however the ordinal regression can still be executed when these variables are treated with more caution in the regression (Kleinbaum & Klein, 2010).
Table 7: Statistics of two binomial logistic regressions
B (parameter estimates) Exp(B) (Odds Ratio) Independent variable Cat1 Cat2 Cat1 Cat2
Age .027 .022 1.028 1.022 Residence_Type -.132 .276 .876 1.318 Presence_Physicalstore -1.227 -.142 .293 .868 Total_Spending -.001 0 .999 1 Clothes_Size_Large_Small -.362 -.677 .696 .508 Intercept 1.546 6.1 4.695
a. Variable(s) entered on step 1: Age, Residence_Type, Size_LargevsSmall, Totaal_Spending, Presence_Physical_store.
The overall fit of the model is explained by using again the likelihood-ratio test, which looks at the change in model fit when comparing the full model to the intercept-only model. Here, the model statistically significantly predict the dependent variable over and above the intercept-only model, χ2
(5) = 2868.349, p < .001.
Table 8: Model Fitting Information
Model -2 Log Likelihood Chi-Square df Sig. Intercept Only 36565,935
