Time-limited price promotions in the era of multi-device online shopping

(1)

Master Thesis

Time-limited price promotions in the era of multi-device

online shopping

Author: Kay van den Bosch Student number: 11110465

University of Amsterdam

Faculty of Economics and Business

MSc. Business Administration – Digital Business track

Supervisor: dr. Jonne Guyt Date of submission: 23-06-2017 Version: final version

(2)

2

Statement of originality

This document is written by Student Kay van den Bosch 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.

(3)

3

Acknowledgements

I would like to take this opportunity to extend my gratitude towards several people that have made this Master Thesis possible. Firstly, I would like to thank my thesis supervisor Jonne Guyt for his help and feedback along the way. Furthermore, I wish to thank my family and in particular my girlfriend Rosalie for her moral support throughout this endeavour, and my brother Jori for his advice and positive criticism. Lastly, I would like to extend my gratitude towards the owners of the Dutch sunglass brand who gave me the opportunity to analyse their data. Thank you all!

(4)

4

Abstract

This study aimed to examine the influence of time-limited price promotions on online consumer purchase behaviour and the moderating role of the consumers’ purchase device therein. Using a dataset comprising of online store sale transaction data, three indicators for online purchase behaviour were assessed: Probability of Purchase, Purchase Volume and Purchase Concentration. The main results of this study suggest that time-limited price

promotions still are an effective marketing tool to boost sales, even in the era of multi-device online shopping. Both the Probability of Purchase and the Purchase Volume increased significantly during such a promotion. The results shed light on the differences in purchase behaviour between consumers using a smartphone, tablet or PC. While the Probability of Purchase is lowest for smartphones, the amount of online store traffic produced by this device is the highest of all devices and spikes during a promotion. Search costs are seen as a possible explanation for these results. Furthermore, although no differences seem present for Purchase Concentration, differences in search costs do seem to exist between devices, as additional analyses seem to present evidence that the session duration and the number of viewed pages per session rises significantly less for smartphones during a promotion than for the other devices. This study provides useful and practical insights for marketers in this contemporary era of multi-device online shopping. Furthermore, the results contribute to the growing academic literature on cross-media research and provide multiple avenues for future research.

Keywords: price promotions, online shopping, mobile shopping, mcommerce, search costs, purchase behaviour.

(7)

7

1 Introduction

With the adoption of the internet, introductions of new devices like smartphones and tablets and the connected rise in online retail, the marketing and retail landscape has changed

drastically over the last few decades. Ecommerce is growing year by year (eMarketer, 2015b; Forrester Research, 2014). Forrester (2014) estimates that the total online retail sales in the US alone will reach $414 billion by 2018, 11 percent of all retail sales that year. By 2020, online sales are estimated to rise even further and surpass $500 billion. New devices like smartphones and tablets are part of this ecommerce revolution in the shape of mobile

commerce, or mcommerce. In China for example, more than 49 percent of online sales where done using a mobile device. By 2017, this is estimated to be more than 60 percent (eMarketer, 2015a). It is clear that a new multi-device marketing and retail landscape has formed and academics try hard to keep up.

An important marketing tool for every retail platform is the use of time-limited price promotions, whether single or multi-itemed, to boost sales (Aggarwal & Vaidyanathan, 2003). This tool, that has been used in retail for decades, was the topic of many studies over the years (e.g. Neslin, Henderson, & Quelch, 1985; Spears, 2001). However, most of these studies looked at time-limited price promotions in the context of an offline retail setting (e.g. Iyer, 1989). With the rise of ecommerce, a gap in academic research has come to the fore. Where do time-limited price promotions stand in the era of online shopping? Furthermore,

consumers use multiple devices to shop online nowadays. These devices are different in form and function and fulfil different needs of the consumer. Remarkably, it seems that little research has been conducted on how the purchase device like a smartphone, tablet or pc effects online purchase behaviour. Therefore, a second gap in academic research can be identified.

(8)

8 This study will attempt to close these gaps by researching what the effects of time-limited price promotions are on online purchase behaviour and how the consumers’ purchase device moderates this effect. The results of this study could help online marketers better understand online shopping behaviour in the context of time-limited price promotions and more importantly the differences in behaviour between users of desktops, tablets and

smartphones. In turn, this could improve the consumer targeting capabilities of marketers and give better insight in how to allocate advertising budgets between devices. Furthermore, this study will contribute to the growing literature on cross-media research.

The next chapter of this study will review current theory regarding time-limited price promotions and how these affect purchase behaviour. Furthermore, consumer motivations for mobile shopping and differences in shopping behaviour between devices is discussed, as well as three indicators for online purchase behaviour and the hypotheses. Following the literature review, the research question and conceptual framework are discussed. Subsequently, chapter three provides an outline of the data and method, explaining the data origin, chosen sample, variables, indicator operationalisations and statistical methods. In chapter four, the main results are given followed by an extensive discussion of these results in chapter five including managerial implications, study limitations and directions of future research. Lastly, the main conclusions for this study are reviewed in chapter six.

(9)

9

2 Literature review

Time-limited price promotions, both single and multi-itemed, have been a popular part of a marketers’ toolbox to influence consumer purchase behaviour for decades (Neslin et al., 1985). However, over the last years, a major change in the marketing and retail landscape has come to pass with the growth of ecommerce, online retail and mobile shopping (eMarketer, 2015b; Forrester Research, 2014, 2016). Furthermore, multiple devices are used by

consumers for online shopping nowadays. What do all these changes mean for a traditional marketing tool like the time-limited price promotion and what part does the consumers’ purchase device play?

This chapter will firstly review current theory investigating why consumers tend to buy more when confronted with single and multi-itemed time-limited price promotions, how this affects online purchases and what consumers increasingly motivates to use mobile devices for online purchases. This will lead to the further exploration of the differences

between smartphones, tablets and PC’s as platforms for online shopping. Lastly, indicators for online purchase behaviour, the research question and the conceptual framework of this study will be discussed.

2.1 The effect of time-limited promotions on consumer purchase

behaviour

One could argue that by definition the purpose of a promotion is to influence consumer purchase behaviour by for example inducing consumers to buy sooner, in larger quantities or change brand (Aggarwal & Vaidyanathan, 2003). Over the last few decades, multiple studies have been conducted on the topic of sale promotions and their effect on consumer purchase behaviour. Spears (2001) for example concludes that the suggestion of time limitation on the

(10)

10 offer is one important characteristic of sale promotions, confirming earlier research by Iyer in 1989 or Bergadaa in 1990. Inman & Mcalister (1994) studied consumer behaviour in the context of time limitations as well. They concluded that an offer is redeemed in larger

quantities close to the expiration date, signalling accelerated purchases due to time sensitivity. Inman & Mcalister (1994) link their findings to regret theory. They theorise that redemption just before the expiration date spikes because the consumer anticipates regret for missing out on the offer and that these feelings can no longer be postponed as the offer is about to expire. Hence, the probability of offer redemption increases close to the expiration date. An

alternative theoretical explanation given by Inman & Mcalister (1994) is prospect theory, which suggests that consumers are more sensitive to loss than to gain (Kahneman & Tversky, 1979). In other words, if a consumer processes a time-limited offer it would first see it as a potential gain, and then as a potential loss if the expiration date would be missed. This would in turn increase the likelihood of offer redemption and therefore the probability of purchase near the expiration date as consumers try to avoid this loss. All these findings signal the importance of a time limit as part of a promotion.

Most studies have investigated the effect of (price) promotions and time sensitive promotions on consumer purchase behaviour in a brick and mortar retail setting. Aggarwal and Vaidyanathan (2003) for example compared the effects of limited and time-independent promotions on purchase acceleration, intent to continue to search for deals, willingness to buy and attitude towards the deal. The results show that consumers offered a time-limited promotion tend to accelerate their purchase, have less intent to search for other deals, are more willing to buy and have a more positive attitude towards the deal. All this is dependent on the perceived limitedness of the promotion; the more limited, the higher these effects.

(11)

11 The concept of purchase acceleration due to time-limited promotions has been

previously researched by Neslin, Henderson, and Quelch (1985). They identify two forms of purchase acceleration: consumers purchase a product sooner than originally planned, or they purchase multiple items at once, also referred to as “consumer stockpiling”. Neslin et al. (1985) conclude that price cuts are the most effective way to make consumers accelerate their purchase.

A different form of a price promotion is a quantity or multi-item promotion wherein a discount is offered only when more products are purchased at the same time (Foubert & Gijsbrechts, 2010). An example of a quantity promotion is “Buy one, get one free”, where a price discount of 50% is offered, but only when buying two products at once. Although one could assume that multi-item price promotions have similar effects on consumer purchase behaviour as single-item price promotions, some differences seem evident. Foubert and Gijsbrechts (2010) for example found that unconstrained multi-item price promotions lead to a higher increase in revenue than single-item price promotions because of the quantity

minimum. However, the study also underlines that when the purchase of a too high number of items is required before the promotion is applicable, a reverse effect occurs, leading to lower revenue. Furthermore, as the study was conducted for frequently purchased consumer goods, unclear remains what the effects are for durable consumer goods like sunglasses.

While Neslin et al. (1985), Bergadaa (1990), Iyer (1989), Spears (2001), Aggarwal & Vaidyanathan (2003) and Foubert & Gijsbrechts (2010) studied time-limited promotions in an offline retail setting, limited research seems to be conducted in the context of an online retail setting. However, Aydinli et al. (2014) did focus on online retail, mainly that of a daily deals platform. Online daily deals platforms, like Groupon or LivingSocial, primarily offer products or services with high discounts and for a limited time. Aydinli et al. (2014) found that a lowered price in such a time-limited setting can lead to a discouragement of deliberation

(12)

12 during the purchase process because the lowered price lowers the stakes and makes a potential purchase less consequential to the consumer. In other words, when confronted with a price promotion, consumers’ purchase decisions are guided more by affect and less by extensive information processing (rationality) because the potential loss for making a bad purchase decision is less consequential due to the lowered price. This indicates the emotional influence of price promotions in combination with a time limit and further underlines the power of this marketing tool on purchase behaviour. The results are also in line with the findings of

Aggarwal & Vaidyanathan (2003) regarding the lowered intent to search for other deals, but one could argue that a link to regret theory (Inman & Mcalister, 1994) and prospect theory (Kahneman & Tversky, 1979) exists as well as loss aversion seems to be influenced.

Previous research on time-limited price promotions, both single-item and multi-item, makes the effect on purchase behaviour evident. This can be concluded for both an offline as an online retail setting, although research on the latter seems less extensive. To further

examine the effect of time-limited price promotions on online purchase behaviour, assumed is that effects of such a promotion on the indicators of online purchase behaviour are similar as in an offline retail setting. Two of these indicators are Probability of Purchase and Purchase Volume and are discussed in paragraph 2.4 in more detail. The hypotheses for this part of this study are as follows:

H1a: The Probability of Purchase increases during an online time-limited price promotion. H1b: The overall Purchase Volume increases during an online time-limited price promotion.

Although not the subject of this study, it seems that time-limited price promotions do not have the same effect for all types of products and the combination with a high discount is important (Devlin, Ennew, McKechnie, & Smith, 2007). A study by Devlin et al. (2007) suggests that the factor of limited time on itself does not directly affect the perceived value of an offering or search and purchase behaviour. Their study showed that only in combination

(13)

13 with a high discount, a significant relation was visible. Important is to note that the study by Devlin et al. (2007) only looked at these effects in the context of one durable consumer good with a relative high price point: a TV. One could argue that the effect of time-limited price promotions on purchase behaviour is less strong when applied to relative expensive durable consumer goods like television sets.

Besides the influence of time-limited price promotions on online purchase behaviour, it is unclear how these promotions affect specific groups within the overall population of online shoppers, like mobile shoppers, as this has not been studied before. Over the next paragraphs, literature will be reviewed regarding why consumers use mobile ecommerce (mcommerce) and what possible key differences are between PC’s, tablets and smartphones when it comes to online purchase behaviour.

2.2 Influence factors and consumer motivations for mobile shopping

With the rise of mobile ecommerce (mcommerce), both the industry and academics have noticed and recognized mobile shopping as an important research topic (e.g. Shankar et al., 2016; Wang, Malthouse, & Krishnamurthi, 2015). Multiple studies have been conducted on factors driving mcommerce, and which factors can be hurdles for consumers to adopt mcommerce. Pagani (2004) for example concludes that usefulness, ease of use, price and speed of use are the most important factors for consumers to adopt mobile services including mcommerce. This conclusion is strongly related to the technology acceptance model (TAM) first introduces by Davis in 1989. TAM states that the adoption of new technologies is influenced by several factors including perceived usefulness and perceived ease-of-use. In other words, consumers are more prone to use technology if it is perceived as useful and easy to use. Later, TAM was further developed by a number of academics, including Venkatesh & Bala (2008) who emphasized on adopting this model in the context of ecommerce. By

(14)

14 introducing TAM 3, Venkatesh & Bala added the effects of trust and perceived risk to the model. Both the lack of trust and a risk perceived as high are seen as clear hurdles for consumers to accept ecommerce services. Wu & Wang (2005) looked at TAM in a different context as well; that of mobile commerce. Wu & Wang concluded that the factors perceived risk, cost, compatibility and perceived usefulness had significant influence on consumers’ intent to use mcommerce. In contrast to previous studies and unexpected, Wu & Wang found that perceived ease of use had no significant influence on consumers’ intent to use

mcommerce.

Kim, Li, & Kim (2015) took a different approach in identifying factors influencing mobile shopping (m-shopping) use by dividing the value of m-shopping in utilitarian and hedonic values. Utilitarian value consists of practical, purposeful value that m-shopping can have for consumers. In other words, the value of the consumer needing something and buying it to fulfil that need. This is the rational product purchase. Hedonic value on the other hand consists not only of the rational product purchase, but the shopping experience itself is part of it as well; as it were a recreational activity. Kim, Li, & Kim (2015) conclude that factors like personalization, simplicity and connectivity influence both values and thereby influence m-shopping use. Linked to the utilitarian value of mobile m-shopping, Kleijnen, de Ruyter, & Wetzels (2007) found that mobile shopping has more value to a consumer if he or she is time-conscious by appealing to temporal concerns of these consumers. In other words, one could state that consumers who experience utilitarian m-shopping value could perceive m-shopping as more convenient or efficient when making a purposeful time-sensitive purchase. Wang et al. (2015) partly underline these findings as they propose that consumers use mobile devices for m-shopping because the technology provides convenient access. This in turn leads to an incorporation of m-shopping in consumers’ daily and habitual routines (Wang et al., 2015). The findings of Wang et al. (2015) are in line with earlier research by Holmes, Byrne, &

(15)

15 Rowley (2012), who found that smartphones are valued for online shopping as they provide convenient and easy access.

One could argue that by design, time-limited price promotions appeal to temporal concerns of consumers. If an offer is not redeemed before the expiration date, consumers lose the opportunity forever. This time pressure combined with the potential fear of losing out on the offer as theorised by regret theory (Inman & Mcalister, 1994) or the evasion of losing the offers’ opportunity as theorised by prospect theory (Kahneman & Tversky, 1979), could make a mobile device more convenient and appealing to consumers to make a purchase on during an online time-limited price promotion than a PC. A mobile device is useable anywhere, anytime and easily incorporated in daily routines (Wang et al., 2015), making that consumers could decide not to switch to their PC’s to finalise the purchase because of the time

constraints. Therefore, proposed is that the Probability of Purchase, or in other terms the conversion rate, rises significantly more on a smartphone device during a time-limited price promotion than on a tablet or PC. This results in the following hypothesis:

H2a: The effect of a time-limited price promotion on Probability of Purchase is stronger on a mobile device than on a desktop or tablet device.

2.3 Differences between mobile, tablet and PC regarding online purchase

behaviour

In the previous paragraph, influence factors and motivations for using a mobile device for online shopping where discussed, signalling that the smartphone could be the go to device for online purchases during a time-limited price promotion. But what makes a smartphone different from a PC or tablet and how could this influence online purchase behaviour? An important difference between smartphones, tablets and PC’s is the screen size of these devices and the incorporated functionality for a user (Wang et al., 2015). The smaller the

(16)

16 screen, the more difficult it is for a user to get a clear overview of the content provided. For example, results from a study by Ghose, Goldfarb and Han (2013) suggest that mobile users tend to click more frequently on high ranked items when searching online than PC users. These ranking effects are linked to the perceived search costs by a device user. Devices with smaller screens, like smartphones and to some extend tablets, limit the amount of information a user is able to receive because a higher degree of manoeuvring and scrolling is necessary to view all the available content (Shankar, Venkatesh, Hofacker, & Naik, 2010). In other words, the smaller the screen size, the more effort is necessary to retrieve all available information which could be interpreted as a cost to consumers; search costs (Sweeney & Crestani, 2006). One could also state that the difference in input methods for text and scrolling/navigating through content could influence perceived search costs by consumers as well. Entry of text by keyboard could be seen as more convenient than using a touchscreen based keyboard. The same could be said for scrolling through content using a mouse instead of touch. If search costs lead to differences in online searching behaviour across devices, one could argue that a difference in search costs could influence the search process preceding an online purchase as well, and therefore influence online purchase behaviour.

As previously discussed, research by Holmes et al. (2012) found that in general, consumers value the convenience and accessibility when using a smartphone for an online purchase. This is in line with findings by Ozok & Wei (2010), who found that the ability to shop anywhere and anytime is one of the strongest plus points for consumers when using mobile shopping. Other results by this study indicate the importance of search costs to the consumer once more as the study concludes that consumers in general prefer to use a PC for online shopping. This is the case because interface limitations and input difficulties due to small screen size and input methods decrease the usability of mobile devices and therefore make them less attractive (Ozok & Wei, 2010).

(17)

17 Although the differences between smartphones and PC’s in the context of online search and purchase behaviour appear to have been studied extensively, little empirical research seems to have been conducted on another device within the mobile category: the tablet. One could describe a tablet as an intermediate device between a smartphone and a PC, sharing characteristics with both device groups. For example, in general a tablet has a larger display than a smartphone, leaning more towards a PC. Oppositely, a tablet shares the touch input capability of a smartphone and in general has a comparable mobile operating system. Research by Adobe Digital Insights (2016), eMarketer (2015) and Forrester (2014) suggests a significant difference between online purchase behaviour on mobile phones and tablets. For example, although total revenue is higher on smartphones, the conversion rate on tablets is higher and more in line with PC’s. Furthermore, a study by Adobe Systems Inc. (2013) suggest that tablets are more frequently used for online shopping than smartphones and while smartphones are used anywhere, tablets are more often used at home, similar to a PC

(Forrester Research, 2013).

Although statistical evidence seems present for the conclusion that there is a difference in consumer purchase behaviour between smartphones and tablets, there seems to be no or too little empirical research and thereby empirical evidence. Therefore, one could conclude that more empirical research is needed to test whether differences between purchase devices exist and what they are. Moreover, what about the differences between tablets and desktops?

Although limited research on the topic of differences between tablets and PC’s in the context of purchase behaviour seems to exist, Brasel & Gips (2014) empirically studied how the use of touchscreens for online shopping increases perceived ownership for a product. This is a relevant difference between both devices as all tablets have touchscreens, while that is less common for laptops and desktops alike. Brasel & Gips (2014) found that by virtually touching an object on a touchscreen, the perceived ownership of that object rises. This causes

(18)

18 people to ascribe more value to the object just because they subconsciously feel they own it. This phenomenon, whereby an object due to (perceived) ownership rises in value to the owner, is also known as the endowment effect. Brasel & Gips (2014) further conclude that the endowment effect and linked willingness to pay increases even more if the device that serves as a platform for online shopping is owned by the consumer. This effect is greater when the device is a touchscreen device compared to a touchpad device like a laptop. The findings by Brasel & Gips (2014) underline the importance of input methods when comparing devices and possible differences in purchase behaviour. The findings could mean that because of a higher willingness to pay, in practice Purchase Volume is relatively higher on a touchscreen device like a tablet than on a non-touchscreen device. As most PC’s do not have, or are not as dependent on a touchscreen for input purposes as a tablet, one could argue that this effect could be primarily visible on tablets or even smartphones. The added time pressure of a time-limited price promotion could further magnify these effects as consumers see the promotion as an opportunity to buy more than they would normally do (Neslin et al., 1985). In practice this would mean that the proposed effects of time-limited price promotions on the overall Purchase Volume per transaction is highest for touchscreen devices like tablets and smartphones, resulting in the following hypothesis:

H2b: The effect of a time-limited price promotion on Purchase Volume is relatively higher on tablets and smartphones than on desktop devices.

In table 1, key attributes of PC’s, tablets and smartphones are shown to further underline the differences between the devices. The attributes screen size and input methods per device type are based on devices available on the Dutch market. The attribute products per page lists the average products visible on the screen of a device and is based on a visual inspection per device on three major online stores in The Netherlands: Bol.com, Coolblue.nl and Amazon.de. The attribute perceived search costs is a combination of the attributes screen

(19)

19 size, input methods and the number of products per page relative to the other device types. If the screen of a device is small, input is limited and so is the number of products per page, one could argue that more effort by the user is necessary to receive all information to make a purchase decision and therefore search costs are higher. Oppositely, one could argue that the perceived search costs are lower on a device with a larger screen, more input methods and more products per page.

Table 1

Key differences between desktops, tablets and smartphones

Attribute PC Tablet Smartphone

Screen size > 12” 7” – 12.9” 3,5” – 6,4”

Input methods Keyboard, mouse, touchpad,

touchscreen

Touchscreen Touchscreen

Products per page 4-8 3-6 1-2

Perceived search costs Low Medium High

As researched by Ghose, Goldfarb and Han (2013), the concept of search costs is linked to the concentration of link clicks on specific platforms. One could argue that this difference in concentration based on the intensity of perceived search costs is applicable to products in an online shop as well. This would mean that the higher the intensity of perceived search costs, the higher the concentration of products purchased is, because a consumer could tend to settle for an easy product choice instead of looking at all alternatives. In practice, this would mean that the product purchase concentration is higher for smartphone users, and to some extend tablet users, and would focus on the top listed products on a particular page. As searching in general is costly for the consumer, the lowered price as part of an online time-limited price promotion could lower the motivation to search for a better deal as well as the stakes for the consumer are less high. Therefore, one could propose that during a time-limited price promotion the indicator Purchase Concentration, that is discussed in paragraph 2.4,

(20)

20 rises in general and the added search costs of using a mobile or tablet increase the effect on this indicator even more. This results in the following hypotheses:

H1c: The Purchase Concentration is higher during an online time-limited price promotion. H2c: The Purchase Concentration is higher on mobile and tablet devices during an online time-limited price promotion than on a desktop device.

2.4 Indicators for online purchase behaviour

Within academic literature regarding offline price promotions multiple indicators for purchase behaviour are used. For example, Zhang and Breugelmans (2012) looked at shopping trip spending and visit frequency to assess the effects of loyalty programs on offline purchase behaviour. Neslin et al. (1985) researched consumer stockpiling under influence of price promotions by looking at the quantity of goods purchased. Quantity was used by Ailawadi et al. (1996) as well to assess the influence of sales promotions on purchase behaviour. One could argue that how much a consumer buys in terms of quantity and spending seems highly related, as a higher quantity would logically lead to higher spending. A combination of both spending and quantity into the construct of Purchase Volume could therefore be seen as suitable.

Aydinli et al. (2014) looked at purchase behaviour in an online setting by assessing purchase likelihood or probability of purchase in regard to price promotions. A decade earlier Spears (2001) did the same to assess the influence of price promotions on purchase behaviour. Therefore, purchase likelihood or Probability of Purchase seems a proven indicator when looking at the effect of time-limited price promotion on online purchase behaviour.

Academic literature regarding differences between purchase devices like search costs and ranking effects seems to be limited. Ghose et al. (2013) looked at the concentration of link clicks to assess possible effects of search costs. One could argue that assessing the

(21)

21 concentration of link clicks to see whether a divers sum or just the top ranked links are

clicked would be applicable to online retail as well. Consequently, Purchase Concentration would be a fitting indicator as this would assess what the influence is of search costs and ranking effects, and therefore device differences, on the distribution of purchases over the product catalogue.

2.5 Research question & conceptual framework

The literature on time-limited price promotions has shown that the concept of a time limit plays an important role in how consumers react to these promotions (Aggarwal &

Vaidyanathan, 2003). Consumers tend to purchase earlier than intended or purchase more, especially for multi-item promotions (Foubert & Gijsbrechts, 2010), and a more emotional and less rational purchase decision is made (Aydinli et al., 2014). The more time-limited the promotion, the greater these effects (Aggarwal & Vaidyanathan, 2003). Although possible psychological mechanism like regret theory (Inman & Mcalister, 1994) or prospect theory (Kahneman & Tversky, 1979) are not a part of this study, they propose relevant background into why such promotions work the way they do. Furthermore, although extensive research has been conducted on how time-limited price promotions affect consumers in an offline retail setting, it seems that the same cannot be said for the online retail channels. For such an important marketing tool within a new era of ecommerce, one could conclude that a relevant gap in academic literature has been identified.

Part of this new era of ecommerce is multi-device shopping. Nowadays, consumers use their smartphones and tablets as platforms for online shopping besides the traditional PC. Several influence factors seem to exist for consumers to use a particular device. Smartphones for example are seen as more convenient (e.g. Holmes et al., 2012; Wang et al., 2015), can be used anywhere at any time and are particularly useful for utilitarian purchases when the

(22)

22 consumer is in a hurry (Kleijnen et al., 2007). Furthermore, smartphones, tablets and PC’s have physical differences such as screen size and input methods. These differences seem to influence the level of search costs, the level of effort it takes a consumer in terms of

manoeuvring, scrolling and searching a webpage to get a complete overview of all necessary information to make a purchase decision (e.g. Ghose et al., 2013; Shankar et al., 2010; Sweeney & Crestani, 2006). These search costs could lead to consumers choosing a higher ranked product on a webpage just because it saves effort and therefore influence the purchase concentration. Moreover, the touchscreen on particular devices could influence purchase behaviour as well as the endowment effect as proposed by Brasel and Gips (2014) could influence willingness to pay and therefore increase purchase probability on these devices. Although differences in purchase behaviour between smartphones, tablets and PC’s can be assumed, little academic research seems to study these three devices next to one another. Therefore, a second gap in academic literature can be identified.

To understand the position and effects of time-limited price promotions in this new era of ecommerce, the differences in devices that serve as platform for online shopping cannot be ignored. This study will address the two identified gaps in academic literature by researching the effect of a time-limited price promotion on online purchase behaviour in the form of Probability of Purchase, Purchase Volume and Purchase Concentration and how the purchase device moderates these effects. This results in the following research question:

What is the effect of time-limited price promotions on online purchase behaviour and how is this moderated by the consumers’ purchase device?

The results of this study could help practitioners better understand the influence of time-limited price promotion on purchase behaviour in an online retail setting combined with the role that a purchase device plays herein. In turn marketers could be enabled to better target

(23)

23 specific devices based on their marketing goals and available advertising budgets. For

academics, this study contributes to the growing literature on cross-media research.

In Figure 1 the conceptual framework visualises the research question by showing the proposed effect of a time-limited price promotion as independent variable on the Probability of Purchase, Purchase Volume and Purchase Concentration as dependent variables. The variable Device type (desktop/mobile/tablet) is the proposed moderator in this framework. Although the direct effect of Device type on the dependent variables is not part of the research question, it will be part of the statistical analysis and therefore discussed in the results.

Figure 1: The effects of time-limited price promotions and purchase device on online purchase behaviour. Time-limited price promotion Probability of Purchase Purchase Volume Purchase Concentration Device type (pc/tablet/mobile) H1 H2

(24)

24

3 Data and method

To study the effect of time-limited price promotions and the consumers’ purchase device on online purchase behaviour, a quantitative and deductive research approach is used. The research question is divided in multiple hypotheses which are tested using quantitative analysis of an existing dataset comprising of online sale transaction data. Over the next paragraphs, the origin of the data and the chosen sample is explained further. Moreover, the operationalisations of the indicators Probability of Purchase, Purchase Volume and Purchase Concentration are discussed and an overview of all variables is shown. Lastly, explained is why three different datasets are created, what alterations have been made and which statistical methods will be used to test the hypotheses.

3.1 Data origin & sample

The dataset for this study is provided by a Dutch sunglass and ski goggle brand which has its own mobile optimised online store in The Netherlands. The dataset consists of transaction data of sales in this online store, but information about the number of sessions (traffic) and the conversion rate per day is provided as well. The online store dataset was automatically

recorded and generated by web analytics service Google Analytics during April 1st_and

September 30th_{, 2016. For every observation (transaction) the device with which the}

transaction was performed (smartphone, tablet or pc) is recorded, as well as the product name of the purchased items, order quantity and total order value (revenue). The number of

sessions, visits to the online store, the mean number of pages viewed per session and mean session duration is recorder per device per day as well. Within the dataset, a distinction is made between normal days and days with a time-limited price promotion. Between April 1st

and September 30th_{, 2016 four multi-item price promotions were offered consisting of a “buy}

(25)

25 fixed discount of 25% and 30%. Both types of price promotions were offered for limited time ranging from between one and three days and were applicable to the total product range. As the number of days with a time-limited price promotion is very limited, both types of promotions are combined in a single variable and analysed without distinction. The data of days with other types of time-limited promotions, like promotions on just a few products or free accessories, were excluded from the dataset to ensure a more reliable distinction between non-promotion and promotion days.

Although the online store offers more items than just sunglasses, this product group is the main focus for this brand. One could argue that most customers will visit the online store in search of new sunglasses. As just a few models exist but the range of colour variations in both frames and lenses is wide, expected is that search costs are most likely to occur within this specific product category. To effectively measure the influence of search cost on

Purchase Concentration, transactions containing other product groups from different parts of the online store are removed from the dataset.

The sample for this study consist of all customers making a purchase or visiting the online store of the sunglass brand during data collection. As the data was purposefully collected during a specific period, the sampling method could be classified as purposive sampling (Saunders, Lewis, & Thornhill, 2012). To study online purchase behaviour, data from a real web shop is crucial. Within the scope of this study and the available resources, the choice for this sample and these data seem justified. Nevertheless, one could argue that the chosen sample and sampling method will have its influence on the generalizability of the results.

(26)

26

3.2 Operationalisations & variables

Over the next paragraphs the operationalisation of the constructs Probability of Purchase, Purchase Volume and Purchase concentration is discussed. Furthermore, the presence and data source of a control variable for sunny weather is explained. Lastly, an overview of all variables is given.

3.2.1 Operationalisation of Probability of Purchase

To measure the effect of a time-limited price promotion and the consumers' purchase device on the Probability of Purchase, one needs to measure if consumers are more likely to make a purchase during specific circumstances involving these two independent variables. The Probability of Purchase can be operationalised by looking at the ratio between all visitors of a store and the total number of transactions for a specific time unit. This ratio is also known as the conversion rate. For this study, the Probability or Purchase is operationalised by dividing the total number of transactions for a day by the number of sessions for that day in the online store. The conversion rate is stored in the variable conversion_rate.

3.2.2 Operationalisation of Purchase Volume

For the indicator Purchase Volume, one can look at the change in spending and order quantity between promotion and non-promotion days and on device level. Therefore, Purchase

Volume can be operationalised by assessing the change in the variables revenue, which is the total spending per transaction, and the variable quantity, which is the total number of items purchased per transaction.

3.2.3 Operationalisation of Purchase Concentration

The indicator Purchase Concentration is based on the idea that search costs and ranking effects have influence on the distribution of sales over the total line of available products. A

(27)

27 higher Purchase Concentration would mean that the diversity of products sold is lower, while a low Purchase Concentration would mean that a more diverse set of products is sold. There are multiple ways to assess diversity within a specific unit. For this study, the choice is made to assess Purchase Concentration by looking at the total number of unique products within a set of ten sold products. A unique product is a product that has no duplicates within the same set. A high number of unique products points to a low purchase concentration, while a low number of unique products would point to the opposite. The number of unique products per set of ten products is stored in the variable unique_products. In paragraph 3.3.3 the statistical procedure for this measurement method is explained in more detail. Given that the amount of available data for some groups within this study is limited, this measure is deemed most suitable as the validity of this measure is least influenced by small samples when compared to alternative measures. A discussion on alternative operationalisations for Purchase

Concentration can be found in Appendix I.

3.2.4 Control variables

One could argue that the sale of sunglasses could be influenced by other factors than promotions or the consumers’ purchase device. For example, the holiday season could positively influence sales. However, as no data is recorded during the holiday season, this possible influence factor is disregarded. Another factor for this specific product group could be the weather, as one could argue that the demand for sunglasses rises during sunny weather conditions. Therefore, a control variable for sunny weather is added to the dataset. This

control variable consists of the number of hours of sunshine on a day expressed in the variable sunshine_hours. This variable is derived from a dataset acquired from the KNMI Data Centre (KDC), which is part of the Royal Netherlands Metrological Institute (KNMI). The weather data is measured and recorded by a weather station in De Bilt, The Netherlands. This station

(28)

28 is chosen because of its representative weather data for the rest of The Netherlands, as it is centrally located. The use of this control variable in the statistical analyses is explained in more detail in paragraph 3.3.

3.2.5 Variable Overview

In Table 2 an overview of all variables is show together with a short description. Table 2

Variable overview Variable Description

device Type of device (desktop, mobile, tablet) used for the transaction(s).

promotion Binary variable for either non-promotion or promotion.

conversion_rate Total conversion rate per device per day.

quantity Total quantity if items purchased per transaction.

revenue Total revenue per transaction.

log_quantity A transformed variant of quantity, log(quantity + 1).

log_revenue A transformed variant of revenue, log(revenue + 1).

unique_products The number of unique products within a set of 10 products (case).

sunshine_hours Total hours of sunshine per day in De Bilt, The Netherlands.

3.3 Datasets & statistical methods

The original dataset as recorded by Google Analytics consisted of 3795 transactions and aggregated data of 181 days in total. To clean this dataset, multiple alterations are made. Firstly, days containing less than five sales transactions are deleted from the dataset, as are days with excluded sale promotions. Consequently, all transactions within these days are deleted from the transaction set as well. Furthermore, all transactions containing other

products than sunglasses are deleted, as are all transactions with a sales quantity of above ten products as this could be viewed as extreme purchase behaviour or possible business-to-business orders. All these alterations are made to strengthen the reliability of the data and reduce possible biases from extreme high or low values, as well as other promotions and products. In total, 903 transactions are manually removed as well as 37 days. To prepare the

(29)

29 data for analysis, three individual datasets are created: (1) Transactions, (2)

Transactions_Days and (3) Product_Sets. Within each dataset six groups are present as the two independent variables promotion and device consist of two and three groups respectively. Over the next paragraphs, the purpose for every dataset is discussed in the context of

hypotheses testing, as well as the removal of outliers and transformation of variables. Furthermore, the chosen statistical methods are addressed.

3.3.1 Transactions_Day

To assess whether Probability of Purchase is higher during a time-limited price promotion (H1a), and if this proposed effect is higher for smartphone devices than tablets and desktops (H2a), the dataset Transactions_Day is created. This dataset contains all transactions

aggregated per device per day, creating three individual cases per day. In the variable device, the device type is stored and in the variable promotion is stored whether a time-limited price promotion is active on that day. For every case, the conversion rate is stored in the variable conversion_rate. Lastly, control variable sunshine_hours is added to the dataset to control for the influence of sunny weather on the Probability of Purchase. The number of cases for all six groups within Transactions_Day is shown in Table 3.

A check of the dataset results in the identification of eight outliers for the dependent variable conversion_rate, one of which is an extreme outlier. As these outliers are not present due to errors in the data or measurement, removing these cases from the dataset would be an extreme measure that could influence the validity of the data (Field, 2014). For the extreme outlier however, a form of extreme or uncommon purchase behaviour could be present. Therefore the choice is made to remove only this extreme case from the dataset as statistical tests can be sensitive to outliers (Field, 2014).

(30)

30 Given that multiple groups are compared, the independent variables are categorical, the dependent variable is continuous and a continuous covariate is considered, a Factorial ANCOVA is chosen as this would be the most suitable analysis to test the hypotheses (Hair, Black, Babin, & Anderson, 2010).

Table 3

Dataset case overview Transactions_Day

Promotion/Device Desktop Mobile Tablet Total

Non-promotion 138 138 137 413

Promotion 6 6 6 18

Total 144 144 143 431

3.3.2 Transactions

To assess if the Purchase Volume rises during a time-limited price promotion (H1b), and if this effect is stronger for smartphones and tablets (H2b), the dataset Transactions is created. This dataset contains all transactions after the initial cleaning of the data. For every

transaction (case), the number of items purchased is stored in the variable quantity. Furthermore, the revenue per transaction is stored in the variable revenue, the device with which the transaction was performed is stored in the variable device and in the variable promotion is stored whether the transaction was performed during a time-limited price promotion. Lastly, control variable sunshine_hours is added to the dataset as sunny weather could influence sale performance.

To improve the normality of the dependent variables revenue and quantity, the

variables are transformed to log_revenue (log(revenue+1)) and log_quantity (log(quantity+1)) respectively. Furthermore, a total of 53 extreme outliers are removed to improve the

reliability of the data. Non-extreme outliers are kept as the outliers are not a result of errors in measurement or missing data, but assumed cases of unusual purchase behaviour. An overview of the cases per group within the dataset are shown in Table 4.

(31)

31 The construct Purchase Volume consists of the dependent variables revenue and quantity. Both dependent variables are theorised to by highly related, and are highly

correlated as well (r(2837)=.657, p < .05). The correlation matrix can be found in Appendix II. Given that the two continuous dependent variables seem highly related, two categorical independent variables and a continuous control variable are present, a Factorial MANCOVA is considered a suitable analysis to test the hypotheses (Hair et al., 2010).

Table 4

Dataset case overview Transactions

Non-promotion 1078 632 191 1901

Promotion 537 338 63 938

Total 1615 970 254 2839

3.3.3 Product_Sets

To assess if Purchase Concentration is higher during a time-limited price promotion (H1c), and if this effect is higher on mobile and tablet devices (H2c), the dataset Product_Sets is created. To create this dataset, all purchased products per group in the original dataset are randomly divided in sets of ten products. Then, all duplicates in these sets are removed, leaving a set of unique products. The count of all remaining products in a set is stored in the variable unique_products. In the variable device is stored on what device the products in the set are purchased and in the variable promotion is stored if the products in the set are

purchased during a time-limited price promotion. Given that products are randomly assigned within the groups, information on the day of purchase is lost. Therefore, the control variable sunshine_hours is not added to the dataset as this variable contains data per day. Every case in the dataset represents a set. As the number of products purchased per group is not equal, the number of cases is unequal between groups as well. An overview of all cases within the dataset is shown in Table 5.

(32)

32 Given that the dependent variable unique_products is continuous and two independent categorical variables are used, a Factorial ANOVA is the most suitable analysis to test the hypotheses (Hair et al., 2010).

Table 5

Dataset case overview Product_Sets

Non-promotion 147 81 27 255

Promotion 98 57 14 169

(33)

33

4 Results

Over the next paragraphs the results of the statistical analyses are shown, starting with the effect of the independent variables promotion and device on the indicator Probability of Purchase. Secondly, the effect of the independent variables on Purchase Volume will be revealed. Lastly, the results of the statistical analysis involving the indicator Purchase Concentration are addressed.

4.1 The effect of promotion and device on Probability of Purchase

Table 6 Descriptive Statistics

Promotion Device Type M SD N

Non-promotion Desktop .041 .018 138 Mobile .013 .006 138 Tablet .039 .033 137 Total .031 .025 413 Promotion Desktop .115 .026 6 Mobile .035 .011 6 Tablet .105 .034 6 Total .085 .044 18 Total Desktop .044 .024 144 Mobile .014 .008 144 Tablet .041 .035 143 Total .033 .028 431

Dependent Variable: conversion_rate

The mean value of the dependent variable conversion_rate, standard deviation and sample size per group are shown in Table 6. Clear is that sample sizes are unequal between groups. Furthermore, the mean conversion rate seems higher for desktops and tablets than for mobile devices, both in promotion and non-promotion circumstances. In Table 7the main results of the Factorial ANCOVA are shown for the effect of the independent variables promotion and device on the dependent variable conversion_rate. The results reveal a significant but

relatively small interaction effect between promotion and device, F(2, 424) = 9.37, p < .001, η2 = .042. The profile plot in Figure 2sheds some light on this effect, and seems to reveal that

(34)

34 the Probability of Purchase increases more during a time-limited price promotion for desktops and tablets than for smartphones. The results of a Bonferroni post hoc test collaborate these findings, as the differences between mobile and desktop and mobile and tablet are significant (p < .001). However, no significant difference is visible between desktops and tablets. All post hoc results can be found in Table 16 in Appendix III.

Both promotion and device seem to have a strong and significant effect on the Probability of Purchase. On days with a promotion, the conversion rate rises significantly, F(2, 424) = 83.09, p < .001, η2 = .164. Furthermore, between devices significant differences seem to exist when it comes to the conversion rate, F(2, 424) = 42.32, p < .001, η2 = .166. The post hoc test and profile plot reveal that smartphones have an overall significantly lower conversion rate than the other devices. Lastly, the influence of the control variable sunshine_hours seems weak but significant, signalling that sunny weather does have some effect on the Probability of Purchase, F(1, 424) = 6.06, p < .001, η2 = .014).

Table 7

Results Factorial ANCOVA

Source SS DF MS F Sig. η2 sunshine_hours .003 1 .003 6.06 .014 .014 promotion .04 1 .040 83.09 .000 .164 device .04 2 .020 42.32 .000 .166 promotion * device .009 2 .004 9.37 .000 .042 Error .203 424 0 Total .82 431

(35)

35

Figure 2: Effect of promotion and device on conversion_rate

4.2 The effect of promotion and device on Purchase Volume

Dependent Variable: revenue & quantity

An overview of the mean value, standard deviation and sample size for every group within the dataset for the untransformed variables revenue and quantity is shown in Table 8. The table makes evident that sample sizes differ greatly between groups. Furthermore, the mean

revenue per transaction seems lower for transactions during a promotion, while the quantity of items purchased per transaction seems higher.

0.0% 2.0% 4.0% 6.0% 8.0% 10.0% 12.0% Non-promotion Promotion Con v ersi on Rate Promotion Desktop Mobile Tablet Table 8 Descriptive Statistics Revenue Quantity Promotion Device N M SD M SD Non-promotion Desktop 1078 57,21 25,85 1,33 0,67 Mobile 632 51,74 21,86 1,22 0,51 Tablet 191 58,39 25,60 1,41 0,72 Total 1901 55,51 24,70 1,30 0,63 Promotion Desktop 537 53,44 14,33 1,75 0,58 Mobile 338 51,88 14,31 1,67 0,56 Tablet 63 57,11 16,66 1,79 0,60 Total 938 53,12 14,53 1,73 0,58 Total Desktop 1615 55,96 22,74 1,47 0,67 Mobile 970 51,79 19,55 1,38 0,57 Tablet 254 58,07 23,68 1,50 0,71 Total 2839 54,72 21,90 1,44 0,64

(36)

36 The multivariate results of the Factorial MANCOVA for the transformed dependent variables log_revenue and log_quantity are shown in Table 9. The results reveal that there seems to be no significant interaction effect between promotion and device for Purchase Volume using Pillai’s Trace, V = .003, F(4, 5664) = 1.95, p > .05, η2= .001. Table 18 in Appendix IV contains the results for all statistics.

However, a moderate and significant effect seems visible for promotion, signalling that there is a significant difference between non-promotion and promotion transactions when it comes to Purchase Volume, V = 1.03, F(2, 2832) = 163.35, p < .001, η2= .103. The effect of device on Purchase Volume seems is very weak, though significant, V = .009, F(4, 5664) = 6.33, p < .001, η2= .004. Furthermore, the multivariate results seem to suggest that the covariate sunshine_hours does have a significant but weak effect on Purchase Volume.

The univariate test results, as shown in Table 10, shed some light on the multivariate results by looking at the effects on both the dependent variables log_revenue and log_quantity separately. The effect of promotion on log_revenue in not significant and seems to suggest that there is no significant difference in the revenue per transaction on non-promotion and promotion days, F(1, 2833) = 1.16, p > .05, η2= .000. However, a moderate and significant effect on log_quantity is visible, partially explaining the multivariate results, F(1, 2833) = 213.85, p < .001, η2= .070. The profile plot in Figure 3 clarifies these results and shows a rise for the variable log_quantity for transactions during a promotion. The effect of device is weak

Table 9

Factorial MANCOVA: multivariate results

Effect Statistic Value F Hypothesis

DF

Error DF

Sig. η2 sunshine_hours Pillai's Trace 0,010 14,92 2 2831 .000 .010 promotion Pillai's Trace .103 163.35 2 2831 .000 .103 device Pillai's Trace .009 6.33 4 5664 .000 .004 promotion * device Pillai's Trace .003 1.95 4 5664 .099 .001 Dependent Variables: log_revenue & log_quantity

(37)

37 but significant for both log_quantity (F(2, 2833) = 9.07, p < .001, η2= .006) and log_revenue (F(2, 2833) = 11.49, p < .001, η2= .008) signalling that differences are present between devices when it comes to both the revenue per transaction and the quantity of items purchased.

The results of a Bonferroni post hoc, as shown in Table 19 in Appendix IV, reveal that the quantity of items purchased is significantly lower for mobile devices compared to tablets and desktops (p < .05) when looking at non-promotion transactions. For promotion transactions however, no significant difference is found. Furthermore, the post hoc results seem to reveal that the revenue per non-promotion transaction is significantly lower for mobile devices compared to the other devices as well (p < .001). For transactions during a promotion, no significant difference is found (p > .05). Lastly, the univariate results suggest that the covariate sunshine_hours only has a significant but weak effect on the quantity of items per transaction, F(1, 2833) = 19.50, p < .001, η2= .007.

Table 10

Factorial MANCOVA: univariate results

Source Dependent Variable SS DF MS F Sig. η2

sunshine_hours log_revenue .002 1 .002 .10 .748 .000 log_quantity .168 1 .168 19.50 .000 .007 promotion log_revenue .028 1 .028 1.16 .281 .000 log_quantity 1.847 1 1.847 213.85 .000 .070 device log_revenue .545 2 .273 11.49 .000 .008 log_quantity .157 2 .078 9.07 .000 .006 promotion * device log_revenue .113 2 .057 2.38 .093 .002 log_quantity .003 2 .001 .15 .864 .000 Error log_revenue 67.208 2832 .024 log_quantity 24.460 2832 .009 Total log_revenue 8444.725 2839 log_quantity 428.250 2839

(38)

38

Figure 3: The effect of promotion and device on log_revenue & log_quantity

4.3 The effect of promotion and device on Purchase Concentration

Table 11 Descriptive Statistics Promotion Device M SD N Non-promotion Desktop 9.29 .759 147 Mobile 9.36 .658 81 Tablet 9.37 .884 27 Total 9.32 .740 255 Promotion Desktop 9.27 .856 98 Mobile 9.26 .813 57 Tablet 9.36 .633 14 Total 9.27 .822 169 Total Desktop 9.28 .797 245 Mobile 9.32 .725 138 Tablet 9.37 .799 41 Total 9.30 .773 424

Dependent Variable: unique_products

The mean value, standard deviation and sample size per group within the dataset is shown in Table 11. Clear is that the sample sizes are unequal and the sample is smallest for tablet devices. The mean values show the mean number of unique products within a set of 10 products. Some minor differences seem present between non-promotion and promotion groups and between devices. These apparent differences are illustrated in the profile plot in Figure 4 as well. However, the results of the Factorial ANOVA, as shown in Table 12, reveal

1.65 1.66 1.67 1.68 1.69 1.7 1.71 1.72 1.73 1.74 1.75 1.76 Non-promotion Promotion log_ rev en u e Promotion Desktop Mobile Tablet 0.25 0.27 0.29 0.31 0.33 0.35 0.37 0.39 0.41 0.43 0.45 Non-promotion Promotion log_qu an ti ty Promotion Desktop Mobile Tablet

(39)

39 no significant interaction effect between promotion and device, F(2, 418) = .064, p > .05, η2 = .001. Furthermore, no significant effect of promotion on the number of unique products per set seems present, F(1, 418) = .106, p > .05, η2 < .001. This signals that no significant differences in Purchase Concentration seem to exist between non-promotion and promotion circumstances. Moreover, no significant effect of device on the dependent variable seems present either, suggesting that no significant differences exist between devices when it comes to Purchase Concentration, F(2, 418) = .148, p > .05, η2 = .001. Additional test results can be found in Appendix V.

Figure 4: The effect of promotion and device on the number of unique products per set

Table 12 Factorial ANOVA Source SS DF MS F Sig. η2 promotion .106 1 .106 .176 .675 .000 device .296 2 .148 .245 .783 .001 promotion * device .128 2 .064 .106 .899 .001 Error 252.283 418 .604 Total 36921 424

Dependent Variable: unique_products

9.25 9.27 9.29 9.31 9.33 9.35 9.37 9.39 Non-promotion Promotion Un iqu e P rod u cts Promotion Desktop Mobile Tablet

(40)

40

5 Discussion

In this chapter, the main findings of this study are discussed for the three indicators of online purchase behaviour: Probability of Purchase, Purchase Volume and Purchase Concentration. Furthermore, the managerial implications and limitations of the study results are addressed. Lastly, avenues for future research are suggested.

5.1 Discussion

The results have provided some insights into how Probability of Purchase, Purchase Volume and Purchase Concentration as indicators of online purchase behaviour are influenced by time-limited price promotions and what the moderating role is of the consumers’ purchase device. An overview of the hypotheses is shown Table 13.

For the Probability of Purchase, the effects of time-limited price promotions were extensively researched in an offline retail setting (e.g. Aggarwal & Vaidyanathan, 2003; Neslin et al., 1985) and suggested a positive influence on revenue, items purchased and purchase likelihood. Given these findings, similar results were foreseen for an online retail setting and expected was that time-limited price promotions would increase the likelihood of a purchase (H1a). The results seem to suggest that a time-limited price promotion in fact does have a significant positive influence on the conversion rate, and therefore the Probability of Purchase. Furthermore, past research found that smartphones are very convenient for online purchases because of their accessibility (Holmes et al., 2012; Wang et al., 2015) and that time conscious consumers value mobile more than other devices (Kleijnen et al., 2007). This combined with the time pressure and time-limited price promotions (Spears, 2001), a greater rise in the purchase likelihood was expected for smartphones (H2a). However, the results show that the Probability of Purchase rises for all devices during a promotion, but contrary to

(41)

41 the hypothesis, more for desktops and tablets when compared to mobile. A possible

explanation comes from the differences in online store traffic between devices. Remarkably, as reviewed in Appendix VI, an additional Factorial ANCOVA shows that the number of sessions per day is significantly higher for mobile devices in general and spikes during a time-limited price promotion. In fact, the mean number of sessions for mobile is almost twice that of desktop during a promotion. As the conversion rate is the ratio between the number of session and the number of transactions on a day, a spike in sessions would drastically

influence this ratio if transactions do not rise in a same manner. An explanation for the spike in traffic could be the convenience and accessibility of the smartphone as online shopping platform. Additionally, it is possible that the smartphone could very well be the first channel consumers learn about the promotion, for example by ways of social media or email. Why this rise in traffic does not result in a comparable spike in transactions seems a mystery. However, a possible explanation could lie in the higher search costs for mobile compared to the other devices as research by Ozok & Wei (2010) among others. They concluded that this could lead to a higher preference for desktop devices as input difficulties due to limited input methods and smaller screen size decrease the usability of mobile devices (Ozok & Wei, 2010). As a result, consumers could abandon the mobile platform when it comes to the actual purchase transaction, or when consumers wish to have a better overview of the product itself. Given the hedonic nature of a product group like sunglasses, the latter reason could be equally important. However, future research is needed to provide a less speculative answer.

Based on the same research as for the Probability of Purchase (e.g. Aggarwal & Vaidyanathan, 2003; Neslin et al., 1985), during a time-limited price promotion an increase in Purchase Volume was expected as well (H1b). The results suggest some support for this premise as the quantity of items purchased within a transaction does increase significantly. However, in this analysis, no significant increase in revenue per transaction is visible. Given

Time-limited price promotions in the era of multi-device online shopping

Master Thesis