“Seek and you shall find”

(1)

0 | P a g e

“Seek and you shall find”

An exploratory study on the effects of consumer in-store search

behavior on purchase decision, total amount spent, and number

of products purchased.

(2)

1

Title: “Seek and you shall find”

Subtitle: An exploratory study on the effects of consumer in-store search behavior

on purchase decision, total amount spent, and number of products purchased.

University: Rijksuniversiteit Groningen

Faculty: Economics & Business

Program: Msc in Marketing, Profile Marketing Intelligence

Qualification: Master’s Thesis

Completion date: 20

th

_{of February 2014}

Author’s name: Christie J.A. Quarton

Author’s address: Verlengde Willemstraat 7a, 9725AW, Groningen

Phone number: 0639073433

E-mail: cquarton@gmail.com

Student number: S2085712

(3)

2

Management Summary

Previous research regarding purchase decision has mainly focused on studying the effects of demographics, environmental variables and marketing decisions on purchase decisions. These studies have found that situational factors such as crowding, time pressure and time duration have an effect on purchase. The demographics of a consumer also influenced purchase choice. Survey data and scanner data was mainly used to infer purchase likelihood which then ignored the in-store consumer product consideration process. The limited number of studies on in-in-store product consideration was largely due to the challenge of collecting in-store consideration data, but recent breakthroughs with RFID equipped carts, handheld barcode scanners, and video-tracking

facilitated the process of examining the in-store shopping process a consumer engages in. Several studies began quantifying the effect of in-store product consideration on purchase, but most remained on investigating specific effects such as impulse purchasing, the effectiveness of shopping paths, or were descriptive. However, a key finding was that deeper product considerations increased purchase likelihood (Hui et al., 2013).

This study seeks to incorporate the behavioral data in the models to see if a. there are constructs in the fourteen different actions listed; b. if they significantly predict purchase and c. if these actions further predict components of purchase – total amount and number of products purchased. The data was collected by researcher Julien Schmitt who followed 190 shoppers in a French beauty store. Customer movements were tracked by the second, as they conducted different actions in-store such as walking, stopping, looking at a shelf, grabbing a product, and handling the product.

(7)

(8)

7

Preface

“Onderschat jezelf niet”

- Prof. dr. Peter Leeflang

These are appropriate words to preface one of the most ambitious projects I have ever taken on; embarking on gaining a master’s degree, and in the process switching fields from marcomms to marketing research. Two years ago, I wrote in my admission essay that I would be honored to be part of this program, so I could finally gain the tools that would enable me to go from “intuition” (black-box) marketing, to being able to make a solid case for why I make decisions as a marketer. I never thought it would be so challenging and yet so rewarding. My purpose was to obtain a master’s degree, what I walk away with is a newfound passion & love for data analysis (who knew!). I want to thank all of my professors as they have given me one of the greatest gifts you can give someone on this earth - knowledge. A special thank you to professor Julien Schmitt of Aston University in the UK for sharing his data with me and being so patient in searching out the smallest details so I could merge the databases. My sincere gratitude to Professor Wieringa and Professor Leeflang for being so involved in this process and inspiring me to do more! 

(9)

8

1. Introduction

1.1.1 Retailing & the consumer

The age-old question for retailers has been: “How do we convert consumers into buyers while in the shopping environment and make them buy as much possible?” Retailing is a global business where sales are in the trillions of dollars each year. Supermarkets which are part of Fast Moving Consumer Goods industry (FMCG) dominate the top 250 retailer list. Walmart Inc. (2013), the biggest global retailer, had sales in excess of $443 billion dollars worldwide last year – which is more than 2/3 of the GDP of the Netherlands in 2012. The competition within FMCG is fierce so insights into the consumer buying process are crucial to remain competitive in the marketplace. With the advent of customer in-store data available through store scanner data and radio

frequency tracking devices installed on shopping carts; researchers are increasingly able to model in-store customer behavior more accurately. The consumer’s travel path in-store including

shopping duration time can now be graphed and coupled with checkout data from which different variables can calculated such as deliberation time, time pressure, store crowding, shopping basket analysis (Briesch, Chintagunta, Fox, 2009; Kamakura, 2012).

Retailers in turn can use these variables to segment shoppers into different clusters which can be targeted more effectively through feature and displays, aisle locations, and product placements on shelves, (Larson, Bradlow, &Fader, 2005; Hui, Inman, Huang, & Suher, 2013), ultimately with the goal of maximizing purchases even if a consumer’s path is short on distance traveled and time duration. Contrary to popular belief, consumers rarely, if ever, shop the entire store or pass

(10)

9

1.1.2 The Consumer In-Store Decision Process

Research has mainly focused on decomposing the consumer’s in-store shopping process into different subsets that influence purchase decision such as promotion, category, store, and

consumer characteristics (Baker, Parasuraman, Grewal, & Voss, 2002; Inman, Winer, & Ferraro, 2009). Store characteristics consist of the store design and store atmosphere such as design, scents, music, temperature, and lighting (Baker et al, 2002; Kaltcheva and Weitz, 2006; Wakefield & Baker, 1998; Summers and Hebert, 2001). What makes a store is its product offerings which is why product assortment (variety, uniqueness, price and quality) plays such a huge role in what type of people come to the store and what is bought (Sloot, Fok, & Verhoef, 2006; Gooner, Morgan, & Perreault, 2011). In-store stimuli such as feature and displays, LCD screens, and promotional offers help to attract the attention of the consumer to products to encourage a sale (Baker et al, 2002). Category characteristics is comprised of the type of goods, the hedonicity of the product, and the utility of the product for a consumer (Hui et al.,2013). The characteristics of a consumer such as age, gender, and income may also influence what they ultimately purchase (Inman et al, 2009).

Lately, more studies have been conducted on the store path that a consumer takes within the store incorporating store and customer characteristics to discover the deliberation process a consumer undergoes in-store (Hui et al., 2003). Larson et al. (2005) were one of the first to use the RFID data provided by shopping carts to segment shoppers into clusters of short, medium, and long shopping trips based on time duration. Their research dispelled the former assumptions about aisle travel, and showed that depending on time duration consumers displayed distinctive travel paths in-store. The study further expanded upon where the consumer’s spent most of their shopping time, such as the aisles or the racetrack.

(11)

10 1.2 Problem Statement & Research Questions

Based on the previous discussion of the absence of actions in the consumer in-store decision process the problem statement of this research is as follows:

Does incorporating consumer in-store behavioral data improve modeling purchase behavior?

To further investigate the problem statement these three research questions will be used to guide the study

1. What is the effect of in-store actions done by a consumer on purchase vs. non-purchase? 2. What is the effect of in-store actions undertaken on the total amount spent by a consumer? 3. What is the effect of in-store actions a consumer undertakes on the number of products

purchased?

These research questions are based conditional on purchase, how many products does a shopper purchase and how much do they spend?

1.3 Research Contribution

Prior research has inferred purchase intentions and shopping behavior by analyzing factors such as scanner data, grocery paths, price elasticities, time duration, pre-shopping factors, products in the basket, and more through the final purchases at the checkout (Guadagni & Little, 1983; Kamakura, 2012; Inman et al., 2009). This study in contrast will incorporate the actions a consumer undertakes during their in-store deliberation in order to infer their likelihood of purchase, total amount spent, and number of products purchased. This research hopes to shed some more light on what actions play a crucial role in furthering the conversion of a consumer from visiting to purchasing the merchandise in-store.

The most important contribution of this study will be to incorporate consumer in-store action with final consumer purchase decisions. Till now there has been very limited research off-line about the actions a consumer undertakes in-store that influences their purchasing decisions. This has been investigated more extensively in online shopping as it is easier to track and see the deliberation process of a consumer, and what product is purchased. In turn, retailers can use this information to adjust their shopping environment to entice consumers to be more actively

(12)

11 Retailers would benefit greatly from knowing how a consumer interacts with products before they select them for purchase. This could lead to placing appropriate stimuli to trigger cues within the shopper or the adjustment of shelving and the aisles to facilitate these actions (Inman et al., 2009). The managerial implications could be vast for stores as the findings could help managers optimize their stores further based on the knowledge of which actions increase purchase probability, and to place strategic shelving and feature & displays to maximize and engage the customer more in their deliberation process. This study differs from the Hui et al.(2013) study as its main focus is to infer purchase likelihood, total amount spent, and number of products purchased from the effect of in-store actions and other shopper characteristics instead of focusing on the difference between planned and unplanned purchases.

1.4 Setup of the Report

(13)

12

2. Theoretical Framework

2.1 The In-store Shopping Process

The key to having an effective retail strategy which maximizes profit per customer lies in understanding and satisfying customer needs better than the competitor. The buying process can be divided into several crucial stages consumers go through when making purchase decisions. It begins with the consumer recognizing an unsatisfied need – either hedonic (pleasure) or

utilitarian (accomplishing a specific task). The unsatisfied need is recognized when the

consumer’s desired state of satisfaction differs from their present level of satisfaction. Once the consumer is aware of their need, the next step is to search for information on products, services, and retailers which may fulfill their need (Levy & Weitz, 2012).

Consumers draw information from internal sources such as memory, previous shopping experiences, and images. When a consumer assesses that their internal information is not

satisfactory, they refer to external sources such as advice from friends, family (reference groups), and in-store sales representatives (Hoyer & McInnis, 2010). The extent to which they engage in information search is directly related to the degree that the additional information will improve their purchase decision. Search costs depend on the amount of time and money that a consumer invests in the process. In addition, the information search process is also influenced by (1) the characteristics of the individual consumer and (2) the buying situation in which the purchase is completed (Levy & Weitz, 2010).

Often, consumers resort to gathering information on the product by going to the retail

(14)

13

2.1.1 The In-store conversion process

Sorensen further dissected the in-store decision making process into two components “visit-to-shop” and “shop-to-buy” referred to as the “double conversion” process. A consumer either is in a state of “visiting” as they are browsing through the aisles, and convert when they interact with the products (“shopping”), and then convert to buying when they make a final purchase decision. Hui et al. (2009) expanded on the Fader et al.

(2005) study by including the “double

conversion” process into the shopping path and purchase behavior of the consumer in-store. The aim of the studies were to discover what factors influence the consumer’s ability and willingness to visit, shop, and buy. Their model included three aspects of in-store shopping: (1) where the shoppers visited and their zone-to-zone transitions, (2) the duration a consumer visited and shopped in each zone, and (3) what product categories were purchased.

Shoppers travel a path within store where they engage in the purchase decision process which consists of the different stages - “visit, shop, and “buy” (Sorensen, 2003; Hui et al, 2009). Figure 1 shows that the consumer makes their first “visit” decision when they are at the store’s entrance; here they decide which section of the store they want to visit next. A consumer converts from “visiting” into “shopping” when they engage in actions with the products such as looking at the shelf, searching the shelf, handling the product. Shopping converts into “buying” when a consumer selects the product and puts it into their basket or shopping cart.

Shoppers can be either in a “browsing” state or a “deliberative” latent state; in each state they exhibit different “search” behavior (Montgomery et al., 2004). Consumers tend to walk through the “race track” (perimeter) of the store only entering aisles which contain the specific products they are looking for (Fader, 2005; Hui et al., 2009). When they enter these aisles, the “double conversion” process happens when they encounter the product category they were looking for. In order for a sale to be completed a consumer first needs to visit the area where the product is

(15)

14 located that they were searching for. A consumer will typically search the shelf facings to find the brand they were looking for and use external cues to select the product they want (Baker et al., 2002). Then they may complete actions such as grabbing a product, examining the product, trying the product and if it is satisfactory putting it in their basket (“buying”), or returning it back to the shelf (Sorensen, 2003). This is more than simply just browsing through the aisles (“visiting”); consumers stop at a particular area, examine the product (“shopping”) through interacting with the merchandise and finally select a product and place it in their basket (“buying”) (Sorensen, 2003).

This process is repeated several times throughout their time in-store until they proceed to the checkout (Hui et al., 2009). Thus it can be hypothesized that a consumer may exhibit different actions while in the process of “visit-to-shop” in comparison to when they are converting from “shop-to-buy”. Key findings were that the longer the duration of a shopping trip, the more

purposeful a consumer became in their shopping behavior (shop and buy), and the less likely they were to explore (Hui et al., 2009). However, how shoppers travel in-store was found quite

inefficient as they did not take the shortest overall route for the different items they purchased (Hui et al., 2009)

Retailers know that a lot of product consideration happens in-store and try to encourage it further by placing features & displays, brochures, price-check scanners, and having sales personnel present in order to inform the consumer. These tools are all part of the retailer’s objective to limit the consumer’s information search to its store and to get them to make a purchase with them instead of with a competitor.

(16)

15 2006; Grohman et al., 2007). Shoppers often consider several alternatives of the product they are evaluating, and based on their gathered information (through search & possibly trial) finally arrive at a final decision of purchasing it or not (Hui et al., 2013). The conversion rate captures how successful a retailer is in converting a shopper into a buyer, as it is the percentage of consumers who enter the store and subsequently buy a product at this store” (Levy & Weitz, 2010).

2.1.2 Product consideration through consumer actions

Limited studies exist exploring the in-store product consideration behavior of a consumer. This has relegated many studies to depend mostly on survey and scanner data (Beatty and Ferrle, 1988; Inman, Winer and Ferraro, 2009; Bell, Corsten, and Knox, 2011). However, there are some descriptive studies where researchers shadowed shoppers during their time in-store and noted how many times a consumer handled items, stopped, and how many products they purchased (Granbois, 1968; Cobb and Hoyer, 1985). Areni and Kim (1994) did some preliminary quantification by examining number of items examined or handled, coupled to where the products were displayed on the shelves. Some descriptives were included regarding the average number of products purchased, the total amount spent, and time spent in-store. Other studies have given shoppers a handheld scanner to record the barcode of each product they placed in their carts (Stilley, Inman, and Wakefield, 2010). Hui et al. (2013) take it further by introducing video-tracking, equipping each shopper with a portable video camera, to record the shopper’s in-store deliberation process and understand their behavior in a field setting. Their aim was to investigate incidence, category propensity, behavioral characteristics, and purchase outcome. The result was a descriptive analysis to investigate purchase conversion versus walking away in unplanned consideration (Hui et al., 2013). Shoppers who looked at fewer shelves & displays and stood closer to shelf, were more likely to purchase. They concluded that deeper product considerations are more likely to result in purchase (Hui et al., 2013).

(17)

16 shopper, and makes them more likely to purchase (Hui et al., 2013). Not all retail environments allow consumers to engage in direct haptic experiences with the products thus possibly affecting purchase likelihood. Certain shoppers were less confident in their evaluations of products when touch was unavailable. Deeper considerations are more likely to result in purchase (Hui et al., 2013).This leads to our first set of hypotheses:

H1: There is a positive effect on purchase probability, total amount spent, and number of

products purchased when a consumer is more engaged in the product consideration process.

2.2 Time Factors

Time variables interplay into the whole shopping process of the consumer which also may affect the actions that shoppers undertake in-store (Cobb & Hoyer, 1985). It is not only the actions that the consumer undertakes in-store but different factors like how much time they’ve allotted for the trip and how much time pressure they are under amongst other factors that may interplay with the actions they may undertake in-store (Eroglu & Harrell, 1986). For the purposes of this study they will be limited to “time” – time duration and time pressure.

2.2.1 Time Duration and Time Pressure

The in-store decision process is also characterized by the consumer spending time searching for products in-store, traveling from one zone to another, and when they arrive at the product, spending time examining the products available (Monga & Saini, 2009). Time can be subdivided into two concepts: time duration and time pressure. Suri and Monroe (2003) defined perceived time pressure as “a limitation of the time available to consider information or to make a decision on consumer’s judgments of prices and products”. Thaler (1999) posited that consumers have a sort of mental accounting perspective with regards to time; which means that a consumer allots a certain amount of time for the shopping trip. This “shopping time budget” depletes in-store the more time the consumer spends in-store. Shoppers who spend more time in-store tend to have larger basket-sizes and more final purchases than those with shorter time duration (Hui et al., 2009).

(18)

17 the more time they spend in-store, the less they are likely to explore (Harrell et al., 1980). Thus they travel the shortest distance to the products they are interested in (Hui et al., 2009). Shoppers with shorter shopping paths want to find the products that they came for as quickly as possible, typically due to time pressure (Larson et al., 2005). Furthermore, the longer consumers are in-store the more likely they are to convert to shopping and buying (Hui et al., 2009). Hence consumers may become more purposeful in their actions and less exploratory – thus engaging in actions that only lead to purchase. This leads to the second hypothesis of this study:

H2a: Time duration positively affects purchase, number of products purchased, and total amount

spent.

H2b: Time pressure positively affects purchase, number of products purchased, and total amount

spent.

H2c: The effect of time duration on purchase is positively moderated by time pressure. 2.3 Social Interaction

(19)

18

2.3.1 “Social Companion”

Shopping companions influence consumers to shop longer in-store and to spend more. Shoppers who have people accompany them during their in-store shopping process, particularly members of the same household, leads to a higher incidence of need recognition (Inman et al., 2009). Verhoef et al. (2009) discussed how other shoppers can assume roles such advisors helping the consumer in their in-store decision process. They are more engaged in the in-store decision making process than they would be alone. As they are spending more time, we can infer that they are engaging in more exploratory shopping which may lead them to engage more with products, thus perform more actions. Leading to the next hypothesis:

H3 Shopping with a companion positively affects purchase, number of products purchased, and

total amount spent.

2.3.2 “Crowding level”

Crowding can be defined as both a physical and a psychological state of the shopper (Stockels, 1972; Harrell et al, 1980). It is a combination of an increasing number of people in the same space which may produce excessive stimulation from social sources (Desor, 1972; Harrell et al, 1980). It is a state of psychological stress that happens when a person’s demand for space exceeds supply (Stokols, 1972). The experience of crowding results from physical, social and other shopper related factors which inform the individual to the actual or potential space constrains (Eroglu & Harrell, 1986). Shoppers can affect each other directly or indirectly by being too close to others creating anxiety (Verhoef et al., 2009).The social presence of different shoppers in a particular area of a store has been found to increase the attractiveness of that area for shoppers (Hui et al., 2009). Furthermore, social influence of other shoppers has a positive effect on the consumer’s shop, buy, and visit conversion (Hui et al., 2009). The “crowding level” of a store may work as a double-edged sword when it comes to the shopping process of its

customers. Eroglu, Machleit, and Barr (2005) found that higher levels of crowding decreases shopper satisfaction.

(20)

19 crowding tend to influence shoppers to become prevention-focused, leading to a shift in

preference towards choice options (Maeng, Tanner, and Soman, 2013). This may cause the shopper to refrain from exploring and taking more actions during their in-store decision process. In retail settings, environmental cues play a crucial role in the individual’s perception of

crowding. The density of shoppers in an area and the degree of enclosure may encourage

perceptions of it being crowded in-store. Additionally, shoppers who are under time pressure are negatively impacted by crowding (Eroglu & Harrel, 1986). Based on these past research findings the next hypotheses can be inferred:

H4a: The more crowded it becomes in-store, the less likely a consumer is to purchase, reducing

total amount spent, and number of products bought.

H4b: Crowding is negatively moderated by time pressure.

2.3.3 “Interaction with Sales Personnel”

The interaction between a consumer and sales personnel in-store is key in the consumer decision-making process and satisfaction (Verhoef et al., 2009). Various literature has covered the sales person interaction defined as the interpersonal exchange between a buyer and a seller (Menon & Dubé, 2000). Service is the cornerstone of the retail/service customer experience (Meuter et al., 2000). The presence of salespeople can influence how much time a consumer may spend searching for products and also shopper satisfaction (Baker et al., 2002).Previous literature has shown that the presence of service from sales personnel have a positive effect on a consumer’s decision-making process (assuming that the consumer is satisfied with the contact). Contact with salesperson also may have an effect on the crowding level that a shopper feels (Eroglu & Harrell, 1986; Wicker, 1986) as the degree of staffing being available in a retail setting indicates the level of crowding to shoppers.

For the purposes of this study it will be limited to the indication that the shopper had contact with a sales person. Thus it can be hypothesized that:

H5a: Contact with a salesperson has a positive influence on purchase, number of products

bought, and total amount spent.

(21)

20 2.5 Customer Characteristics

2.5.1 Store Knowledge

Store knowledge also plays into the buying decision process. Depending on their familiarity with the store, a shopper may display different search behavior in-store. Consumers who have visited a particular store several times, become familiar with the general layout of the store and are more able to navigate the store to find what they are looking for. Their familiarity may influence the in-store shopping process in two opposing ways. When a shopper is in an unfamiliar in-store, they tend to rely more on the external cues to figure out where particular items are, which in turn increases their exposure to in-store stimuli (Inman et al., 2009). They tend to be more engaged in the

shopping process and are more likely to engage in exploratory behavior in order to find out where things are. These consumers tend to display more in-store search behavior than those who are more familiar with the store and have more store knowledge (Levy & Weitz, 2010) Shopping more aisles has been shown to have a positive effect on the probability of making purchases (Inman et al., 2009). With regards to actions undertaken with products, we can hypothesize that consumers who are less familiar with the store will tend to display more actions during the double-conversion process as they are building up their knowledge and also engaging in purchasing decisions.

H6a: Less store knowledge will have a positive effect on purchase decision, total amount spent,

and number of products purchased as consumers who are less familiar with the store will be more dependent on store cues.

On the other hand, when a shopper is familiar with their store surroundings they tend to fall into a routine, thus solely focusing on the task at hand, which limits the amount of store cues they will notice.

H6b: More store knowledge has a negative effect on total amount spent and number of products

(22)

21

2.5.2 Gender

Socialization theory posits that gender influences that each gender perceives the shopping process and store patronage (Noble et al., 2006). Males and females differ in beliefs towards shopping. Campbell (1997) found that the male perspective on shopping is mostly needs-driven as they are mainly motivated by the completion of the purchase. In contrast, females are most likely to see shopping as an experience to be enjoyed from which they derive satisfaction from purchasing products for themselves and others (Campbell, 1997). Females are more prone to browse and socialize with retail patrons and sales clerks than male while males are more likely to seek specific product information than females (Noble et al., 2006). Males are also more time-conscious as each gender perceives time differently (Grewal et al., 2003; Krishnan and Saxena, 1984).

Research has shown that most shopping trips are done by women with young children. Retailers have made many adjustments to the shopping environment in order to accommodate their needs, such as coloring corners for kids, and changing aisles to contain items that young families need. However, males are doing more grocery shopping trips due to partners that work and a more equal division in household tasks (Mortimer & Clarke, 2012). Otnes & McGrath (2001) have categorized male shopping behavior and have found that men browse infrequently, engage in deliberate in-store search, and limit price-comparison shopping.

Noble et al. (2006) found that male shoppers seek to gain information, attainment and convenience in contrast to women who placed more importance on uniqueness, assortment-seeking, and social interaction. Men take less time to complete the shopping task and purchase fewer items than women, but paid a higher price per item. When comparing the time they spent with the number of items they purchase, men seem to take less time to make final purchase decisions than women (Mortimer & Clarke, 2012). Thus we can infer that men engage in less exploratory shopping, and are more goal-directed when they are in-store. This may affect their actions as they typically spend less time in store they may be less inclined to engage in many actions during the “double conversion” process. Based on the above we can hypothesize:

H7a: Gender exerts a significantly larger positive effect on purchase decision, total amount, and

total number of products purchased when a shopper is a female rather than a male.

(23)

22 Furthermore, research by Mortimer & Clarke (2012) also shows that friendly and helpful in-store staff was not crucial for the in-store shopping process for men, as they seek to complete the shopping task as quickly as possible. Thus we can infer that an interaction effect may be present between gender and the action of reaching out to sales personnel in-store.

H7c: The number of times a consumer interacts with a sales person is positively moderated by

gender (when female).

2.5.3 Age

Literature regarding age and its effect on the in-store search behavior has been quite limited. The study by Arnold & Reynolds (2003) held some key insights with regard to shopping motivations and the age of the shopper. It showed that middle-aged men tended to score low on hedonic shopping motivations (pleasure-shopping) and shopped with more utilitarian motives (Arnold & Reynolds, 2003). This finding is quite in line with the literature discussion on gender where males were found to be very goal-directed during the shopping process (Arnold & Reynolds, 2003). These shopping motivations have the nature to be more engaged in the shopping process than those who are simply out shopping with utilitarian motives (to the point). Middle-aged women referred to as “Providers” are most likely to be more engaged in the shopping process and were either shopping for others or for value (Arnold & Reynolds, 2003). Young women on the other hand score high on all hedonic motivations and the shopping process itself is an adventure where they have high engagement with others, and with the retail environment (Arnold &

Reynolds, 2003). Young men can also be motivated by hedonic aspects of shopping thus more for the purposes of gathering information. Through this we can hypothesize that older men would tend to be less engaged in the shopping process, and thus would perform fewer actions during the in-store process. This is in contrast to middle-aged women who are engaged in the in-store shopping process. For the young men and women the effect of their age on actions could be hypothesized to be stronger as they are also very actively involved with the in-store decision making process. Thus this leads to our final hypotheses

H8: There is a positive influence of age on number of actions completed when a consumer is

(24)

23 2.6 Conceptual Model

Based on the literature framework discussed in the previous sections, the following conceptual model was constructed highlighting the main effects. Figure 2 shows the relationships between the independent variables of in-store process, time, social interaction and shopper characteristics and the dependent variables of purchase decision, total amount spent, and number of products purchased. The interaction effects will be tested against this main model to see if they add any significant contribution to the main model.

(25)

24

Figure 3: Store Layout

Figure 4: PDA First Screen

3. Research Design

This research study is divided in three different parts which investigate the different effects that the in-store shopping process, time, social interaction, and shopper characteristics have on purchase decision, total amount spent, and number of products purchased. In this chapter, the characteristics of the dataset are explained and the way that measures are operationalized for further analysis.

3.1 Dataset Description

The dataset is quite unique as it is one of the first to not only track the path of a customer in-store, but also to track the actions that a consumer

undertakes while considering & purchasing a product. The data was collected by French researcher, Julien Schmitt, who followed 190 people in a French beauty-care store. The store sells products such as perfume for men and women, hair care products, body lotions, pharmaceutical products. The layout of the store is as follows – there are two main entrances and different types of shelves and a checkout counter.

A special PDA program was used by the research in order to record all the shoppers’ movement in-store. The program consisted of three different screens which enabled the researcher to accurately record what the consumer was doing in-store.

The first screen displayed the store map which the researcher regularly pressed to accurately reflect the shopper’s location in-store and the corresponding time-stamp. Figure 4 shows the two buttons on-screen “Stop” and “Lâche chariot”. Any time the shopper abandoned

(26)

25

Figure 6: PDA Third Screen

pressed again to continue.

Once a shopper stopped in front of a shelf the “Stop” button was pressed and the second screen appeared. This screen in figure 4 gave different options to record what the shopper was doing in front of the shelf. Each action was individually registered and time-stamped by the researcher by recording the beginning and the ending of the action.

The actions were divided in two main categories:

a. “Global Actions”: looking at the shelf, searching for a product on the shelf, deliberating with a companion, looking at their shopping list, speaking with a salesperson, and an “Other” box.

b. “Product Actions”: grabbing a product, investigating the product, trying the product, putting the product back on the shelf, and buying the product.

Once the shopper resumed walking again the research selected the button of “Terminé” which in return made the first screen reappear. Thus these two screens allowed the capture of a shopper’s behavior during their whole shopping trip. Once the tracking of the consumer ended a third screen enabled the collection of additional data about the shopper such as age, gender, number of companions, time pressure, and store

knowledge.

The final step in the data collection was to export all this data to the computer interface of the software which for example allowed the graphing of zones on the store map. The software also generated a database according to each shopper’s observed behavior.

(27)

26

3.1.2 Database Transformation

For this research study, the databases were merged into one single dataset which contained all the previously mentioned information for each customer. The shopper movement database was aggregated to show the total number of actions per action type and how many seconds they spent shopping in total. The types of in-store actions were recoded from the original values to

categories numbered from 1 to 14 (which can be seen in table 3.2). This was then matched to the Customer ID’s of the additional shopper information, and the purchase data. The merging carried out, uncovered inconsistencies across the datasets which resulted in missing values. The first 11 observations of the dataset were deleted by the researcher due to some technical issues. The merging of the databases revealed that 11 customer ID’s were missing total information which was due to different issues which happened during recording. Thus the sample size went from 190 cases to 168 cases with complete information. Of these 168 cases; there are 104 consumers who purchased a product and 64 that did not purchase any product.

3.2 Initial Summary Statistics

Table 3.2 shows differences between purchasers and non-purchasers. Shoppers who convert into buyers spend nearly twice as much time in-store than those who don’t purchase. They are

(28)

27

Table 3.2: Sample Characteristics of Purchasers vs. Non-Purchasers

Variables Purchaser Non- Purchaser

Number of Prod Purchased 2.79 0.00

Total amount of Purchases (in euros)

73.15 0.00

Promotional Pressure 1.35 0.00

Age 42.02 40.86

Gender Female (.88) Female (.84)

Shopping companion 0.29 0.38

Crowding level 2.09 2.64

Store knowledge 3.25 3.36

Subjective time pressure 2.92 3.39

Shopping duration in min 9.227 5.782

(29)

28

Table 3.3: Sample Characteristics of in-store actions of Purchasers vs. Non-Purchasers In-store actions* Purchaser Non-Purchaser

Browse 53.98 43.98 Stop 8.58 7.28 Restart walking 8.577 7.28 Handle 7.53 3.20 Examine 4.75 1.92 Choose 2.76 0.00 Put back 4.77 3.22 Tries Product 2.14 1.53 Look 9.32 7.09 Search 1.05 0.39 Deliberate w/ companion 0.58 0.56 Check list 0.11 0.03 Consult salespeople 2.50 0.86 Other 0.17 0.13 Total actions 106.81 77.48

(30)

29 3.4 Variable Creation

Due to the high number of variables explaining the different types of actions a consumer undertakes during their buying process, it was necessary to create new variables to be able to capture the same effects while being able to keep the final model as parsimonious & complete as possible (Little, 1960). Also other variables were created to be able to test the hypotheses

aforementioned in chapter 2.

3.4.1 Factor analysis

The dataset contained 14 variables related to the in-store actions that consumers undertake while evaluating a purchase. These 14 search actions correlated very highly with each other which would later on cause problems with modelling as when tested they showed VIF scores greater than 300. This could be expected as these actions are interrelated. To find if there are common constructs PCA was conducted on these variables as they share common variance, (Malhotra, 2010). When the variables of “stop” and “walking” were included, PCA returned a “negative identity matrix” meaning that factor analysis could not be completed. Once these two variables were excluded, data reduction was possible through PCA, it initially showed three factors. The solutions were rotated by using Varimax rotation, communalities checked, and reliability analysis done to see if variable deletion would improve the Cronbach alpha. The result after reliability analysis are two factors labelled “Search” and “Trial & Return” based on Bartlett’s test of

sphericity (p-value of 0.00) and KMO measure of 0.633, which are appropriate for this data (Hair et al., 2010).

Table 3.4: Principal Components Analysis Results

PC1 PC2 Communality

In-store action Search Trial & Return

Handling a product 0.696 0.688 0.958

Examining a product 0.741 0.504 0.803

Put back product on shelf 0.891 0.935

Choose a product 0.832 0.692

Looks at the shelf 0.679 0.935

Searches the shelf 0.786 0.680

Tries the product 0.910 0.621

Explained variance 60.88% 18.03%

(31)

30 The Search factor consists of the actions of “look at shelf, search shelf, examine, handle, and choose the product”. The Trial factor has the highest loadings from “try product” and “put back” – however one may note that “examine” and “handle” also load on the factor. But there is clear distinction between the factors as one is directed more towards searching for the product, and the second towards trying the product (which precedes previous handling & examining of the

product). The factors were pretested together in a regression & logistic model which showed that these variables had high explanatory power with regards to predicting purchase (p-value: 0.00) and multicollinearity did not seem to be an issue (VIF <4, Condition Indices <30).

Another variable that was created was Promotional Pressure (on basket). The aim of this variable is to account for the discount applied to the different products. Time duration was transformed from seconds into minutes as this is easier for interpretation. To account for the different actions that a customer completes, a variable was created that counted how many of the 14 different actions a shopper completed. A complete list of new variables and their descriptions can be found below.

List of New variables:

o Compl_Act_actions : total number of different actions a shopper completed o Time duration_min: total shopping time

o Promotional Pressure: Total promotional pressure on basket o Promo_dummy: Indicator if discount was applied to purchases o Search: construct derived from factor analysis

(32)

31

3.4.2 Variable Testing

An ANOVA (table 3.5) was done with Purchase decision as the dependent variable to help determine the final variables that will go into the model. As the sample is limited, it is key to keep the model as parsimonious as possible, because when the sample size is smaller, the

statistical power of the results are not as strong as when based on a sample with a large amount of observations (Hair et al., 2010). These variables were also pretested in a regression and similar results were found except that shopping companion was borderline significant, thus it is included in the final mode

Table 3.5: ANOVA of Purchasers vs Non-Purchasers (dv: Purchase_decision)

Variables Significant P-value

Promotional Pressure + 0.012

Age - 0.591

Gender - 0.570

Shopping Companion (adult) - 0.246

Crowding level + 0.023

Store knowledge - 0.597

Subjective time pressure + 0.037

Shopping duration time + 0.001

Complete number of different actions completed

+ 0.000

3.5 Measures

After factor analysis, the ANOVA, and pretesting all the variables in a regression the following measures were operationalized for continued analysis. The variables which measured shopper characteristics were removed as they consistently were found not to exert a significant effect with regards to purchase. This already leads to the first conclusion for this study that shopper

(33)

32

Table 3.6: Description of the dependent and independent variables for this study

Variables Description

Dependent variables:

Purchase Decision indicator if a shopper made a purchase decision (0: no 1:yes)

Total Amount The total value of the goods purchased by the consumer in euros

Number of Products purchased The total number of products purchased by consumer Independent variables:

In-store Shopping Process variables:

Search Factor of searching a shelf, looking at shelf, handling a product, examining, and choosing a product

Trial & Return Factor of trying a product and putting it back on the shelf

Number of actions completed The total number of fourteen different actions that a consumer completed (in-store)

Time variables:

Time pressure Measure of consumer’s subjective time pressure (5 point scale)

Time duration Total shopping duration of consumer in minutes

Social Interaction variables:

Shopping with an adult companion indicator if a consumer is accompanied by an adult during their shopping trip

Interaction with sales personnel The number of times a shopper consulted a sales person

(34)

33

4. Purchase Decision

4.1 Introduction

The first research question delves into discovering what the effect is of different variables, the focus being on in-store actions, is on a consumer’s purchase decision. Approximately 64% of the 168 shoppers made a purchase decision in this dataset. To answer the research question of what affects purchase decision, the method of logistic regression was used.

4.2 Logistic regression

The dependent variable of purchase decision was coded as a binary dependent variable with 1 indicating purchase, and 0 as no purchase based on the variable of total amount spent. Logistic regression is the primary choice to model a binary dependent variable (Hair et al., 2010) and was thus applied to investigate the effects of the IV’s on the probability of consumeri making a purchase decision.

Different models with all the variables were tested including interaction effects. Store knowledge, age, and gender did not have any explanatory effect in this dataset and were subsequently

dropped from the model. This also eliminated the bulk of the moderation effects that were hypothesized. In Appendix B.3 the results of the interaction effects can be seen (testing them against the main effects model to see if they provided additional explanatory power). The significance levels of the χ2 test were all above 0.1.

4.2.1 Model Notation

The model is therefore specified as P(purchase) representing the probability that consumeri purchases a product.

𝐏(𝐩𝐮𝐫𝐜𝐡𝐚𝐬𝐞)𝐢 = F(Ui) =

exp (U_i)

1+exp (Ui) (4.1)

where

(35)

34

Model key

𝐴𝑑𝑢𝑙𝑡𝑎𝑐𝑐𝑜𝑚𝑝𝑖 =being accompanied by an adult 𝐶𝑜𝑛𝑠𝑆𝑎𝑙𝑒𝑠𝑖 = consulting a sales person

𝑆𝑒𝑎𝑟𝑐ℎ𝑖 = looking at shelf, searching the shelf, examine, handle, and choose a product 𝑇𝑟𝑖𝑎𝑙_𝑅𝑒𝑡𝑢𝑟𝑛𝑖 = try product and put back

𝐶𝑜𝑚𝑝𝑙_𝑎𝑐𝑡𝑖 = total number of the 14 different actions (see appendix) completed by consumeri. 𝑇𝑖𝑚𝑒𝑝𝑟𝑒𝑠𝑠𝑖 = subjective time pressure indicated by shopper on a scale 1-7

𝐶𝑟𝑜𝑤𝑑𝑖𝑛𝑔𝑖 = crowding level of the store (indicated by the store supervisor at the end of shopping trip of consumeri )

𝑇𝑖𝑚𝑒_𝑑𝑢𝑟𝑖 = total shopping time duration of consumeri 𝜀𝑖 = error term

4.2.3 Multicollinearity

In marketing research, it is common for variables to correlate with each other, which may affect the proposed model (Mahajan et al., 1977). To assess the effect of multicollinearity on the model, the bivariate correlations and variance inflation factors were investigated. VIF cut-off values of 2.5 were used (Wieringa and Verhoef, 2007). The bivariate correlations between time duration and 5 other regressors correlated above 0.4 at a p-level of 0.05, causing significant

multicollinearity (see Appendix A: Table 1 bivariate correlations 4.1). Time duration also had the highest VIF value above 4; hence the decision was made to eliminate it from the model. Although time duration can be seen as a crucial variable for this model, it can be argued that all the

remaining variables capture the effect of time as they are performed over a length of time. After the deletion of shopping time, the VIF scores and bivariate correlations were again checked. Table 4.2 shows that the bivariate correlations between the regressors are now less than 0.4 in 71% of the cases, and the VIF scores are all below 2. Hence multicollinearity does not appear to be an issue of concern in the final model without time duration.

Thus the final model of the utility of consumeri is specified as

(36)

35

Table 4.2: Bivariate correlations and VIF values (model notation 4.2)

4.3 Findings

The estimated results of the final model are displayed in table 4.3 The LL of the full model is -67, the χ2 _{is 87.65 (p-level: 0.00), therefore the variables affect purchase decision significantly. The} model fit is good with a McFadden R2square of 0.533. The percentage of correctly specified values in this case is 81%. The fraction of observations of y=1 correctly specified is 87.5% (sensitivity) and the ratio of non-purchase correctly classified is 70.5% which is relatively high. So the model performs slightly better at predicting purchase (y=1) than non-purchase (y=0). A global F-test was also conducted to see how the constrained (3) model performs compared to the original model (2). The resulting χ2 is 1.94 at a significance level of 0.1639 which means that the removal of time duration is justified. Next, the interaction effects were tested via a global f-test (model with the interaction effects vs. model without the interaction effects). The result showed a χ2 of 0.22 at a p-value of 0.6425 deeming the interaction insignificant, thus the moderation variable does not add any additional explanatory power to the model, consequently it was removed from the model.

Bivariate correlation 1 2 3 4 5 6 7 VIF 1Accomp_adult 1 1.18 2Crowding .016 1 1.13 3Comple_Act ,175* -,291** 1 1.96 4Search -.116 -,167* ,424** 1 1.47

5Trial & Return ,193* -,179* ,485** .000 1 1.62

6Consulting Sales _{-.096 -,181}*

,404** ,215** ,390** 1 1.42

7Time_Press -.130 ,270** -,499** -,362** -,411** -,410** 1 1.62

Note: *. Correlation is significant at the 0.05 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed).

(37)

36

Table 4.3 Results of Final Model Logistic Regression

Variable Coefficient Standard error Exp(B) Constant -7.33*** (1.773) 0,000 Search 1.319*** _(0.153) _3,741

Trial & Return -.558* (0.304) 0,572 Complete_act 0.794*** (0.175) 2,213 Consult Salespeople 0.279 *** (0.113) 1,321 Crowding 0.007 (0.153) 1,007 Accomp_adult*(1) -0.806* _(0.488) _0,447 Time_press 0.495** (0.205) 1,641 R2_square (McFadden) 0,553 % correctly predicted 81% Sensitivity 87.5% Specificity 70.3% significant at *0,1 - **0,05 -***0,01

*base category of accompaniment adult is 0 (not accompanied)

All the variables except crowding have a significant effect on a consumer’s purchase decision with p values less than 0.1. The variables that are highly significant at a p-value of 0.01 are search, completing different type of actions, consulting a sales person, and time pressure.

(38)

37 pressure (5,585). Trial & Return has the largest negative effective (3,380) followed by shopping with an adult companion (2,275).

4.2 Discussion

These findings suggest that purchase probability is primarily influenced by number of completed actions (how many of the fourteen in-store actions a consumer completed), but is negatively influenced by being trying a product in-store as it is most likely to be returned to the shelf instead of purchased. In contrast to what was hypothesized, trying a product does not benefit the

consumer’s purchase probability as it actually decreases their likelihood of purchasing. Contact with a salesperson, searching in-store, and time pressure also play a key positive role in

influencing a consumer to complete a purchase decision. Two of these findings are according to what was hypothesized but in contrast to what was hypothesized, time pressure actually exerts a positive instead of negative effect – meaning that time pressure actually propels consumers to make a purchase decision. The crowding level of the store did not have an effect on the purchase decision but this may have to do with the matter of assessment, as the store manager rated the store crowding of the entire store, instead of crowding being measured at the places where the consumer traveled in-store.

(39)

38

5. Total Amount Spent

5.1 Introduction

The logistic regression revealed some interesting insights into what influences people to convert from shoppers to buyers. But it made no distinction between how much shoppers spent, as one euro and 280 euro were deemed the same, a purchase. The minimum amount spent was 3.90 euros and the maximum amount 281.84 with a conditional mean of 73.14 euros (for those who purchased). It is salient to see how variables influence the total expenditure of a consumer. Hence, this part of the research delves into the research question, conditional on purchase, how do in-store actions (and of course other variables related to time and social interaction) affect the total amount spent by a consumer? This means that the data is censored to only include those consumers who have a total amount spent greater than 0.

5.1.1 Variables

The use of censored regression models is salient to further expound on the influence of the explanatory variables used in the logit model. The dependent variable in the censored regression model is Total amount. The advantage of using a censored regression model is that it

automatically censors either the lower limit, the upper limit, or both when specified. For the purposes of this study, it will be the lower limit which pertains to those who have not purchased (amount spent =0) as the interest lies in investigating the explanatory power of the variables conditioned on purchase. For the subsequent models, the explanatory variables used previously in the logit model are used interchangeably. These are: time search, trial & return, time pressure, adult shopping companion, consulting a sales agent, and total number of different actions completed.

First, the distribution of the dependent variable Total_amount was investigated as the assumption of normality is crucial for the Tobit I model. Table 5.1 shows that the distribution of the values indicates that the variable of Total_amount is very heavily skewed to the right tail of the

(40)

39

Table 5.1: Distribution of the dependent variable Total Amount Spent Total_amount Measurements Untransformed (n=168) Censored (zeros dropped) (n=104) Log Tranformation (n=168) Mean 45.28 73.14 2.98 Standard Deviation 57.69 57.78 1.44 Variance 3328.38 3338.51 2.08 Skewness 1.71 1.50 0.04 Kurtosis 6.00 5.11 1.44 5.1.2 Lognormal transformation of DV

Since the assumption of normality is violated in this dataset, a transformation of the dependent variable may be helpful. Expenditure data tends to violate the assumption of normal distribution and is better suited to be modeled as lognormal (Cameron & Trivendi, 2010). The data

transformation yields the dependent variable lny instead of y, and the (lower) threshold equals the minimum uncensored value of lny. The censored values of lny are set to a value equal to or less than the minimum uncensored value of lny. The variable of dy is also created for use in the selection equation, this variable indicates where Total_amount > 0.

(41)

40

5.1.3 Tobit I model specification

The Tobit 1 regression is then specified as an unobserved latent variable of y*

y* = exp (β1Adultaccompi+ β2ConsSalesi+ β3Searchi+ β4Trial_Returni+ β5Compl_acti+ β6Time_pres i + β7Crowdingi + εi), i = 1, …..N (5.1)

where

εi ~ N(0,σ2) , and xi indicates (Kx1) vector of exogenous and fully observed regressors.

The observed variable of yi is linked to the latent variable of yi* through the observation rule that y = y* if lny* > γ and 0 if lny* ≤ γ

5.3 Validation

To check if the model is correctly specified various components of the test statistic are computed and stored. The inverse of the Mills’ ratio, λi,, and other related variables are calculated, this includes also the generalized residuals (Cameron & Travendi, 2010). The generalized residuals need to sum up to zero in the sample (Cameron & Travendi, 2010). The residuals test in table 5.3 shows that the mean of residual 1 is -0.43956 thus it is concluded that the model is not correctly specified. To try to remedy this problem, variables were selected for deletion. The first variable deleted was time pressure as it had been consistently insignificant.

Table 5.3: Results of Tobit I model specification test

Variable Mean Std. Dev Min. Max residual -0.440 0.414 -1.642 0.433

After the removal of the variable of time_pressure, once again the residual tests were run to verify correct model specification. Again, it showed that the residual did not sum up to zero so the variable of complete_actions was removed (as it is one variable that correlated the most with the other variables) as when it was added to the model, it caused significant changes in the betas. After the removal of this variable, the residual did not sum up to zero so the next variable

(42)

41 removed, the mean did sum up to zero (-2.23e-08) from which it can be assumed that the Tobit I model is now correctly specified with the independent variables of search, trial & return,

promo_dummy, and consulting sales people (see Appendix B: Table 4). To further validate this model, crucial assumptions relating to normality and homoscedasticity were tested to investigate whether they held.

5.3.2 Test of normality

Cameron & Travendi (2010) specify how to implement the test of normality for Tobit I by computing and storing various components of the test statistic. This consists of the inverse of the Mills ratio, i,, and related variables are calculated, including the generalized residuals.

The test outcome with a p-value of 0 (1.381e-25) shows a very strong rejection of the normality hypothesis though the dependent variable of Total_amount was log-transformed (refer to Appendix B: Item 3 for formula used).

5.3.3 Test of homoscedasticity

The outcome of this test with a p-value of 0 (7.7407e-28) also indicates a very strong rejection of the null hypothesis of homoscedasticity against the alternative hypothesis that the variance is of the form specified (refer to Appendix B: Item 3 for formula used)

5.4 Two-part censored regression model

These two diagnostic tests reveal weaknesses with the Tobit I model which was assumed to be satisfactory (after the first residuals test). The Tobit I model asserts a strong assumption that the same probability of mechanism generates both the zeros and the positives. The failure of meeting the assumptions, have serious implications for the Tobit I model thus another approach is

(43)

42 The type-2 Tobit model is used as it considers the possibility of allowing dependence between the two parts of the conditional equation, and the outcome equation based on the observed outcome. In the Tobit I model, the outcome is observed if y2* (the outcome variable) is greater than zero. The type-2 Tobit model introduces a second latent variable, y1*, and the outcome y2* is observed if y1* > 0. In this research study, y1* determines if a consumer has purchased, y2* the level of expenditure, and y1*≠ y2*.

(1) The selection equation is as follows for y1 = { 1 if y1* > 0 and 0 if y1* ≤ 0 } (2) The outcome equation for y2 = { 1 if y1* > 0 and -- if y1* ≤ 0 }

(3) Consequently y2 is only observed when y1* > 0.

5.4.1 Model specification

A pretest was done with a Heckman model without exclusion restrictions to find an appropriate variable for use as an exclusion restriction. Cameron & Travendi (2010), state that the exclusion variable should have a substantial impact on the probability of selection, but should not affect the outcome directly. The only variable to meet these requirements was time pressure. As it is

significant in predicting purchase but when it came to predicting total amount it was found insignificant (see appendix B: table 5). Thus the model for the Heckman regression with exclusion restriction can be specified as

y₁∗ = x₁′β₁+ ε₁ (5.4)

y₂∗ = x₂′β2+ ε2 (5.5)

where x₁′_β

1= β1Adultaccompi+ β2ConsSalesi+ β3Searchi+ β4Trial_Returni+ β5Compl_acti+ β6Time_pressi

x2′β2 = β1Adultaccompi + β2ConsSalesi+ β3Searchi+ β4Trial_Returni+ β5Compl_acti

5.4.2 Findings

The Wald Chisquare of the full model is 22.81 at a p-level of 0.0004, from which can be

(44)

43 between the errors is not significantly different from zero, thus the null hypothesis that the two parts are independent cannot be rejected.

Table 5.4: Parameter estimates and standard errors for selection and total amount equations (5.4) and (5.5)

Variable dy (selection) lny (total amount)

Coefficient Standard Error Coefficient Standard Error Constant -4.315*** 0.921 1.618** 0.821 Search 0.773*** 0.254 0.043 0.080

Trial & Return -0.385** 0.168 -0.112 0.107 Complete_act 0.450*** 0.091 0.234*** 0.082 Consult Salespeople 0.181*** 0.627 0.066** 0.030 Time_press 0.317*** _0.111 _- _- Accomp_adult (1) -0.321 0.276 -0.414** _0.197

LR test of independent equations(rho= 0) χ2 = 2.13 p-level: 0.1449

5.4.2 Selection Equation

(45)

44 on predicting purchase in contrast to the other parameters. This is also to be expected as those who undertake more search actions are generally more highly engaged in the shopping process, and more likely to purchase the product after consideration.

5.4.3 Conditional Equation

Interestingly when it comes to predicting the total amount spent, search and trial and return do not exert a significant effect (in this dataset). They are highly insignificant; however, if a shopper completes one more action, their total amount spent increases by 23.3%. In contrast with the selection equation, being accompanied by an adult may not exert influence on a shopper and if they will purchase, but with regards to the amount they spend it does have significant negative influence (𝛽̂₁ = -0.414 ). If they are accompanied by an adult their total amount spent decreases by 41.4%. This is the largest effect on total amount spent by a parameter, and also the largest negative effect thus it is actually detrimental to the total expenditure if a shopper is accompanied. Finally, if a consumer increases one more unit by consulting a salesperson, their total amount increases by 6.6%. Hence, it can be concluded that search and trial predict purchase, but they do not exert any explanatory power when it comes to the total amount spent, conditional on

purchase.

5.5 Predictive Power

5.5.1 Tobit I

(46)

45

Table 5.5: Comparison of predicted values of Tobit I model (uncensored and censored)

yhat all observations 168 observations

Variable Mean

Std.

Dev Min. Max

y 45.28 57.69 0 281.84

yhat 267.10 1649.39 7.58 20042.97

yhat when ytrunchat dy if dy>1 104 observations

Variable Mean

Std.

Dev Min. Max

y 73.15 57.78 3.9 281.84

yhat 409.54 2087.25 19.34 20042.97 ytrunchat 428.50 2084.83 37.50 20043.79

5.5.2 Heckman with exclusion restriction

The Heckman model with exclusion restriction however, predicts very well. Comparing the means of y (45.28) and yhat (47.49) of the whole sample – the difference is only 2.19 euros. The actual minimum and maximum amount versus the predicted are also close in range. When the sample is limited to only purchases, the difference is also small with 4.35 euros (see table 5.6 below), confirming that this model is the best fit for this dataset (compared to the Tobit I model).

Table 5.6: Comparison of predicted values of Heckman model (5.4) with excl. restrictions (censored and uncensored)

yhat all observations

168 observations

Variable Mean

Std.

Dev Min. Max

yhatheck 47.49 43.35 0.0033514 270.998

y 45.28 57.69 0 280.84

probpos 0.615 .326 0.000006 1

(47)

46

yhat when probpos dy if dy > 1 104 observations

Variable Mean

Std.

Dev Min. Max

yhatheck 77.49 38.30 23.8449 270.9986

probpos 0.783 0.232 0.1125 1

dy 1 0 1 1

y 73.15 57.78 3.9 281.84

5.6 Discussion

Two models were tested in order to predict the effect of search, trial & return, consulting a sales person, being accompanied by an adult during shopping, and time pressure on the total amount spent. The Tobit I model proved to be a poor fit for the datasets it severely overestimated the y compared to the actual spend. The poor prediction was in line with the violation of the strict assumptions of normality and homoscedasticity upon which the Tobit I model relies. In order to remedy this problem, a second model was applied to the data set, a Tobit II model and a

Heckman model with exclusion restriction. This provided the necessary relaxing of the assumption as it consists of a selection equation, and a conditional equation. Thus the

mechanisms of selection are different from the ones which predict the total amount. This proved to be a better fit for the data, as can be seen from the predicted values which are in close

proximity.

Search and trial & return predicted if a consumer would purchase, however these variables did not have any additional explanatory power on total amount spent. This is in accordance with the previous analysis of purchase decision where the in-store shopping process variables were found to have a significant effect (the signs of the effects are also in accordance with the purchase decision model). Being accompanied by an adult on the other hand, did not predict purchase but affected the total amount spent by a consumer significantly. From this result it can be concluded that for the sake of expenditure, it is better if a shopper is not accompanied. Completing the different type of actions in-store is significant in predicting if a consumer purchases, and also has a positive effect on total expenditure, as the more actions they complete, the more their

(48)

“Seek and you shall find”