• No results found

How much will a consumer spend next?

N/A
N/A
Protected

Academic year: 2021

Share "How much will a consumer spend next?"

Copied!
71
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)
(2)
(3)

Executive summary

Paying money in exchange for goods and services is extremely common in modern society. Despite being so common, making payments is a dispassionate affair and evokes an emotional response from the consumer. Recent technological advancements made it possible to track a consumer’s shopping trip on purchase level, allowing the researcher to investigate the sequence of purchases. Previous research shows that shopping dynamics exist, meaning that subsequent decisions at least in part depend on previous decisions. This is not yet studied in the context of spending decisions. That is why this research examines the drivers of spending dynamics in a grocery store. Combining literature on spending, temporal sequence and shopping emotions several drivers are identified that could possibly affect spending decisions made by a consumer.

(4)

Preface

(5)
(6)
(7)

1. Introduction

(8)

(Murray, Talukdar and Gosavi, 2010) and the effects of price reframing on consumer perceptions (Shirai, 2017).

A topic that has not received much attention yet is how the prices paid by a consumer evolve throughout a shopping trip and how this is affected by previous purchases within that trip. Research shows that previous purchase decisions influence following purchasing decisions (Bell, Corsten and Knox, 2010). Additionally, research has shown that consumer decisions are dependent on the stage of the shopping trip (Lee and Ariely, 2006). Also, certain types of shopping behavior, such as unplanned buying occur dynamically throughout a shopping trip, based on the shopper’s previous decisions (Stilley, Inman and Wakefield, 2010). Furthermore, Sheehan (2015) finds that relative spending evolves nonlinearly over a single shopping trip. Hence, previous research suggests that shopping dynamics exist.

(9)

1. What are the drivers of spending decisions during a shopping trip? 2. What are the drivers of the average spending level of a shopping trip? This study extends current research in several ways. First, this is the first study on this topic to use handheld scanner data. This has two main benefits: the consumers are observed in their natural environment and the sequence in which the purchases are made is known. By studying consumers in their natural environment, the risk at biases is reduced. Knowing the sequence of purchases allows us to examine the effect of previous purchases on later purchases. Previous research barely paid any attention to this as it mainly focused on cross-sectional analyses of end-of-trip variables, such as basket composition. Second, this research provides insight on how emotions, such as feelings related to paying affect subsequent purchases within a shopping trip. As mentioned before, people can both enjoy or dislike spending money. Finding out what evokes these emotions will generate new insights on consumer decision making in grocery stores.

1.2 Outline

The remainder of this study will first present an overview of the conceptual model, followed by an overview of the relevant academic literature that will be linked to the conceptual model, which results in several hypotheses. The subsequent chapter presents the methodology section, which includes the models that are used to test the hypotheses stated in second chapter. The next chapter provides some descriptives of the data. The following chapter discusses the results, which are subsequently placed in the broader context of the literature, followed by the implications of the outcomes of this study. Lastly, the limitations of this study and recommendations for future research are provided. After the final chapter, the reference list and appendices are included.

(10)

2. Theoretical framework

This chapter reviews the literature on which the hypotheses are based that help to answer the research question. First, the conceptual model is presented which is a visual representation of the variables under research and their expected relationships. Second, relevant academic research related to the variables included in the conceptual model is discussed and the hypotheses are formulated. Furthermore, several psychological theories are discussed that help to explain the relationships within the model. The chapter ends with an overview of all the hypotheses.

(11)

2.1 The basics of spending money

When a consumer starts a shopping trip in the grocery store, he or she usually plans on spending money. In its simplest form, spending money is not more than an exchange: one person makes a payment to another and receives something in return. Spending money is not a random occurrence and it is done with the intent to achieve or create a particular outcome (Carter, 2014). In the grocery store, this outcome is to buy food and other products that help to fulfill human needs. The decision to spend the money is based on the believe that the benefits of spending the money to purchase these products and fulfilling these needs are larger than spending the money on possible alternatives (Mellers and McGraw, 2001). In addition to the expected benefit, spending money also involves costs. This holds the price of the products, but also an opportunity cost, since the money used to pay for the purchases can no longer be used for other purposes. So, spending money has two effects: the loss of money and opportunity, and the gains related to beliefs about future hedonic states (Carter, 2014). The decision to spend or not to spend the money is thus a result of the interplay of those two and can be seen as a sort of cost-benefit analysis. To arrive at the decision to spend money, a consumer evaluates the different alternatives and predict how each alternative will make him or her feel.

In theory, more options should lead to better outcomes for consumers, since the chance that there will be a product perfectly suited to their preferences increases (Kahn and Lehmann, 1991; Shugan, 1980). However, in many instances, the number products to choose from has extended far beyond what is good and comprehensible for consumers (Schwartz, 2004), which demotivates people to start the decision-making process (Iyengar and Lepper, 2000). This means that in practice, consumers do not have the cognitive ability or the willingness to process the large choice sets which leaves them feeling confused and unconfident (Haynes, 2009). As a result, consumers can resort to a heuristic-based decision-making process (Wansink and Chandon, 2014), where they rely on cues to make a quick decision instead of truly outweighing all the different alternatives.

(12)

2.2 Spending dynamics

Previous marketing and economic models assume that the spending evolves linearly during a shopping trip (Bell et al., 2010). These models implicitly state that the relative spending, defined as the price of the item relative to the mean price of the product category, is constant throughout the shopping trip. That would mean that within each product category, the shopper picks an item that has the same relative price as the previous as well as the following item. Cases exist where a shopper might deviate a little, by selecting some items that are relatively more expensive than others, but in general it is assumed that spending dynamics are minimal (Wakefield and Inman, 2003).

The decisions of a consumer during a shopping trip should not be viewed in isolation.

Research suggests that preferences amongst alternatives might be affected by consumer’s prior decisions (Novemsky and Dhar, 2005). To illustrate, Dhar, Huber and Khan (2007) found that shoppers go through a process which is called shopping momentum. Shopping momentum occurs when a consumer has made an initial purchase. This initial purchase enhances the purchase of a second, unrelated product. This theory is based on work of Gollwitzer’s (1990), who developed a theory about implementation and deliberation mind-sets. Based on this theory, the shopping momentum states that a consumer shifts from a deliberative to an implemental mind-set, thereby driving subsequent purchases.

When a consumer spends money on a product during a shopping trip, this could have

(13)

of what-the-hell (Cochran and Tesser, 1996). Here a bad decision is followed by another bad decision, such as two expensive products in a row. The customer feels like he or she is already ‘lost’ and the motivation to engage in good behavior decreases. Within this study, spending decisions will be discussed in terms of relative spending. Hence, the price of a product will be compared to the average price of the product category by using an index. With this method, the issue of large absolute price differences between categories is removed. Relative spending, relative price and price index will be used interchangeably throughout the remainder of this study.

2.2 Temporal sequence related drivers of price dynamics during a shopping trip

2.2.1 The primacy effect: price index of the first purchase Previous research has already studied temporal sequence effects. Crowder (2014) states that experiences that are found early in a sequence of events tend to have a stronger impact on the overall evaluation of the event. Furthermore, Montgomery and Unnava (2009) argue that when people need to memorize a list of items, they are likely to better memorize the first items. This phenomenon is called the primacy effect. An explanation for this is that the first experiences or items have less other items to compete with for the limited memory capacity held by humans (Waugh and Norman, 1965). When applying this to spending dynamics, this indicates that the first purchase is likely to influence the purchases later in the sequence. Shopping momentum, as described earlier in this chapter, occurs when the purchase of an item increases the likelihood of purchasing another, unrelated product (Dhar, Huber and Khan, 2007). From a normative perspective, this is not a very logical consequence. Since the goods are not complements, the consumer should assess the value of each purchase individually by making a utility-maximizing choice for each item. This means that there should not be any systematic increase from initiating one purchase to the likelihood of buying other items. As a result of budget and income constraints, spending the money on one item could even potentially decrease the likelihood of subsequent purchases.

(14)

switch to an implementation mind-set, as a consequence of the first purchase. This implementation mind-set evokes feelings of commitment to purchase by reducing the psychological barrier to action. Chandran and Morwitz (2005) suggest that a price participation (e.g. auction) exercise leads to an implementation focus that in turn leads to a greater purchase intent, than a fixed price offer for the same product. That is why Dhar, Huber and Khan (2007) state that an initial purchase can induce an implementation focus and that this focus, consistent with Chandran and Morwitz (2005) can lead to greater subsequent purchase. Based on the research on temporal sequence effect, there is reason to believe that the first purchase affects purchases later in the sequence. When combining that with the theory of shopping momentum, which states that the initial purchase leads to a greater subsequent purchase, this results in the following hypotheses: H1a: The price index of the first purchase is positively related to the index of the next purchase H2a: The price index of the first purchase is positively related to the total basket average price index 2.2.2 The recency effect: price index of the previous purchase Another finding by Kahneman et al. (1993) is that the end of an experience is more heavily weighted in the evaluation of the experience than that what came before. The temporal sequence literature calls this phenomenon recency (Greene, 1986). When asked right after an event took place, the parts of the experience that are most recent are remembered well and hence show a disproportionately large effect on subsequent actions than other parts of the experience that are remembered less well. In a pricing context, this means that a consumer is more aware of the price of the most recently bought product compared to products the customer picked at other, earlier moments of the shopping trip. Consequently, the most recently bought purchase is expected to affect the subsequent purchases. Thus, there is a reason to assume that the previous purchase affects the purchase coming right after it. The consumer is expected to clearly remember the price of the previous purchase and will take this into account when considering the next purchase.

(15)
(16)

the weighted average of the most intense part of the movie as these were most clearly remembered. Montgomery and Unnava (2009) also studied peaks and evidence for found peak intensity indeed affecting evaluation. Since peaks are well-remembered by a consumer, they are highly likely to affect subsequent choices. When a consumer has reached a peak or a dip, it is expected that he or she will start to balance out this peak or dip using the subsequent purchases. The same mechanisms described in section 2.2.2, licensing and guilt, are expected apply here. It is expected that after a dip, the consumer feels that he or she saved money and uses this dip to justify subsequent expensive purchases due to the previously mentioned licensing effect (Khan and Dhar, 2006). Adversely, after reaching a peak, the consumer is expected to continue with cheaper products due to feeling of guilt. This results in the following hypotheses: H1c: The previous peak during the shopping trip has a negative influence on the price index of the next purchase

H1d: The previous dip during the shopping trip has a negative influence on the price index of the next purchase

The expected effect is slightly different with regards to the average price index of the total basket. The first model considers the previous peak and dip, meaning that the consumer already made the decision to buy this product. Within the second model, the sequence is not taken into account, resulting in the fact that the peak and dip could be anywhere throughout the shopping trip. After a peak or a dip, it is expected that guilt or licensing will guide subsequent purchases. The effect of an anticipated peak or purchase is however less clear. Besides, as a rule, an increase in one of the price indices increases the average price index of the total basket while a decrease in one of the indices leads to a decrease in the average price index of the basket. This would mean that a positive relation is expected between the peak and dip and the average price index. However, taking licensing and guilt into account, it is possible that because of these shopping emotions a consumer offsets the postive effect, resulting in a negative effect. Hence, the influence of peaks and dips are unclear. This leads to the following hypotheses:

H2b1: Peaks during the shopping trip have a positive influence on the total basket average price index

(17)

H2c1: Dips during the shopping trip have a positive influence on the total basket average price index H2c2: Dips during the shopping trip have a negative influence on the total basket average price index

2.3 Other drivers of price dynamics during a shopping trip

2.3.1 Basket share of private label brands Another factor to consider here is the shopping basket that the consumer has created during the trip. There has been a rise in the emergence of private label brands over the recent decades. Private label items are products owned and branded by organizations whose primarily objective is the distribution of products rather than production (Schutte, 1969). With the emergence of the large retail chains, store brands have emerged as a weapon in the battle between manufacturers and retailers over channel control and customer loyalty (Patti and Fisk, 1982). The growth of private label brands has been attributed to improved product quality, increased retailer power and decreased national brand innovation and advertising (Hoch and Banerji, 1993; Steenkamp and Dekimpe, 1997). Another factor named by Sinha and Batra (1999) is the increasing price consciousness by consumers. Price consciousness has been defined in the marketing literature in different ways. Definitions vary from a buyer’s “unwillingness” to pay a higher price for a product to “the exclusive focus” on paying low prices (Lichtenstein, Ridgway and Netemeyer, 1993, p. 235). Since this study is concerned with the relative price within a product category, the definition used in this study is “a consumer’s reluctance to pay for the distinguishing features of a product if the price difference for these features is too large” (Monroe and Petroshius, 1981, p. 44). This definition highlights the trade-off that must be made by consumers: the higher price accompanied with higher potential benefits such as quality increase versus a lower price and a possible lower quality.

(18)

price. To conclude, private labels have extended beyond the typical low-price low-quality product but remain cheaper than national brands. It can hence be expected that private label brands are most often bought by shoppers who favor good value for money and are price conscious. That means that if a consumer’s basket for a large part consists of private label brands, it is likely that the consumer again decide a lower priced product. This results in the following hypotheses: H1e: The basket share of private labels has a negative influence on the price index of the next purchase

H2d: The basket share of private labels has a negative influence on the total basket average price index

2.3.2 Presence of health labels

When it comes to food choices, many considerations impact the decision making, with price and health as the most important ones. Haws, Reczek and Sample (2016) argue that consumers lay beliefs about the relationship between the healthiness of a food and its price. Lay beliefs are the commonsense explanations people use to understand their environment. They reflect people’s understanding of the world and hence do not necessarily have to be objectively true, also called a heuristic (Furnham, 1988). One of these beliefs is the healthy-is-expensive intuition (Haws, Reczek and Sample, 2016). Although this belief is true in some cases, such as with organic eggs and free range chicken and eggs (Oberholtzer, Greene and Lopez, 2006), this belief is often extended to other products where it is not true. It is difficult for consumers to know the exact relationship between price and health in a product category (Carlson and Frazão, 2012). It requires significant research, something in which consumers are unlikely to engage before a shopping trip. Food decisions, as consumers eat more than 1,000 meals per year, are heavily influenced on heuristic-based decision making (Wansink and Chandon, 2014). Thus, consumers tend to rely on heuristics when shopping for food, which in turn draw on existing cognitive notions like lay theories (Chaiken, 1987; Chaiken and Eagly, 1983). Other studies also demonstrate that consumers use lay theories to make inferences about missing attributes from known attributes (Raghunathan, Naylor and Hoyer, 2006; Luchs, Naylor, Irwin and Raghunathan, 2010). When relying on this belief, consumers might infer on a product’s characteristics that it is expensive, while it in fact is not.

(19)
(20)

2.3.3 Self-control

When a person makes a decision, a level of self-regulation is required (Baumeister and Heatherton, 1996). Self-regulation can be viewed as the extent to which one is able to control his or herself. Self-regulation is needed to make good decisions that benefit a person in the long term and not only focus on short term gain. Self-regulation can be viewed as a resource and is limited (Baumeister and Heatherton, 1996). This means that this resource can be depleted. A person is simply not able to regulate everything and the extent to which a person is able to do this differs per individual. Furthermore, making many decisions can lead to exhaustion which temporarily reduces one’s ability to regulate themselves.

(21)

2.3.4 Interaction effects with the stage of shopping

The number of purchases is also expected to affect other variables. First, the effect that the first purchase has on the following purchase is expected to decrease as the shopping trip continues. Where the shift from a deliberative to an implementation mind-set first leads to more spending, this effect is expected to reduce with every purchase. This because the mind of the consumer has already shifted and the marginal increase due to the shift in mind-set decreases. Besides, other factors such as the price of the previous purchase, peaks and dips will be more recent and the first purchase will gradually fade from the memory of the consumer (Crowder, 2014) and hence will its effect. This leads to the following hypotheses: H1h: The number of purchases negatively moderates the effect of the price index of the first purchase on the price index of the next purchase H2g: The number of purchases negatively moderates the effect of the price index of the first purchase on the total basket average price index Since the basket share of private label brands is believed to be a proxy for price consciousness. Thus, if private label brands make up a large part of the consumers’ basket, the consumer tends to prefer relatively cheap products. The further along the consumer is in a shopping trip, the stronger the predictive value of the basket share becomes. In case of a basket size of 10 products with 80% being private labels, it is highly likely that subsequent purchases are no private label brands and the basket share decreases with each additional purchase. However, in the case of a basket size of 25 products with 80% being private labels, it is more likely that the consumer is indeed price conscious and that it is not merely a coincidence that this shopper selected that many private label brands. This leads to the following hypotheses: H1i: The number of purchases positively moderates the effect of the basket share of private label brands on the price index of the next purchase

H2h: The number of purchases positively moderates the effect of the basket share of private label brands on the total basket average price index

(22)

2.4 Overview of hypotheses

TABLE 1. Overview of hypotheses

1. What are the drivers of spending decisions during a shopping trip? 2. What are the drivers of the average spending level of a shopping trip?

1a The price index of the first purchase is positively related to the index of the next purchase +

2a The price index of the first purchase is positively related to the total basket average price index +

1b The price index of the previous purchase is negatively related to the index of the next purchase - 1c The previous peak during the shopping trip has a negative influence on the price index of the next purchase - 1d The previous dip during the shopping trip has a negative influence on the price index of the next purchase - 2b1 The peak has a negative positive on the total basket average price index + 2b2 The peak has a negative influence on the total basket average price index - 2c1 The dip has a positive influence on the total basket average price index + 2c2 The dip has a negative influence on the total basket average price index - 1e The basket share of private labels has a negative influence on the price index of the next purchase -

2d The basket share of private labels has a negative influence on the total basket average price index - 1f The presence of a health label on the previous purchase has a negative influence on the price index of the next purchase - 2e1 The basket share of health labels has a positive influence on the price index of the total basket average price index + 2e2 The basket share of health labels has a negative influence on the price index of the total basket average price index - 1g The number of purchases has a positive influence on the price index of the next purchase + 2f The number of purchases has a positive influence on the total basket average price index + 1h The number of purchases negatively moderates the effect of the price index of the first purchase on the price index of the next purchase - 2g The number of purchases negatively moderates the effect of the price index of the first purchase on the total basket average price index - 1i The number of purchases positively moderates the effect of the basket share of private label brands on the price index of the next purchase +

(23)

3. Methodology

(24)

TABLE 2. Overview of weeks and national promotions

Week National promotion Details

9 50% 50% discount on selected products 10 No national promotion 11 “Vandaag extra laag” Each day different products at steep discounts 12 “Wijnsale” Steep discounts on wine 13 “Wijnsale” “Waardebonnen” Steep discounts on wine Coupons send to consumers’ homes that provide discounts on selected products 14 No national promotion

3.2 Sample and criteria screening

(25)

3.3 Operationalization of variables

To test the hypotheses that are stated at the end of chapter two, it is important that the concepts described in chapter 2 are operationalized. That means that for each concept a measurement should be defined. 3.3.1 Price indices Central to this study are the price indices and the dynamics therein during a shopping trip. To calculate the price index for each product, the product category to which the product belongs is used. For each product category, the average price per week is calculated. It is decided to calculate the average price per week, as product prices change over time. To calculate the price index per product, the price of the product is divided by the average price of the product category. This number illustrates how relatively expensive or inexpensive the product is within its category and can be used to model how the price indices evolve throughout a shopping trip. To summarize, the following two calculations were made: 1. Average price of the product category = 56789: ;< =>> ?@9 56;AB8?: C7?@7D ?@9 8=?9E;6F G;?=> DBHI96 ;< 56;AB8?: C7?@7D ?@9 8=?9E;6F 2. Price index of product 𝐽 = O6789 ;< 56;AB8? P QR96=E9 56789 ;< =>> 56;AB8?: 7D ?@9 8=?9E;6F ;< 56;AB8? P When the price index is higher than one, this indicates that the product is relatively expensive whereas an index smaller than one indicates that the product is relatively cheap. Since the second research questions focusses on the average price index of the total shopping basket, the following calculation is made as well: 3. 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑝𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥` = 56789 7DA789: ;< =>> A987:7;D: H=A9 IF 8B:?;H96 ` G;?=> DBHI96 ;< A987;D: H=A9 IF 8B:?;H96 ` This calculation is repeated to calculate the average price of all previous purchases by taking the sum of all price indices before time tn and dividing it by the number of products bought by the customer before time tn, where tn is the current observation.

For the category “statiegeld”, this is not appropriate. Statiegeld is a reward consumers receive when recycling soda and beer bottles. It has a monetary value and can be used to pay for (a part of) the groceries. As these are also part of the shopping trip it is decided to give them a 0 as a price index, indicating that it is an extremely low value. The same was done for other products the customer received for free: ‘inzamelzakken’, ‘Spaarkaart Joseph’ and ‘Pluspunten spaarboek’.

(26)

The price indices calculated in the previous step are also used for the operationalization of several other variables. Note that there are two peak variables, as the model contains both the highest price index (peak) and the lowest price index (dip). TABLE 3. Operationalization of variables using price index Concept Mathematical notation Operationalization Primacy effect

Price index`abc Price index at t

1 of customer i

Recency effect

Price index`abdec Price index of the product bought at tn-1 by customer

i

Peak Peak`Î{b|bibd} The highest price index of customer before time tn

Dip Dip`Î{b|bibd} The lowest price index of customer before time tn

3.3.2 Other independent variables The first of the other variables relates to the presence of private health labels. Plus has several private label brands. A categorical variable was created to indicate the different private label brands and other brands. To calculate the basket share of private labels the following formula was used: 5. Basket share of private labels` = Number of private label products bought by customer 𝑖 Total number of products bought by customer 𝑖 The second set of variables relates to the Dutch health label ‘het vinkje’. To create a variable that indicates whether the previous purchase contains ‘het vinkje’ the variable indicating whether a product had ‘het vinkje’ was lagged. To calculate the basket share of private label products the following formula was used:

6. Basket share of health labels`=Number of products with a health label bought by customer 𝑖Total number of products bought by customer 𝑖

It should be mentioned that Plus has several private label brands at different price levels. All these brands are included in this calculation.

(27)

3.3.3 Control variables

In addition to the variables described above various other variables are included in the models. These are included because they are expected to have an effect but are not the main focus of this research. It is important to add these variables as they cause some variation that otherwise would have been caught by the independent variables in the model and excluding them would lead to an omitted-variable bias (Leeflang, Wieringa, Bijmolt and Pauwels, 2015). First, a variable is added that represents the different weeks. This variable is categorical and indicates in which week the shopping trip took place. This variable is important as it captures the differences in promotional activities of the different weeks. There are also two weeks where there was no national price campaign, so it would be interesting to see if this has any effect. Second, a dummy variable was added that states if the shopping trip took place on a weekday or on the weekend, to control for differences attributable to these two moments. Third, a variable is created that captures the time of day when the shopping trip took place. The store that is used for this research is open from 08.00 to 20.00 Monday through Saturday and from 12.00 till 18.00 on Sundays. Four categories are created: from 08.00-11.00, 11.00-14.00, 14.00-17.00, 17.00-20.00. This variable is interesting as it helps determine if consumer decisions differ throughout the day.

3.4 Research method

(28)

which holds for all level 2 variables. Thus, level 2 variable are variables that only differ between subjects and not within subjects. To evade this problem, it has been proposed to estimate within effects in random-effects models (Allison, 2009; Rabe-Hesketh and Skrondal, 2008; Wooldridge, 2010). To do this, the level 1 variable is decomposed into a between 𝑥` and a cluster (𝑥`b − 𝑥`) component. The estimate before (𝑥`b − 𝑥`) provides the within-estimate, that is the FE estimate (Mundlak, 1978), whereas the estimate before 𝑥`

obtains the between-estimate. It is not necessary to include the cluster mean (𝑥`) to obtain the estimate of (𝑥`b− 𝑥`), but its inclusion ensures that effect estimates of the level 2 variables are corrected for

between cluster differences in 𝑥`b. This model, called the hybrid model (Allison, 2009) will be

used to estimate the first model.

For the second model, the data is aggregated on basket level. There are no longer multiple observations per shopper and the assumption of independence is no longer violated. Hence, an ordinary least squares (OLS) regression will be used to estimate the second model. 3.4.1 Evaluating the quality of the data After cleaning the data and calculating and adding new variables it is important to check the quality of the panel data at hand. One of the most important issues here is consistency of the unit of analysis. If each observation is not equivalent any analysis based on the data is likely not to be reliable. Such criteria are that the entities being researched are consistent, that the time periods should be consistent, that each entity has the same number of observations per time period and that the measurements are consistent. These are problems a researcher can encounter when combining different data from different suppliers. However, since this dataset was created specifically for this purpose by the researcher itself no problems were encountered. However, it should be mentioned that this is an unbalanced panel since not all shoppers have the same number of purchases and thus differ in the number of observations.

(29)

3.4.2 Model 1

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abd= 𝛽z+ 𝛽|𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abc+ 𝛽}𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abdec+ 𝛽~𝑃𝑒𝑎𝑘 𝑖𝑛 𝑃𝐼`Î{b|bibd} + 𝛽𝐷𝑖𝑝 𝑖𝑛 𝑃𝐼`Î{b|bibd}+ 𝛽ƒ𝐵𝑎𝑠𝑘𝑒𝑡 𝑠ℎ𝑎𝑟𝑒 𝑝𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙`bdec

+ 𝛽Š𝐻𝑒𝑎𝑙𝑡ℎ 𝑙𝑎𝑏𝑒𝑙`abdec+ 𝛽Œ𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠`bdec + 𝛽’𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠`bdec∗ 𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abc

+ 𝛽” 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠`bdec∗ 𝐵𝑎𝑠𝑘𝑒𝑡 𝑠ℎ𝑎𝑟𝑒 𝑝𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙`bdec+ 𝜀`a Where:

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abd = The price index of a single product of category j by

customer i at time tn

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abc = The price index of product j chosen by customer i at

time t1

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abdec = The price index of the product j chosen by customer i at

tn-1 𝑃𝑒𝑎𝑘 𝑖𝑛 𝑃𝐼`aÎ{b|bibd} = The peak in the price index of product j at time t where t<tn in the basket of customer i 𝐷𝑖𝑝 𝑖𝑛 𝑃𝐼`aÎ{b|bibd} = The dip in the price index of product j at time t where t<tn in the basket of customer i 𝐵𝑎𝑠𝑘𝑒𝑡 𝑠ℎ𝑎𝑟𝑒 𝑝𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙`bdec = The basket share of private label products purchased by customer i at tn-1

𝐻𝑒𝑎𝑙𝑡ℎ 𝑙𝑎𝑏𝑒𝑙`abdec = Indicating if customer i purchased a product containing

(30)

3.4.3 Model 2 Within the second model the data is no longer viewed as panel data as the analysis is on basket level. All the variables were aggregated from individual purchase level to basket level. Since this removes the dependence of the variables, the independence assumption by Leeflang et al. (2015) is no longer violated and an Ordinary Least Squares regression can be used. 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑃𝐼` = 𝛽z+ 𝛽|𝐹𝑖𝑟𝑠𝑡`a+ 𝛽}𝑃𝑒𝑎𝑘`a+ 𝛽~𝐷𝑖𝑝`a+ 𝛽𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙`+ 𝛽ƒ𝐻𝑒𝑎𝑙𝑡ℎ 𝑙𝑎𝑏𝑒𝑙` + 𝛽Š𝑁𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠`+ 𝛽Œ𝐹𝑖𝑟𝑠𝑡`∗ 𝑁𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎 𝑒𝑠`+ 𝛽𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙𝑠` ∗ 𝑁𝑟 𝑜𝑓 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑠`+ 𝜀` Where: 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑃𝐼` = The total basket average price index for customer i 𝑃𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥`abc = The price index of product j chosen by customer i at time t1

𝑃𝑒𝑎𝑘 𝑖𝑛 𝑃𝐼`a = The peak in the price index of product j in the basket of

customer i

𝐷𝑖𝑝 𝑖𝑛 𝑃𝐼`a = The dip in the price index of product j in the basket of customer

i

𝐵𝑎𝑠𝑘𝑒𝑡 𝑠ℎ𝑎𝑟𝑒 𝑝𝑟𝑖𝑣𝑎𝑡𝑒 𝑙𝑎𝑏𝑒𝑙` = The basket share of private label products purchased by

(31)

4. Results

This chapters starts with some graphs and statistics to explore the data. First, the dependent variable price index will be discussed. Boxplots will be used to spot outliers and several figures will show the changes in price index over time. Three moments are selected: per week, per day and throughout the shopping trip. Next, two tables displaying the descriptive statistics of the main variables of model 1 and 2 will be included. Second, this chapter shows the results of model 1 and model 2. This section will start with checking the assumptions for each of the models and end with an interpretation of the results.

4.1 Exploratory analysis

4.1.1 Price index

(32)
(33)

0,9 0,92 0,94 0,96 0,98 1 1,02 1,04 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 Pr ic e in de x Purchase sequence 4.1.3 Price index throughout the shopping trip

To gain more insight into the development of the average price index throughout the shopping trip, the average price per purchase in the shopping trip is presented. This was done by calculating the average price index for all first purchases, all second purchases and so on. The average basket contained 20.51 products with a standard deviation of 10.86. It is decided to show the average basket plus one deviation in the graph, resulting in 42 purchases. Figure 6 shows the average price index of a sequential shopping trip. Since all prices are corrected for the average price of the category the order in which customers visited the different product categories of the store does not matter. The trend line shows a slightly increasing line from the first purchase to the 17th purchase, after which it slightly declines until the 23rd

purchase. After the 23rd purchase the line becomes more volatile with peaks around the 29th

and 35th purchase and dips round the 33th and 37th purchase. This could be some preliminary

(34)

here represents the amount of purchases that the customer has made before arriving at time tn. This ranges from 0 to 92.

TABLE 4. Descriptive statistics of the main variables of model 1

Mean deviation Standard Minimum Maximum

Price index`abd .994 .473 0 9.057

Price index`abc .984 .531 0 8.078

Price index`abdec .995 .469 0 9.057

Peak in price index`Î{b|bibd} 1.181 .797 0 9.057

Dip in price index`Î{b|bibd} .467 .291 0 8.078

Basket share private label`bdec .655 .239 0 1

Health label`abdec 1.2% = health label

(35)

TABLE 5. Descriptive statistics of main variables of model 2

Mean deviation Standard Minimum Maximum

Average price index` .989 .144 .464 1.874 Price index` bc .958 .539 0 8.078 Peak in price index` 2.070 .816 .784 9.057 Dip in price index` .354 .190 0 1.046 Basket share private labels` .495 .162 0 1 Basket share health label𝑠` .0120 .029 0 .307 Number of purchases` 20.53 10.858 10 93

4.2 Model 1: predicting the next purchase

As explained in section 3.4, a hybrid model is used to estimate the first model. To validate the regression model, four assumptions are assessed: multicollinearity, autocorrelation, heteroscedasticity and normality (Leeflang et al., 2015). Multicollinearity occurs when at least two independent variables are highly correlated, meaning that one can be predicted from the others to a certain extent. This reduces the accuracy of the coefficient estimates for the independent variables. Autocorrelation, also called serial correlation, describes a situation in which the error terms over time are correlated and this can cause errors in the prediction. Heteroscedasticity, looks at the size of the error term differs across values an independent variable. Ideally, data is homoscedastic, meaning that the error term is similar over observations. If this is not the case, the coefficient estimates will be less precise as the observations with larger disturbances more heavily affect the estimation than the observations with a smaller disturbance. The final assumption, Normality requires a normal distribution of the error term. 4.2.1 Multicollinearity To test the assumptions, the model must be estimated. StataSE is used to estimate the model. Before the model can be estimated, the within-cluster (𝑥`) averages are calculated as well as

(36)

time-invariant variables, which are all the control variables and the price index of the first purchase, since they do not differ within clusters. The time-invariant variables, together with the newly created variables are added to the model and a random-effects model is estimated. Since this model does not allow for the opportunity to check multicollinearity, the same model is estimated, now using an OLS regression to calculate the VIF scores. The outcomes can be found in appendix 1.1. The results show that none of the variables have VIF scores > 5 and a tolerance < 0.2. Thus, no variables are violating the assumption of multicollinearity in this model. Next, the full model is estimated, which also contains the interaction effects. Again, the variables are saved and an OLS is run to calculate the VIF scores. This time there are VIF scores for variables included in the interaction which violate acceptable VIF levels (appendix 1.1). To reduce the effect of multicollinearity, two additional models are created (e.g. model 1.2 and model 1.3), both containing one interaction effect. Possible multicollinearity will then probably only be limited to the direct effects and not the interactions. Using this method, the interactions can be safely interpreted. Thus, the model containing only the direct effects should be used for interpreting the direct effect, while the other models can be used to interpret the interaction effects. The four models specified at this step can be found in appendix 1.1.

4.2.2 Autocorrelation and heteroscedasticity

Autocorrelation can occur when there is a time structure present in the data. When autocorrelation is present, it means that there is a relationship over time in the residuals which results in an incorrect estimate of the variance (Leeflang et al., 2015). The Wooldridge test (Drukker, 2003) can be used to test for autocorrelation. This test does not have the ability to include categorical or nominal variables. However, this is not a problem within this research as the dummy variables, which are all the control variables, are all constant throughout a single trip and hence cannot cause autocorrelation. The Wooldridge tests for the four models can be found in appendix 1.2. All tests are significant, indicating that autocorrelation is present in the models. This means that the residual at tn-1 can be used to partly predict the value of the residual at time tn (Leeflang et al., 2015).

(37)

significant for all models (appendix 1.2). Hence, it can be concluded that both autocorrelation and heteroscedastic errors are present in the models.

(38)

4.2.4 Interpretation

The model statistics can be found in table 6. The table containing all the results can be found in appendix 1.3. The first model only contains the direct effects while the second model contains the direct effects and the interaction between the number of purchases and the first price index. The third model includes the direct effect and the interaction between number of purchases and the basket share of private labels. The fourth model is the full model, with the direct effects as well as both interaction effects. The results show that all four models are significant (p<0.01). Each model provides three different types of R squares. The ‘R square within’ indicates how much of the variance within the clusters is accounted for by the model. The ‘R square between’ indicates how much of the variance between the separate clusters in the model is accounted for. The last one, ‘R square overall’ is a weighted average of the ‘within’ and ‘between’ R squares. To assess the performance of the model, the ‘R square overall’ should be compared (Wooldridge, 2010). The overall R square of the first model is .1003, which means that the model explains 10.03% of the variance in the dependent variable is explained by the variables in this model. Adding the moderators only slightly improves the model. Since 10.03% is quite low, this indicates that important drivers are missing in this model. In addition to the entire model being significant, several variables are also significant. None of the control variables are significant and are for that reason left out of table 7, which can be found on the next page. Several variables have both a significant within- as well as a between-estimator as can be seen in table 7. For the interpretation, model 1.1 is used for the direct effects and models 1.2 and 1.3 for the interaction effects. To test the similarity of these estimates, a Walt test can be used (Schunck, 2013).

The results of the Wald test (appendix 1.4) indicate that the within- and between-estimators are significantly different, which can be considered as evidence against the between estimator, which is the random effects estimator (Schunck, 2013). Also, only the between-estimators of the price index of the previous purchase and the interaction effect of number of purchases and the price index of the first purchase remained significant after

TABLE 5. Model statistics

Model 1.1 Model 1.2 Model 1.3 Model 1.4

Wald Chi Square 133090.14** 134846.95** 113861.73** 136317.28**

R square within .0270 .0290 .0274 .0294

R square between .9223 .9317 .9223 .9317

(39)

be carefully interpreted. Since the within-estimators represent the dynamic effects and for reasons given above, the within-estimators will be used for interpretation.

TABLE 6. Effects for interpretation

Within-estimator Std. Err. Robust estimator Between- Std. Err. Robust

Price index`abc -0.050** .001

Price index`abdec 0.065** .004 1.019** .004

Peak`Î{b|bibd} -0.188** .007 -0.004** .001 Dip`Î{b|bibd} -0.153** .009 -0.007* .003 Basket share private label`bdec 0.041** .009 -0.003 .002

Health label`abdec -0.003 .011 -0.012 .013

Number of purchases`bd 0.004** .000 -0.000 .000

Number of purchases`bdx Price index`abc -0.005** .000 0.003** .000

Number of purchases`bdx Basket share private label`bdec 0.006** .001 0.000 .001 * p<0.05; ** p<0.01 The price index of the first purchase has a negative, significant influence on the price index of the current purchase (ß = -0.050, p<0.01). This means that when the price index of the first purchases increases, the price index of the current purchase decreases. Thus, hypothesis 1a is not supported, since it expected a positive effect between the first price index and the current price index. This shows that if the price index of the first purchase increases with 1, the price index of the current purchase decreases with 0.05. Second, the effect of the price index of the previous purchase on the current purchase is positive and significant (ß =0.065, p<0.01), meaning that when the price index of the product bought before the current purchases increases with 1, the price index of the current purchase increases with 0.065. Hypothesis 1b is thus not supported. It stated that the effect would be negative due to balancing, as a result from the experienced pain of paying. Instead, the model shows a positive, yet small relationship.

The effect of the peak on the effect of the current price index is negative and significant (ß =-0.188, p<0.01), thereby supporting hypothesis 1c. Thus, when the price index of the

(40)

decrease of 0.188 in the price index of the current purchase. When the highest price index decreases with 1, this leads to an increase of 0.188 the price index of the current purchase. Hypothesis 1d is also supported, since the effect of the dip is negative and significant

(ß =-0.153, p<0.01). This means that the lower the dip, the higher the price index of the current purchase and vice versa.

The basket share of private label before time tn has a positive, significant effect on the price index of the current purchase ((ß =0.041, p<0.01). This means that as the basket share of private label brands increases, the price index of the current purchase also increases, whereas when the basket share decreases the price index of the current purchase also decreases. This finding does not support hypothesis 1e, which stated that there would be a negative effect. It was expected that the basket share of private label is an indication of a price consciousness consumer. However, evidence provided by this model suggests otherwise.

The presence of a health label on the previous purchase was expected to decrease the

(41)

The final variable discussed here is the interaction effect between the basket share of

private label brands and the number of purchases. This effect is positive and significant, hence providing support for hypothesis 1i. This means that the effect of the basket share of private label brands on the price index of the current purchase increases with each additional purchase and thus becomes a stronger predictor. Again, the interaction is plotted and the result can found in figure 8. This clearly shows that when combing the effect of basket share of private label with the number of purchases already done, the effect becomes stronger. FIGURE 4. Plotted direct and interaction effect: basket share private label brands As mentioned earlier, none of the control variables are significant. The dummy variable for weekend is significant after estimation with a robust error but this effect disappears when estimating the model using bootstrapping. A possible explanation for their insignificance could be that by adding the means of all the independent variables to the model, the average differences between moments are captured in this effect. To illustrate, it could be possible that the average price is indeed higher on weekends in comparison to weekdays. But since the mean and the mean deviation of the time-varying the independent variable are both added in the estimation model, they already partly capture this effect.

2nd purchase 26th purchase 50th purchase

Direct effect basket share private label brands Interaction with number of purchase

2nd purchase 26th purchase 50th purchase

Direct effect first price index Interaction with number of purchase

(42)
(43)

that heteroscedasticity can be thought of as a sign that a model is misspecified. But as MacCallum (2003) points out, all models are wrong to some extent. Hence, it is decided to continue working with this model.

An approach to solving this problem is weighted least squares (WLS) (Draper and Smith, 1981; Cook and Weisberg, 1983). WLS weights each case differently in the derivation of the sum of squared residuals. When the variance of errors is related to at least one predictor by a constant multiplier, weights can be assigned to produce parameter estimates that are more efficient. A popular alternative is that of reducing the effects of heteroscedasticity by using a heteroscedasticity-consistent standard error (HSCE) estimator of the parameter estimates (Hinkley, 1977; White, 1980; MacKinnon and White, 1985; Long and Ervin, 2000). The benefit of this method is that unlike WLS it does not require the knowledge or a model of the functional form of the heteroscedasticity. This second approach is chosen and a new model is estimated. As can be seen in table in appendix 2.3 the coefficients do not change but the standard errors are adjusted for the possibility of heteroscedasticity. The models as specified with the robust standard errors can be found in appendix 2.3.

4.3.4 Normality

To test for normality, the residuals are saved and are plotted in a histogram (appendix 2.4). The residuals do not seem normally distributed but to be sure this needs to be tested statistically. The Kolgomorov-Smirnov test indicates that the residuals are not normally distributed. To account for this problem, the same regressions are performed using bootstrapping, which was performed 1.000 times. The results of the bootstrap can be found in appendix 2.4. The outcomes do not differ in significance levels or directions compared to the model with the robust errors that was estimated at the end of section 5.2.3. This means

(44)

Table 7. Final models for interpretation

10-21 item baskets 21+ item baskets 10-42 item baskets

Beta Robust Std. Err. Beta Robust Std. Err. Beta Robust Std. Err. First price index 0.025 .004 0.015 0.004 .020 .003 peak 0.497** .003 0.405** .003 .474** .003 dip 0.310** .009 0.225** .014 .296** .008 % private label -0.032** .009 -0.023 .017 -.034** .008 % health label -0.020 .050 0.019 .089 -.009 .044 Nr of purchases 0.002 .000 -0.027 .000 -.002 .000

Week 9 BASE BASE BASE

Week 10 -0.012 .007 -0.023 .010 -.009 .006 Week 11 -0.028 .006 0.014 .008 -.021 .005 Week 12 -0.038* .006 0.019 .008 -.021 .005 Week 13 -0.018 .006 -0.006 .008 -.011 .005 Week 14 -0.032* .006 0.014 .008 -.013 .005 Morning – 08:00-11:00

BASE BASE BASE

Lunch – 11:00-14:00 0.007 .005 0.005 .006 .002 .004 Afternoon – 14:00-17:00 0.016 .005 0.056* .006 .021 .004 Evening – 17:00-20:00 0.020 .006 0.011 .008 .013 .005

Weekday BASE BASE

(45)

4.3.5 Interpretation

Table 9 on the previous page shows the final estimation result for model 2, which are the models specified with the robust errors. The first column shows the parameter estimates of the model estimated on the complete dataset. Since the basket size of shoppers varies a lot (ranging from 10 up to 92 products) the model is also estimated on baskets containing 10 to 42 purchases, which is the mean plus two standard deviations. Comparing the models provides an indication of how sensitive the results are to the changes in basket size. When comparing the models, it becomes evident that the results slightly differ in their significance levels. To start, the effect of the basket share of the private label brands is significant wen estimated on the small sample. But this effect becomes insignificant when estimated the larger sample and significant again on the 10-42 sample. This indicates that this variable is sensitive to the sample. Furthermore, the dummy variables representing week 12 and week 14 are also significant for the sample containing the small baskets, but are not significant in the other two models. Lastly, the dummy for afternoon is only significant (p<0.05) when estimated on the sample containing the larger baskets. Since the models show that some variables differ between them and hence are not very robust, they should be carefully interpreted.

As mentioned before, the models as a whole are significant (F=105.94, p<0.01; F=32.34,

p<0.01; F=124.56, p<0.01). The adjusted R2 of the first model is .359, which means that the

model explains 35.9% of the variance of the total basket average price index. The second model explains less of the variance, only 21.0% whereas the third model explains 31.4%. The table shows that the price index of the first purchase does not influence the total basket average price index, since this effect is positive but not significant (p >.05). This means that hypothesis 2a is not supported. The effects for peak and dip are significant and more or less constant through all three models. As described in chapter 2, the direction of these effect was unclear. The effect of peak (ß = .497, p<0.01; ß = 0.405, p<0.01; ß = 0.474, p<0.01) as well as the effect of dip (ß = .310, p<0.01; ß = 0.225, p<0.01; ß = 0.296, p<0.01). This means that support is found for hypotheses 2b1 and hypothesis 2c1 and none for hypothesis 2b2 and hypothesis 2c2.

Basket share of private label has a negative effect on the average price index and this effect is significant in model 1 (ß = -.032, p<0.01) and in model 3 (ß = -.034, p<0.01). This means that hypothesis 2d is partially supported since the effect is found significant, but only

(46)

The effect of basket share of health labels is not significant in any of the models, indicating that there seems to be no relationship between the amount of product containing health labels bought and the average price index of the basket. This means that both

hypothesis 2e1 and 2e2 are not supported. The effect of the number of purchases on the price index is also not significant, which means that hypothesis 2f is not supported. The number of purchases does not affect the average price index of the total basket, which could be considered evidence against the depletion of self-control. Since the initial model contained multicollinearity, the model was estimated on two different samples. By estimating on two different samples the effects of smaller and larger baskets can be compared. It was expected that the number of purchases would negatively moderate the relationship between the price index of the first purchase and the average price index. For this to hold, the effect of price index on the average price index should have been stronger in for small baskets compared to larger baskets. But since the price index of the first purchase does not have any influence in any of the models, no support is found for hypothesis 2g. For the second interaction effect, between number of purchases and the basket share

of private label, a positive effect was expected. Meaning that the larger the number of purchases and the higher the basket share, the more of an indication this would be of a price sensitive consumer. For this to hold, the effect of basket share of private labels should have been stronger in the model estimated on the larger baskets. However, no significant relationship was found within this model and hence hypothesis 2h is not supported.

(47)

5. Conclusion

The objective of this study was to find the drivers of relative spending of a consumer in a retail environment. Two models were estimated. One model to find the drivers of the relative spending of the next purchase and a second model to find the drivers of the average price index of a basket. An overview of all the hypotheses drafted to test these objectives and their outcome can be found below. TABLE 6. Overview of hypotheses 1. What are the drivers of spending decisions during a shopping trip? 2. What are the drivers of the average spending level of a shopping trip?

Hypothesis Support? Comment

1a The price index of the first purchase is positively related to the index of the next purchase Opposite direction

2a The price index of the first purchase is positively related to the total basket average price index Not significant

1b The price index of the previous purchase is negatively related to the index of the next purchase Opposite direction

1cc The previous peak during the shopping trip has a negative influence on the price index of the next purchase Yes

1dd The previous dip during the shopping trip has a negative influence on the price index of the next purchase Yes

2b1 The peak has a positive on the total basket average price index Yes

2b2 The peak has a negative influence on the total basket average price index Opposite direction

2c1 The dip has a positive influence on the total basket average price index Yes

2c2 The dip has a negative influence on the total basket average price index Opposite direction

1e The basket share of private labels has a negative influence on the price index of the next purchase Opposite direction

2d The basket share of private labels has a negative influence on the total basket average price index Partial Effect becomes insignificant as basket size increases

1f The presence of a health label on the previous purchase has a negative influence on the price index of the next purchase Not significant

2e1 The basket share of health labels has a positive influence on the price index of the total basket average price index Not significant

2e2 The basket share of health labels has a negative influence on the price index of the total basket average price index Not significant

1g The number of purchases has a positive influence on the price index of the next purchase Yes

2f The number of purchases has a positive influence on the total basket average price index Not significant in any model

1h The number of purchases negatively moderates the effect of the price index of the first purchase on the price index of the next purchase Yes

2g

The number of purchases negatively moderates the effect of the price index of the first purchase on the total basket average price index Not significant in any model

1i The number of purchases positively moderates the effect of the basket share of private label brands on the price index of the next purchase Yes

2h

The number of purchases positively moderates the effect of the basket

(48)

5.1 Discussion

This section discusses the outcomes of both models while considering existing research on this topic. The research questions as formulated in chapter one will be discussed individually in the paragraphs below. 5.1.1 Drivers of spending decisions during a shopping trip The objective of the first model was to answer the following research question: what are the drivers of spending decisions during a shopping trip? The self-scanner data made available by

Plus contained 8.184 shopping trips that met the criteria as described in chapter 3, with basket sizes ranging from 10 up to 93 products. Figure 6 in section 4.1.3 shows that the relative spending throughout a shopping trip is not consistent. This is a preliminary indication that spending dynamics exist. Based on the literature on temporal sequence and pricing literature several concepts were found that were expected to affect the spending level of subsequent decisions. According to the temporal sequence literature, the first event in a sequence is remembered very well (Montgomery and Unnava, 2009). This research finds support for that claim, as the relative spending on the first purchase does indeed affect the relative spending on a subsequent purchase. However, contrarily to the theory on shopping momentum (Dhar, Huber and Khan, 2007), a negative relationship was found. According to the shopping momentum, the first purchase would increase subsequent spending since the shopper’s mind-set would have shifted towards one that enhances spending. An explanation would be that the shopping emotions resulting from the first spending decision affect the remainder of the shopping trip (Khan and Dhar, 2006). If the first purchase is relatively expensive, the consumer apparently tries to economize in the remainder of the shopping trip due to guilt. When the first purchase is relatively cheap, this allows the consumer to justify future more expansive purchases due to licensing. This effect becomes less strong as the shopping trip continues, indicating that the first purchase gradually fades from memory (Montgomery and Unnava, 2009). Other drivers based on the temporal sequence literature are the previous purchase,

(49)

versa for the dip due to licensing. The negative effect of the peak is stronger than for the dip, which is line with previous research. As described in the prospect theory (Kahneman and Tversky, 1979), from the same reference point, losses are felt more strongly than gains (Kahneman and Tversky, 1984; Tversky and Kahneman, 1991). Consumers thus tend to put more weight on the most expensive purchase, since it produces the highest loss, leading to a larger negative effect for peak than dip. AS a result, the consumers more strongly feel the need to save money after a loss compared to the need to spend more money after a cheaper product.

However, a positive relationship is found between the relative spending on the previous purchase and the spending on the current purchase. This suggests that when the price index of the previous purchase does not reach an extreme value, it is possible that the pain of paying remains limited. Zellermayer (1996) states that when a price can be justified, the pain of paying is reduced. This would mean that the consumer buys a product for which the price-quality trade-off is correctly balanced. Thus, when a consumer buys an average product for an average quality the pain of paying is expected to be less than when he or she buys an expensive product and is not entirely sure whether the quality fully justifies the higher price. Consequently, the outcomes of this study suggest that the shopping emotions licensing and guilt are only triggered when the price index of the previous purchase reaches an extreme value, since the expected negative effect is found for peaks and dips.

Referenties

GERELATEERDE DOCUMENTEN

With a simple example of a hospital with two primary wards, we have shown that open- ing an EOA Ward results in an increase of urgent patient admissions, but at the same time in

Instead, modal mineralogy information on a num- ber of samples is used to build a quantitative multi- variate partial least squares regression (PLSR) model that links the mineralogy

Effect of yeast strain on &#34;medicinal&#34; odour intensity ratings and p-vinyl guaiacol (PVG) concentrations in Kerner wines. 2) Average of duplicate

One independent variable; The introduction of a premium private label, one moderator; Hedonic level of the product, three mediators; Perceived product quality,

This paper will present a research on the perceived quality and purchase intention differentials for luxury and standard cars when the country of origin

For example, when a customer would purchase a product from the Electronics category and a product from the Garden category at the same time, what are the mutual influences regarding

Based on the main idea of the basket balance hypothesis that additional products are able to counterbalance the perceived threat to the public identity coming from the

It was expected that the percentage of healthy and unhealthy purchase decisions would both increase during the health promotion week compared to the other weeks in the data, and