The effectiveness of display advertising on sales during Christmas
Is it really the most wonderful time of the year?
By Marieke Meisner
1
The effectiveness of display advertising on sales during Christmas
Is it really the most wonderful time of the year?
By Marieke Meisner
15-01-2017 S2985470 Soephuisstraatje 18-8, 9712BZ Groningen mariekemeisner93@hotmail.nl 0630317062Master thesis Marketing Management & Marketing Intelligence
University of Groningen Faculty of Economics & Business
Department of Marketing PO Box 800, 9700 AV Groningen
2 ABSTRACT
Every year, Christmas draws a lot of media attention. Retailers and manufacturers are always trying to make the holiday season as profitable as possible. Most retailers already start planning their holiday investments, in July or earlier (Knight, 2016). A lot of money is spent during this period. For example, Walmart spends around $66.4 million in holiday adds alone, so before and during the Christmas period. The biggest department stores in the United States, spent about $869.7 million on holiday-themed national ads in 2016, together (Lynch, 2016). However, does this really increase sales in supermarkets? Or would it be better for retailers to avoid advertising in this period? Only a little research is done regarding the effectiveness of display advertising. This study extends the current literature by investigating if retailers and/or brand managers should invest in display advertising to increase sales, at different periods before Christmas. This is done by analyzing 43 FMCG categories. The findings reveal that display advertising does increase category sales, and that this effect is even stronger in the weeks before Christmas. The results implies that the sales of the whole product category can increase, while this is not due because of brand differentiation. However, the findings also reveal that there is a negative lasting effect of display advertising, which makes it a less attractive marketing tool to use.
Keywords: display advertising, product categories, brand sales, category sales, brand
3 PREFACE
“Research is to see what everybody else has seen, and to think what nobody else has thought” (Albert Szent-Gyorgyi)
4
TABLE OF CONTENT
ABSTRACT ... 2 PREFACE ... 3 1. INTRODUCTION ... 6 2. THEORETICAL FRAMEWORK ... 82.1 INCREASED ADVERTISING EFFECTIVENESS ... 9
2.2 DECREASED ADVERTISING EFFECTIVENESS ... 10
2.3 IN-STORE ADVERTISING ... 10
2.4 BRAND SALES VS. CATEGORY SALES ... 13
3. METHODOLOGY ... 14 3.1 VARIABLES ... 15 3.2 CONTROL VARIABLES ... 16 3.3 MODEL SPECIFICATION ... 17 3.3.1 BASIC MODEL ... 17 3.3.2 TESTING MODERATION ... 18
3.3.3 ACCOUNTING FOR CONTROL VARIABLES ... 18
3.3.4 LAGGED SALES ... 19
3.3.5 THE LASTING EFFECT OF DISPLAY ADVERTISING ... 19
5
6.2 FINAL CONCLUSION ... 38
6.3 MANAGERIAL IMPLICATIONS ... 39
6.4 LIMITATIONS AND FURTHER RESEARCH ... 40
REFERENCES ... 41
APPENDIX ... 47
A. REGRESSION ANALYS OUTPUT PER PRODUCT CATEGORY ... 47
6
1. INTRODUCTION
Before, and during the Christmas period, companies invest a lot of money into advertising in order to seduce people to buy more. A lot of advertising campaigns during the December month are about Christmas and you cannot ignore that it is ‘the most wonderful time of the year’. These advertisements are of course not for free, for example John Lewis, a chain in high-end
department stores in the UK, spent £7 million on its Christmas campaign of 2014. The biggest part of this investment was allocated to in-store technology in order to keep the attention of children to the in-store advertisements. One of the things they put in their stores was an Iceberg Station (Morrell, 2014). Furthermore, British retailers spent around $2.2 billion on TV slots alone in 2015, and $3.3 billion on online advertising, during the Christmas period (Howe, 2015). In 2016 this even went up to $5.6 billion in total, even though the Brexit was a big thing in the UK, Christmas was apparently more important (Sweney, 2016).
However, it has never been proven that this is an effective strategy. It might be the case that consumers do not pay any attention to the advertising messages, because there are so many of them. Probably consumers are more likely to get involved into impulse buying, which makes the in-store stimuli much more interesting than ‘normal advertising’. According to the Popai & Du Pont Consumer Buying Habits study (1977), approximately 65% of all supermarket purchase decisions were made in-store with over 50% of these being unplanned. Furthermore, Johnson and Williams (1984) found that 20% of the purchase decisions were made inside the store. Finally, Kollat and Willet (1984), found that on average, of all the (supermarket)products that consumers buy, 50,5% is unplanned. These results highlight the importance of in-store advertisements, like the use of a display. However, there is only little research concerning in-store advertising, while it is likely that this type of advertising could be very important and has a lot of potential. When looking at the expenses to in-store advertising by John Lewis, it seems that it is might even more important to invest in this kind of advertising during the Christmas period in order to capture the attention of children and tell a story. This led to the following research question: ‘To what extent do in-store stimuli influence category and/or brand sales and is this even enhanced by the Christmas period?’.
7
1. ‘To what extent does display advertising influence sales?’
2. ‘Is the effect of display advertising on sales different during the Christmas period in comparison to other periods?’
3. ‘Is there a difference in the effect of display advertising on sales, when looking at the period of three weeks before Christmas and the period of one week before Christmas?’ 4. ‘What is the reason of an increase in sales, is this due to brand differentiation or because
of an increase of the whole category?’
This study investigates whether there is a difference in display advertising intensity around the Christmas period and, if so if this also results in more sales. Furthermore, it focuses not only on sales in total, but it also makes a distinction between an increase in brand sales or an increase in category sales.
Manufacturers spent a lot of money on advertising each year, in order to create more awareness and market share (Abratt, & Goodey, 1990). However, it is important to know if this is an good strategy. This study is interesting for manufacturers, because it gives them insights in how to increase their sales by using in-store advertising messages for their brands. Besides that, this research increases the knowledge about the effectiveness of using displays, in particular when it is most effective to use them. For example, this can be right before Christmas, or maybe a few weeks before it. By knowing this, it becomes possible for manufacturers to divide their Christmas budget more effectively and to influence consumer purchasing by using in-store advertising. Furthermore, the study gives a clear indication if the in-store stimuli might only lead to
consumers choosing another brand or to an increase in sales of the whole category. This makes it also an interesting study for retailers because, if the whole category increases it means that their profits will increase. However, if it only results in switching brands, this won’t have that much influence on the profits of a supermarket. So, this research contributes to the existing literature by extending the knowledge about the effectiveness of display advertising in a particular period. Furthermore, this research has an explorative aspect in it because it also investigates if there is an lasting effect of display advertising. This implies that the research investigates how an investment in display advertising in one week affects the sales in the upcoming week.
8
4 presents the data. The results appear in section 5, and section 6 presents a discussion of the key insights, including the managerial implications and the limitations of this research.
2. THEORETICAL FRAMEWORK
This study aims to investigate whether and to what extent in-store advertisement messages result in more sales in comparison to not using these messages. Next, this research tends to find out if the in-store stimuli have more effect three weeks before Christmas or if they are more effective when used one week before Christmas. Of course, the research also has to compare the Christmas period (December) to other months, in order to check if the buying behaviour is different this month. Which means that the difference in sales is the dependent variable and the independent variable is the number of in-store advertisement messages by the different categories.
Furthermore, the research should account for the fact that consumers might buy more in December compared to other months, even if there is no in-store advertisement in the
supermarket at all. This leads to the Christmas period being a moderator. The expectation is that this period has a positive effect on the relation between in-store stimuli and sales. Sales has been divided into category and brand sales. This is because, it is still quite unknown if promotions lead to an expansion of the category or if it only results in brand differentiation.
Figure 1 Conceptual Model
The model needs to account for the action of competitors, like the price of their products and their advertising promotions. Besides that, the model needs to account for other marketing mix variables of the product/brand itself, in order to be sure that the effect measured is due to display advertising.
In-Store Stimuli Sales
Christmas period
Category
Brand
9
2.1 INCREASED ADVERTISING EFFECTIVENESS
Because of the emotional and warm feelings people have before and during the Christmas period, advertising is expected to become more effective. This is because, companies can use these positive feelings, consumers have, in order to transfer it to their brands. In this sense, consumers also get positive feelings about the brands that are advertised (Bagozzi, Gopinath, & Nyer, 1999). When consumers have positive feelings about a brand, it is likely that it will result in higher purchases (Morris, Woo, Geason, & Kim, 2002). Following this line, it could be expected that the more companies use advertisements, the higher the sales will be during the Christmas period. However, this is likely to be the case for ‘normal’ advertising, but it would be very interesting to study if this might also be the case for in-store advertising.
Furthermore, people are consuming much more during the Christmas period because this is the way they express themselves and make clear what they feel and think. In the UK, an average person consumes nearly 6,000 calories on the 25th of December, which is three times the normal
daily amount a women needs (Hosie, 2016). According to a study of 2,000 people who celebrate Christmas, 37% crack open a bottle every day of the festive week. These people ate and drink that much during Christmas, because they feel like it is their ‘get out of jail free’ card, which means that day can eat whatever they want for a few days, and then get back to normal (Hosie, 2016). So, the market kind of increases because of the increased consumption, which makes the potential to attract additional sales through the same investments in advertising larger, this raises the advertising’s potential effectiveness. However, this study was only about food and drinks, which means that there is only evidence about the increase in sales of these categories during the Christmas period. It is likely that this does not hold for household care and personal care. So, there will be a difference in the increase of consumption and sales between categories like food and drinks, and categories like personal care, during the Christmas period. It is expected that, the food and drinks categories experience a higher increase in sales in comparison to the other categories.
Previous research on in-store stimuli during the Christmas period found that when a retailer has consistency between an ambient Christmas scent and Christmas music, it will lead to more favourable evaluations of the store, its merchandise and the store environment. Also, the
10
& Sprott, 2005). Consequently, it would be interesting to find whether this finding also extends to a) actual purchases, and b) the use of in-store advertisements.
2.2 DECREASED ADVERTISING EFFECTIVENESS
During the month December, there are a lot of companies that invest in advertising. This means that the competition to attract the attention of consumers, is much higher than during other months. This can possibly lead to clutter and interference for the ‘normal’ types of advertising (Burke, & Srull, 1988). The own-advertising effectiveness can be negatively affected by
messages of competing brands in the same category (Danahar et al., 2008). It is also very likely that the overload of advertisements will result in brand confusion, because of similar themes that are used (Poiesz, & Verhallen, 1989). For the ‘normal’ types of advertising this is likely to result in an decrease of sales, this period. However, it would be interesting to find if the same holds for in-store advertising.
One advantage of all the different advertisements that can be found in the supermarket during the Christmas period, is that it can result in ego depletion. This means that consumers experience a temporary reduction in the willingness to make volitional choices, which is caused by a prior exercise of volition, like resisting temptations (Baumeister, Bratslavasky, Muraven, & Tice, 1998). Consumers, have to resist all different advertisements in-store constantly, which makes that the self-resources get depleted, this makes the consumer more vulnerable to impulse buying because, one experiences less self-control (Baumeister, et al., 1998). In short, what Baumeister et al. (1998) suggested in their research is that the self-regulation of people reduces when they are exposed to high arousal or over-stimulation. Consequently, this would create an environment that is very influential for impulse buying (Wirtz et al., 2007). Which is the case when the whole supermarket is filled with Christmas advertising for all kind of different products.
2.3 IN-STORE ADVERTISING
11
messages in order to increase impulse buying. When a consumer walks by a relevant visual stimulus, like a display, the buying impulses can be set off (Piron, 1991). This results in an urge to acquire the product by the consumer because of the environmental stimuli (I see I want to buy) (Rook, & Hoch, 1985). So, in-store stimuli can be very effective, and they are might even more effective in the weeks before Christmas.
To make the need for research into in-store advertising more obvious, Phillips (1993) said that the intent to purchase can be modified all the time until the point of purchase, because it is not fixed. This is even more the case in events of unplanned purchases than planned purchases, which makes unplanned purchasing a big factor.
In his paper about the influence of advertising during sports event, Gijsenberg (2014) researched the effectiveness of out-of-store advertising, like TV commercials. According to that paper, on average the own advertising investments have more than 50% less direct impact on sales of brands around sports events. However, in-store advertising was not used in the before mentioned research, while this marketing tool might have a lot of potential. By in-store stimuli the
promotional techniques that are placed in the store in order to increase unplanned purchases of products, are meant. Techniques that can be used for this purpose are in-store siting, price-off promotions, on-shelf positions, point-of-purchase displays, sampling, in-store demonstrations, and coupons (Abratt, & Goodey, 1990). In this study display advertising will have the main focal attention.
Related to in-store advertising, there is already prove that the retail atmosphere plays a significant role in the decision-making of consumers. Retailers that are willing to invest valuable resources into the physical appearance of the store, can see a positive impact of this on purchasing
behaviour of consumers (Babin, & Attaway, 2000).
12
(Kollat, & Willet, 1969). Second, unplanned purchasing can partly be attributed to the incomplete measurement of the purchase plans. So, part of the impulse purchases are because of in-store stimuli (Kollat, & Willet, 1969). This makes that in-store advertisements could be very helpful in order to influence the choices consumers make. According to Peak, and Peak (1977) and Quelch (1983) displays can increase unplanned purchases in retail stores, this is because consumers mostly focus their attention at the eye level. More studies found that the sales of items can be increased by the use of displays (Chevalier, 1975).
Again according to Chevalier (1975), display can account for a sizeable percentage of supermarkets total sales. Furthermore, price promotions, feature and display advertising can result in an increase of brand sales and an decrease in sales from competing brands in-store and in competing stores (Kumar , and Leone, 1988). Next, display and feature are able to increase brand choice, because of the following theories (Zhang, 2006):
The ‘price-cut proxy effect’, consumers can interpret an in-store promotion marker as a proxy for a price cut.
The ‘consideration-set formation effect’, this explains that in order to form consideration sets, display and feature promotions are likely to be used.
These theories could imply that an display could also be effective, without a price promotion. Because consumers apparently make a lot of in-store decisions, the expectation is that in-store advertising messages can influence these decisions in a positive way. This implies that sales can increase because of in-store advertising. This is also studied by Abratt and Goodey (1990), who found that in-store stimuli significantly influence unplanned purchasing. In the same study they found that 41% of the respondents spend more on groceries than they were intended to do. So, the first thing that is expected to happen in this research is that, sales can go up when using in-store stimuli, like display advertisements.
13
stimuli. This results in depletion of the self-resources, which consequently results in consumers who are more likely to get involved into impulse buying. Besides that, consumers have time to browse the supermarket, which also results in more impulse buying. However, one week before Christmas, this effect is likely to vanish because, consumers are used to the advertisements which means that it does not result in ego depletion anymore. Furthermore, consumers are under time pressure the week before Christmas, which implies that they are probably less likely to get involved into impulse buying. So, the expectation here is, that the use of in-store stimuli will result in an increase of sales, two and three weeks before Christmas, but this effect is likely to vanish one week before Christmas. However, this hypothesis can also be contradicted because, many consumers actually do have more time than usual, because of the Christmas break. In this case, they are not pressured by time and they might also get involved into impulse buying.
2.4 BRAND SALES VS. CATEGORY SALES
The final topic that needs to be discussed is if in-store advertising only results in consumers switching to another brand or if it also leads to an increase of category sales. According to Mela, Gupta, and Jedidi (1998), advertising increases brand differentiation. This is also proved by Nijs, Dekimpe, Steenkamp, and Hanssens (2001), who found that within the category, advertising creates differentiation among brands. Related to this, they also found that category demand is predominantly stationary, around a deterministic trend. Which indicates that advertising only results in consumers choosing another brand within the same category. The sales of the whole category does not increase because of advertising.
14
this implies that all brands in a category are showing peaks and throughs at the same time. If this is not the case, it implies that some brands win and some brands loose (Gijsenberg, 2017). By using these correlations, it becomes clear if the sales of the whole product category increases or if consumers only switch to another brand when they are exposed to display advertising.
Again, this is likely to differ per product category, because according to Gijsenberg (2017), depending on the product, the extent to which intrayear cycles can affect category demand can differ markedly. He gives the example of product categories like washing machine products, that can be expected to be in relatively stable demand patterns. However, strong cycles are likely to occur in the demand for categories like sun preparations, here the strongest demand peak is concentrated in the summer.
Other research found that brand differentiation accounts for more than 84% of the sales increases that is the result of promotion (Gupta, 1988). The same was found by Totten, and Block (1987), who indicated that brand switchers are the primary reason for the promotional volume increase. However, Vilcassim, and Chintagunta (1992), found the opposite and claim that the increase in promotional volume is because of category expansion. This is also consistent with observations about price elasticities. Bemmaor, and Mouchoux (1991), found that own-price elasticities are bigger than cross-price elasticities, this indicates that an increase of promotional volume of the own brand is not because of a decline in the sales of another brand.
So, there are two different perspectives on this issue, one explanation for this could be that it depends on the characteristics of the category (Blattberg, & Wisniewski, 1987). Blattberg, and Wisniewski (1987), explain that the potential for increased consumption varies between category products.
So, there is no clear expectation regarding a sales increase. Is this mainly due to brand switching or to an expansion of the whole category? All research done in the past about this subject is contradictory. The only thing that can be said about it, is that it probably depends on the product category. So, the expectation is that an increase in sales is due to brand switchers in some categories, and is due to category expansion in other categories.
3. METHODOLOGY
15
approach should allow advertising effects to differ in different periods during the year. Second, the model should be able to handle contemporaneous correlations in sales among brands and categories.
3.1 VARIABLES
The main goal of this study is descriptive, i.e., to describe the relationship between display advertising and sales, rather than predicting the volume of sales. The following variables are included in the model: brand sales, category sales, week, and the expenditures on display. The various variables are provided in table 1. For all variables, the natural logarithm is taken, except for the values of the dummy variables.
Table 1 Variables used to measure the construct
Variable Variable Type Measurement
Brand sales – The unit sales of
brand j in week t
Interval (0 - ∞)
> 0: amount of sales for brand j in week t
0: no sales
Category sales – The unit sales
of category j in week t
Interval (0 - ∞)
> 0: amount of sales for category j in week t
0: no sales
Display – The investment on
display in a specific week for a specific brand/category
Interval (0 - ∞)
> 0: expenditures for display advertising for brand/category j in week t
0: no expenditures on display
Display high – The effect of an
investment in display in week t-1 on sales of week t
Binary (0 – 1)
1: Displayt-1 > Displayt
0: Displayt > Displayt-1
Week – To indicate if it is three
weeks before Christmas, one week before Christmas, or a ‘normal week’
Binary (0 – 1)
1: for normal week 0: for other weeks Binary
(0 – 1)
1: for 2 or 3 weeks before 0: for other weeks
Binary (0 – 1)
16
The variable ‘High Display’ will be used as an extension to this research, and is only included in the last model. This variable explains if there is a lasting effect of display advertising. The
variable for week is split into three different dummy variables, the ‘normal’ weeks, the ‘one week before Christmas’, and the ‘three weeks before Christmas’ variable. However, the first variable will not be shown in the model because it is the benchmark.
3.2 CONTROL VARIABLES
In order to reduce the error term, multiple control variables are included to capture effects which might influence the construct of interests – sales (Becker, 2005). Table 2 provides an overview of the control variables that are included. For all variables the natural logarithm is taken.
Table 2 Control variables used during this research
Variable Variable Type Measurement
Advertising – The use of other
forms of advertising (no display)
Interval (0 - ∞)
> 0: amount of money spend on advertising for brand/category j in week t
0: no money spend on advertising
Competitor actions – The use of
a display by competitors
Interval (0 - ∞)
> 0: amount of money spend on display advertising for brand j in week t, by competitors
0: no expenditures on display by competitors
Competitor actions – The use of
other forms of advertising by competitors
Interval (0 - ∞)
> 0: amount of money spend on advertising for brand j in week t, by competitors
0: no expenditures on advertising by competitors
Price – Price of a specific brand
in a specific week
Interval (0 - ∞)
> 0: price of brand j in week t
17 Competitor Price – Price of
competitor brands in a specific week
Interval (0 - ∞)
> 0: price of competitor brand j in week t
0: competitor brand j was not available in week t
Lagged Brand Sales – The
volume of sales per brand in week t-1
Interval (0 - ∞)
> 0: amount of sales for brand j in week t-1
0: no lagged sales for brand j
Lagged Category Sales – The
volume of sales per category in week t-1
Interval (0 - ∞)
> 0: amount of sales for category j in week t-1
0: no lagged sales for category j
Some control variables are only used in the model that is specifically focussed on brands. These variables are: competitor price, own price, both the competitor actions, and of course the lagged brand sales. This is done because, it would be very complicated to use these variables on category level.
3.3 MODEL SPECIFICATION
To allow for differential effects of display investments, three weeks before Christmas, one week before Christmas, and other weeks, a set of event conditions j is defined. Unfortunately, a multiplicative model cannot allow for dummy variables, so these have to be turned into
multipliers. The before condition 23, refers to two and three weeks before Christmas, the other before condition 1, refers to one week before Christmas.
3.3.1 BASIC MODEL
The basic model includes the independent variables Display Advertising, and Week. Model 1 can be specified as follows:
ln(𝑆𝑗𝑡) = 𝛽0𝑗+ 𝛽1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛽2𝑗𝑊1𝑡+ 𝛽3𝑗𝑊23𝑡+ 𝜀𝑗𝑡
Sjt = Sales volume for brand/category j, in week t, measured in # of products.
β0j = An unknown constant (intercept)
lnDjt = The use of display advertising for brand/category j, in week t.
18
0 for all other cases.
W23t = Multiplier for week two and three. The 1 indicates when it is the case of two or three
weeks before Christmas, 0 for all other cases. εjt = An error term
3.3.2 TESTING MODERATION
A variable that has an impact on the strength and/or direction of the relationship between an independent and an dependent variable is called a moderator (Baron, and Kenny, 1986). When it is the case that there exists a moderating effect, there is an interaction effect present between the independent variable and the moderator. The interaction effect of display advertising and
Christmas is presented as a moderation effect in the second model. In order to create this interaction variable, the variables display advertising and Christmas have become a
multiplication, for each case. This new model will be compared to the basic model in order to test whether the moderation effect is significant. Model 2 can be specified as follows:
ln(𝑆𝑗𝑡) = 𝛽0𝑗+ 𝛽1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛽2𝑗𝑊1𝑡+ 𝛽3𝑗𝑊23𝑡 + 𝛽4𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛽5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡 + 𝜀𝑗𝑡
lnDjtW1t = The use of display advertising for brand j one week before Christmas, if this is not the
case, the variable will take the value 0 (after taking the LN).
lnDjtW23t = The use of display for brand j two or three weeks before Christmas, if this is not the
case, the variable will take the value 0 (after taking the LN).
3.3.3 ACCOUNTING FOR CONTROL VARIABLES
As mentioned before, some control variables are included in order to reduce the error term. These variables are: Advertising, Competitor Actions, and Price Discount. Lagged category and lagged brand sales are included in the next paragraph. Model 3 can be specified as follows:
ln(𝑆𝑗𝑡) = 𝛽0𝑗 + 𝛽1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛽2𝑗𝑊1𝑡+ 𝛽3𝑗𝑊23𝑡 + 𝛽4𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛽5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡 + 𝛽6𝑗𝑙𝑛𝐴𝑗𝑡 + 𝛽7𝑗𝑙𝑛𝐶𝐷𝑗𝑡+ 𝛽8𝑗𝑙𝑛𝐶𝐴𝑗𝑡+ 𝛽9𝑗𝑙𝑛𝑃𝑗𝑡+ 𝛽10𝑗𝑙𝑛𝐶𝑃𝑗𝑡+ 𝜀𝑗𝑡
lnAjt = Use of other forms of advertising for brand/category j, in week t.
lnCDjt = Use of display by a competitor for brand j, in week t.
lnCAjt = Use of other form of advertising by competitors for brand j, in week t.
lnPjt = Price (ƒ.) of brand j, in week t.
19 3.3.4 LAGGED SALES
Marketing is dynamic in essence. For example, when consumers or competitors anticipate a marketing stimulus and adjust their behaviour, one can observe a sales response. Such an effect is called a lagged effect (Leeflang, et al. 2015).
Because lagged sales effects are likely to occur in the dataset used in this study, these effects needs to be accounted for. Model 4 is composed which include the lagged sales effects of one week. Model 4 can be specified as follows:
ln(𝑆𝑗𝑡) = 𝛽0𝑗 + 𝛽1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛽2𝑗𝑊1𝑡+ 𝛽3𝑗𝑊23𝑡+ 𝛽4𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛽5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡 + 𝛽6𝑗𝑙𝑛𝐴𝑗𝑡 + 𝛽7𝑗𝑙𝑛𝐶𝐷𝑗𝑡+ 𝛽8𝑗𝑙𝑛𝐶𝐴𝑗𝑡+ 𝛽9𝑗𝑙𝑛𝑃𝑗𝑡+ 𝛽10𝑗𝑙𝑛𝐶𝑃𝑗𝑡+ 𝛽11𝑗𝑙𝑛𝑆𝑗𝑡−1+ 𝜀𝑗𝑡
lnSjt-1 = The lagged sales of brand/category j, in week t-1.
Because, some variables cannot be used when measuring on category level, the following model is especially specified for the analyses on category level:
ln(𝑆𝑗𝑡) = 𝛾0𝑗 + 𝛾1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛾2𝑗𝑊1𝑡+ 𝛾3𝑗𝑊23𝑡+ 𝛾4𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛾5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡+ 𝛾6𝑗𝑙𝑛𝐴𝑗𝑡 + 𝛾7𝑗𝑙𝑛𝑃𝑗𝑡+ 𝛾8𝑗𝑙𝑛𝑆𝑗𝑡−1+ 𝜀𝑗𝑡
3.3.5 THE LASTING EFFECT OF DISPLAY ADVERTISING
As an extension to the current research, there will also be checked if there is a lasting effect of display usage. So, for example in week 33 there is an investment done in display advertising, and in week 34 the investment was lower. It would be interesting to see if the investment of week 33 still has an effect on sales in week 34. So, model 5 can be specified as follows:
ln(𝑆𝑗𝑡) = 𝛽0𝑗 + 𝛽1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛽2𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛽3𝑗𝑊1𝑡+ 𝛽4𝑗𝑊23𝑡+ 𝛽5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡 + 𝛽6𝑗𝑙𝑛𝐴𝑗𝑡
+ 𝛽7𝑗𝑙𝑛𝐶𝐷𝑗𝑡+ 𝛽8𝑗𝑙𝑛𝐶𝐴𝑗𝑡+ 𝛽9𝑗𝑙𝑛𝑃𝑗𝑡+ 𝛽10𝑗𝑙𝑛𝐶𝑃𝑗𝑡+ 𝛽11𝑗𝑙𝑛𝑆10𝑗𝑡−1+ 𝛽12𝑗𝐷𝑗𝑡+ + 𝛽13𝑗𝐷𝑗𝑡+𝑊1𝑡+ 𝛽14𝑗𝐷𝑗𝑡+𝑊23𝑡+ 𝜀𝑗𝑡
D+jt = 1: When brand/category j invested more display advertising in weekt-1 in comparison with
week t. 0: If this is not the case.
D+jtW1t = If the investment in displays for brand j in weekt-1 is higher than the investment in week
t (one week before Christmas), if this is not the case, the variable will take the value 0. D+
20
week t (two/three weeks before Christmas), if this is not the case, the variable will take the value 0.
Because, some variables cannot be used when measuring on category level, the following model is especially specified for the category measurements:
ln(𝑆𝑗𝑡) = 𝛾0𝑗 + 𝛾1𝑗𝑙𝑛𝐷𝑗𝑡+ 𝛾2𝑗𝑊1𝑡+ 𝛾3𝑗𝑊23𝑡+ 𝛾4𝑗𝑙𝑛𝐷𝑗𝑡𝑊1𝑡+ 𝛾5𝑗𝑙𝑛𝐷𝑗𝑡𝑊23𝑡+ 𝛾6𝑗𝑙𝑛𝐴𝑗𝑡 + 𝛾7𝑗𝑙𝑛𝑃𝑗𝑡+ 𝛾8𝑗𝑙𝑛𝑆𝑗𝑡−1+ 𝛾9𝑗𝐷𝑗𝑡++ 𝛾10𝑗𝐷𝑗𝑡+𝑊1𝑡+ 𝛾11𝑗𝐷𝑗𝑡+𝑊23𝑡+ 𝜀𝑗𝑡
So, this last model only includes the ‘own’ sales and lagged sales of the whole category, the ‘own’ use of display by the category, the ‘own’ use of advertising by the whole category, and the ‘own’ mean price of the category. Besides that, it also includes the explorative aspect, namely the lasting effect of display advertising.
4. DATA DESCRIPTION
4.1 DATASET DESCRIPTION
In order to empirically examine the expectations proposed in section 2, one data set will be used. The dataset contains weekly data on 560 FMCG categories. The dataset include sales, price, advertising, promotion, distribution, feature and display, as well as market share information. The data in the data set is quantitative, aggregate, and secondary data because it was gathered for another purpose than this research (Malhotra 2010, p. 132). The data can be formulated as panel data, because it consists of a time series of each cross-sectional member in the data set, and the units are followed over a given time period (Leeflang, Wieringa, Bijmolt, and Pauwels, 2015, p. 66). More specifically, the data is aggregated on week level. The data set covers data collected over a period of four years and is collected from week 29, 1994, until week 28, 1998. This means that there is ‘Christmas data’ over four years (1994, 1995, 1996, 1997). In total, this data set contains 208 weeks.
21
categories, can be found. Of all categories, the three brands with the highest market share have been used. So, ultimately there will be two datasets. One dataset contains information about the brands within the sub-categories, and the other dataset only contains information about the six big categories.
Table 3 Categories that will be used
Main Category
Number of categories
Sub-Categories
Food 16 Spice Mixes Spaghetti
Dry Soup Complete Meal Sauces (wet)
Complete Meals Sauces (dry) Peanut Sauce
Cake Mix Jam
Sprinkles Candy bars
liquorice Fruit gums
Cake Pizza
Chocolate Biscuits Yogurt (natural) Alcoholic
Drinks
4 Beer Red Wine
White Wine Rose
Drinks 7 Ground Coffee Tea For One
Cola Orange
Ice tea Apple Juice
Orange Juice Household
Care
4 Batteries Paper Towel
Toilet Paper Detergent (White) Personal Care 6 Bath- and Shower Foam Toothpaste
Toilet Soap Shampoo
Deodorant Hand- and Body Milk
Relaxation products
6 Newspapers Sports Magazines
Book pockets Puzzle Magazines
22 4.2 MODEL-FREE INSIGHTS
In the table below (table 4), there is more information about the sub-category sales per week, average price, and use of display (portion of stores that used a display).
Table 4 Data Descriptives
Category N Sales
Min
Sales Max Sales Mean Price Mean (ƒ.) Display Mean Tea bags 624 264 5 425 2,004 23.84 0.13 Jam 624 3 871 132 208 53 281 5.37 1.64 Sprinkles 416 1 034 83 827 28 650 8.36 0.84 Candy bars 624 946 65 369 14 662 16.13 4.14 Liquorice 416 416 14 857 3 895 5.99 0.87 Fruit gums 416 642 15 306 2 636 11.82 0.97 Cake 624 53 23 664 5 592 9.51 0.85 Chocolate biscuits 624 1 886 93 193 16 206 8.28 3.47 Spice mixes 624 14 880 156 930 67 360 1.79 0.68 Yoghurt 624 5 668 69 394 21 844 4.25 0.16 Pizza 416 2 442 84 606 18 918 13.08 0.25 Spaghetti 624 23 130 190 921 75 061 3.38 2.42 Dry soup 624 28 848 517 104 119 490 2.53 2.13 Toilet soap 624 10 365 75 427 25 708 5.32 2.41 Bath- and shower foam 624 11 161 65 119 22 690 4.61 1.43 Deodorant 624 5 170 21 498 12 439 6.36 0.92 Hand- and body
24 4.3 DATA CLEANING
The first step in cleaning the dataset was to check for brands that had a lot of zero sales during the period. In total there were nine categories which had a brand that contained a lot of weeks with zero sales in it. In these cases, the brand was probably not available or was introduced later in the period. However, these brands have been removed out of the dataset. Furthermore, there were four brands that contained a few weeks with zero sales. In these cases, only these specific weeks of the brand have been removed. After this, there were still 24.930 observations left. After creating the interaction effects, it became clear that there are 208 cases in which there is a display used during the period of one week before Christmas. In the two and, three weeks before Christmas, a display was used 470 times. In 8.341 weeks the investment in display was higher in week t-1 in comparison with weekt.
5.
MODEL-BASED INSIGHTS
5.1 CATEGORY-LEVEL MODELS
First, the models on category level will be discussed. In section 5.2, the brand-level models will be discussed.
5.1.1 MODEL DIAGNOSTICS
Before discussing the results of the analysis, the quality of models 4 and 5 for categories, will be presented. For these analyses, a variable has been made which represents the six different
categories: food, drinks, personal care, household care, relaxation, and alcohol. When estimating the models, a small value had to be added to the (log)display, (log)advertising, and
(log)Week*Display, because they contained zero’s. Both a pooled and a partially pooled model are not allowed according to the chow test, which results in using unit-by-unit models.
25
compare both models by using the AIC, BIC, and RMSE scores. These values can be found in the table below.
Table 5 AIC and BIC Scores on Category Level AIC
Food Drinks Personal Care Household Care Relaxation Alcohol Model 4 2155.503 1941.204 2351.165 2274.6 2325.383 2712.705 Model 5 2135.709 1941.19 2351.623 2273.97 2318.521 2717.961 BIC
Food Drinks Personal Care Household Care Relaxation Alcohol Model 4 2218.737 2004.515 2414.398 2337.911 2388.616 2776.016 Model 5 2217.913 2023.494 2433.826 2356.274 2400.724 2800.265
When looking at the AIC and BIC scores, model 4 and 5 do not differ that much because the values are comparable. It is different per category which model performs better. So, the fit of both models is quite the same. In order to investigate further which model is the best model for the data, the RMSE scores are displayed in the table below.
Table 6 RMSE Scores on Category Level RMSE
Food Drinks Personal Care Household Care Relaxation Alcohol Model 4 0.3136 0.3051 0.3211 0.3175 0.3201 0.3347 Model 5 0.3125 0.3047 0.3209 0.3172 0.3196 0.3346
26
This might implies that the data is not from a normal distributed population. However, when using such a big dataset as is the case here, it most of the times results in a non-normal
distribution. So, it does not have to be a problem because in large samples, linear regression is still valid for every distribution (Lumley, Diehr, Emerson, & Chen, 2002).
To conclude and to go on to the next section, model 5 is the model that will be used to do the final analyses.
5.1.2 RESULTS
Table 7, shows the overall across category parameter estimates together with the associated Z-scores (Rosenthal, 1991), and the P-values. Variable ‘Week 1’ is the variable indicating if it is one week before Christmas. Consequently ‘Week 23’ indicates the period of two or three weeks before Christmas.
Table 7 Overall across-category parameter estimates
Weighted β Z-Score P-Value
Intercept 0.37712 18.617 < 0.0001*** Display 0.02882 8.994 < 0.0001*** Display*Week1 0.05326 2.188 0.029* Display*Week23 0.03426 2.251 0.024* Week 1 0.12682 5.535 < 0.0001*** Week 23 -0.01949 -1.65 0.172 Advertising 0.00297 5.612 < 0.0001*** Price -0.03969 -15.902 < 0.0001*** Lagged Sales 0.96561 20.141 < 0.0001*** High Display -0.02195 -4.706 < 0.0001*** Week1 *High Display -0.09543 -2.589 0.0096** Week23 * High Display 0.03293 1.449 0.147 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
5.1.2.1 THE EFFECT OF DISPLAY ADVERTISING ON CATEGORY SALES
27
category sales, this is in line with the expectations regarding in-store stimuli. It implies that when there is an investment of 1 guilder in display advertisement within a product-category, the sales of the whole category goes up by 0.02882 in that week. This finding is in line with previous studies that found that a prominent display can significantly influence sales (Curhan, 1974; Chevalier, 1975; Wilkinson, Mason, and Paksoy, 1982; Gagnon and Osterhaus, 1985). In the appendix, the regression results per category can be found, these results implies that display advertising has a very significant and positive result for all categories, except for the relaxation category, which includes magazines etc. This is also what Chevalier (1975) found, there are wide differences of the effect of displayed products among product groups. It is interesting to see that display advertising has the highest effect on products in the food category. This implies that consumers are most vulnerable to impulse buying when shopping for food.
5.1.2.2 THE EFFECT OF DISPLAY ADVERTISING ON CATEGORY SALES DURING CHRISTMAS
When looking at the use of display advertisement in the weeks before Christmas, it becomes clear that both variables are significantly influencing sales. However, the effect of investing in display advertising one week before Christmas, is bigger (weighted β = 0.05326). This is in contradiction with the expectations. It was expected that the effectiveness of display advertisement would decrease one week before Christmas because consumers might got used to the advertisements or they are under time pressure to get their groceries fast. However, this is apparently not the case. Consumers do have enough time to do groceries and are consequently likely to get involved into impulse buying.
Investing in display advertisement two or three weeks before Christmas also has a significant and positive effect on sales, although this effect is smaller (weighted β = 0.03426). This result is in line with the expectations, consumers are influenced by the in-store stimuli, a few weeks before Christmas. They might get overwhelmed, which results in ego-depletion, and consequently consumers are more likely to get involved into impulse buying. Or, they simply just have more time to browse the supermarket, which also results in more impulse buying. Either way, using display advertising two or three weeks before Christmas will result in an increase of sales. When comparing the use of display advertising during Christmas with ‘normal’ periods, it
28
the period of two or three weeks before Christmas, display advertising has a quite stronger effect on sales in comparison with other weeks. This implies that using display advertisements is a good idea anyway, however, it works even better during Christmas.
Next thing to look at are the dummy variables. There is a highly significant effect of the dummy that represents the period of one week before Christmas. In this week the sales of the overall product-categories increase a lot (weighted β = 0.12682). This is logical, consumers simply need more products during the holidays, because they consume more these days. Contrary to the expectations, sales for the relaxation category and personal care category, do increase during Christmas. This might implies that consumers are not only focussed on preparing a good
Christmas meal, but also care about themselves. Surprisingly, the sales of alcohol do not increase, which is against all expectations since Christmas is for a lot of people an excuse to drink alcohol (Appendix A).
The overall category sales do not increase in the two or three weeks before Christmas (p > .05). This implies that consumers do not buy more products on beforehand. However, it would make sense to invest in display advertising these weeks because that does have a positive effect on sales.
5.1.2.3 THE EFFECT OF CONTROL VARIABLES
All control variables (advertising, price, and lagged sales) are highly significant, with price influencing sales in a negative way. Lagged sales and advertising are influencing sales positively. These results are logical, because when price goes up, most of the time the amount of consumers decrease and consequently, the sales goes down (Clancy). So, it is not a big surprise that price has a negative effect on sales for these product-categories. Furthermore, advertising results in an increase of sales, which is also logical. When investing in advertising, consumers get more aware of the product and are more likely to buy it. Finally, lagged sales also positively influence sales. This implies, that an increase in the sales of week t-1 has a positive effect on the current sales.
5.1.2.4 THE LASTING EFFECT OF DISPLAY ADVERTISING
29
them in week t-1. An increased inventory significantly reduces purchase incidence and consumers can learn to buy higher quantities on fewer occasions when they get used to
promotions (Mela, Jedidi, & Bowman, 1998). According to the results of this research, the same holds for display advertising. Consumers buy more, but on fewer occasions. This effect is
strongest for the food category, which might implies that consumers buy food products more often in bigger amounts in comparison to other product categories.
The negative lasting effect also holds for the interaction between investing more in display in week t-1 and less in the week before Christmas. It seems that sales in the week before Christmas will decrease when there was investment in display advertising in the period of two weeks before Christmas. The interaction between display advertisement in week t-1 and the period of two and three weeks before Christmas is not significantly influencing sales. This implies that there is no effect of investing in display advertising in week t-1, two or three weeks before Christmas.
5.2 BRAND-LEVEL MODELS
From now on, the models on brand-level are discussed.
5.2.1 MODEL DIAGNOSTICS
First, the variables concerning competitor actions were created. These were created for
competitor price, competitor advertising, and competitor display advertising. The final variables indicate the difference between competitor actions and price, and the own brand actions and, in order to measure the influence on own brand sales. When estimating the first model, it was necessary to add a small value to most (log)variables, because they contained zero’s. So, a 1 was added. A loop was created in order to estimate the model for the three brands per category. So, the model was used for all 43 categories and three times per category, because there were three brands. In order to know which model would be better to use, a pooled, a partially pooled, or an unpooled model, a chow test was performed. It became clear, that a fully pooled model is not allowed to use in this study. Also, partially pooling is not allowed. So, a unit-by-unit model is the best model to use in this case.
Model 4 also includes the lagged brand sales. Lagged sales itself is very significant for all brands and influences sales in a positive way.
Next, model 5 is being estimated. This model includes the lasting effect of display advertising for a particular brand, in comparison with the previous week.
30
which model has the best fit to the data, both models are compared using their AIC and BIC scores. These scores are the mean scores per brand per model (table 8). So, over all 43 categories, the mean AIC and BIC per brand are measured. The AIC scores of brand 2 are very low because there were a lot of negative AIC values.
Table 8 AIC and BIC Scores
AIC BIC
Brand 2 Brand 3 Brand 4 Brand 2
Brand 3 Brand 4
Model 4 1.997 112.347 132.161 45.168 152.397 172.378
Model 5 -3.716 121.219 156.432 48.677 175.467 156.406
According to these scores, model 4 and 5 do not differ that much, because the AIC and BIC scores are comparable. However, most scores for model 4 are a little bit lower than the scores for model 5. The adjusted R-squares for both models are also comparable and for all brands, the values are very high. Both models explain around 97% of the variability in the data. So, only a very small amount is unexplained, which is why both models fit the data quite well. When looking at the VIF scores, it becomes clear that the variables (log)advertising, and
(log)competitor advertising, express a little bit of multicollinearity, because according to Wieringa (2016), when the VIF-value exceeds the limit of 5, the issue of multicollinearity is present. However, according to Disatnik and Sivan (2016), this is not an issue because it is just a matter of interval scaling. Consequently, the variables will be kept in the model in the same way. All the other variables have VIF-values below 5. In order to get more insights into the fit of both models, also the RMSE scores are measured. This is done in the same way as with the AIC and BIC scores. So, it is the average score of all categories per brand, per model. The values are presented in table 9.
31
Table 9 RMSE Scores on Brand Level RMSE
Brand 2 Brand 3 Brand 4
Model 4 0.2311 0.3136 0.3641
Model 5 0.224 0.3035 0.3346
In order to check if the residuals in model 5 are autocorrelated, the Shapiro-Wilk and the Jarque-Bera test are performed. Both tests showed that the data is not from a normally distributed population. This implies that the data is nonnormally distributed. Subsequently, a plot of the residuals for all 43 categories per brand have been drawn. These plots indicate that there are a few extreme values in the data, which would not be expected when the residuals came from a normal distribution. However, for most categories the plot follows a normal distribution. The nonnormality issue is probably because the sample size is very large, which implies that it is not a big issue.
To conclude, the model that is going to be used is a unit-by-unit model, because the regression has to be done for each category separately per brand. Furthermore, the type of model that is being used is Model 5, so the model including the variable accounting for the lasting effect of display advertising.
5.2.2 RESULTS
Table 10, shows the overall across brand parameter estimates together with the associated Z-scores (Rosenthal, 1991), and the P-values. Variable ‘Week 1’ is the variable indicating if it is one week before Christmas. Consequently ‘Week 23’ indicates the period of two or three weeks before Christmas.
Table 10 Overall Across-Brand Parameter Estimates
Weighted β Z-Score P-Value
Intercept 0.36178 18.284 < 0.0001***
Display 0.01756 2.840 0.005**
Display*Week1 0.01991 0.628 0.530
Display*Week23 -0.00267 -0.133 0.894
32 Week 23 -0.01163 -0.808 0.419 Advertising 0.00333 2.229 0.026* Price -0.03988 -13.471 < 0.0001*** Comp. Display -0.00966 -1.517 0.129 Comp. Advertising 0.00035 0.221 0.825 Comp. Price -0.0052 -1.647 0.099 . Lagged Sales 0.96831 90.068 < 0.0001*** High Display -0.01225 -16.788 < 0.0001*** Week1*High Display -0.10132 -2.312 0.021* Week23*High Display 0.03137 1.201 0.231 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
In order to get additional insights into the significant (< .05) parameters per unit-by-unit model, table 11 is presented. In this table the underlying insights per parameter can be found. Also, the parameters that are only marginal significant (p < .1) are included to get more complete insights.
Table 11 Percentage Significant Parameters
33 p < 0.1 10% | 3.33% 4.62% | 4.65% 10% | 5% 16.22% | 0% Price p < 0.1 0.83% | 21.67% 1.67% | 32.5 0% | 11,63% 0% | 30.23% 0% | 30% 2.5% | 37.5% 2.7% | 24.32% 2.7% | 29.73% Competitor Display p < 0.1 4.17% | 6.67% 5.83 | 9.17% 4.65% | 6.98% 4.65% | 9.3% 0% | 2.5% 2.5% | 5% 8.11% | 10.81% 10.81% | 13.51% Competitor Advertising p < 0.1 3.33% | 5.8% 4.17% | 6.67% 6.98% | 2.33% 9.3% | 2.33% 2.5% | 5% 2.5% | 7.5% 0% | 10.81% 2.7% | 10.81% Competitor Price p < 0.1 1.67% | 1.67% 1.67% | 5% 0% | 0% 0% | 4.65% 2.5% | 0% 2.5% | 2.5% 2.7% | 5.41% 2.7% | 8.11% Lagged Sales p < 0.1 100% | 0% 100% | 0% 100% | 0% 100% | 0% 100% | 0% 100% | 0% 100% | 0% 100% | 0% High Display p < 0.1 1.67% | 39.17% 3.33% | 50% 0% | 60.47% 0% | 72.09% 0% | 45% 0% | 65% 5.41% | 8.11% 10.81% | 8.11%
Week 1 *High Display
p < 0.1 0% | 4.17% 0% | 5% 0% | 0% 0% | 0% 0% | 2.5% 0% | 5% 0% | 10.81% 0% | 10.81%
Week 23 * High Display
p < 0.1 3.33% | 3.33% 6.67% | 3.33% 2.33% | 6,98% 2.33% | 6.98% 0% | 2.5% 2.5% | 2.5% 8.11% | 0% 16.22% | 0%
5.2.2.1 THE EFFECT OF DISPLAY ADVERTISING ON BRAND SALES
As can be seen in table 10, display advertising is significantly (p < .01) influencing brand sales. When investing in display advertising, it will influence sales positively (weighted β = 0.01756). This result is in line with the expectations of display advertising. However, when looking at table 11, it becomes clear that only 10.83% of all brands experiences a positive effect of display advertising (p < .1). This picture is quite stable for brands 3 and 4, with the strongest effect for brand 3, on which display advertising has a positive effect 30% of the time (p < .1). For brand 2, the positive and negative effects are the same (9,3%). This might result in an overall
nonsignificant effect of display advertising on sales, for brand 2 over the 43 different categories.
5.2.2.2 THE EFFECT OF DISPLAY ADVERTISING ON BRAND SALES DURING CHRISTMAS
34
The distribution of the positive and negative effects (table 11), indicates that overall, brands experience positive and negative effects of display advertising in the weeks before Christmas, in the same proportion. For the variable of one week before Christmas, only brand 3 experiences positive (0%) and negative (1%) effects of display advertising in the same proportion. Brand 2 only experiences positive effects of display advertising (9.3%), and brand 4 only experiences negative effects of display advertising (5.41%). This probably explains the overall nonsignificant effect of display advertising one week before Christmas, because they might neutralize each other. The effect of display advertising two or three weeks before Christmas is only positive for brand 2 (6.98%). Both brands 3 (12.5%) and 4 (5.41%) experience more negative than positive effects of display advertising in these weeks (p < .1). This implies that the opposite of the
expectation is true, display advertising probably has a negative effect on brand sales, two or three weeks before Christmas. However, this is not significantly proven, so it is hard to draw
conclusions based on table 11.
As expected, brand sales is positively influenced by the period of one week before Christmas (p < .01). This simply implies that consumers buy more products than they do normally, this week (weighted β = 0.15199). Table 11 also supports this, there is no negative effect shown of this week on sales for all brands. Table 10 indicates that there is no significant effect found for the two or three weeks before Christmas on brand sales. This implies that there does not change a lot in terms of brand sales, these weeks in comparison to other weeks. According to table 11, the distribution of positive and negative effects is slightly in favor of the positive effects, across brands. However, this differs per brand. For brand 2, the distribution is even (p < .1), for brand 3 there are more positive effects (7.5% against 2.55%), and for brand 4 there are only negative effects (2.7%). It might be the case that these effects neutralize each other, which can be the reason of the nonsignificant effect. However, it is hard to prove this because the p-value for this variable is far from being significant.
5.2.2.3 THE EFFECT OF CONTROL VARIABLES
35
-0.03988).
Next, what is the influence of competitor price and advertising activity on the own brand sales? In their paper about price and advertising effectiveness over the business cycle, van Heerde et al. (2013), found that the own brand sales are higher if the prices of competing brands increase. In this research the opposite is found, however the effect is only marginal significant (p < .1). This implies that competitive pricing has a negative effect on own brand sales (weighted β = -0.0052). In the same research, they found that competitive advertising has a positive effect on own brand sales. However, in this research this is not the case (p > .05). Also, when looking at table 11, brands experience a more negative effect of competitor advertising instead of a positive effect. Only brand 2 shows a positive effect (9.3%, p < .1). According to Schultz and Wittink (1976), the form of advertising effects that is present in this research is the ‘primary sales effect’, this implies that when a brand increases its advertising, this only affects its own sales, without affecting competitive sales. The same holds for competitor usage of display advertising, this does not affect own brand sales (p > .05). Table 11, shows that own brand sales is more often negatively affected by competitor display usage, which can also be seen when looking at the estimate (weighted β = -0.00966). However, it seems that for the categories chosen in this research, the competitor actions are of low influence on own brand sales.
The last control variable is the influence of lagged sales on brand sales. It seems that an increase in the own brand sales in week t-1 has a positive effect (weighted β = 0.96831) on the current sales (p < .01). This might implies that consumers are brand loyal, because they are likely to buy the same brand again, the following week. The same is found by Bendixen (1993), who shows that products and services which have a low consumer involvement in the purchase decision, like foodstuffs and household goods, are characterized by brand loyalty effects.
5.2.2.4 THE LASTING EFFECT OF DISPLAY ADVERTISING
36
The same holds for the lasting effect of display advertising on the week before Christmas, which is also significant (p < .05) and negative (weighted β = -0.10132). However, there is no lasting effect of display advertising for the period of two or three weeks before Christmas (p > .05).
6. DISCUSSION
6.1 SUMMARY
37
advertising that is used one week before Christmas, does not result in brand differentiation. For the period of two and three weeks before Christmas, the data shows no significant increase or decrease in both category and brand sales. Apparently, consumers do not feel the need to go shopping two or three weeks before Christmas. An explanation for this can be that they postpone the Christmas groceries till the Christmas break starts. This can be emphasized by the fact that sales do increase one week before Christmas.
However, when display advertising is used in the period of two and three weeks before
Christmas, it results in an increase of category sales this period. This implies that when investing in display advertising these weeks, it will boost sales.
According to the results, the positive effect of display advertising not only holds for the two or three weeks before Christmas, but also for other weeks during the year. Both on category and on brand level, display advertising affects sales positively. So, investing in display advertising does result in an increase of the category sales, which is not due because of brand switching.
Consumers simply buy more of all brands. Brand switching results when there is only an increase shown in brand sales, without showing an increase on category level. The findings are in line with the expectations and with previous research. When a consumer walks by a relevant visual stimulus, like a display, the buying impulses can be set off (Piron, 1991). This results in an urge to acquire the product by the consumer because of the environmental stimuli (I see I want to buy) (Rook, & Hoch, 1985). Besides that, consumers tend view displays as special bargains, which is the reason that they often buy products that are placed on a display. Most of the time consumers did not even had the intention of buying these products (Chevalier, 1975). So, display advertising works as a trigger for consumers to buy products without thinking it through.
Furthermore, the extent to which display advertising influences sales is much stronger in the Christmas period in comparison to other periods. This implies that consumers are more
influenced by in-store stimuli this period in comparison to other periods. An explanation for this can be that consumers transfer the positive feelings they have this period on to products they see on a display. In this sense, consumers also get positive feelings about the brands that are
38
effect on category sales during the Christmas period.
Contrary to the expectations, display advertising has a stronger effect on category sales in the period of one week before Christmas, than it has on the period of two or three weeks before Christmas. This implies that consumers are more influenced by in-store stimuli one week before Christmas than they are two or three weeks before Christmas. It might be the case that consumers are more likely to get involved into impulse buying one week before Christmas, because they take time to do the groceries and stay longer in the supermarket. When consumers stay longer in the supermarket, the chance on impulse buying is greater. Most consumers do have more time this period because of the holidays. It seems that this results in more impulse buying because consumers do not feel the rush to do their groceries fast, but instead take their time.
Finally, the lasting effect of display advertising has been researched. The data showed that for both category and brand sales, investing in display advertising results in an decrease of sales in the upcoming week. This implies that there is a long-term effect of display advertising, however this effect is negative. Consumers might stock the products that they bought in one week, which results in buying less of these products the next week. This also holds for the period of one week before Christmas. When there is invested in display advertising two weeks before Christmas, it results in an decrease of sales on both category and brand level. This implies, that consumers also stock products before Christmas.
So, it is an effective strategy to invest in display advertising because it does result in an increase of sales on the short run. It is especially useful to invest in display advertising two or three weeks before Christmas because this boost the sales in that period. However, just like using price
promotions, display advertising has a negative effect on the long term. The upcoming week, sales will decrease. Though, the positive effect of display advertising is stronger than the negative effect for both category and brand sales.
6.2 FINAL CONCLUSION
This research was conducted with the aim of answering the following question: ‘To what extent do in-store stimuli influence category and/or brand sales and is this even enhanced by the
39
be increased by using a display. This might indicate that consumers are vulnerable to impulse purchases this period. The effect of using a display one week before Christmas is higher in comparison when using it two or three weeks before Christmas. This is contradicting to the expected influence of using a display these weeks. It was expected that it would be more effective to use display advertisement two or three weeks before Christmas.
On the other hand it can be seen as very effective to use a display two or three weeks before Christmas, because otherwise the sales in this period do not increase. In this sense, it might be more effective to invest in display advertisement in one of these weeks instead of investing in it one week before Christmas, because in that week the sales will go up anyway.
In a ‘normal’ period, it is possible to increase sales by using a display. It seems that consumers view display advertisement as special bargains which results in more impulse buying. Display advertising has a bigger effect on category sales than it has on brand sales. This implies that an increase of brand sales is not only because of brand differentiation but also because of an increase of the whole product category. This makes display advertising an effective strategy for both retailers and brand managers.
Unfortunately for both retailers and brand managers, there is no positive lasting effect of display advertising, which makes it a less effective strategy for the long run. It seems that display
advertising is effective especially on the short term. On the longer term it is not effective, because it results in an decrease of sales in the next week. However, the positive effect is greater than the negative lasting effect, which overall implies that display advertising can be seen as an effective strategy on both category and brand level.
To conclude, display advertising does influence sales significantly for both brand and category sales and this effect is even enhanced during the Christmas period for category sales.
6.3 MANAGERIAL IMPLICATIONS
