THE EFFECT OF MEDIA CONTACT ON COMPETITOR SALES AND THE EFFECTS ON PRIMARY AND SECONDARY DEMAND

(1)

THE EFFECT OF MEDIA CONTACT ON COMPETITOR SALES AND THE EFFECTS ON PRIMARY AND SECONDARY DEMAND

by

Ruben Oosterhoff

Rijksuniversiteit Groningen Faculty of Economics and Business

MSc Marketing Intelligence MSc Marketing Management

20-06-2016

Huygensstraat 111 9727 JC Groningen

0631314917

r.s.oosterhoff@student.rug.nl Student number 2123878

Abstract

This study examines the effect of Company X advertisements on competitor purchases and

secondary demand. The effects are estimated using a GLS type of regression and a logistic

regression to be able to determine quantities and chances of purchasing with competitors and

to be able to determine the effects on secondary demand in a nested type of model. It is found

that some variables lead to increase of either competitor or secondary demand, where other

variables lead to decreases of these demand. Especially the use of folder seems to have

negative effects on competitor demand and positive effects on primary demand. Finally, also

primary demand is estimated to be able to compare the found effects. A discussion of the

results of this study, managerial implications, limitations and recommendations are given.

(2)

2 Acknowledgements

Firstly I would like to thank my supervisor Alec Minnema for his useful support in writing my master thesis, especially the feedback and the discussions during the meetings helped me to come with new insights. I would also like to thank my fellow students for the feedback and support given during the meetings. I want to thank the second supervisor Dr. Keyvan

Dehmamy for his time to read and comment on the thesis. Finally I want to thank my family

for their support and understanding throughout my study and the process of writing my master

thesis.

(3)

3 Table of contents

Management Summary ... 6

Introduction ... 7

Problem statement... 9

Contribution to literature ... 10

Relevance for practitioners ... 10

Conceptual model ... 11

Literature review ... 12

Effect of offline media contact ... 12

Effect of online media contact ... 13

Customers purchasing online, purchase more from competitors than from focal brand ... 14

Media contact leads to more online purchases ... 15

Online and offline advertising have cross-channel effects ... 15

Data description ... 16

Source ... 16

Dependent binary variables ... 16

Dependent sales (regression) variables ... 17

Dependent variables ... 17

Independent variables ... 17

Demographics ... 18

Variables used ... 19

Methodology ... 21

Model specification ... 21

Model criteria ... 21

Pooling ... 22

Specification ... 22

Effect of media contact on overall online purchases ... 22

Secondary demand model... 22

Final model ... 23

Initial analysis ... 23

Competitor effects model ... 24

Primary demand model ... 25

Estimation ... 25

Results - Regression ... 25

Effect of media contact on online purchase spending (H4a, H5) ... 25

Competitor effects (H1a, H2a, H3a, H5a, H5b, H5c) ... 26

(4)

4 Secondary demand (H1b, H2b, H3b, H5a, H5b, H5c) ... 27

Primary demand ... 28

Results – Logit model ... 29

The effect of media contact on competitor demand (H1a, H2a, H5) ... 29

The effect of media contact on primary demand ... 30

The effect of the inclusive values on secondary demand (H1b, H2b) ... 30

Post-Estimation ... 31

Hausman test for fixed versus random effects ... 31

Basic assumptions for GLM ... 31

Hypotheses reviewed ... 31

Discussion ... 32

Media contact & online purchases ... 33

Secondary demand - regression ... 33

Competitor demand – Regression ... 34

Primary demand - Regression ... 35

Logit models ... 35

Conclusions ... 36

Managerial implications ... 37

Limitations & Suggestions for further research ... 37

References ... 38

Appendices ... 40

Appendix 1 – Example Company X folder ... 40

Appendix 2 – Figures Dependent Variables ... 40

Appendix 3 – Explanation variables used in model specification ... 43

Appendix 4 – Overview of estimation results ... 44

(5)

5 Figures

Figure 1: Primary & secondary demand ... 9

Figure 2: Conceptual model ... 11

Figure 3: Nested logit model (graphically represented)... 24

Figure A1: Average spend per week ... 40

Figure A2: Total number of purchases per week ... 40

Figure A3: Number of competitor purchases ... 41

Figure A4: Spend with competitors per week ... 41

Figure A5: Spend with Company X per week ... 42

Figure A6: Average media contact per week ... 42

Tables Table 1: Example coding of dependent variables ... 17

Table 2: Independent variables ... 18

Table 3: Independent variables after recoding outliers ... 18

Table 4: Summary of demographics ... 19

Table 5: Explanation of variables used ... 20

Table 6: Results regressions online purchases ... 26

Table 7: Results regressions competitor demand ... 27

Table 8: Results regression secondary demand ... 28

Table 9: Results regressions primary demand ... 28

Table 10: Competitor purchases ... 29

Table 11: Company X purchases ... 30

Table 12: Secondary demand effects ... 30

Table 13: Review of hypotheses with explanation ... 32

(6)

6 Management Summary

Through studying literature and subsequently analysing a dataset containing panel data over more than ##### households, different effects of customers’ contact with Company X

advertising were found. Below follows a short summary of some of the most important results found.

First of all, folder advertising seems an effective medium to decrease the chance of customers purchasing with competitors and increase the chance customers’ purchase with Company X.

To drive overall online purchases, it was found that radio, the interaction between display banners with television and the interaction between folder with television have positive effects.

The variables that have an effect on the amount purchased with competitors are similar to all variables that lead to secondary demand. Which indicates that certain media variables might lead to an increase in competitor purchases through a possible increase in total category demand.

Print ads lead to an overall increase in secondary demand, this indicates that not only Company X benefits from the advertising done by print ads, but the whole industry benefits (category expansion).

Customers purchasing with competitors have a lower chance to increase secondary demand, and thus might ‘steal’ away sales from Company X or other competitors. On the other hand, it was found that customers purchasing at Company X have a higher chance to increase the secondary demand and are thus likely to increase the market size for the category.

Furthermore, research on demographics was done from this it was found that difference

between regions exists, especially the southern region purchases more from competitors. Also

customers aged above 65 purchase more with competitors

(7)

7 Introduction

Most companies advertise, either to gain awareness, to promote certain campaigns, or to drive customer sales. The importance of advertising can be seen from the amount of money spent on this marketing activity. According to eMarketer (2015) the worldwide total ad spending reaches $569.65 billion. Considering this large amount of yearly spend, it is important to research the many different aspects of advertising. Also, with the increase in online

advertising spending, the role of traditional marketing is reduced. It is generally accepted that spending on advertising leads to increased brand awareness, increased sales and through this to higher revenues. Clark, Doraszelski, & Draganska (2009) support this by stating that advertising has positive effects on brand awareness.

A customer goes through different sequences before purchasing a product. A simple

representation of this process is given by Peterson, Balasubramanian, & Bronnenberg (1997).

These authors state that the first step can be either the customer choosing a brand to buy from or the customer starts with a category choice (e.g. “I need a new computer”). If the first is the case, the customer will not purchase from the competitor. In the second case, the customer will look for information and prices across retailers in either online or offline channels and subsequently makes a brand choice decision or searches for information in alternative

channels or the other way around. Finally, the customer will make an acquisition decision for a certain brand. From this framework defined by Peterson et al. (1997) it is shown that there are many moments in the purchase journey of a customer where he/she decides to purchase a product from a competitor instead of from the focal company. In the case of Company X, where goods are purchased infrequently, brand choice is likely to occur after searching on- and offline. For tangible goods, the authors comment, the “need for product inspection may strongly influence the decision process” (Peterson et al., 1997), this might lead to the conclusion that customers in this category purchase more offline.

Nijs, Dekimpe, Steenkamp, & Hanssens (2001) state that there are different effect of price promotions. They argue that there is an immediate sales increase but also post promotion dips.

Therefore, the weeks after a major promotional event should also be analysed. Nijs et al.

(2001) also state that there are category expansions effects of price promotions. These effects are about increasing demand for the whole category through a price promotion. Advertising creates more loyalty to a brand and reduces price sensitivity (Krishnamurthi & Raj, 1985).

This finding might be contrasting to what Lewis (2006) finds, when looking at acquiring customers through price advertising. Price advertising leads to more price sensitive customers, thus making them more susceptible to prices from different sources. Company X is

considered as a low cost chain which advertises with low prices. They also send out folders every week, advertise online and have television and radio commercials.

When looking at online advertising in particular, Degeratu, Rangaswamy, & Wu (2000) find

that brand names online are more important when there is information on fewer attributes

available. Adding to that, they find that sensory attributes of products are more important

offline where factual information is more important online. They state that the combined

effect of promotion and discount is smaller online than it is offline. For Company X this

would imply that advertisements online should contain much more factual information than

offline, where offline should be more focused on sensory elements.

(8)

8 Another way to look at elements of advertising is given by Nijs et al. (2001), they state that retailers sometimes insist on increasing promotional spending to stimulate category demand.

Deemphasizing price promotions may weaken the market positions of these companies.

Increasing non-price-promotional advertising therefore is an “effective strategy to reduce category-level price-promotion effectiveness” (Nijs et al., 2001). Thus communicating

benefits, brand name, USP (unique selling points) etcetera will lead to less focus on price and, through this, to a better position in the market for companies who do not focus on using price as a competitive advantage. As was stated before, Company X communicates mainly with their price. Competitors of Company X that communicate with other USPs, rather than price, may deemphasize the effect of the price promotions of Company X and therefore the effect of the price promotions on demand.

Looking at different effects of advertising which are found in marketing literature, among which are the purchase acceleration effect, the post-promotion dip, category expansion and stockpiling. All these effects have different implications for companies. For some companies some effects might be more relevant than for other companies. Stockpiling for example might be a less relevant concept for *** retailers as customers usually only buy 1 or a few items which they then keep/use for a longer period. Post-promotion dips are however more relevant as customers who wanted to purchase a television for example will accelerate their purchase when it is in promotion and many customers will not buy that product anymore after the promotion is over.

Dinner, Van Heerde, & Neslin (2011) state that (back then) researcher have only just begun to find cross-channel effects of online and offline advertising. It is thus a research topic of the last decade, making it important to address these cross-channel effects. They also find that if competitor advertising has the function of recognizing a need, it might be beneficial for the focal company (Dinner et al., 2011). It can however be argued the other way around; if the advertising of the focal company makes the customer recognize a need, it might be beneficial for the competition. For Company X specific this means that if the media contact of a

customer leads to the awareness of a certain product and the customer purchases this product with a competitor, there is an effect of media contact, however not the desired effect for the focal company.

With regard to the sales effects of advertising, Schultz & Wittink (1976) argue that there is a primary sales effect and a primary demand effect of advertising. The first effect refers to the increase in own sales as an effect of advertising. The second effect refers to the increase in own and competitor sales as a result of advertising. The focus of this research will be on the part of the secondary demand effect that leads to the purchases with competitors.

Where primary demand is only about the increase of sales of the focal company, secondary demand is about the increase of demand for the whole category, i.e. the ‘pie’ becomes larger.

A result of advertising of the focal company might not be the increase solely at competitors, but rather an increase of the sales of the whole category. These effects are graphically

represented in figure 1. In the first situation the ‘blue’ company advertises and as an effect of

this advertising gets a larger share (primary demand effect). In the second situation the shares

remain the same, but the total category demand increases (secondary demand effect).

(9)

9 Problem statement

While increasingly more companies are advertising online, the positive effects are well documented. Also the negative effects of online advertising can be found in literature.

However recent research does not show the negative effects of advertising in general and specifically when online advertising of a focal company is measured in the effects on competition. Therefore, in this paper it will be researched whether increased advertising contact leads to more purchases at competitors of Company X and whether this effect is affected by overall online purchases.

Since many companies advertise their products online and offline nowadays, it is important to have insights in the effects of advertising on competition. The effects of advertising have been researched by many scholars. The effect of advertising of a focal company on competitor advertising has however not yet been addressed.

Adding to this, there will also be looked at secondary demand effects, as explained in the introduction. While it might seem that media contact leads to competitor purchases it could be that this effect is caused by increased category spending. Therefore, the found effects on primary demand and competitor demand are compared to the effects on secondary demand to see if results are due to in- or decreases in secondary demand.

20%

60%

35%

20%

45%

20%

60%

20%

60%

v

Figure 1: Primary & secondary demand

Market size: €100 million Market size: €100 million

Market size: €100 million

Market size: €150 million Primary demand effect

Secondary demand effect

(10)

10 The main focus of this paper is online advertising and the effect of this advertising channel on competitor purchases and on secondary demand, therefore the problem statements are

formulated as follows.

Does on- and/or offline media contact lead to more purchases at competitors of Company X and is this effect strengthened online?

Does on- and/or offline media contact lead to more category purchases?

Contribution to literature

The Marketing Science Institute define different research priorities, for 2014-2016 the MSI has divided these priorities in different tiers, with tier 1 being the highest priorities and tier 3 consist of items that are still important, but have the lowest priority.

The first item in the tier 2 list is about “measuring and communicating the value of marketing activities and investments” (Marketing Science Institute, 2016). This research will contribute to this topic as it will shine new light on how the marketing budget will be used and affect the results of competitors.

This research can also be related to the first item in the tier 1 list, which is about new customer behaviours in the multi-media and multi-channel environment. As this research focusses on both the online and offline channel and on traditional and “new” types of media, results derived from it can be supporting to this research priority.

Since the online channel is developing very fast, it is important to get insights in what drives customers to purchase from a company or not and what drives customers to purchase from a competitor. Also adding the combination and the separate effects of the on- and offline media will contribute to current literature as this has only recently been researched. Adding to this, research in this area needs to be done on a regular basis as the internet channel develops so fast.

Relevance for practitioners

For managers it is important to know what the effect is of where they spend their media budget on competitors’ sales. If this study shows that increased spending in media budget leads to higher purchases with a competitor, either on or offline, this information has to be incorporated in decisions on marketing budget allocation. Also taking into account the on- and offline media contact enables managers to better tailor decisions to either on- or offline media to achieve the desired results.

Generalizable results can also be interpreted the other way around so that manager can make

use of the results from this paper based on the advertising done by their competitors. The

potential effects of own media spending on competitor sales can then also be compensated by

the effects of competitor media advertising on own sales.

(11)

11 Conceptual model

Since this research aims at providing more clarity on whether or not the advertising done by Company X leads to more or less competitor purchases and the effect of the advertising done by Company X on secondary demand. The conceptual model is displayed in figure 2, the different variables will be explained in the next section. The conceptual model is based upon the literature found in the literature review and on expectations.

Figure 2 explains the conceptual model of this study. From the conceptual modal it can be seen that contact with media for Company X consists of contact with offline media and contact with online media, which in their turn have an interaction effect. The contact with media leads to more overall online purchases, secondary demand and competitor purchases. It might also lead to primary demand. While the primary demand effects are not of focal interest here, they are estimated to be able to compare difference between primary demand,

competitor demand and secondary demand. The online purchases are assumed to moderate the effect of media contact on secondary demand and competitor purchases. Further

explanation of the hypotheses and variables used in this study are given below.

Next, based on the conceptual model represented above, the literature is reviewed. Based on literature hypotheses are developed. These hypotheses are related to the interaction effects of

Figure 2: Conceptual model Contact with media

Contact with Offline media

Contact with Online media

H1a H4

Secondary Demand /

‘category’ demand

Competitor Purchases H2a

Primary Demand /

‘own’ purchases + +

Online purchases H5

H3

H1b H2b

+ +

(12)

12 online and offline media contact, the direct effects of media contact, online purchases and the effects of the above mentioned on demand.

Literature review

Below a review of current literature is given to find issues in literature and findings related to the conceptual model. The review is build op as follows; first the effect of offline advertising done by Company X on secondary and competitor demand is given. Next the online effects are explained as found in literature. Then the effects of purchasing online and the effects of contact with media on purchasing online are explained. Finally, it is argued that on- and offline channels have cross-channel effects. After this, the methodology of the research that is executed is explained.

Effect of offline media contact

Lewis (2006) finds that acquiring customers through price promotion leads to more price sensitive customers who are more likely to purchase at competitors if the price is lower. It can thus be argued that if the advertisements are price promotions, the customers are more likely to purchase at competitors. When looking at a folder of Company X (see appendix 1), the prices are presented in a large font, thus making customers more aware of the price.

DelVecchio, Henard, & Freling (2006) find no significant effect of price promotion on post promotion brand perception. They argue however that it might not be the case for every brand or category so this needs to be taken into account. Following this line of reasoning, price promotions might detract from brand perception and subsequently might increase competitor sales.

Communicating low prices might lead to a perceived lower quality of products. Völckner &

Hofmann (2007) find that the perceived quality of a product is lowered when the price is lowered. It can be argued that if this applies to products, the same could be applied for a complete brand. Therefore, low price might signal less quality and therefore increase competitor purchases. Although Clark et al. (2009) find that advertising has no effect on perceived quality, making customers more aware of where they purchase their products through advertising and signalling a low price may reduce purchases with the focal brand through reduced perceived quality.

It can be reasoned through a classic AIDA model funnel that customers might purchase at competitors. If the customer first sees a product in a television commercial, his/her awareness is created. Then the customer might also gain interest in the product and does some research in print folders. The customer decides he wants the product and then takes action to purchase it from the company advertised in one of the folders. If the focal company raised the

customer’s awareness but the customer purchased somewhere else, it can be stated that the ad

was more successful in generating awareness for the product but was not successful in letting

the customer perform the action. This is also relevant in researching the effects of media

contact on secondary demand. As raising awareness leads to increased interest and desire for a

product, the actions (i.e. purchases) in this category will probably be higher.

(13)

13 According to van Heerde, Gupta, & Wittink (2003) there are effects of promotions done by a company on secondary demand due to brand switching. They also argue that other brands might benefit from other brand’s promotions as they might benefit more from the category expansions due to the promotion of the focal brand due to category expansions as compared to the loss they suffer from the brand switching effect. For Company X this would imply that they have a positive effect of their own advertising due to primary demand, buy also positive effects of their promotions due to brand switching customers (secondary demand). For certain competitors of Company X it could imply that the brand switching they suffer due to

promotional activities of Company X is compensated for by the category expansion effect as a result of these same promotional activities.

Lewis & Reiley (2014) find that online advertising has a strong effect on offline sales in their research on this effect for a retailer. Since there apparently can be strong cross channel effects of advertising on sales (on-/offline) it could be that offline media has a strong effect on

competitor sales.

H1a: Offline media contact with the focal brand leads to more purchases with the competitors H1b: Offline Company X media contact leads to increased category demand

Effect of online media contact

An internet motive is defined by Rodgers & Thorson (2000) as “an inner drive to carry out any online activity”. Where according to the authors the word “drive” is about the effort the user has to put in. Rodgers & Thorson (2000) argue that customers go through certain stages when forming their consumer behaviour. First consumers will find information in

commercials they see, next they will form attitudes towards the product or brand and intentions to (or not to) purchase, based on the information they find in those ads combined with what they already know.

The finding of Lewis (2006) are also applicable to the online medium when price promotions are used. Combining this with the fact that customers can much easier compare products online through comparison sites might lead to the conclusion that customers are much more price sensitive on the internet already and are therefore even more likely to purchase from competitors (i.e. the company that offers the lowest prices.

Banners are a major form of online advertising research has shown that banners cause

increased awareness, when potential customers actually click on these banners, their positive attitude towards these banners increase and purchase intentions become stronger, where larger banners have more clicks than their smaller equivalents (Rodgers & Thorson, 2000).

Since the online medium is much faster in spreading information, comparing competitors and actually buying from other companies the same applies for the online media effect as for the offline media effect, only the relation is expected to be stronger. Also not having to physically go to other stores (physically move) might be strengthening the effect.

Price comparison advertising has an effect on acquiring new customers (Breuer & Brettel,

2012). The authors find that price comparison type of advertising outperforms other types of

advertising when acquiring new customers. This form of online advertising is thus successful

(14)

14 to acquire new customers with a company, but less successful in generating conversions from current customers.

The online medium has many different types of advertising possibilities. Banners are mainly used for creating awareness, while text type advertisements in search engines are more aimed at generating a conversion. These different type of ads probably all have different ways of affecting the customer in his/her purchasing decision. When online, memorizing the product from other advertisements might lead to competitor purchases when encountering their advertisements. In this case increased media contact with the focal company might lead to more competitor purchases. This effect might also be visible when looking at the secondary demand effect, where the increased media contact might lead to more demand in the whole category.

The type of task a customer is doing online could have an effect on the effectiveness of online advertisements. If customers are performing a demanding kind of task, they are probably less involved or only processing the message of the ad superficially. When the task the customer performs is less demanding however, the ads could be more thoroughly processed and the ad performs better for the focal company.

Nijs et al. (2001) find that price promotions have an effect on secondary demand. Where they make a distinction between short-run and long-run category expansion effects, it was found that price promotions overall have a large effect on category expansion, but the effect is only short term. The introduction of a new product, however, has according to these authors a more long-lasting effect on category expansion. As prices are more prominent online (in banners, comparison websites etc.) it can be argued that category expansion effects exist for Company X online, but the effects can only be seen in the short run.

H2a: Online media contact with focal brand advertisements leads to more purchases with the competitors

H2b: Online media contact with focal brand advertisements leads to more purchases within the category

Customers purchasing online, purchase more from competitors than from focal brand Lynch & Ariely (2000) did research on online comparison possibilities of wine. They found that price sensitivity online increased when cross-store comparison was made easy. It can be argued that for Company X the same applies, when customers come into contact with

advertisements online of Company X, they are much abler to compare the statements in the ad when compared to offline. The internet as a medium has many possibilities to easily compare products, companies and information. It is expected here that customers who purchase online also have more chance come into contact with these comparisons and are therefore more likely to purchase from competitors. Consistent with this, Breuer & Brettel (2012) find that PCA (Price Comparison Advertising) is very effective in getting new customers to purchase from a company.

The nature of the internet is different from the traditional/offline media, which is an

information-rich environment, ease of searching, organizing and disseminating information, able to provide searched-for information on demand and the ability to provide

visual/perceptual experiences beyond offline media (Peterson et al., 1997). This makes it

(15)

15 easier for customers to check out other companies their products, prices and promotions.

Having to physically move to another store while being offline might discourage customers to compare products, prices and promotions with competitors.

Thus the more customers purchase online, the more they purchase from competitors due to the ease of comparing different competitors and products. Therefore, there is an expected

moderating effect of online purchases on the relation between media contact and number of competitor purchases.

H3a: Purchasing online moderates the effect of media on competitor purchases, online purchases are done more with competitors than offline purchases

H3b: Purchasing online moderates the effect of media on secondary demand Media contact leads to more online purchases

Based on a combination of the above effects of online and offline media contact, it can be concluded that customers who come into contact with media, either on- or offline leads to more online purchases.

It can be argued that there might be a repetition effect of advertising contact with an ad offline and seeing it again online. If a customer first encounters an ad message offline and then goes online to look for a product, the effect of the offline and online ad contact is higher than when the customer saw only one of the ads either online or offline.

H4a: Advertising leads to more purchases online

H4b: Online purchases mediates the effect of advertising on competitor purchases (online media causes customers to purchase more from competitors)

Online and offline advertising have cross-channel effects

Dinner et al. (2011) argue that advertisements have an “own-effect” and a “cross-effect”. The first refers to the direct effect of advertisements drive sales in the same type of medium.

Where the latter is about driving sales in the other medium. They find that the cross-effect is larger than initially was expected, i.e. online advertising has a large effect on offline sales. It can thus be argued that customers who come in contact with online advertisements, are not online more likely to purchase online, but also their offline purchase likelihood increases.

Similarly, Abraham (2008 IN Dinner et al., 2011) find that, the largest effect of combined search and display advertising was the effect on offline sales. This underlines the importance of researching the own and cross channel effects as explained above when researching the impact of the different media channels.

Remembering the advertisement might strengthen the effect of online and offline

advertisements. As people have seen a similar ad before, they might remember the product or the message of this advertisement. Oliphant (1983) finds in his research on word repetition and recency tasks that there is a recency effect in remembering words the respondents have seen before. He also finds that recency does not influence the effect of repetition in this task.

Following the line of reasoning of this research, it can be possible that certain repetition

effects also exist in the remembering of advertisements and the effect this advertisement has

on the customers.

(16)

16 H5a: online advertising leads to larger effects of offline advertising

H5b: offline advertising leads to larger effects of offline advertising H5c: the combined effects are larger than the individual effects

Next follows the description of the dataset that is being used for this study. First the source of the data is given, next the dependent, independent and descriptive variables are explained.

This is then followed by the methodology section where the model is specified and

explanations are given on the type estimations that are done. The estimations and the results are then explained next, followed by a discussion in the last section.

Data description

The data description is given first by describing the source of the data. Next the dependent variables, the independent variables and demographics that are being used in this study are explained and descriptive statistics are given. After the data is described, the methodology of executing the research is explained.

Source

The data used in this study is gathered from the Company X by Company Y. Company Y gathered this data from a panel of customers by analysing the purchases and their contact with media over a period of ## weeks. The households who opted in for the panel living in the Netherlands and are related to purchases done at Company X in *. The data from this panel is then added in a file which is used in the analyses executed in this study.

It is found that the first 4 available weeks in the dataset do not contain any sales. These weeks are deleted because they are unable to explain any effects. As problems might arise when creating lagged variables, first the lagged effects variables are generated and after that, the weeks are deleted. Doing it this way ensures that all data can be used. The same pattern, as was explained above, can be seen from the nominal variable indicating whether there was a purchase yes or no. The first few weeks there are no purchases done, the most purchases in a week are ## purchases in week ## (see for examples appendix 2).

Dependent binary variables

Different dependent variables are used in this study as different models are estimated. The first dependent variable is the competitor purchase variable, which is a bi-nominal (yes/no) variable indicating whether a customer bought with a competitor in a certain week. Statistics on this variable are given in table 3. This shows that most purchases are done with

competitors of Company X. In the dataset, #### purchases are done with competitors and

#### are done with Company X. The second dependent variable of interest is the variable indicating secondary demand. This variable gets the value 1 if a purchases is made within the category (####=####+#### occurrences), either with Company X or with a competitor. If no purchase was made within the category, this variable is coded as a 0 (###### occurrences).

Finally, the Company X purchase is included to indicate whether a purchases was made with

Company X (1; #### occurrences) or not (0; #### occurrences).

(17)

17 Table 1 shows the different coding of the binary variables. These variables will be estimated using a binary choice model. The first variable is an indicator of whether a purchase is done with a competitor. The second variable, is used to estimate the effects of media contact on Company X own sales. And the third variable estimates an increase in secondary demand.

The way the dependent variables are coded indicates that customers first decide to buy a product and then decide where to buy it. This is the case since the variables are indicated as missing when no purchase was made at all, and only are coded as 0 when there was a purchase made, but not in the variable of interest.

Week Amount spend category

Amount spend competitor

Amount spend Company X

With competitor

?

With Company

X?

Secondary Demand

48 - 7 . . . No (0)

8 39000 39000 No (0) Yes (1) Yes (1)

9-23 . . . No (0)

24 7500 7500 Yes (1) No (0) Yes (1)

25 . . . No (0)

26 15000 15000 Yes (1) No (0) Yes (1)

Table 1: Example coding of dependent variables Dependent sales (regression) variables

For the regressions, dependent sales variables are used. These variables indicate the amount spend in euros, in cents (table 4). The variables that are used are the amount spend in the category, which is the secondary demand, the amount spend with Company X and the amount spend with competitors. These variables are all included as dependent variables in different regressions, as explained later. The amounts are all denoted in cents, so a value of 39000 should be interpreted as €390,00.

Dependent variables

It can be seen from figure 1 in Appendix C that the spending differs per week. The competitor purchases show the same pattern, there are however more purchases done with competitors in the dataset for competitors than there are for Company X.

Independent variables

The independent variables that are being used in the study are listed in table 1. The

independent variables all indicate media contact per week. The averages are thus the average media contact per week and the same goes for the minima and maxima. The independent variables are measured using a reach, recency, frequency method. Researchers following this method ask panellists what they read/watched or listened to, when they did this for the last time and how often they show this behaviour normally. Based on this, a chance is determined they saw an advertisement. The chances can be higher than 1, indicating someone saw

multiple ads. From the table some numbers are extreme values these are identified as outliers

and replaced with the maximum numbers that seem possible.

(18)

18

Variable Avg. Min. Max.

Folder_Company X #### #### ####

Variable_B_Company X #### #### ####

Banner_overig_Company X #### #### ####

Google_masthead_Company X #### #### ####

Gdn_Company X #### #### ####

Print_Company X #### #### ####

Radio1 #### #### ####

Tv1 #### #### ####

Table 2: Independent variables For GDN (Google Display Network) this means the extreme value of #### (week ##, 20##) is replaced with ###. For the banner variable, the highest value is more than twice the value of the second highest value. Maintaining the value of ### would indicate contact with about ##

banners per day, which seems very much. It is therefore also reduced to the second highest value of ##. This does not go for the ‘variable_B’ variable as it is not so much an outlier.

Variable Avg. Min. Max.

Folder_Company X #### #### ####

Variable_B_Company X #### #### ####

Banner_overig_Company X #### #### ####

Google_masthead_Company X #### #### ####

Gdn_Company X #### #### ####

Print_Company X #### #### ####

Radio1 #### #### ####

television1 #### #### ####

Table 3: Independent variables after recoding outliers

Demographics

The dataset contains weekly household level data covering ## weeks from 20## to 20##. The dataset consists different variables among which different demographics, advertising data and sales data. Since there is no data on products/product categories purchased, the analysis will be on store level. There is also no identifiable information on competitors, therefore

competitors are mentioned as a group rather than specific stores. The dataset contains panel data. The variables folder, print, radio and television are measured using a questionnaire, asking panellists what they listened to/watched. Based on these questions a probability is calculated for these variables that determines the chance a customer came into contact with an advertisement. It has to be noted here that probabilities can be higher than 1 as people can read multiple print advertisement.

There are no households excluded from the dataset. Some variables are recoded from binary coded;1,2 values to 0,1 binary coded to be able to better estimate the binary choice model.

Next, a description of the dataset follows, with a description of the recoded variables after.

(19)

19 Variable # of occurrences Reference group?

Households #####

Children Do have children ####

Don't have children ####

Region South ####

West ####

North ####

East ####

Agglomeration #### Reference group Income

Income €0 - €1.100 #### Reference group Income €1.100 - €2.100 ####

Income €2.100 - €3.100 ####

Income €3.100 - €4.100 ####

Income > €4100 ####

Income Not indicated ####

Education

Lower education #### Reference group Academic education ####

Secondary education ####

Special education ####

Vocational education ####

Age of Household

<29 #### Reference group 30 - 39 ####

40 - 49 ####

50 - 64 ####

>65 ####

Table 4: Summary of demographics

Variables used

Below (table 6) is a summary of the variables that are being used in this study with their type, description of the variable and whether the variable varies per household, time or both. This information is used in specifying the models.

Variable Type Description Varies

per…

hhid nominal Unique id of the household Household Folder_Company X Ratio Contact with a folder of

Company X

Household Variable_B_Company X Ratio Contact with variable B (i.e.

number of contacts with mastheads on non-Google

Household

& week

(20)

20

Variable Type Description Varies

per…

sites per household per week)

Banner_overig_Company X Ratio Contact with banners Household

& week Google_Masthead_Company

X

Ratio Contact with masthead Household

& week GDN_Company X Ratio Contact with banners in

Google Display Network (GDN)

Household

& week Print_Company X Ratio Contact with Print Household

& week Radio1 Ratio Contact with radio Household

& week Tv1 Ratio Contact with television Household

& week Concurrent_aankoop Nominal Indicator of whether

purchase was done with competitor (1=yes, 0=no)

Household

& week Betaal_winkel_ykw Ratio Amount of money spend

with Company X in cents

Household

& week Price_sum_kopers Ratio Total purchased by

household

Household

& week Company X_aankoop Nominal Whether the purchase was

made at Company X or elsewhere (1 = competitor, 2

= Company X)

Household

& week

Competitor_Purchases_Reg Ratio Amount spend with competitors

Household

& week Competitor_Purchases_Binary Nominal Whether a purchase was

made with a competitor

Househdold

& week Lag variables of media effects Ratio 1 period lag Household

& week Dummies (inkom_X) Nominal Dummy indicating whether

the category of income applies to the household, compared to low education level

Household

Dummies region (reg_X) Nominal Dummy indicating whether the region category applies to the household, compared to the 3 large cities indicated in the data

Household

Hhtype Nominal 0 = no children, 1 = children Household lvtHVJ Ratio Age of the housewife in

years

Household hhage Categorical Age of the household,

derived from stratmep

Household

Table 5: Explanation of variables used

(21)

21 Methodology

Since the aim of this research is to estimate customer purchases with the competitor and the total category effects, the best way to describe the relation is by a standard regression. A GLS regression will be specified to estimate the ‘secondary demand’ effect, another GLS

regression type of model will be specified to estimate the effects on competitors purchases as an effect of Company X media contact. Also a binary choice model is estimated to find effects of Company X media contact on competitor purchases as the function of a choice. A second binary choice model is estimated to find effects of media contact on secondary demand. For both types of regressions, the Primary demand is also estimated to test for the effects and to be able to compare these effects to those of the secondary demand and the competitor purchases.

It is chosen to use two types of models, ‘normal’ regression and logistic regression to be able to measure different effects. The regression model is used to estimate the effect of media contact on competitor purchases in terms of monetary value, e.g. do customers purchase more value from competitor when they came into contact with Company X advertisements.

The logistic regression model is used to estimate the effect of online purchases as a nested model and the effect of media contact on the decision of whether a customer purchases with Company X or with a competitor. The logistic regression is also better suited to estimate a nested model than a ‘normal’ regression. In the logistic regression it is first estimated what the effect of media contact is on the decision to purchase online, next, the online purchase

decision is modelled as the independent variable on the decision to purchase with a competitor, purchase in category or purchase with Company X.

The difference in the models lies thus in what the aim of the study is, the regression is about the value, where the logit is about the decision a customer makes. In the next section the model specification and criteria is explained and the model is specified.

Model specification Model criteria

According to Leeflang, Wieringa, Bijmolt, & Pauwels (2015: 26) models should be simple,

“built in an evolutionary way”, “complete on important issues”, are “adaptive” and “robust”.

The specified models are linear to accommodate the “simple” criterion of a model, as Leeflang et al. (2015) state that models should be parsimonious which might indicate that a linear model is preferred to a non-linear model. The model contains only observed values which adds to the simplicity criterion as the model user does not have to combine variables.

Completeness is an important issue for model builders as models should be complete to incorporate all predictor effects. This model is complete on important issues as the variables relevant to the research question are all incorporated in the model.

The model is adaptive in the sense that new media contact variables can be added relatively

easily. Another factor that accommodates this criterion is that the model can be re-estimated

when new data is available to obtain more recent values for the different variables.

(22)

22 Robustness is defined as a “quality characteristic which makes it difficult for a user to obtain bad answers” (Little, IN Leeflang et al., 2015: 33). This means that the model should not produce results which are logically inconsistent. For the binary choice model it would for example be impossible to obtain value higher than 1 or lower than 0. For the regression model it would be impossible to obtain values below 0. This has to be accounted for in the model.

Pooling

Leeflang et al. (2015) specify four ways to account for heterogeneity. Since the interest of this study is to investigate the differences between individuals, a unit-by-unit model is used. This type of model allows for maximal heterogeneity between entities (Leeflang et al., 2015: 49).

Since the individual user is of interest here and there is plenty data to estimate the individual level effects, it is not preferred to pool or aggregate the model.

Specification

The model is specified in the order of the ‘flow’ of the conceptual model. So first the effect of media contact on overall online purchases is specified. The results from estimating this model will be used in subsequent models. Next the effect of media contact on competitor purchases is specified. Lastly the effect of media contact on secondary demand is specified

Effect of media contact on overall online purchases

Since the model will be estimated in a sequential way (see estimation), the first model is the effect of media contact on overall online purchases.

𝑂𝑃

_𝑖𝑡

= 𝛽

₀

+ 𝛽

₁

𝐹𝑂𝐿

_𝑖𝑡−1

+ 𝛽

₂

𝑆𝑃𝐶

_𝑖𝑡−1

+ 𝛽

₃

𝐵𝐴𝑁

_𝑖𝑡−1

+ 𝛽

₄

𝑀𝐴𝑆𝑇

_𝑖𝑡−1

+ 𝛽

₅

𝐺𝐷𝑁

_𝑖𝑡−1

+ 𝛽

₆

𝑃𝑅𝐼

_𝑖𝑡−1

+ 𝛽

₇

𝑅𝐴𝐷

_𝑖𝑡−1

+ 𝛽

₈

𝑇𝑉

_𝑖𝑡−1

+ 𝜀

_𝑡

Equation 1: effect of media contact on online purchases

OP = Online purchases of household i at time t in eurocents β

0…k

= unknown slope parameters

FOL

it

= folder contact for household i at time t

SPC

it

= special banner contact for household i at time t BAN

it

= banner contact for household i at time t MAST

it

= masthead contact for household i at time t GDN

it

= display contact for household i at time t PRI

it

= print contact for household i at time t RAD

it

= radio contact for household i at time t TV

it

= television contact for household i at time t ε

t

= unknown error term

Secondary demand model

Secondary demand is about an increase in total category demand across all competitors in a

category. A secondary demand model is specified to estimate the (total) category effects of

media spending by Company X. First a model is specified that explains the demand based on

explanatory (‘media contact’) variables.

(23)

23 Since it is unlikely that media contact causes direct action of a customer and because there is no information about the day of the week at which the customer had contact with the media in the dataset, it is assumed that contact with media leads to effects in the next week. For folder advertising in particular, it is the case that folders of Company X are sent out to consumers a week before the promotions start. Also the variable indicating someone saw a folder is not varying over time, but is fixed per household. Therefore, a model is specified with only lagged effects. Also because of reasons of simplicity it is chosen to only use lagged variables and not include the ‘normal’ effects as well, this would lead to a complicated model due to moderating effects and problems with correlating variables.

The first model that is specified is the effect of lagged media contact on sales. This is defined by:

𝑆𝑃

_𝑖𝑡

= 𝛽

₉

+ 𝛽

₁₀

𝐹𝑂𝐿

_𝑖𝑡−1

+ 𝛽

₁₁

𝑆𝑃𝐶

_𝑖𝑡−1

+ 𝛽

₁₂

𝐵𝐴𝑁

_𝑖𝑡−1

+ 𝛽

₁₃

𝑀𝐴𝑆𝑇

_𝑖𝑡−1

+ 𝛽

₁₄

𝐺𝐷𝑁

_𝑖𝑡−1

+ 𝛽

₁₅

𝑃𝑅𝐼

_𝑖𝑡−1

+ 𝛽

₁₆

𝑅𝐴𝐷

_𝑖𝑡−1

+ 𝛽

₁₇

𝑇𝑉

_𝑖𝑡−1

+ 𝜀

_𝑡

Equation 2: lagged media contact

The above model only explains sales as an effect of media contact, it is however likely that other variables have an influence on the sales. Therefore, the control variables, as indicated already above, will be included in the model. Also as was explained, moderator variables are included in the model. The final model thus will be as follows:

Final model

𝑆𝑃

_𝑖𝑡

= 𝛽

₁₈

+ 𝛽

₁₉

𝐹𝑂𝐿

_𝑖−1

+ 𝛽

₂₀

𝑆𝑃𝐶

_𝑖𝑡−1

+ 𝛽

₂₁

𝐵𝐴𝑁

_𝑖𝑡−1

+ 𝛽

₂₂

𝑀𝐴𝑆𝑇

_𝑖𝑡−1

+ 𝛽

₂₃

𝐺𝐷𝑁

_𝑖𝑡−1

+ 𝛽

₂₄

𝑃𝑅𝐼

_𝑖𝑡−1

+ 𝛽

₂₅

𝑅𝐴𝐷

_𝑖𝑡−1

+ 𝛽

₂₆

𝑇𝑉

_𝑖𝑡−1

+ 𝛽

₂₇

RWEST

_𝑖𝑡

+ 𝛽

₂₈

RZUID

_𝑖𝑡

+ 𝛽

₂₉

ROST

_it

+ 𝛽

₃₀

RNRD

_it

+ 𝛽

₃₁

OHBO

_it

+ 𝛽

₃₂

OWO

_it

+ 𝛽

₃₃

OMBO

_it

+ 𝛽

₃₄

OMID

_it

+ 𝛽

₃₅

OAND

_it

+ 𝛽

₃₆

HAGE1

_it

+ 𝛽

₃₇

HAGE2

_it

+ 𝛽

₃₈

HAGE3

_it

+ 𝛽

₃₉

HAGE4

_it

+ 𝛽

₄₀

INK1

_it

+ 𝛽

₄₁

INK2

_it

+ 𝛽

₄₂

INK3

_it

+ 𝛽

₄₃

INK4

_it

+ 𝛽

₄₄

𝐼𝑁𝐾5

_𝑖𝑡

+ 𝛽

₄₅

MFLBN

_it

+ 𝛽

₄₆

MFLSPC

_it

+ 𝛽

₄₇

MFLMST

_it

+ 𝛽

₄₈

MFLGDN

_it

+ 𝛽

₄₉

MFLPRT

_it

+ 𝛽

₅₀

MFLRAD

_it

+ 𝛽

₅₁

MFLTV

_it

+ 𝛽

₅₂

MSPCBAN

_it

+ 𝛽

₅₃

MSPCMST

_it

+ 𝛽

₅₄

MSPCGDN

_it

+ 𝛽

₅₅

MSPCPRT

_it

+ 𝛽

₅₆

MSPCRAD

_it

+ 𝛽

₅₇

MSPCTV

_it

+ 𝛽

₅₈

MBANMST

_it

+ 𝛽

₅₉

MBANGDN

_it

+ 𝛽

₆₀

MBANPRT

_it

+ 𝛽

₆₁

MBANRAD

_it

+ 𝛽

₆₂

MBANTV

_it

+ 𝛽

₆₃

MMSTGDN

_it

+ 𝛽

₆₄

MMSTPRT

_it

+ 𝛽

₆₅

MMSTRAD

_it

+ 𝛽

₆₆

MMSTTV

_it

+ 𝛽

₆₇

MGDNPRT

_it

+ 𝛽

₆₈

MGDNRAD

_it

+ 𝛽

₆₉

MGDNTV

_it

+ 𝛽

₇₀

MPRTRAD

_it

+ 𝛽

₇₁

MPRTTV

_it

+ 𝛽

₇₂

MRADTV

_it

+ 𝛽

₇₃

OP

_it

+ 𝜀

_𝑖𝑡

Equation 3: final model

Explanations of the different variables that are specified in the models are explained in Appendix 3. For the logit model, the model specification is similar. However, a variable is added indicating the inclusive value. The concept of the inclusive value will be explained in the estimation of the logit model.

Initial analysis

The moderator variable of special banners with radio (interaction effect) is deleted since the

interaction always results in a 0. The same goes for the interaction variable of banner with

(24)

24 radio. These variables are thus excluded from analysis as they have no explanatory value.

Based on analysis of the Variance Inflation Factors (VIF) it shows that some variables cause multicollinearity and are therefore removed from further analysis.

The lagged interaction effect of banner with television has the highest VIF, it is therefore decided to delete it from the analysis. The lagged effect of banner, masthead, other banners and GDN are also deleted due to high multicollinearity. Lastly the lagged interaction effect of folder with masthead and special banners with television are deleted.

Competitor effects model

The competitor effects model will be very similar to the above model, it has however only a different dependent variable.

The binary choice model is a logit type of model with two possible outcomes, yes or no. In the case of this research this means a customer bought with a competitor, yes, or a customer did not buy from a competitor. A model is specified, similar to the regression model to estimate the effects of Company X media contact on competitor purchases. The logit model used in this study is a nested binary logit model, this type of model is explained by Leeflang et al. (2015) as follows. In essence, the nested logit model is based on the assumption that the choice a customer makes is based on the choice a customer made before. For this study, the nested model is portrayed in figure 3. It is assumed here that a customer first decides to buy a product, next decides to buy it with a competitor or with Company X and next an increase or decrease in either secondary or primary demand should be seen.

Where does the customer purchase?

The effect of the purchase is due to:

Figure 3: Nested logit model (graphically represented) Purchase with

competitor

Purchase with Company X

Increase in secondary demand

Increase in

primary demand

(25)

25 Primary demand model

With a primary demand model, the effect of media contact on the demand of Company X is modelled. Again the independent variables are modelled the same as above, however the dependent variable is differences, since the interest here is in the effect on primary demand instead of secondary or competitor demand.

Estimation

Estimation of the specified models will be done similar to the method of Gordon, Goldfarb &

Li (2012). They specify a nested multinomial logit model which is estimated using a sequential strategy. First the effect of media contact on online purchases is estimated (equation 1). This effect of estimating equation 1 is used in the final model (equation 3). To estimate the models, random effects type of estimations are used as choosing other estimation types e.g. fixed effects do not use any within group variation and are therefore less useful for estimating the specified models. The fixed effects model is also not suitable for the scope of this study as within panel variances are of interest.

To check for multicollinearity, a check on VIF (variance inflation factors) values was performed, from this it was found that there high multicollinearity between some variables existed, these variables were therefore deleted from this analysis. The lagged effect of print interacting with radio and the lagged effect of print showed high VIF values and were therefore excluded from this analysis.

Next follows an overview of the different estimations done. First the results of the regression models will be explained and then the results of the logit models will be given. All

estimations are done for primary demand, competitor demand and secondary demand. The primary demand regressions are not directly within the scope of the research, but are added for completeness as the secondary demand consists of primary demand and competitor demand. After the results section, the hypotheses will be reviewed and this is followed by a discussion of the results.

Results - Regression

A description of the results is given below, the different estimations are linked to the hypothesis they examine and whether they are confirmed. A summary of the hypotheses confirmation is given after the results sections, which is the followed by a discussion.

All regression models are estimated using a robust type of regression. The robust estimation treats issues with misspecification of the model as long as the observations are independent of each other. As the different panels are independent of each other, the robust option is

appropriate to use.

Effect of media contact on online purchase spending (H4a, H5)

When estimating the effect on online purchases with robust variance estimators, it shows the

lagged effect of radio (β=294450; p=<0,01), the lagged effect of television (β=-750; p<0,01),

the interaction effect of GDN and television (β=750; p<0,01) and the interaction effect of

(26)

26 folder and television (β=78750; p<0,01) are all significant. While the significances changed, the estimates that were used in subsequent analysis remained the same as with the non-robust regression.

Regression

^β

P-value Online purchases

Radio 294450 <0,01 Tv -750 <0,01 GDN Tv 750* <0,01 Folder Tv 78750* <0,01 Table 6: Results regressions online purchases

To be able to say more about the existing effects, the data was split by age, from examining this data, it shows that only the age group of 40-49 does have enough data to be able to give some information. From this it can be seen that the combination of folder and television have a high impact (p<0,01). Customers that saw an ad in the previous week on television, have a lower chance to purchase online, customers who saw and ad on television and also saw the folder, have a higher chance to purchase online. It could be concluded based on this that age might be an important factor in whether someone purchases online. Also when comparing age groups of above 40 with below 40, the above 40 group only gives some results where the below 40 group does not reveal any results.

For subsequent analysis there are moderator variables created (=online purchases * media variables) to be able to estimate the effects of online purchases on the relation between media contact and competitor purchases / secondary demand. These moderator variables are

included in the regressions and logits. By analysing these newly created variables, it was found that they have such little information, i.e. the multiplication necessary to create these variables almost always results in values of 0, that they are excluded from further analysis.

For the hypothesis H4a the results of this study show that indeed advertising with certain media (GDN & television, Folder & television, and radio) lead to more online purchases. Also for some variables there is an interaction effect of online and offline advertising, related to hypothesis H5, especially for the GDN combined with television seems to have positive effect on online purchases, thus confirming H5a with a side note that it does not hold for all

variables. Using only television as advertising medium, reduces the amount purchased online, according to this study.

Next follows a regression on secondary demand and competitor demand using the online media purchases as an additional regressor in the estimations.

Competitor effects (H1a, H2a, H3a, H5a, H5b, H5c)

After performing the regressions with robust variance estimators, the following results were found. Customers living in the southern region have a marginally significant (β=13453,48;