Optimizing media strategies:
Short- and long-term cross-channel- and cross synergy effects of
offline and online advertising on offline and online sales using weekly
household level data.
By
Mark Buursma
2
Optimizing media strategies:
Short- and long-term cross-channel- and cross synergy effects of
offline and online advertising on offline and online sales using weekly
household level data.
By
Mark Buursma
University of Groningen
Faculty of Economics and Business
MSc Marketing Intelligence
Master Thesis
June 2017
Oostersingel 3a 9713 EW, Groningen +31648114691 markbuursma@hotmail.com Student number: S23792953 Abstract
The possibility to collect individual level data enables companies to gather more detailed information about the effectiveness of their marketing actions. Although a lot of studies examined the effectiveness of the different advertising channels on sales, no study has yet used this type of individual level data to examine the main-, cross-channel- and cross synergy effects of advertising on offline as well as online sales simultaneously. This study uses weekly household level data from a Dutch retail organization including the first 26 weeks of 2011. After conducting an error correction model to investigate the effectiveness of the different advertising channels, both positive as well as negative advertising effects have been found. Results show a positive short-term (print advertising) and negative short- and long-term (TV advertising) effect, a positive short-term cross-channel effect (Google masthead banners) and a negative short-term cross synergy effect (TV and Google masthead banners) on offline sales. No advertising effect has been found on online sales. These findings, together with managerial implication, limitations and future research opportunities are being discussed.
Keywords
4 Preface
Before you lies the master thesis: Optimizing media strategies: Short- and long-term cross-channel- and cross synergy effects of offline and online advertising on offline and online sales using weekly household level data. It has been written to fulfill the MSc Marketing Intelligence at the University of Groningen in the period from January to June 2017.
Writing my master thesis has been a great experience from which I have learned a lot. The topic, optimizing media strategies, together with the use of panel data from Gfk, made that I have learned a lot about the opportunities and problems regarding analyzing advertising effectiveness with the use of big data.
5 Table of Contents Abstract 3 Preface 4 1. Introduction 6 2. Literature review 7
2.1 Own channel advertising effects 7
2.1.1 Traditional (offline) media 7
2.1.2 Online advertising 8
2.2 Cross channel effect 9
2.3 Synergy effects 10
2.4 Short-term and long-term effects 11
3. Conceptual model 12
4. Methodology 12
4.1 Model choice 13
4.2 Error correction model 13
4.3 Endogeneity 15
4.4 Model Estimation Procedure 16
5. Data description 16
5.1 Offline advertising 16
5.2 Online advertising 18
5.3 Sales 19
6. Model results 20
6.1 Quality of the model 20
6.2 Model results 22
6.2.1 Offline sales 22
6.2.2 Online sales 24
7. Conclusion 25
7.1 Summary 25
7.2 Discussion and managerial implications 25
7.2.1 Offline sales 25
7.2.2 Online sales 27
7.3 Limitations and future research 28
8. References 29
9. Appendix 33
9.1 Appendix A: Estimation offline sales 33
9.2 Appendix B: Estimation online sales 34
9.3 Appendix C: variance-covariance matrix offline sales (1) 35
9.3 Appendix C: variance-covariance matrix offline sales (2) 36
9.4 Appendix D: variance-covariance matrix online sales (1) 37
6 1. Introduction
To reach potential customers, companies have the possibility to use a large variety of marketing channels (e.g. Anderl, Schumann & Kunz, 2016). They can choose several online marketing channels (e.g. search engines, banners, email and social media) on different devices (e.g. desktop, tablet and mobile) as well as traditional offline media channels (e.g. television, catalog, newspapers and outdoor banners) (Verhoef, Kannan & Inman, 2015). Due to this large variety of marketing channes, it becomes not only important to know the effectiveness of each channel (e.g. Zhang et al. 2010; Wiesel, Pauwels & Arts, 2011; Dinner, van Heerde & Neslin, 2014), but also how these channels interact (e.g. Naik & Raman, 2003; Nottorf, 2014). Moreover, due to the increasing pressure on the marketing departments as a whole, marketing managers need to be able to account for the impact of their decisions on firm performance (Verhoef & Leeflang, 2009). One can achieve this accountability by formulating performance metrics to measure the effect of marketing actions on performance indicators, such as return on investments (Verhoef & Leeflang, 2009; Zhang et al. 2010; Dinner et al. 2014). The availability of individual level data enables companies to track individual customer journeys, which in turn helps them in making richer analyses of customers’ purchase decisions and the different factors that influence this decision (Anderl et al. 2016). Since most of the research is based on aggregated level data, the possibility to use individual level data can create new and potentially better insights for managers in the effectiveness of the different marketing channels (Danaher & Dagger, 2013; Nottorf 2014; Anderl et al. 2016).
7
‘What are the short- and long-term cross-channel effects and cross synergy effects of offline and online advertising exposures on offline and online sales?’
This paper will have several theoretical contributions. First, research on cross-channel effects on sales mainly uses aggregated level data, in which online and offline marketing actions are measured based on the marketing expenditures per channel (e.g. Wiesel et al. 2011; Dinner et al. 2014). The papers using individual level data (e.g. Danaher & Dagger, 2013 and Nottorf, 2014), do not make a distinction between own- and cross channel sales. These papers use either overall sales (e.g. Nottorf, 2014) or other dependent variables, such as brand consideration (Naik & Peters, 2009) or website visitations (Hoban & Bucklin, 2015). This study will use weekly household level data, where the effect of the different advertising channels is being measured based on the amount of exposures instead of the total amount of expenditures. Furthermore, recent studies on cross channel sales (e.g. Wiesel et al. 2011; Dinner et al. 2014) do not take cross synergy effects into account. This paper contributes to this field of research by being the first to research both cross channel effects as well as cross synergy effects on sales while using weekly household level data. From a managerial perspective this paper may be valuable since it gives an overview of the effectiveness of the different marketing channels on own- and cross-channel sales as well as the interactions between the different channels.
The remainder of this paper is structured as follows. First, a literature review will discuss the relevant literature on the own channel effects, cross channel effects, cross synergy effects and short- and long-term effects. Second, a conceptual model will be presented. Third, the data and the methodology will be described. Finally, this paper ends with a conclusion, discussion, managerial implications, research limitations and future research possibilities.
2. Literature review
Literature in this field of research can be divided into four major parts: the own channel effects, cross channel effects, synergy effects between the channels and the short- and long-term effects.
2.1 Own channel advertising effects 2.1.1 Traditional (offline) media
8 has an average short-term elasticity 0.22 and an average long-term elasticity of 0.46. Sethuraman, Tellis and Briesch (2011) conducted a similar meta-analysis and found a short-term elasticity of 0.12 and a long-short-term elasticity of 0.24. According to Sethuraman et al. (2011), these results are different due to the increasing competition and globalization, the emergence of the internet and the consumers’ ability to opt out of TV advertising. Rubinson (2009) tried to statistically confirm the declining effect of offline media but did not find any evidence, hence, he concludes that especially TV advertising is still effective. More studies found that, despite the decline in influence, traditional offline media still has a considerable positive effect on sales (Winer, 2009; Danaher & Dagger, 2013), which leads to the first hypothesis:
H1: Traditional (offline) media advertising has a positive effect on offline sales.
2.1.2 Online advertising
The effect of online advertising has been studied extensively (e.g. Rutz & Bucklin, 2012; Manchanda, Dubé, Goh & Chintagunta, 2006). The study of Braun and Moe (2013) found a direct relation between online advertising and online conversions. They state that this direct relation can be improved, when the banner adapts to a consumers’ ad impression history. Goldfarb and Tucker (2011) found that matching a banner to the website content and increasing an ads obtrusiveness will increase consumers’ purchase intention. However, combining these two banner characteristics have a negative effect on consumers’ purchase intention due to privacy concerns. Furthermore, Manchanda et al. (2006) used a constant piecewise hazard model to see which advertising covariates influence the relation between online advertising and the probability of a customer repurchase. They found a direct advertising elasticity of 0.02 of online advertising on customer repurchase. In addition, they found that the number of exposures, number of websites and number of pages are positively affecting the probability of customer repurchase, whereas the number of unique creatives negatively influences this relation.
9 attitudes towards a brand which in turn has a negative effect on consumers’ purchase intentions (Chatterjee, 2008). Additionally, Chatterjee (2008) also states that this negative effect is relatively small in comparison to the positive effect on other banner performance measures, such as brand recall and brand recognition. This also introduces another stream of literature, which argues that online sales is bad indicator of banner performance. Accordingly, Rutz and Bucklin (2012) argue that advertisers’ main goal of online advertising is brand building as opposed to online sales. Drèze and Hussherr (2003) confirm this by stating that although a lot of individuals are not consciously aware of the online banner, they do not directly click on the banner and make a purchase, it does have an impact on other brand measures (e.g. brand awareness and brand recall). This in turn has a significant impact on online sales.
Dinner et al. (2014) combine the findings of previous mentioned research and looked into the direct effect of online advertising on online sales as well as the indirect effect. They found a significant direct relationship between online advertising and online sales, whereas they did not find a significant relation between online advertising and more brand building matrices (e.g. search impression shares). This leads to the second hypothesis:
H2: Online advertising has a positive effect on online sales.
2.2 Cross channel effect
Previous research measuring the cross channel effects including the effect of offline advertising on online sales as well as online advertising on offline sales is scarce (Dinner et al., 2014; Nottorf, 2014). Research that does investigate both cross channel advertising effects mainly use aggregated level data and focus on one specific type of advertising (Nottorf, 2014). Moreover, there are only a few studies that investigate the cross channel effects on sales (Dinner et al., 2014).
10 marketing effects on online funnel metric and online funnel metric on offline purchases. Additionally, they found that online customer initiated contacts (e.g. e-mail and catalog) have a higher profit than online firm-initiated contacts (e.g. paid search advertising). Dinner et al. (2014) elaborated on the work of Wiesel et al. (2011) by adding competitor effects into the model. They found that the cross effects elasticities are almost as high as the own effects elasticities which indicates that it is too naïve to measure a channel’s profitability by only looking at the own channel sales. Their results show a positive cross channel effect on sales for offline advertising as well as online advertising. This leads to the following hypotheses:
H3: Traditional (offline) advertising has a positive effect on online sales H4: Online advertising has a positive effect on offline sales
2.3 Synergy effects
According to Stammerjohann, Wood, Chang and Thorson (2005), there are three main theories that explain the existence of media synergy effects. First, the encoding variability theory of Tavassoli (1998) suggests that when a message is communicated through multiple channels, it will be encoded in a person’s memory in a clearer, stronger and more accessible way in comparison to when only one type of channel is being used. This way the likelihood that a person will recall the information accurately will increase. Second, Schumann and Petty (1990) explain the repetition variation theory in their study, which suggests that to encounter the diminishing effect of repeating the same ad, communicating different messages via different channels will increase the preference for a product. The third theory is the selective attention theory of Kahneman (1973), which explains that people give most attention to ads that are either complex and familiar or simple and novel. To increase familiarity, one can use different channels and different formats to increase complexity. This will increase a customer’s attention to an advertisement (Stammerjohann et al. 2005). These theories suggest that using multiple channels will have a larger effect on advertising effectiveness than using one channel (Stammerjohann et al. 2005).
11 (banners and search ads). Several scholars focus on the cross synergy effects. Chang and Thorson (2004) found a positive cross synergy effect on sales. This effect is the result of an increase in consumer attention and message credibility. Furthermore, Bollinger, Cohen and Jiang (2013) state that cross synergy effects will create positive attitudes towards a brand, which in turn positively affects the purchase intention of a consumer. This study will focus on the cross synergy effects which leads to the following hypotheses
H5: There is a positive cross synergy effect of traditional (offline) media and online advertising on offline sales
H6: There is a positive cross synergy effect of traditional (offline) media and online advertising on online sales
2.4 Short-term and long-term effects
Previous research shows that advertising has a short- and long-term effect on customer behavior (Braun and Moe, 2013). The majority of research finds a positive short-term advertising effect on sales (e.g. Ataman, Van Heerde & Mela, 2010). However, they find different outcomes concerning the long-term advertising effects on sales (Chatterjee, Hoffman & Novak, 2003). In their meta-analysis on offline advertising effectiveness, Tellis (2009) makes a distinction between two possible long-term advertising effects: wear-in and wear-out effects. An advertising wear-in effect occurs when the effect of an advertisement keeps increasing as the advertisement is repeated over time, whereas an advertising wear-out effect occurs when the effectiveness of an advertisement decreases as it is exposed to consumers multiple times over time (Tellis, 2009). The study results indicate that wear-in effects occur quickly or do not happen at all, whereas wear-out effects occur more often when an advertising campaign runs long enough. Therefore, managers should test their advertisements for wear-in and wear-out effects in order to decide how long they should run their advertising campaign (Tellis, 2009).
12 individual stops to respond to the online ads. This is also known as the saturation effect of advertising (Dekimpe & Hanssens, 2007). Since there are different outcomes concerning short- and long-term advertising effects, it is important to take these into account (Braun & Moe, 2013) Therefore, the six hypotheses mentioned before are tested for the short-term as well as the long-term in order to account for their different impact on sales.
3. Conceptual model
Figure 1 shows the conceptual model of this study. It consists of the direct effects of online and offline advertising on both own channel sales as well as cross channel sales. Also, the cross synergy effects between online and offline advertising are included.
Figure 1: Conceptual model
Next to the hypotheses mentioned before, two control variables are included in the conceptual model. These control variables are expected to have a significant effect on both offline and online sales and therefore included in the model.
4. Methodology
13 wear-out effects of advertising. Furthermore, the cross synergy effects between the different types of offline and online advertising should be accounted for and their effect on both offline as well as online sales.
4.1 Model choice
Error correction models have been used extensively in literature, since they make a clear distinction between the short- and long-term effects (e.g. Fok, Paap, Horváth & Franses, 2006; Pauwels, Srinivasan & Franses, 2007). The error correction models differ from other models in the sense that they measure the long-term effects as being the impact of, in this case, advertising on future buying behavior by accounting for changes in individual behavior. This differs from other models since the majority of the models capture long-term effects by adding them as model parameters (Fok et al., 2006). Since this study uses weekly household level data, the first thing that has been investigated is whether it is appropriate to use a random effects panel error correction model. This random effects model treats the individual characteristics between the households as being random and uncorrelated with the explanatory variables (Greene, 2003, p.285). To test whether a random effects model is appropriate a Hausman test (1978) can be conducted. Results of this test were significant (p<0.05), which indicates that the individual characteristics are correlated with the explanatory variables and thus violate the main assumption of a random effects model. Therefore, a fixed effects panel error correction model should be used, since this model accounts for these correlated individual characteristics by including these effects in a dummy variable for each household (Greene, 2003, p.285). However, due to the large amount of households in the data, this approach is less feasible. Therefore, a pooled error correction model has been conducted in this study.
4.2 Error correction model
Equation 1 shows the specification of the pooled error correction model used in this study.
∆𝑙𝑛𝑆𝑎𝑙𝑒𝑠𝑡 = 𝛼 + 𝛽1𝑃𝑒𝑎𝑘𝑡+ 𝛽2𝐻𝐻𝑡𝑦𝑝𝑒𝑡+ 𝛽3𝐼𝑛𝑐𝑜𝑚𝑒𝑡+ 𝛽4𝑠𝑡∆𝑙𝑛𝑇𝑉𝑡+ 𝛽5𝑠𝑡∆𝑙𝑛𝑃𝑅𝐼𝑁𝑇𝑡+ 𝛽6𝑠𝑡∆𝑙𝑛𝐺𝐷𝑁𝑡+ 𝛽7𝑠𝑡∆𝑙𝑛𝑀𝐴𝑆𝑇𝑡+ 𝛽8𝑠𝑡∆𝑙𝑛𝑇𝑉𝑡∗ ∆𝑙𝑛𝐺𝐷𝑁𝑡+ 𝛽9𝑠𝑡∆𝑙𝑛𝑇𝑉𝑡∗ ∆𝑙𝑛𝑀𝐴𝑆𝑇𝑡+ 𝛽10𝑠𝑡∆𝑙𝑛𝑃𝑅𝐼𝑁𝑇𝑡∗ ∆𝑙𝑛𝐺𝐷𝑁𝑡+ 𝛽11𝑠𝑡∆𝑙𝑛𝑃𝑅𝐼𝑁𝑇𝑡∗ ∆𝑙𝑛𝑀𝐴𝑆𝑇𝑡+ 𝜋 [lnSalest−1 −
(β4ltlnTV
t−1+ β5ltlnPRINTt−1+ β6ltlnGDNt−1+ β7ltlnMASTt−1+ β8ltlnTVt−1∗ lnGDNt−1+
β9ltlnTVt−1∗ lnMASTt−1+ β10lt lnPRINTt−1∗ lnGDNt−1+ β11lt lnPRINTt−1∗
14 where
∆ First difference operator
lnSalest Volume of sales in week t
α Inctercept
Peakt Dummy variable indicating a sales peak when 1
HHtypet Control variable type of household
Incomet Control variable income
TVt Amount of exposures to TV advertisements in week t PRINTt Amount of exposures to print media in week t
GDNt Amount of exposures to Google Display Network banners in week t MASTt Amount of exposures to Google Masthead banners in week t
β𝑠𝑡 Short-term advertising elasticity β𝑙𝑡 Long-term advertising elasticity
π Adjustment effect Since the sales and advertising variables are expressed in natural logs, the effects can be interpreted as elasticities1. 𝛽𝑠𝑡represents the term advertising elasticities and the short-term cross synergy effects between the different types of advertising. Furthermore, 𝛽𝑙𝑡
represents the long-term advertising elasticities and the long-term cross synergy effects. In order to capture the advertising effects on cross channel sales, the model has been estimated twice: once to assess its effect on offline sales and once to measure the impact on online sales. Before estimating the results, a Levin, Lin and Chu (2002) panel unit root test has been conducted to test whether the (log transformed) series are stationary. In this case, when the individual household level series are stationary, one can interpret the long-term effects as being the permanent effects of permanent changes in the advertising variables as well as being the cumulative effects of temporary changes (Gijsenberg, 2014). The results of this test show that
1Since the data that is being used contained a lot of 0 values, 0.001 has been added to all the variables before
15 for sales and advertising series the null hypothesis should be rejected at the 5% level, which indicates that all series are stationary.
4.3 Endogeneity
A common problem of using descriptive models to estimate the impact of the different marketing mix instruments on performance is the existence of endogeneity (Ebbes, Papies & van Heerde, 2011). Dekimpe and Hanssens (2007) mention this endogeneity problem of advertising and sales in their research. They argue that not only advertising is driving sales, but sales is also driving advertising. These so called feedback effects imply that firms may respond to an increase in sales, by spending more on advertising (Wiesel et al., 2011; Srinivasan, Rutz & Pauwels, 2016). Moreover, from a consumer perspective endogeneity can exist as well. This study measures advertising effects based on the amount of weekly exposures per household. Therefore, it might be that when a household intends to buy a new product, it will be more aware of the offline advertising exposures as well as online advertising exposures (Braun & Moe, 2013). Furthermore, when a person is actively searching for a product online, retargeting causes an increase in online advertising banner exposures (Lambrecht & Tucker, 2013). Therefore, endogeneity can be present from both a firm perspective as well as a consumer perspective.
One should account for endogeneity in order to avoid biased estimates (Ebbes et al. 2011). Instrumental variables can be used to predict the endogenous variables in the model, which in turn removes the endogenous variation in the explanatory variables (Ebbes et al. 2011). However, this approach is not appropriate in every situation. According to Ebbes et al. (2011), predicting endogenous variables by using instrumental variables is only appropriate when exogenous variables are being used to predict the endogenous variables. They show that when endogenous variables are being used to predict other endogenous variables, these predicted values will increase the bias in the estimates, which makes this approach inappropriate (Ebbes et al., 2011).
16 4.4 Model Estimation Procedure
In order to obtain the initial estimates of the error correction model (equation 1) an OLS regression has been conducted (see appendix A and B). To obtain the long-term advertising parameters, the adjustment parameter π has been multiplied with the term between square brackets. Subsequently, the initial estimates are being divided by −π to obtain the long-term advertising elasticities (e.g. 𝛽4𝑙𝑡
−π ). The standard errors of the long-term advertising parameters
have been calculated by using the delta method (Greene, 2003, p.175). Input for the delta method can be obtained from appendix A t/m D.
5. Data description
The data includes weekly household level data from a Dutch retail company for the period from 29/11/2010 to 03/07/2011 (31 weeks), which has been provided by GfK. In total, 11.672 households are included in the study. However, data about offline advertising has been measured for 9.934 households. To get a complete overview of the interaction between offline and online advertising, only the households with data for both channels (offline and online) are included. The data contains: offline advertising efforts, online advertising efforts and a household’s online and offline purchase behavior.
5.1 Offline advertising
Data collected to measure offline advertising exposures contains: print media advertising, folder advertising, radio advertising and TV advertising. Descriptive statistics of the offline advertising media are included in table 1.
Variable N Minimum Maximum Mean St.dev.
Print 307954 0 17.882 1.745 3.112
Folder 307954 0 1.0 0.329 0.420
Radio 307954 0 28.0 0.422 2.011
TV 307954 0 57.771 2.311 3.988
Table 1: Descriptive statistics offline advertising media.
17 questions about the channels they listen to/watch, at what time and for how long. This way, the chance that a household heard/saw a specific ad at a specific time has been determined and used to obtain the amount of advertising exposures per household. Data concerning print media (advertisements in newspapers) has been collected in a similar way. In a survey, people were asked about which magazines they read (reach), how many (frequency) and the last time they read a magazine (recency). Data concerning folder advertising has been collected in a different way. Based on postal codes and yes/no, no/no mailbox stickers, it has been determined if a household is in the possession of a folder. With this data a dummy variable has been created, which is 1 when a household is in the possession of a folder or 0 when a household does not have a folder. The frequency of the offline advertising exposures over times are shown in figure 2.
Figure 2: Offline advertising exposures over time
As figure 2 indicates, TV advertising has the biggest contribution to the overall offline advertising exposures (51.61%) in comparison to printed media (38.96%) and radio advertising (9.42%). The advertising exposures concerning folder advertising are constant over time, since it is only measured if the household is in possession of the folder, not the amount of times a person looked at the folder. Since TV advertising and printed advertising have a bigger contribution to the overall offline advertising exposures in comparison to radio and folder advertising, this study uses TV advertising and printed advertising to determine the effect of offline advertising exposures.
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 29-11-2010 29-12-2010 29-1-2011 28-2-2011 31-3-2011 30-4-2011 31-5-2011 EXP O SURE S DATE
18 5.2 Online advertising
To measure the online advertising exposures, data on four different types of banners has been collected:
· Google Display Network (GDN) banners: banners on a Google site (site that used google advertising platform)
· Google Masthead banners2: masthead on Google site (mostly YouTube) · Special banners: non-Google masthead banners
· Other banners: non-Google banners
Figure 3: Online advertising exposures over time
Figure 3 shows that GDN banners have the biggest contribution to the total amount of banner exposures (83.27%). Furthermore, in several weeks there were no online advertising campaigns and in others there were. As figure 2 shows, there were 3 weeks in which consumers were exposed to Google masthead banners, 2 weeks to special banners and 1 week to other banners. Since the GDN and Google masthead banners have the largest contribution to the overall online advertising exposures in comparison with the special and other banners, they are being used in this study to determine the effect of online advertising exposures.
2 Masthead banners are ad types which you can see at the top of the YouTube homepage (Google, 2017).
0 2000 4000 6000 8000 10000 12000 14000 16000 29-11-2010 29-12-2010 29-1-2011 28-2-2011 31-3-2011 30-4-2011 31-5-2011 EXP O SURE S DATE
19 5.3 Sales
The purchase data in this dataset indicates per household: if they made a purchase, where this purchase took place took place (online or offline), if the purchase was at the Dutch retail company or at a competitor and how much was spent on the purchase. The total amount of purchases per week can be found in figure 4.
Figure 4: Amount of purchases per week
20
Figure 5: Number of sales per household
The sales variable that is being used as a dependent variable in this study is the monetary value of the purchases instead of the sales volume. The reason for this is the low amount of purchases per household. As figure 5 shows, the majority of the households did not make a purchase and the household that did make a purchase, did this once in the time frame of the study. Therefore, when using volume of sales as a dependent variable, it can almost be interpreted as a binary dependent variable. Since this study aims to use a model that needs a continuous dependent variable, the monetary value of the sales (in eurocents) has been used.
6. Model results
6.1 Quality of the model
21
Model LL R² Adjusted R² P-value AIC BIC.
Model without cross synergy effects (offline)
-376118.8 0.4846 0.4845 0.00*** 752265.5 752411.4
Full model (offline) -376112.5 0.4846 0.4845 0.00*** 752268.9 752498.2 Model without cross
synergy effects (online)
-1195.74 0.4574 0.4574 0.00*** 2419.5 2565.4
Full model (online) -1194.89 0.4574 0.4574 0.00*** 2433.8 2663.1
p < 0.10 *, p < 0.05 **, p<0.01 *** Table 2: Model fit
Table 2 shows, that there is hardly any difference between the models without the cross synergy effects and the full model. According to the AIC and the BIC the models without cross synergy effects are preferred since the have lower values in comparison to the full model. Furthermore, the R2 and the adjusted R2 are almost similar for each model and the same between models. An explanation for this could be that the large amount of observations causes that adding more parameters (and therefore adding degrees of freedom) has little impact on the R2 and adjusted
R2. The similarity of the R2 and adjusted R2 of the full model and the models without cross synergy effects could also be an indication that adding cross synergy effects does not have a big impact on the overall model fit. This is supported by the insignificance of the likelihood ratio test for the full model and the models without cross synergy effects for offline (p=0.13) as well as online (p=0.99). The insignificance of these tests show that the models with cross synergy effects are not significantly better for the overall model fit compared to the models without cross synergy effects.
22 6.2 Model results
6.2.1 Offline sales
The parameter estimates of the different advertising effects on offline sales are shown in table 3.
Variable Beta Std.Error t-statistic p-value
Intercept 𝛼 -6.626 0.0609 -108.73 <2e-16*** Promotion dummy 𝛽1 0.0455 0.0086 5.303 0.0000*** Household type 𝛽2 0.0170 0.0047 3.621 0.0003*** Income 𝛽3 0.0013 0.0004 3.616 0.0003*** ST TV 𝛽4𝑠𝑡 -0.0375 0.0131 -2.857 0.0043*** ST Print 𝛽5𝑠𝑡 0.0144 0.0078 1.834 0.0666* ST GDN 𝛽6𝑠𝑡 0.0021 0.0016 1.289 0.1974 ST MAST 𝛽7𝑠𝑡 0.0126 0.0061 2.057 0.0397** ST TV * GDN 𝛽8𝑠𝑡 -1.573e-05 0.0003 -0.048 0.9621 ST TV * MAST 𝛽9𝑠𝑡 -0.0055 0.0019 -2.876 0.0040*** ST Print * GDN 𝛽10𝑠𝑡 0.0001 0.0003 0.366 0.7144 ST Print * MAST 𝛽11𝑠𝑡 0.0018 0.0011 1.626 0.1039 LT TV 𝛽4𝑙𝑡 -0.0353 0.0210 -1.682 0.046** LT Print 𝛽5𝑙𝑡 0.0139 0.0120 1.156 0.2477 LT GDN 𝛽6𝑙𝑡 0.0029 0.0032 0.922 0.3567 LT MAST 𝛽7𝑙𝑡 0.0112 0.0094 1.191 0.2336 LT TV * GDN 𝛽8𝑙𝑡 0.0004 0.0022 0.168 0.8669 LT TV * MAST 𝛽9𝑙𝑡 -0.0055 0.0035 -1.592 0.1115 LT Print * GDN 𝛽10𝑙𝑡 0.0003 0.0022 0.150 0.8807 LT Print * MAST 𝛽11𝑙𝑡 0.0019 0.0027 0.717 0.4736 Adjustment π -0.9789 0.0020 -482.85 <2e-16*** p < 0.10 *, p < 0.05 **, p<0.01 *** Table 3: Offline sales parameter estimates
23 (p<0.01) whereas printed advertising (𝛽5𝑠𝑡) shows a positive elasticity of 0.0144 (p<0.10).
Furthermore, there is a significant negative long-term elasticity of TV advertising (𝛽4𝑙𝑡 = -0.0353, p<0.10). This shows the short-term effect of TV advertising has a larger negative influence on sales in comparison to its long-term effect. Thus, this indicates that hypothesis 1, which indicates a positive effect of offline advertising on offline sales, is supported by printed media, but should be rejected according to TV advertising.
Furthermore, there is one significant cross channel advertising effect on offline sales. The short-term advertising elasticity of Google masthead banners (𝛽7𝑠𝑡) shows a positive short-term elasticity of 0.0126 (p<0.05). Based on this result, hypothesis 4 is supported.
Next to the main and cross channel effects, there is also one significant cross synergy effects. The cross synergy effect of TV advertising and Google masthead banners (𝛽9𝑠𝑡) shows a negative short-term elasticity of -0.0055 (p<0.01). Thus, hypothesis 5, that indicates a positive cross synergy effect on offline sales should be rejected according to this result.
Next to the all the advertising effects, there is also a significant negative relationship between the adjustment parameter and offline sales (π=-0.9789, p<0.01). This adjustment parameter can be interpreted as being the carry-over sales from the week before.
Moreover, both control variables have a significant impact on offline sales. Household type and income show a significant positive effect on offline sales (respectively 0.0170, p<0.01; 0.0013, p<0.01). The dummy that accounts for the sales weeks is also significant (𝛽1= 0.0455,
24 6.2.2 Online sales
The parameter estimates of the different advertising effects on online sales are shown in table 4.
Variable Beta Std.Error t-statistic p-value
Intercept 𝛼 -6.766 0.0198 -341.812 <2e-16*** Promotion dummy 𝛽1 0.002315 0.001891 1.224 0.2209 Household type 𝛽2 0.003113 0.001038 3.000 0.0027*** Income 𝛽3 -0.000030 0.000082 -0.367 0.7136 ST TV 𝛽4𝑠𝑡 -0.000053 0.002897 -0.018 0.9853 ST Print 𝛽5𝑠𝑡 0.000083 0.001728 0.048 0.9616 ST GDN 𝛽6𝑠𝑡 -0.000305 0.000363 -0.840 0.4011 ST MAST 𝛽7𝑠𝑡 -0.000387 0.001336 -0.290 0.7721 ST TV * GDN 𝛽8𝑠𝑡 0.000004 0.000073 0.060 0.9520 ST TV * MAST 𝛽9𝑠𝑡 0.000014 0.000419 0.032 0.9741 ST Print * GDN 𝛽10𝑠𝑡 -0.000038 0.000073 -0.525 0.5995 ST Print * MAST 𝛽11𝑠𝑡 0.000034 0.000248 0.139 0.8897 LT TV 𝛽4𝑙𝑡 -0.000264 0.004828 -0.055 0.9565 LT Print 𝛽5𝑙𝑡 -0.000194 0.003381 -0.057 0.9543 LT GDN 𝛽6𝑙𝑡 -0.000535 0.002292 -0.233 0.8154 LT MAST 𝛽7𝑙𝑡 -0.000715 0.002970 -0.241 0.8098 LT TV * GDN 𝛽8𝑙𝑡 -0.000047 0.002238 -0.021 0.9833 LT TV * MAST 𝛽9𝑙𝑡 0.000065 0.002319 0.028 0.9777 LT Print * GDN 𝛽10𝑙𝑡 -0.000088 0.002237 -0.040 0.9685 LT Print * MAST 𝛽11𝑙𝑡 0.000077 0.002265 0.034 0.9729 Adjustment π -0.9784 0.00214 -457.28 <2e-16*** p < 0.10 *, p < 0.05 **, p<0.01 *** Table 4: Online sales parameter estimates
25 (π = −0.9784, p < 0.05). This carry-over sales from period t-1, is highly significant since there are few online purchases in the data. Therefore, in most of the observations a 0 in week t-1 is followed by a 0 in week t and therefore it makes sense that this variable is highly significant. The lack of sales could also be an explanation why all of the advertising variables are insignificant. Since there are 54 online sales, it might be that those sales are more at random instead of being influenced by the different advertising efforts.
7. Conclusion
7.1 Summary
Due to the possibility for companies to track customers on individual level, new and more detailed information becomes available. No research has yet used this type of individual level data to assess the combined effectiveness of three different types of advertising effects combined, namely: main effects, cross synergy effects and cross channel effects. This study aimed to provide an overview of the effectiveness of these three advertising effects using weekly household level data of a Dutch retail organization the first 26 weeks of 2011. Results show the existence of all three types of advertising effects on offline sales. However, contrary to the hypotheses, advertising could also have a negative effect on sales. Moreover, none of the advertising effects had a significant influence on online sales.
7.2 Discussion and managerial implications 7.2.1 Offline sales
26 influence. Rotfeld (2006) states that the future of advertising is hiding the sales message in the ad and/or adjust the advertisement’s content to the content of the advertising medium that is being used. He found that this form of advertising increased the volume of sales for various companies. For ads in a newspapers (printed media), it might be easier to integrate an ad into the content of the newspaper in comparison to TV advertising. This could be an explanation why consumers get less bored and tedious when seeing an ad in a newspaper.
Furthermore, no long-term effect of printed advertising has been found. This is in line with Lodish et al. (1995) who found that, for well-established brands, the saturation effect of advertising is reached very quickly. Since the Dutch retail store of which the data has been used is a well-established firm it could be that this in the reason why there is only one long-term advertising effect (TV advertising). Furthermore, the number of sales in comparison with the total sales data collected is very small. This could also be a reason why only one long-term advertising effect has been found.
Next to the main effects, the short-term cross channel effect of Google masthead banners is in line with the hypothesis 4. However, there is no significant relation between the GDN banners and offline sales. Drèze and Hussherr (2003) found that the size of a banner is an important driver of banner visibility which in turn leads to banner effectiveness. This might explain why Google masthead banners have an influence on offline sales and GDN banners not, since Google masthead banners are in general larger than GDN banners.
Furthermore, the negative short-term cross synergy effect of TV advertising and Google masthead banners is inconsistent with hypothesis 5. Rotfeld (2006) state that when consumers are exposed to a lot of advertising it will create a feeling of overwhelming mass media spam which has a negative influence on sales. Therefore, in order to make cross synergy effects between two types of advertising work, one should carefully consider the amount of exposures to consumers to avoid having negative consequences.
27 7.2.2 Online sales
This study found no significant relation between any type of offline and online advertising exposures on online sales. However, it is unlikely that this is because none of the relations exist. Many researchers have found significant impact of the different advertising channels on online sales (e.g. Dinner et al. 2014; Nottorf, 2014). An alternative explanation would be that due to the weekly household level data, combined with the low amount of online sales (54) makes that it hard to find significant relations between the different types of advertising and these 54 online sales. Wedel & Kannan (2016) mention this problem in their paper, stating that although this new kind of data potentially can provide companies with very detailed and specific information, many models that are being used in marketing literature cannot handle these large volumes of data. They introduce the bias-variance trade-off as an explanation for the decision on the aggregation level that needs to be used as input for an analysis. This bias-variance trade-off says that more aggregated level data is an inconsistent representation of the real world, which will result in biased results. However, large amounts of data reduces the variance in the data. Models that analyze these large amount of data tend to over-fit, so that they capture random errors instead of tracking the variance that can be explained by the variables in the model. Therefore, to fully capture the information in these large amount of data, more complex and advanced models need to be used (Wedel & Kannan, 2016). Because of this trade-off, Wedel and Kannan (2016) argue that the data and the model that are being used should be aligned with the research objective in order to get the desirable results. Since the dataset of this study contained 54 online sales of the in total 248705 sales data points that have been collected, one can argue that more aggregated level data could have been used to determine the different advertising effects on online sales. This might be the reason that no significant relation has been found with the different advertising exposures.
This has also important implications for managers. Managers should not use detailed level data with the assumption that it will give them more detailed information. There are situations, mostly with descriptive and diagnostic analysis, that there is no need for this detailed information. Therefore, managers should carefully align the research objective with the data that is being used to be able to develop a useful model (Wedel & Kannan, 2016).
28 of 26 weeks. This might have had an impact on the results, since only few products per household were sold during the time period of this study. Therefore, managers should next to the objective of their analysis also take the setting of the research into account when making a decision about the level of detailed information they need.
7.3 Limitations and future research
29 8. References
Anderl, E., Schumann, J. H., and Kunz, W. (2016). Helping firms reduce complexity in multichannel online data: A new taxonomy-based approach for customer journeys. Journal of Retailing, 92(2), 185-203.
Assmus, G., Farley, J.U and Lehmann, D.R, (1984). How Advertising Affects Sales: Meta-Analysis of Econometric Results. Journal of Marketing Research, 21(2), 650-674. Ataman, M.B., Van Heerde, H.J. and Mela, C.F. (2010). The Long-Term Effect of Marketing
Strategy on Brand Sales. Journal of Marketing Research, 47(5), 866-882.
Bollinger, B., Cohen, M., & Jiang, L. (2013). Measuring asymmetric persistence and interaction effects of media exposures across platforms. University of British Columbia.
Braun, M. and Moe, W.W. (2013). Online Display Advertising: Modeling the Effects of Multiple Creatives and Individual Impression Histories. Marketing Science, 32(5), 753-767.
Chatterjee, P. (2008). Are unclicked ads wasted? Enduring effects of banner and pop-up ad exposures on brand memory and attitudes. Journal of Electronic Commerce Research, 9(1), 51-61.
Chatterjee, P., Hoffman, D.L. and Novak, T.P. (2003). Modeling the Clickstream: Implications for Web-Based Advertising Efforts. Marketing Science, 22(4), 520-541.
Cho, C.H. and Cheon, H.J. (2004). Why do people avoid advertising on the internet? Journal
of Advertising, 33(4), 89-97.
Chang, Y. and Thorson, E. (2004). Television and Web Advertising Synergies. Journal of
Advertising, 33(2), 75-84.
Danaher, P.J. and Dagger, T.S. (2013). Comparing the Relative Effectiveness of Advertising Channels: A Case Study of a Multimedia Blitz Campaign. Journal of Marketing
Research, 50(4), 517-534.
Dekimpe, M.G. and Hanssens, D.M. (2007). Advertising response models. In Teller, G.J. and Ambler T. (Eds), Handbook of advertising (pp. 247-263). Londen: Sage.
Dinner, I.M., Van Heerde, H.J. and Neslin, S.A. (2014), Driving Online and Offline Sales: The Cross-Channel Effects of Traditional, Online Display, and Paid Search Advertising,
Journal of Marketing Research, 51(5), 527-545.
Drèze, X. and Hussherr, F.X. (2003). Internet Advertising: Is Anybody Watching? Journal of
30 Ebbes, P., Papies, D.. and van Heerde, H.J. (2011). The Sense and Non-Sence of Holdout Sample Validation in the Presence of Endogeneity. Marketing Science, 30(6), 1115-1122.
Edwards, S.M., Li, H. and Lee, J.H. (2002). Forced exposure and psychological reactance: antecedents and consequences of the perceived intrusiveness of pop-up ads. Journal of
Advertising, 31(3), 83-95.
Fok, D., Horváth, C., Paap, R. and Franses, P.H. (2006). A Hierarchical Bayes Error Correction Model to Explain the Dynamic Effects of Price Changes. Journal of Marketing
Research, 43(3), 443-461.
Goldfarb, A. and Tucker, C (2011). Online display advertising: Targeting and obtrusiveness.
Marketing Science, 30(3), 389-404.
Gijsenberg, M.J. (2014). Going for gold: Investigating the (non)sense of increased advertising around major sports events. International Journal of Research in Marketing, 31(1), 2-15.
Google. (n.d.) YouTube Masthead. Retrieved June 2017, from Doubleclick by Google: https://support.google.com/richmedia/answer/2560792?hl=en.
Greene, W.H. (2003). Econometric analysis (5th ed.) Englewood Cliffs, NJ: Prentice Hall. Hausman, J.A. (1978). Specification Tests in Econometrics, Econometrica, 46(6), 1251-1271. Hoban, P.R. and Bucklin, R.E. (2015). Effects of Internet Display Advertising in the Purchase
Funnel: Model-Based Insights from a Randomized Field Experiment. Journal of
Marketing Research, 52(3), 375-393.
Kahneman, D. (1973). Attention and effort (Vol. 1063). Englewood Cliffs, NJ: Prentice-Hall. Lambrecht, A. and Tucker, C. (2013). When Does Retargeting Works? Information Specificity
in Online Advertising. Journal of Marketing Research, 50(5), 561-576.
Levin, A., Lin, C.F. and Chu, C.S.J. (2002). Unit root tests in panel data: asymptotic and finite-sample properties. Journal of Econometrics, 180(1), 1-14.
Lodish, L.M., Abraham, M., Kalmenson, S., Livelsberger, J., Lubertkin, B. Richardson, B., Stevens, M.E. (1995). How T.V. Advertising Works: A Meta-Analysis of 389 Real World Split Cable T.V. Advertising Experiments. Journal of Marketing Research, 32(2), 125-139.
Manchanda, P., Dubé, J.P., Goh, K.Y. and Chintagunta, P.K. (2006). The Effect of Banner Advertising on Internet Purchasing. Journal of Marketing Research, 41(1), 98-108. Montgomery, A.L., Li, S., Srinivasan, K. and Liechty, J.C. (2004). Modeling Online Browsing
31 Naik, A. and Peters, K. (2009). A Hierarchical Marketing Communications Model of Online
and Offline Media Synergies. Journal of Interactive Marketing, 23(4), 288-299.
Naik, A. and Raman, K. (2003). Understanding the Impact of Synergy in Multimedia Communications. Journal of Marketing Research, 40(4), 375-388.
Nottorf, F. (2014). Modeling the clickstream across multiple online advertising channels using a binary logit with Bayesian mixture of normals. Electronic Commerce Research and
Applications, 13(1), 45-55.
Pauwels, K., Srinivasan, S., & Franses, P. H. (2007). When do price thresholds matter in retail categories?. Marketing Science, 26(1), 83-100.
Rethans, A.J., Swasy, J.L. and Lawrence, J.M. (1986). Effects of Television Commercial Repetition, Receiver Knowlegde, and Commercial Length: A test of the Two-Factor Model. Journal of Marketing Research, 23(1), 55-61.
Rotfeld, H.J. (2006). Understanding advertising clutter and the real solution to declining audience attention to mass media commercial messages. Journal of Consumer
Marketing, 23(4), 180-181.
Rubinson, J. (2009). Empirical evidence of TV advertising effectiveness. Journal of
Advertising Research, 49(2), 220-226.
Rutz, O.J. and Bucklin, R.E. (2012). Does banner advertising affect browsing for brands? Clickstream choice model says yes, for some. Quantitative Marketing and Economics, 10(2), 231-257.
Sethuraman, R., Tellis, G. J., and Briesch, R. A. (2011). How well does advertising work? Generalizations from meta-analysis of brand advertising elasticities. Journal of
Marketing Research, 48(3), 457-471.
Schumann, D.W. and Petty, R.E. (1990). Predicting the Effectiveness of Different Strategies of Advertising Variation: A test of the Repetition-Variation Hypothesis. Journal of
Consumer Research, 17(2), 192-202.
Stammerjohann, C., Wood, C.M., Chang, Y. and Thorson, E. (2005). An Empirical Investigation of the Interaction between Publicity, Advertising, and previous Brand Attitudes and Knowledge. Journal of Advertising, 24(4), 55-67.
Srinivasan, S., Rutz, J.O. and Pauwels, K. (2016). Paths to and off purchase: quantifying the impact of traditional and online consumer activity. Journal of the Academy of Marketing
Science, 44(4), 440-453.
32 Tellis, G. J. (2009). Generalizations about advertising effectiveness in markets. Journal of
advertising research, 49(2), 240-245.
Verhoef, P. C., and Leeflang, P. S. (2009). Understanding the marketing department's influence within the firm. Journal of Marketing, 73(2), 14-37.
Verhoef, P.C., Kannan, P.K. and Inman, J.J. (2015). From Multi-Channel Retailing to Omni-Channel Retailing: Introduction to the Special Issue on Multi-Omni-Channel Retailing.
Journal of Retailing, 19(2), 174-181.
Wedel, M. and Kannan, P.K. (2106). Marketing Analytics for Data-Rich Environments.
Journal of Marketing, 80(6), 97-121.
Wiesel, T., Pauwels, K. and Arts, J. (2011), Marketing’s Profit Impact Quantifying Online and Off-line Funnel Progression, Marketing Science, 30(4), 604-611.
Winer, S. (2009). New Communications Approaches in Marketing: Issues and Research Directions. Journal of Interactive Marketing, 23(2), 108-117.
33 9. Appendix
9.1 Appendix A: Estimation offline sales
Call:
lm(formula = deltasalesoffline ~ deltatv + deltaprint + deltaGDN + deltagoogle_masthead + deltainttvgdn + deltainttvmasthead + deltaintprintgdn + deltaintprintmasthead + laggedsalesoffline + laggedtv + laggedprint + laggedGDN + laggedgoogle_masthead + laggedinttvgdn + laggedinttvmasthead + laggedintprintgdn + laggedintprintmasthead + promotiondummy + inkomHH + HHtype, data = MMdata_analyse)
Residuals:
Min 1Q Median 3Q Max -0.5354 -0.0806 -0.0677 -0.0567 19.6433 Coefficients:
Estimate Std. Error t value Pr(>|t|) (Intercept) -6.626e+00 6.094e-02 -108.730 < 2e-16 *** deltatv -3.751e-02 1.313e-02 -2.857 0.004278 ** deltaprint 1.437e-02 7.833e-03 1.834 0.066616 . deltaGDN 2.122e-03 1.646e-03 1.289 0.197377 deltagoogle_masthead 1.246e-02 6.055e-03 2.057 0.039665 * deltainttvgdn -1.573e-05 3.312e-04 -0.048 0.962110 deltainttvmasthead -5.466e-03 1.900e-03 -2.876 0.004025 ** deltaintprintgdn 1.199e-04 3.276e-04 0.366 0.714361 deltaintprintmasthead 1.828e-03 1.124e-03 1.626 0.103897 laggedsalesoffline -9.789e-01 2.027e-03 -482.845 < 2e-16 *** laggedtv -3.456e-02 1.858e-02 -1.860 0.062899 . laggedprint 1.364e-02 1.102e-02 1.238 0.215565 laggedGDN 2.854e-03 2.235e-03 1.277 0.201660 laggedgoogle_masthead 1.095e-02 8.569e-03 1.278 0.201423 laggedinttvgdn 3.574e-04 4.907e-04 0.728 0.466357 laggedinttvmasthead -5.417e-03 2.689e-03 -2.015 0.043917 * laggedintprintgdn 3.185e-04 4.494e-04 0.709 0.478509 laggedintprintmasthead 1.891e-03 1.592e-03 1.188 0.234919 promotiondummy1 4.546e-02 8.573e-03 5.303 1.14e-07 *** inkomHH 1.349e-03 3.731e-04 3.616 0.000299 *** HHtype1 1.703e-02 4.703e-03 3.621 0.000294 *** ---
34 9.2 Appendix B: Estimation online sales
Call:
lm(formula = deltasalesonline ~ deltatv + deltaprint + deltaGDN + deltagoogle_masthead + deltainttvgdn + deltainttvmasthead + deltaintprintgdn + deltaintprintmasthead + laggedsalesonline + laggedtv + laggedprint + laggedGDN + laggedgoogle_masthead + laggedinttvgdn + laggedinttvmasthead + laggedintprintgdn + laggedintprintmasthead + promotiondummy + inkomHH + HHtype, data = MMdata_analyse)
Residuals:
Min 1Q Median 3Q Max -0.3915 -0.0050 -0.0035 -0.0017 18.2381 Coefficients:
Estimate Std. Error t value Pr(>|t|) (Intercept) -6.766e+00 1.977e-02 -342.314 <2e-16 *** deltatv -5.330e-05 2.897e-03 -0.018 0.9853 deltaprint 8.315e-05 1.728e-03 0.048 0.9616 deltaGDN -3.050e-04 3.632e-04 -0.840 0.4011 deltagoogle_masthead -3.869e-04 1.336e-03 -0.290 0.7721 deltainttvgdn 4.399e-06 7.307e-05 0.060 0.9520 deltainttvmasthead 1.361e-05 4.193e-04 0.032 0.9741 deltaintprintgdn -3.794e-05 7.227e-05 -0.525 0.5995 deltaintprintmasthead 3.441e-05 2.480e-04 0.139 0.8897 laggedsalesonline -9.784e-01 2.140e-03 -457.276 <2e-16 *** laggedtv -2.580e-04 4.099e-03 -0.063 0.9498 laggedprint -1.894e-04 2.430e-03 -0.078 0.9379 laggedGDN -5.236e-04 4.931e-04 -1.062 0.2883 laggedgoogle_masthead -6.995e-04 1.890e-03 -0.370 0.7114 laggedinttvgdn -4.574e-05 1.083e-04 -0.422 0.6727 laggedinttvmasthead 6.344e-05 5.932e-04 0.107 0.9148 laggedintprintgdn -8.646e-05 9.915e-05 -0.872 0.3832 laggedintprintmasthead 7.489e-05 3.513e-04 0.213 0.8312 promotiondummy1 2.315e-03 1.891e-03 1.224 0.2209 inkomHH -3.021e-05 8.231e-05 -0.367 0.7136 HHtype1 3.113e-03 1.038e-03 3.000 0.0027 ** ---
9.3 Appendix C: variance-covariance matrix offline sales (1)
(Intercept) deltatv deltaprint deltaGDN deltagoogle_masthead deltainttvgdn deltainttvmasthead deltaintprintgdn deltaintprintmasthead laggedsalesoffline laggedtv
36 9.3 Appendix C: variance-covariance matrix offline sales (2)
laggedprint laggedGDN laggedgoogle_masthead laggedinttvgdn laggedinttvmasthead laggedintprintgdn laggedintprintmasthead promotiondummy1 inkomHH HHtype1
(Intercept) 3,67E+02 1,76E+01 4,91E+02 3,13E-01 -6,74E+01 2,19E+00 5,14E+01 9,97E+00 -1,29E+00 -1,52E+01
deltatv -1,49E+01 1,13E-01 -3,35E+01 4,54E-01 2,46E+01 -1,42E-01 -2,03E+00 -2,51E-01 -3,11E-03 -2,35E-01
deltaprint 6,07E+01 8,65E-01 2,58E+01 -1,54E-01 -2,00E+00 3,32E-01 8,49E+00 -1,04E+00 -1,96E-02 -3,96E-02
deltaGDN 1,13E+00 2,49E+00 -8,96E-01 2,18E-02 -6,96E-03 3,09E-01 -1,24E-01 3,70E-01 2,55E-03 -2,40E-01
deltagoogle_masthead 2,55E+01 -1,12E+00 3,67E+01 4,34E-02 -4,87E+00 -1,48E-01 3,86E+00 3,70E-01 1,54E-02 -5,53E-01
deltainttvgdn -1,78E-01 5,21E-02 -2,79E-02 1,19E-01 -4,77E-02 -3,82E-02 1,01E-02 -1,45E-01 5,25E-04 -7,92E-03
deltainttvmasthead -2,02E+00 -3,30E-02 -4,82E+00 -4,51E-02 3,61E+00 1,38E-02 -3,07E-01 -8,43E-02 -1,20E-03 -3,18E-02
deltaintprintgdn 4,12E-01 2,96E-01 -1,08E-01 -3,72E-02 1,24E-02 1,01E-01 -3,46E-02 1,18E-02 -6,13E-04 -7,22E-03
deltaintprintmasthead 8,42E+00 -1,38E-01 3,85E+00 1,10E-02 -3,08E-01 -4,38E-02 1,27E+00 -1,64E-04 -1,78E-03 -2,92E-03
laggedsalesoffline -9,24E-02 -1,25E-02 -1,06E-01 -4,96E-04 7,88E-03 -1,41E-03 -1,27E-02 -5,51E-02 -4,62E-03 -5,87E-02
laggedtv -2,98E+01 1,62E-01 -6,67E+01 8,85E-01 4,92E+01 -2,80E-01 -4,05E+00 5,95E-01 -6,38E-03 -4,74E-01
laggedprint 1,21E+02 2,06E+00 5,13E+01 -3,32E-01 -4,04E+00 7,54E-01 1,69E+01 3,04E-01 -3,95E-02 -5,94E-02
laggedGDN 2,06E+00 5,00E+00 -2,07E+00 7,66E-02 -4,52E-02 6,02E-01 -2,63E-01 2,10E-01 5,58E-03 -4,77E-01
laggedgoogle_masthead 5,13E+01 -2,07E+00 7,34E+01 2,16E-02 -9,69E+00 -2,62E-01 7,71E+00 1,28E+00 3,01E-02 -1,13E+00
laggedinttvgdn -3,32E-01 7,66E-02 2,16E-02 2,41E-01 -9,55E-02 -7,33E-02 2,10E-02 1,21E-01 1,07E-03 -1,22E-02
laggedinttvmasthead -4,04E+00 -4,52E-02 -9,69E+00 -9,55E-02 7,23E+00 2,53E-02 -6,15E-01 -2,59E-01 -2,44E-03 -6,73E-02
laggedintprintgdn 7,54E-01 6,02E-01 -2,62E-01 -7,33E-02 2,53E-02 2,02E-01 -7,95E-02 1,15E-01 -1,15E-03 -1,64E-02
laggedintprintmasthead 1,69E+01 -2,63E-01 7,71E+00 2,10E-02 -6,15E-01 -7,95E-02 2,54E+00 -7,70E-02 -3,66E-03 -5,00E-03
promotiondummy1 3,04E-01 2,10E-01 1,28E+00 1,21E-01 -2,59E-01 1,15E-01 -7,70E-02 7,35E+01 -1,94E-03 1,74E-01
inkomHH -3,95E-02 5,58E-03 3,01E-02 1,07E-03 -2,44E-03 -1,15E-03 -3,66E-03 -1,94E-03 1,39E-01 -2,91E-01
37 9.4 Appendix D: variance-covariance matrix online sales (1)
(Intercept) deltatv deltaprint deltaGDN deltagoogle_masthead deltainttvgdn deltainttvmasthead deltaintprintgdn deltaintprintmasthead laggedsalesonline laggedtv
38 9.4 Appendix D: variance-covariance matrix online sales (2)
laggedprint laggedGDN laggedgoogle_masthead laggedinttvgdn laggedinttvmasthead laggedintprintgdn laggedintprintmasthead promotiondummy1 inkomHH HHtype1
