Analysis the effect of online and offline advertising on consumer
purchase behavior in the Dutch electronic market
by
Hanying Zhou
Analysis the effect of online and offline advertising on consumer
purchase behavior in the Dutch electronic market
by
Hanying Zhou
University of Groningen
Management Summary
Preface
This thesis has been written to fulfill the graduation requirements of the Marketing Intelligence master program at the Groningen University of Faculty of Economics and Business. This thesis offers a very brief view of Dutch electronics market. The research question was formulated together with my supervisor, Edwin Kooge. The research was difficult, but data that provided by Dutch electronic markets has allowed me to answer the question that we identified.
I would like to thank my supervisor Edwin Kooge for his excellent guidance and support during the process. Without his encouragement, advices and motivation, this thesis would not be completed successfully.
I would like to thank my parents and my best friend Wei Zhou for being helpful and supportive during my time studying.
I also would like to thank my boyfriend Bochao Deng and my lovely dog Shidan for being always on my side.
Table of Contents
1INTRODUCTION ... 6
2THEORETICALFRAMEWORK ... 10
2.1 The Customer Buying Decision Process ... 10
2.2 Advertisement ... 10 2.3 Conceptual model ... 14 3METHODOLOGY ... 15 3.1 Sample ... 15 3.2 Data manipulation ... 15 3.3 Variables ... 16 3.4 Cluster analysis ... 19 3.5 Model specification ... 19 4RESULTS ... 22 4.1 Dataset preparation ... 22 4.2 Consumer segmentation ... 22 4.3 Analysis results ... 25
5CONCLUSIONANDRECOMMENDATIONS ... 30
5.1 Conclusions and Discussion ... 30
5.2 Implications ... 32
5.3 Limitations ... 32
5.3 Recommendations for future research ... 33
REFERENCES ... 34
APPENDICES ... 39
Appendix A Consumer demographic variables explanation ... 39
Appendix B Consumer demographic variables frequencies ... 40
Appendix C Consumer demographic variables description ... 41
Appendix D Online, offline and cross-media advertising variables boxplot ... 42
Appendix E K-mean cluster analysis: Iteration history change ... 44
Appendix F K-mean cluster analysis: final cluster centers ... 45
Appendix G K-mean cluster analysis: relative sizes of cluster solutions ... 46
Appendix H ANOVA test results ... 47
Appendix I Most important variables for each segment ... 48
1 INTRODUCTION
Since the advent of the Internet, every aspect of our lives in this modern era has been constantly changing in terms of advertising, communication, commercial activities. For example, by 2014, only ten years after it began, the social media website Facebook counted 1.3 billion active users. And, there are an additional 860 million login users who use Facebook to communicate every day (Zephoria,2014). To add to these remarkable figures, the search engine Google records 3.5 billion worldwide searches per day (Internet Live Stats, 2017). The influence of the Internet is reflected in the increasing marketing budgets of many companies: spending on online advertising is expected to reach $129.23 billion by 2021 (eMarketer, 2017). Even though spending on TV advertising is still growing in the US, spending on digital advertising surpassed TV advertising for the first time at the end of 2016 (eMarketer, 2016). Furthermore, global spending on newspaper print advertising has decreased by 8.7% to 52.6 billion in 2016 (Statista,2017). Some people therefore conclude that traditional advertising can be ignored as people perceive little value in it nowadays. Nevertheless, traditional advertising has not fallen. Take the time people spend watching television as an example. Americans still spend five hours watching TV per day (Nielsen, 2010): 27% of Americans search online for the information about products they saw in a TV advertisement (Nielsen, 2012). Additionally, many people still listen to the radio on the computer or through a traditional radio set; thus, expenditures on radio advertising are, accordingly, still rising (RAB, 2011). This increase assumes that traditional advertising still plays an important role in generating awareness, knowledge and interest in products.
exponential rate. Nowadays, many companies generally adopt both online and traditional (offline) advertisement in various forms to promote their products (Dekimpe & Hanssens, 2007). In the e-commerce era, it has become common practice for consumers to apply cross-media information checking (Dinner et al. 2011, Dinner et al, 2014); for example, one consumer can browse for product information online but experience the product in a nearby retail shop. Alternatively, the consumer checks the promotions online and later makes the purchase based on product information that he or she received from TV or newspapers.
As a consequence of the rapid development of e-commerce and online advertising in the Netherlands, it is of both academic and practical significance to understand the cross-channel effects of modern advertising that are exerted on consumers’ purchasing behavior. In spite of the fact that there are diverse kinds of media emerging through these cross-channel effects, not all media are equally effective. Therefore, using different media at the same time would most likely generate a synergistic effect. Synergy in media arises when the combined effect of a number of media activities has a greater impact on sales than a single media effect (Schultz et al, 2012). One way to achieve this is cross-media advertising, which is to combine traditional advertising with the interactivity and service capabilities of online communication (Shimp & Andrews, 2012). As Ernst & Young foresee (2015), cross-channel purchasing will be more and more popular among customers; consequently, the boundary between, and concept of, online and offline retailing are becoming increasingly vague. As a result, it is necessary for companies to develop cross-media strategies in order to adapt to changing consumer behaviors. Naik and Peters (2009) have examined a German car company and found that cross-media between offline and online media-groups promotes sales. Furthermore, based on Wisesl et al.’s research (2011), online and offline cross-media affect the path of consumer purchase in the business to business (B2B) domain. Chang and Thorson (2004) have explored cross-media conditions, such as display and Internet advertising followed by TV advertising, and found that individuals pay more attention to this synergetic condition.
on individual reactions to specific advertising forms in isolation. These studies thereby offer little guidance either on the general performance or the cross-media effects of advertising on individual purchase and general sales. Current literature focuses either on one particular form of advertising (Manchanda et al, 2006; Agarwal et al, 2011; Kim & Ko’ 2012; Braun & Moe, 2013; Skiera et al, 2013) or on one product category at a time (e.g. Li & Kannan, 2014, Pauwels et al, 2013). In some cases, research on multi-media advertising tends to omit offline marketing (Li and Kannan, 2014).
The aim of this study therefore is to investigate the impact of cross-media on consumer purchase behavior with the focus on one unique market: the Dutch electronics market. The sales turnover of electronics retailers in the Netherlands has grown at an exponential rate since 2005. The report by Statista (2015) showed that electronics retailer sales increased by 2.2% in 2014 alone. The revenues generated in wholesale consumer electronics in the Dutch market reached US$9.7 billion in 2014 (Statista, 2015). Despite these figures, the volume sales of electronic products in the Dutch market began to decline in 2015 and continued to fall in 2016 (Euromonitor, 2016). This fall in sales was surprising because the overall disposable income increased in 2016 with the result that consumers were more willing to purchase electronic products than in 2015 (Euromonitor, 2016). The rise in disposable income, however, was still not sufficient to boost the sales volume in the Dutch electronics market. Hence, this thesis not only includes characterization and segmentation consumers in the Dutch electronics market, but it also investigates whether cross-media marketing activities are able to affect the purchase behavior of consumers and consumer segments. If this is the case, it is important to ascertain which combination of cross-media marketing activities might have the greatest influence on the purchase behavior of different consumer segments in the Dutch electronics market.
Therefore, the central research question for this thesis is:
How do cross-media marketing activities affect the purchase behaviors of different consumer segments in the Dutch electronics market?
electronics market?
2. How do advertising activities affect consumer purchase behavior in the electronics market?
3. What are the cross-media effects of online and offline advertising on consumer purchase behavior in the electronics market?
4. How do consumer segments moderate the effects of online and offline advertising on consumer purchase behavior in the electronics market?
5. How do consumer segments moderate the cross-media effects between online and offline advertising in the electronic market?
2 THEORETICAL FRAMEWORK
2.1 The Customer Buying Decision Process
Based on the theory of customer buying decision process, a full purchase cycle for a consumer usually includes five steps of a decision-making process: need recognition, information search, evaluation of alternatives, purchase decision, and post-purchase behavior (Kotler & Armstrong, 2010). Consumers make such decisions under the constant impact of advertising and these decisions are moderated by the fundamental influence of the consumers’ own characteristics. Advertisements are external stimulators while consumer characteristics are internal moderators, both of which work together and drive the the decision-making process when purchasing a commodity (Solomon et al. 2012).
2.2 Advertisement
2.2.1 Online and offline advertisement
Brand building advertising in traditional media, such as TV advertising, is crucial for the generating of awareness, knowledge and interest in products. In addition, TV advertising has a significant impact on increasing short-term sales (Danaher & Dagger, 2013). Another important traditional offline advertising is print advertising, which refers to physically printed media, such as magazines, newspapers and leaflets, that can reach some consumers. Print advertising is used to guide consumers to search further for more information about the products or brands. The characteristics of offline advertising, outlined above, leads to the first hypothesis proposed in this thesis:
H1: Offline advertising has a positive relationship with the consumer’s actual purchase
behavior.
Because only two forms of offline advertising are taken into account in this thesis, TV and print advertising, H1 is broken into two sub-hypotheses:
H1a: TV advertising has a positive relationship with a consumer’s actual purchase behavior. H1b: Printing advertising has a positive relationship with a consumer’s actual purchase
behavior.
has a positive effect on repeating purchase probabilities. Masthead advertising is mainly used in online video websites with customizable size and contents. In this regard, Google owns one of the largest paid e-marketplaces: Google online display network. Google charges the advertisers to run advertisements on the network of websites and thereby provide reports about the performance metric of advertisements on the activities of advertisers. Both masthead advertising and Google display networks are forms of display advertising (Digital Analytics, 2013). Based on Goldfarb & Tucker (2011), a matching and obtrusive advertising display increases consumer purchase intention. Furthermore, according to Bart et al. (2014), a mobile display advertising campaign has a significant impact on consumers' attitudes and purchase intentions. As forms of display advertising, we can therefore expect that masthead and the Google display network also have positive relationship with consumer purchase intention. The characteristics of online advertising leads to the second hypothesis proposed in this thesis:
H2: Online advertising has a positive relationship with a consumer’s actual purchase
behavior.
Three forms of online advertising are therefore taken into account in this thesis: banner advertising, masthead advertising and the Google display network. Thus, this hypothesis composed of three sub-hypotheses:
H2a: Banner advertising has a positive relationship with a consumer’s actual purchase
behavior.
H2b: Masthead advertising has a positive relationship with a consumer’s actual purchase
behavior.
H2c: The Google display network has a positive relationship with a consumer’s actual
purchase behavior.
2.2.2 Cross-media advertisement
cross-channel effect of advertisement, research findings suggest that the use of cross-media integration strengthens the perceived media engagement of advertising messages (Tang et al, 2007); subsequently, multi-channel communication tends to enhance the advertising message credibility. Greater advertising message credibility results in more positive thoughts through a heuristic thinking process (Maclnnis & Jaworski, 1989). Combined with advertising credibility, positive thoughts induce more positive brand credibility and increase the purchase intention (Ajzen & Fishbein, 1975; Lim et al, 2015). Chang and Thorson (2004) examined cross-media conditions, conditions involved in the display of an Internet advertisement followed by TV advertising, and found that individuals pay more attention to this synergetic condition.
The following hypotheses imply that a combination of online advertising and offline advertising may have a greater impact on consumer purchase. The cross-media advertising effect in the following hypotheses denotes the interactive effect of two certain offline and online advertising channels. Since there are two forms of advertising for offline advertising and there are three forms of advertising for online advertising, this leads to six sub-hypotheses:
H3a: The cross-media advertising effect of banner advertising and TV advertising has a
positive relationship with a consumer’s actual purchase behavior.
H3b: The cross-media advertising effect of banner advertising and print advertising has a
positive relationship with a consumer’s actual purchase behavior.
H3c: The cross-media advertising effect of masthead advertising and TV advertising has a
positive relationship with a consumer’s actual purchase behavior.
H3d: The cross-media advertising effect of masthead advertising and print advertising has a
positive relationship with a consumer’s actual purchase behavior.
H3e: The cross-media advertising effect of Google display advertising and TV advertising has
a positive relationship with a consumer’s actual purchase behavior.
H3f: The cross-media advertising effect of Google display advertising and print advertising
2.2.3 Consumer segments
Segmenting is often useful in predicting how likely it is for a group of people to purchase a specific kind of product or service. Additionally, segmenting typically includes demographic, psychological behavioristic and geographic characteristics which cover cultural, social, personal and psychological aspects (Park and Kim 2003). Backman (1994) stated that segmenting heterogeneous customers into homogeneous sub-groups was based on the similar characteristics of consumers, which allows marketers to identify the differences between customer groups. In light of these points, market segmentation can be informative for marketing managers to better understand the market and further develop appropriate advertising for various distinct target segments. In addition, in this thesis, Dutch electronics market consumers are classified according to consumers’ demographic information.
H4: For different consumer segments, there are differences between the effects of online
advertising, offline advertising and cross-media advertising on a consumer’s actual purchase behavior.
2.3 Conceptual model
3 METHODOLOGY
3.1 Sample
For analyzing the consumer purchase behavior in the Dutch electronics market, a household level dataset from the Dutch electronics market are obtained. The frequency of the data set is at the weekly level and spans from 2010 (week 48) to 2011 (week 26), with a total of 361,832 observations. The main data include four sub-domains and cover consumer characteristics, online advertising data, offline advertising data and consumer purchasing behavior. With respect to consumer characteristics, the first sub-domain data incorporates the following variables: housewife age, income level, living distinct, number of children, lifecycle and household type (Appendix A-C). The second sub-domain consists of online advertising data, including banner advertising, masthead advertising and the Google display network. The value of these variables demonstrates the actual online advertising contact number that an individual has. The third sub-domain of offline advertising data includes TV advertising, print advertising, flyers and radio advertising. The value of the variables here indicates the individual’s offline channel exposure amount. Following from this, consumer purchase behavior domain consists of four variables: consumer internet purchase intention, consumer actual purchase behavior, consumer purchase behavior involving competitors and consumer purchase spending. The data provider has coded the first three consumer purchase behavior variables as dummy variables in the dataset in order to manifest consumer purchase intention and actual behavior. In this case, only consumer actual purchase behavior is used as a dependent variable to investigate consumer purchase behavior in the Dutch electronics market.
3.2 Data manipulation
2011). For each advertising variable, we consequently obtain an average media exposure value for eight months. In other words, one advertising variable is split into eight variables, expressing the average media exposure value per month. Furthermore, we combine consumers’ actual purchase behavior with a time indicator to create a new variable: buymonth. This variable indicates in which month households purchased the electronic product.
Secondly, missing values and outliers are examined. Missing values come into existence when data and response are not available in the dataset and might result in an inaccurate analysis of the data (Malhotra, 2008). Based on information given by the data provider, there is a variable called weighting factor in the dataset, showing whether the dataset has a missing value. The weighting factor indicates whether the survey respondents have participated in the research. Therefore, the weighting factor from the dataset is taken into account when searching for missing values. If the value of the weighting factor is 0, this means that the survey did not receive a response from participants and those values have been removed. Furthermore, the outliers of each variable are checked. To find out whether there are outliers in the dataset, boxplots are conducted for all variables (Appendix D). Thirdly, since the housewives in the household who are younger than 18 and older than 65 have less purchasing power, they are also not taken into account. Lastly, each household’s online and offline media effects after the last purchase are not accounted for, since consumer behavior resulting from those media effects cannot be predicted.
3.3 Variables
2010 – July 2011), 562 (7.2%) households made a single purchase in the Dutch electronics market. Therefore, 7275 (92.8%) households from the observations are censored. However, those censored observations are included to avoid selection bias.
Independent variables. Because the effect of an advertisement is not only to influence consumer purchase during the advertising period, but also in the future period. Therefore, we take the marketing dynamics situation into account when deciding independent variables. One type of marketing dynamic is lagged advertising effect. Lagged advertising effect arises from the delay between a consumer receiving the advertising message and a purchase being made (Leeflang et al, 2015). The advertising adstock model can be used to explain advertising lagged effect (Broadbent, 1979). A basic adstock model can be described as follows:
𝐴"= 𝑇"+ 𝜆𝐴"() (𝑡 = 1, … . , 𝑛)
Where 𝐴" refers to the adstock at time t, 𝑇" endnotes the value of advertising variable at time t, and 𝜆 is the lag weight parameter. 𝐴"() refers to the value of advertising at time t-1. A decaying lag structure implies that an advertisement in period t has greatest impact on consumer in the same period. With monthly data, advertising has its peak effect after 1 month and up to 1.3 months (Montgomery & Silk, 1972; Leeflang et al, 2015). In this case, for customers who made a purchase, we include the media value of purchase month and one month prior to purchase to calculate the media exposure value. For those censored consumers, the media values of the last two observation months are used to acquire the media exposure value. In addition, the lag weight parameter is set to 0.2471 (Rufino, 2008). Therefore, five advertising channels are calculated based on the following model:
𝐴" = 𝑎"+ 0.2471𝑎"() (𝑡 = 2,3,4,5,6,7,8) 0 50 100 150 1 2 3 4 5 6 7 8
In order to create independent variables, 40 advertising channel variables are selected from the dataset: TV advertising (1–8 months), print advertising (1–8 months), banner advertising (1–8 months), masthead advertising (1–8 months), and Google display network advertising (1–8 months). The media exposure values indicate the offline and online advertising channel exposure amount that a household had during the observation period. After calculation, five online and offline advertising variables and six online-offline advertising interaction variables are created. Table 1 shows the descriptive statistics of these variables.
Table 1: Descriptive of variables
Variable Label Mean SD
Print advertising P 0.194985 0.923924
TV advertising T 0.136120 0.768876
Banner advertising B 0.001084 0.047073 Masthead advertising M 0.000868 0.020591 Google display network (GDN) G 0.149232 0.673342 Interaction variables Label Mean SD Banner * Print (Cross-media advertising of banner
advertising and print advertising)
BP 0.002543 0.151836 Banner * TV (Cross-media advertising of banner
advertising and TV advertising)
BT 0.000499 0.027873 Masthead * Print (Cross-media advertising of
masthead advertising and print advertising)
MP 0.001638 0.058162 Masthead * TV (Cross-media advertising of
masthead advertising and TV advertising)
MT 0.000960 0.023328 GDN * Print (Cross-media advertising of GDN and
print advertising)
GP 0.045008 0.661554 GDN * TV (Cross-media advertising of GDN
advertising and TV advertising)
GT 0.041538 0.764372
3.4 Cluster analysis
Cluster analysis can be sorted into two primary methods when making distinctions with clusters; namely, the priori segmentation method and the post hoc segmentation method (Bigné et al, 2008). The priori segmentation method pertains to identifying the consumer groups on the basis of the researcher’s previous knowledge of the segments: for example, segmenting the consumer groups with regards to consumer characteristics. The post hoc segmentation method pertains to segmenting the consumer with regards to behavioral variables. This research makes use of the priori segmentation method to categorize Dutch electronics market consumers. Consumer demographic characteristics of Dutch electronics market consumers is selected as active variables to group consumers. Since consumer characteristics have different measurements, all active variables are standardized by z-scores before analyzing. K-means clustering and hierarchical clustering approaches are both selected to determine the amount of cluster. In addition, hierarchical cluster analysis is used to interpret the cluster solution. The distance measure chosen is the squared Euclidean, as it is the most widely used distance measuring method and because it expands distances, which makes it more convenient to spot the cutoff point in the scree plot (Malhotra, 2008; Jurowski & Reich, 2000).
3.5 Model specification
To explore the relationship between offline channels, online channels, the combined effect of offline and online channels and consumer actual purchase decision behavior, we first implement the choice model. The basic choice model demonstrates whether consumers make certain decisions or not (e.g. purchase or no purchase). Logit and probit models are the two basic types of choice model. The logit model is often preferred over the probit due to mathematical convenience and its relative easy in interpreting the parameters (Fok, 2017). The dependent variable consumer actual purchase decision behavior (Y<) has two values, with 0 being no purchase and 1 being purchase:
Consequently, the dependent variable is binomial, which makes the binary choice model a better fit for this study. The binary choice model refers to the basics of the logistic regression model for a binary dependent variable. The basic choice model is as follows (Leeflang at al, 2015):
𝑌< = 𝛼 + 𝛽)𝑥< + 𝜀<
Where 𝛼 is the intercept, 𝑥< refers to a vector of characteristic of product or consumer, 𝛽)refers to a coefficients of vector, and 𝜀< refers to error term.
Therefore, based on the independent variables in this case, the logit model for this analysis can be formulated as follows:
𝑌< = 𝛼 + 𝛽)𝑃<+ 𝛽P𝑇< + 𝛽Q𝐵< + 𝛽S𝑀< + 𝛽U𝐺<+ 𝛽W𝐵𝑃< + 𝛽X𝐵𝑇<+ 𝛽Y𝑀𝑃< + 𝛽Z𝑀𝑇< + 𝛽)[𝐺𝑃<+ 𝛽))𝐺𝑇<+ 𝜀< Where, 𝛼 = intercept 𝛽)… 𝛽)) = estimation parameter 𝑃< = print advertising 𝑇< = TV advertising 𝐵< = banner advertising 𝑀< = masthead advertising
𝐺< = Google display network advertising
𝐵𝑃< = cross media of banner advertising and print advertising 𝐵𝑇< = cross media of banner advertising and TV advertising 𝑀𝑃< = cross media of masthead advertising and print advertising 𝑀𝑇< = cross media of masthead advertising and TV advertising 𝐺𝑃< = cross media of GDN advertising and print advertising 𝐺𝑇< = cross media of GDN advertising and TV advertising
𝜀< = error term
purchase behavior, cross-media variables are introduced. The coefficient of cross-media variables show how increases in one media advertising enhances the effect of other media advertising on consumer purchase behavior. Through a comparison of the coefficient of varying cross-media combinations, we can determine which cross-media combination has the greatest effect on consumer purchase behavior.
Following from this, consumer segments are introduced into the regression. The coefficients of the interaction terms between the consumer segments and independent variables show the difference in the effect of cross-media advertising on consumers’ actual purchase behavior between different consumer segments. The effects of different cross-media combinations on consumers’ actual purchase behavior are then compared to find out which cross-media combination is most suitable for distinct consumer segments.
To obtain addition insights regarding the timing of consumer purchase behavior, we implement another model: the hazard model. The hazard model is used to study the probability that an event (purchase) happens in a certain time interval given it has not happen yet (Leeflang at al, 2015). A cox proportional hazard model is applied, allowing us to obtain a hazard rate from this model. Hazard rate is the conditional probability that the event occurs at time t, given that it has not occurred until time t (Leeflang at al, 2015). Purchase timing is introduced as a dependent variable. This variable indicates at which month consumers make actual purchase decisions during the observation period. The basic choice model is as follows:
ℎ<(𝑡) = ℎ\(t)exp (𝛽)𝑋<)+ ⋯ + 𝛽c𝑋<c)
Where ℎ\(t) is the baseline hazard, 𝑋<)…<c refers to vectors of explanatory variables, and 𝛽)…c is vector of parameters.
Therefore, based on the independent variables in this case, the harzad model for this analysis can be formulated as follows (vectors of explanatory variables are same as logit model):
ℎ<(𝑡) = ℎ\(t)exp ( 𝛽)𝑃< + 𝛽P𝑇<+ 𝛽Q𝐵<+ 𝛽S𝑀< + 𝛽U𝐺< + 𝛽W𝐵𝑃< + 𝛽X𝐵𝑇< + 𝛽Y𝑀𝑃< + 𝛽Z𝑀𝑇< + 𝛽)[𝐺𝑃<+ 𝛽))𝐺𝑇<)
4 RESULTS
4.1 Dataset preparation
After aggregating the dataset into household level, there are 9934 households, each row represents one household. Based on previous chapter, the valid weighting factor should be greater than 0, and there are 1738 observations showing value 0. Consequently, 1738 missing values are found in the dataset and removed. Since housewives who are younger than 18 years and older than 65 and repeat purchase are not taken into account, the dataset totally contains 7837 households.
4.2 Consumer segmentation
4.2.1 Identify number of cluster
Before the cluster analysis is performed, five consumer demographic active variables are determined: number of children (HHtype), living district (distrMEP), household type (stratMEP), lifecycle (lifecycl) and housewife age (lfthvj). Households are classified based on these variables. One demographic variable from the dataset is not included: income level. This is due to that 18.5% of the households did not want to disclose their income level; consequently, a large amount of income level data was missing. Therefore, only five demographic variables have been selected and converted into standardized variables before analysis.
10 iterations, they are still changing, meaning that these two kinds of cluster solutions are not very stable. Therefore, four- and five-cluster solutions are not appropriate approaches in this case. A one-way ANOVA test therefore is performed to check whether the standardized values of the active variables show significant differences (Appendix H). In the two-cluster solution, all active variables are significant except the variable of living district (P-value = 0.414); the F statistic value is also relatively low compared to other variables (F-value = 0.668). In the three-cluster solution, all active variables are significantly different (P-value < 0.05). The variable living district also shows a low F statistic value (F-value = 5.301) in the three-cluster solution, so this variable might be less useful in determining the cluster solution. Based on the iteration history changes and the ANOVA test, we can conclude that compared to the two-cluster solution, the three-cluster solution is stronger and more stable cluster solution in this case.
Secondly, hierarchical clustering is applied to examine the K-mean cluster solution. The scree plot below shows that the scree is at cluster three (Figure 3). Another ANOVA test is performed to see whether those active variables show significant differences when compared with each other (Appendix H). With the exception of living district (P-value = 0.053 > 0.05), all the active variables differ significantly. Since living district adds no information to the clustering, a new hierarchical clustering analysis is performed without taking this variable into account. Both scree plots show that the three-cluster solution is the most appropriate (Figure 4). This is also confirmed by the results of the K-means analysis. Hence, the optimal number of clusters is three; consequently, the variable living district is not taken into account in the following analysis.
4.2.2 Segmentation and profiling
A cluster analysis reveals three distinct segments, which are significantly different in terms of demographic variables. Segment 1 contains 3375 (43.1%) households, segment 2 contains 3260 (41.6%) households and segment 3 contains 1202 (15.3%) households. Based on the ANOVA test, all segments show a significant difference in number of children, household type, lifecycle, and housewife age. The ANOVA outputs also show the importance of the demographic variables for each segment from the mean standardized values. Number of children is the most important variable for segment 1 (mean = 1.15) and segment 3 (mean = -0.87), and lifecycle is the most important for segment 2 (mean= 0.89) (Appendix I).
Segment 1 Children family. This segment differs from the other two segments in number of children, as it is the only segment with children. Therefore, this segment is designed as a “children family.” This segment also differs significant from segment 3 in household type, and has three people in the household with an age ranging from 40 to 49 years old on average. Furthermore, the households from this segment are mainly limited-income families. The segment average housewife age is 44 years old (Appendix F).
Segment 2 Big family. This segment mainly differs from other segments in terms of household type, as it has the largest household size of four people. This segment is therefore designed as a “big family.” It differs significantly in number of kids from segment 1, as this segment does not have children in the household. It also differs from segment 3 in lifecycle: the households from this segment contain breadwinners with partners. The average housewife age here is 53 years old (Appendix F).
4.3 Analysis results
4.3.1 Multicollinearity checking
In order to check whether relations existed between independent variables, the variance inflation factor (VIF) is calculated. Based on Leeflang et al. (2015), multicollinearity arises when the VIF is higher than 4. As we can see from the table (Table 4) below, all VIF scores are lower than 4 except for the following variables: banner advertising, banner*print interaction, banner*TV interaction. In order to solve this problem, mean centering of the variables has been applied. Mean centering the variables involves subtracting a variable’s mean from all its values to decrease its multicollinearity. After mean centering the variables, those three variables’ VIF scores are still higher than 4 (Appendix J). We therefore exclude at least one of those variables. After excluding banner advertising, all the variables have VIF scores below the threshold of 4 (Appendix J). As a result, banner advertising has been removed from the model due to multicollinearity.
Table 4: Multicollinearity checking: VIF scores (Before mean centering)
Variable VIF scores Variable VIF scores
Print advertising 1.519578 Banner * Print 6.812166
TV advertising 1.385253 Banner * TV 10.555625
4.3.2 Hypothesis testing
Table 5: Main estimation results
Logit Model Hazard Model
Coefficient P-value Coefficient P-value
Print advertising 4.432e+00 0.000238 0.056533 0.000816
TV advertising 1.566e+03 0.939041 -0.011247 0.617817 Masthead
advertising
5.571e+02 0.999767 1.755445 0.046139
GDN 1.705e-01 0.300555 0.102157 0.109679 Banner * Print 4.012e+00 0.999409 0.107879 0.148059 Banner * TV 1.014e+03 0.986702 0.355729 0.436501 Masthead * Print 1.690e+03 0.999880 -0.164370 0.584089 Masthead*TV -6.476e+03 0.998584 -0.486495 0.548339 GDN * Print -1.617e-01 0.922745 -0.001973 0.834917 GDN * TV -1.204e+02 0.968095 -0.02321 0.335326
The logit model with a AIC score of 571.31, and McFadden R2 is 0.86, indicate that the model fit is good. The hazard model has close results across Likelihood ratio test (=21.01), Wald test (=25.57) and Score test (=27.5). And P-value of those three tests are highly significant (P-value <0.05), therefore, we can conduct that the model fit of hazard model is quite good.
the results of hypothesis 1a and 1b, we can conduct that hypothesis 1 is partially supported. The online advertising masthead also has a positive effect on purchase timing (P-value = 0.046139). Hazard rate changes 67.28 % (exp(coef) = 1.6728) with each exposure amount increase of masthead advertising. Therefore, hypothesis 2b is partially supported. Google display network advertising, however, do not have significant effects on both purchase probability and timing of purchase (P-value >0.05). Hypothesis 2c is not supported. Therefore, Hypothesis 2 is partially supported.
Table 6: Segment estimation results
Logit Model Hazard Model
Coefficient P-value Coefficient P-value
Segment Children family
Print advertising 1.409e+00 1.68e-05 0.031349 0.275031 TV advertising 1.513e+03 0.960670 0.002895 0.919563 Masthead
advertising
9.451e+02 0.992243 1.550651 0.128778 GDN -8.217e-02 0.798798 0.206386 0.100714 Banner * Print 1.596e+01 0.999740 0.099277 0.204004 Banner * TV 2.655e+02 0.998766 0.228863 0.685833 Masthead * Print 2.020e+02 0.999023 -0.090969 0.773565 Masthead*TV -4.100e+03 0.990016 -0.512093 0.571323 GDN * Print 5.472e-01 0.845260 0.021602 0.576667 GDN * TV -1.586e+02 0.961830 -0.060043 0.138138
Segment Big family
Print advertising 1.737e+00 0.000362 0.094835 0.000177
TV advertising 2.612e+02 0.987992 -0.046785 0.176644 Masthead advertising -3.622e+03 0.998509 4.775855 0.378855 GDN -3.573e-01 0.450816 0.048834 0.721573 Banner * Print NA NA NA NA Banner * TV -3.932e+02 0.999791 6.141616 0.147588 Masthead * Print NA NA NA NA Masthead*TV NA NA NA NA GDN * Print -3.837e-01 0.753661 -0.014021 0.797411 GDN * TV -1.520e+01 0.997999 -0.113224 0.449633 4.3.3 Validation
Table 7: Summary of the hypothesis testing results
Hypothesis Conclusion
H1: Offline advertising has a positive relationship with the consumer’s actual
purchase behavior.
Partially supported
H1a: TV advertising has a positive relationship with consumer actual purchase
behavior.
Not supported
H1b: Printing advertising has a positive relationship with consumer actual purchase
behavior.
Supported
H2: Online advertising has a positive relationship with a consumer’s actual purchase
behavior.
Partially supported
H2a: Banner advertising has a positive relationship with consumer actual purchase
behavior.
Not supported
H2b: Masthead advertising has a positive relationship with consumer actual
purchase behavior
Partially supported
H2c: Google display network has a positive relationship with consumer actual
purchase behavior.
Not supported
H3a: The cross-media advertising effect of Banner advertising and TV advertising
has a positive relationship with consumer actual purchase behavior.
Not supported
H3b: The cross-media advertising effect of Banner advertising and Print advertising
has a positive relationship with consumer actual purchase behavior.
Not supported
H3c: The cross-media advertising effect of Masthead advertising and TV advertising
has a positive relationship with consumer actual purchase behavior.
Not supported
H3d: The cross-media advertising effect of Masthead advertising and Print
advertising has a positive relationship with consumer actual purchase behavior.
Not supported
H3e: The cross-media advertising effect of Google display advertising and TV
advertising has a positive relationship with consumer actual purchase behavior.
Not supported
H3f: The cross-media advertising effect of Google display advertising and Print
advertising has a positive relationship with consumer actual purchase behavior.
Not supported
H4: For different consumer segments, there are differences between the effects of
online advertising, offline advertising, and cross-media advertising on consumer actual purchase behavior.
5 CONCLUSION AND RECOMMENDATIONS
5.1 Conclusions and Discussion
This research investigated consumer segmentation in the Dutch electronics market and whether online channel, offline channel and cross-media marketing activities are able to affect consumer purchase behavior in this market. In doing so, a household level dataset included consumer demographic, advertising exposure amount and consumer purchase behavior was obtained from Dutch electronic market during the period 2010 to 2011. Logit model and Harzard model were used enable us to investigate the consumer purchase behavior regarding to consumer purchase probability and timing of purchasing.
Nevertheless, with the exception of masthead advertising, our results show that online channels do not have a significant impact on consumers’ actual purchase behaviors. Masthead advertising can shorten consumer electronic product purchase timing in the Netherlands. Because masthead advertising is mainly used in online video websites, this result might be explained by the rapid development of online video websites during the observed period (2010-2011). The biggest online video website at this time was YouTube with three billion views per day in 2011: this was a 50% increase from the preceding year (Thenextweb, 2011). The amount of online video website users is growing and the increased time spent on online video websites like YouTube accordingly results in the increased possibility that consumers receive advertising messages. Usually, masthead advertisements can link to the advertiser YouTube channel or website homepage, which can maximize user engagement and encourage consumers to inquire for more product information. Consequently, the rich product information that is received by consumer may reduce the purchase process. It should be noted that masthead advertising only shortens the electronic product purchase timing: it does not result in higher purchase probability. However, banner advertising and the Google display network do not have an impact on consumer purchase behavior. This finding is not in line with previous studies. This could be caused by the fact that the data were collected from 2010 to 2011 and online advertising might not have been widely used by the Dutch electronics industry at that time. The chance that customers saw an advertisement on the Internet is, therefore, relatively small. As a result, online advertising did not have enough influence to encourage purchasing.
very limited, and do not have a greater impact on consumer purchase than using single advertising channel (e.g. print advertising).
Third, consumers in the Dutch electronics market can be divided into three segments: children family, big family and young family. Three segments are significantly difference in print advertising effect. Print advertising has highest impact on segment big family since it not only increase segment consumer purchase probability but also shortens the purchase timing compared to segment young family.
5.2 Implications
This research offers a point of view of segmenting and marketing for the Dutch electronics market. On one hand, Dutch electronics market can be classified into three segments: children family segment, big family segment and young family segment. On the other hand, advertising activities have an impact on consumer purchase probability and timing of purchasing in Dutch electronics market. Such as print and masthead advertising can shorten consumer purchase timing, and print advertising can increase consumer purchase probability. Dutch electronics retailers can learn from these results and have an overview of the purchase behavior of different consumer segments. Moreover, they can better establish different advertising strategies based on distinct segment behavior. In general, Dutch electronic marketer can utilize more print advertising and masthead advertising for Dutch consumers, for instance, newspaper advertisements in the areas where these consumers are living and masthead advertising on online media website.
5.3 Limitations
electronic product is both a low purchase frequency product and high involvement product, consumers usually require more product information and longer processing before purchasing. In addition to this, advertising has a lag effect. These two points combined reveal that a longer observation period for data might have been better to investigate consumer behavior of the Dutch electronics market. Furthermore, online advertising plays a more important role today than compared with the observed period of study (2010-2011). Therefore, outdated data and information might have limited, or even biased, the research results.
5.3 Recommendations for future research
REFERENCES
Zephoria, (2014), “Top 15 valuable Facebook statistics”, (accessed May 26,2017), [available at https://zephoria.com/social-media/top-15-valuablefacebook- statistics/.]
Internet Live Stats, (2017), “Google search statistics”, (accessed May 26,2017), [available at http://www.internetlivestats.com/googlesearch-statistics/]
eMarketer, (2017), “US Ad Spending eMarketer Forecast”, (accessed May 26,2017), [available at
https://www.emarketer.com/Report/US-Ad-Spending-eMarketer-Forecast-2017/2001998] eMarketer, (2016), “US digital ad spending surpass TV this year”, (accessed May 26,2017), [available at
https://www.emarketer.com/Article/US-Digital-Ad-Spending-Surpass-TV-this-Year/1014469 ]
Statista, (2017), “newspaper-advertising-spending-in-the-us”, (accessed May 26,2017), [available at
https://www.statista.com/statistics/272411/newspaper-advertising-spending-in-the-us/] Nielsen, (2010), “What consumers watch”, (accessed May 26,2017), [available at
http://www.nielsen.com/us/en/newswire/2010/what-consumers-watch-nielsens-q1-2010-three -screen-report.html. March,2017]
Nielsen, (2012). “Media: Advertising and Audience Report”, (accessed May 26,2017),
[available at
http://nielsen.com/content/dam/corporate/us/en/reports-downloads/2012-Reports/nielsen-adve rtising-audiences-report-spring-2012. pdf. Accessed April 2014.]
RAB, (2011), “Radio boosts campaign effectiveness”, (accessed May 26,2017), [available at http://www.rab.com/whyradio/pf_faq17.cfm]
Zeff, R. L., & Aronson, B. (1999). Advertising on the Internet. John Wiley & Sons, Inc.. Dekimpe, M. G., & Hanssens, D. M. (2007). Advertising response models (pp. 247-63). Thousand Oaks, CA: Sage Publications.
Dinner, I. M., Van Heerde, H. J., & 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.
Schultz, D. E., Block, M. P., & Raman, K. (2012). Understanding consumer-created media synergy. Journal of Marketing Communications, 18(3), 173-187.
Shimp, T. A., & Andrews, J. C. (2012). Advertising promotion and other aspects of integrated
marketing communications. Cengage Learning.
Ernst & Young (2015) Global Online Retailing Report
Naik, P. A., & Peters, K. (2009). A hierarchical marketing communications model of online and offline media synergies. Journal of Interactive Marketing, 23(4), 288-299.
Wiesel, T., Pauwels, K., & Arts, J. (2011). Practice prize paper—Marketing's profit impact: quantifying online and off-line funnel progression. Marketing Science, 30(4), 604-611.
Chang, Y., & Thorson, E. (2004). Television and web advertising synergies. Journal of Advertising, 33(2), 75-84.
Manchanda, P., Dubé, J. P., Goh, K. Y., & Chintagunta, P. K. (2006). The effect of banner advertising on internet purchasing. Journal of Marketing Research, 43(1), 98-108.
Agarwal, A., Xie, B., Vovsha, I., Rambow, O., & Passonneau, R. (2011, June). Sentiment analysis of twitter data. In Proceedings of the workshop on languages in social media (pp. 30-38). Association for Computational Linguistics.
Kim, A. J., & Ko, E. (2012). Do social media marketing activities enhance customer equity? An empirical study of luxury fashion brand. Journal of Business Research, 65(10), 1480-1486. Braun, M., & Moe, W. W. (2013). Online display advertising: Modeling the effects of multiple creatives and individual impression histories. Marketing Science, 32(5), 753-767.
Skiera, B., & Abou Nabout, N. (2013). Practice prize paper—prosad: a bidding decision support system for profit optimizing search engine advertising. Marketing Science, 32(2), 213-220.
Stacey, E., Pauwels, H., & Lackman, A. (2013). Beyond Likes and Tweets: Marketing, Social Media Content, and Store Performance. The Center for Measurable Marketing.
Statista ,(2015), “Year-on-year change in sales turnover of shops selling consumer electronics in the Netherlands from 2005 to 2015”, (accessed May 26,2017), [available at https://www.statista.com/statistics/440070/netherlands-retail-sales-electronics-shops-turnover -change/]
Euromonitor, (2016), “Consumer electronics in the Netherlands”, (accessed May 26,2017), [available at http://www.euromonitor.com/consumer-electronics-in-the-netherlands/report] Kotler, P., & Armstrong, G. (2010). Principles of marketing. pearson education.
Solomon, M., Russell-Bennett, R., & Previte, J. (2012). Consumer behaviour. Pearson Higher Education AU.
Fernandez, K. V., & Rosen, D. L. (2000). The effectiveness of information and color in yellow pages advertising. Journal of Advertising, 29(2), 61-73.
Lohse, G. L., & Rosen, D. L. (2001). Signaling quality and credibility in yellow pages advertising: The influence of color and graphics on choice. Journal of advertising, 30(2), 73-83.
Rosenberg, J. M. (1995). Dictionary of marketing and advertising (Vol. 13). Wiley.
Fernandez, K. V. (1995). Information processing in directional media: the effectiveness of
selected advertising executional cues in the yellow pages.
Bergemann, D., & Bonatti, A. (2011). Targeting in advertising markets: implications for offline versus online media. The RAND Journal of Economics, 42(3), 417-443.
Danaher, P. J., & 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.
Winer, R. S. (2009). New communications approaches in marketing: Issues and research directions. Journal of Interactive Marketing, 23(2), 108-117.
Strauss, J., & Frost, R. (2014). E-marketing (internat. ed.).
Digital Analytics, 2013, “Youtube and social media adverting”, (accessed May 26,2017), [available at http://dauofu.blogspot.nl/2013/01/youtube-and-social-media-adverting.html] Goldfarb, A., & Tucker, C. (2011). Online display advertising: Targeting and obtrusiveness. Marketing Science, 30(3), 389-404.
Bart, Y., Stephen, A. T., & Sarvary, M. (2014). Which products are best suited to mobile advertising? A field study of mobile display advertising effects on consumer attitudes and intentions. Journal of Marketing Research, 51(3), 270-285.
Tang, T., Newton, G. D., & Wang, X. (2007). Does synergy work? An examination of cross-promotion effects. The International Journal on Media Management, 9(4), 127-134. Maclnnis, D. J., & Jaworski, B. J. (1989). Information processing from advertisements: Toward an integrative framework. The Journal of marketing, 1-23.
Ajzen, I., & Fishbein, M. (1975). Belief, attitude, intention and behavior: An introduction to theory and research.
Lim, J. S., Ri, S. Y., Egan, B. D., & Biocca, F. A. (2015). The cross-platform synergies of digital video advertising: Implications for cross-media campaigns in television, Internet and mobile TV. Computers in Human Behavior, 48, 463-472.
Chang, Y., & Thorson, E. (2004). Television and web advertising synergies. Journal of Advertising, 33(2), 75-84.
Park, C. H., & Kim, Y. G. (2003). Identifying key factors affecting consumer purchase behavior in an online shopping context. International Journal of Retail & Distribution Management, 31(1), 16-29.
Backman, S. J. (1994). Using a person-situation approach to market segmentation. Journal of Park and Recreation Administration, 12(1), 1-16.
Nikkei ,2009. http://adweb.nikkei.co.jp/english/newspapers_in_japan/ Nrc,2014,
https://www.nrc.nl/nieuws/2014/07/14/tmg-stopt-met-spits-krant-gaat-op-in-metro-a1500903 Thenextweb, 2011, “Youtube hits 3 billion views per day”, (accessed May 26,2017), [available at
FIPP, 2014, “Proof of magazine performance”, (accessed May 26,2017), [available at http://fipp.s3.amazonaws.com/media/documents/ef2c2981-3654-4878-8888-58c6ddd86e79.p df]
Malhotra, N. K. (2008). Marketing research: An applied orientation, 5/e. Pearson Education India.
Montgomery, D. B., & Silk, A. J. (1972). Estimating dynamic effects of market communications expenditures. Management Science, 18(10), B-485.
Leeflang, P.S.H., J.E. Wieringa, T.H.A. Bijmolt, and K.H. Pauwels (2015), Modeling Markets: Analyzing Marketing Phenomena and Improving Marketing Decision Making. Springer. Broadbent, S. (1979). One way TV advertisements work. Journal of the Market Research
Society, 21(3), 139-166.
Bigné, E., Gnoth, J., & Andreu, L. (2008). Advanced topics in tourism market segmentation. Tourism management: Analysis, behaviour and strategy, 151-173.
Jurowski, C., & Reich, A. Z. (2000). An explanation and illustration of cluster analysis for identifying hospitality market segments. Journal of Hospitality & Tourism Research, 24(1), 67-91.
Fok, Dennis (2017), Advanced Individual Demand Models
Rufino, C. C. (2008). Lagged Effect of TV Advertising on Sales of an Intermittently Advertised Product.
APPENDICES
Appendix A Consumer demographic variables explanation
Table 8: Demographic variable explanation
Variables Description Levels
HHtype Number of children
0 = Without kids 1 = With kids
distrMEP Living district 1 = 3 grote steden (3 main cities) 2 = Rest west (Rest west)
3 = Noord (North) 4 = Oost (East) 5 = Zuid (South) stratMEP Type of household 1 = 1 person, -29 year 2 = 2 persons, -29 year 3 = 3+ persons, -29 year 4 = 1 person, 30-39 year 5 = 2 persons,30-39 year 6 = 3 persons,30-39 year 7 = 4 persons,30-39 year 8 = 5+ persons,30-39 year 9 = 1 person, 40-49 year 10 = 2 persons,40-49 year 11= 3 persons,40-49 year 12 = 4 persons,40-49 year 13 = 5+ persons,40-49 year 14 = 1 person, 50-64 year 15 = 2 persons,50-64 year 16 = 3+ persons,50-64 year 17 = 1 person ,65+ year 18 = 2+ persons,65+ year lifecycl Lifecycle description 0 = Onbekend 1 = Jonge alleenstaande 2 = Tweeverdieners 3 = Gezin, beperkt inkomen 4 = Gezin, welgesteld 5 = Kostwinner met partner 6 = Alleenstaande 7=Gepensioneerd,beperkt inkomen 8 = Gepensioneerd, welgesteld
Appendix B Consumer demographic variables frequencies
Figure 5: Number of children variables distribution
Figure 7: Household wife age distribution
Figure: 6 Living district distribution
Figure 8: Household type distribution
Appendix C Consumer demographic variables description
Table 9: Descriptive of demographic variables
Variable Description Mean SD
HHtype Number of children 0.430595 0.495191 distrMEP Living district 3.154376 1.425322 stratMEP Type of household 10.545037 4.832443 lifecycl Lifecycle description 3.812452 1.761392 inkomHH Income level 27.020924 34.681666
Appendix D Online, offline and cross-media advertising variables boxplot
Figure 11: Print advertising boxplot
Figure 13: Masthead advertising boxplot
Figure 15: GDN advertising boxplot
Figure 12: Print advertising boxplot
Figure 14: TV advertising boxplot
Figure 17: Banner*TV advertising boxplot Figure 18: Masthead*TV advertising boxplot
Figure 19: Masthead*Print advertising boxplot Figure 20: GDN*Print advertising boxplot
Appendix E K-mean cluster analysis: Iteration history change Table 10: Two-cluster solution iteration history change
Iteration Change in Cluster Centers
1 2 1 2.521 2.456 2 0.094 0.161 3 0.059 0.101 4 0.064 0.122 5 0.046 0.095 6 0.022 0.047 7 0.004 0.008 8 0 0
Table 11: Three-cluster solution iteration history change
Iteration Change in Cluster Centers
1 2 3 1 1.408 1.588 1.674 2 0.437 0.246 0.329 3 0.127 0.132 0.065 4 0.086 0.044 0.069 5 0.015 0.027 0 6 0.005 0.008 0 7 0 0 0
Table 12: Four-cluster solution iteration history change
Iteration Change in Cluster Centers
Appendix F K-mean cluster analysis: final cluster centers
Figure 22: Two-cluster solution final cluster centers
Figure 23: Three-cluster solution final cluster centers
Appendix G K-mean cluster analysis: relative sizes of cluster solutions Table 13: Relative sizes of two clusters solution
Cluster Number of households
1 5265
2 2573
Table 14: Relative sizes of three clusters solution
Cluster Number of households
1 3202
2 1763
3 2872
Table 15: Relative sizes of four clusters solution
Cluster Number of households
1 1541
2 1988
3 2729
4 1579
Table 16: Relative sizes of five clusters solution
Cluster Number of households
1 1437
2 1213
3 1022
4 1519
Appendix H ANOVA test results
Table 17: K-mean cluster analysis two-cluster solution ANOVA test results
Variables F-value P-value
Number of children 17.341348 0.000032 Living district 0.668223 0.413697
Type of household 27627.910187 0.0E0 Lifecycle description 3143.408461 0.0E0 Housewife age 16855.932358 0.0E0
Table 18: K-mean cluster analysis three-cluster solution ANOVA test results
Variables F-value P-value
Number of children 44323.983440 0.0E0 Living district 5.253408 0.005248 Type of household 7424.386139 0.0E0 Lifecycle description 6667.315429 0.0E0 Housewife age 6608.335011 0.0E0 Table 19: Hierarchical cluster analysis ANOVA with variable living district
Variables F-value P-value
Number of children 8.2574E30 0.0E0 Living district 2.913353 0.054352 Type of household 3515.730366 0.0E0 Lifecycle description 8456.237925 0.0E0 Housewife age 3068.038309 0.0E0
Table 20: Hierarchical cluster analysis ANOVA without variable living district
Variables F-value P-value
Appendix I Most important variables for each segment
Table 21: Active variables mean (standardized)
Number of children Household type Lifecycle Housewife age
Segment 1 1.14987 0.150986 -0.367895 -0.112092 Segment 2 -0.869552 0.428822 0.891484 0.611989 Segment 3 -0.869552 -1.587326 -1.385593 -1.345578
Table 22: Active variables mean
Number of children Household type Lifecycle Housewife age
Segment 1 1 11.274667 3.164444 44
Segment 2 0 12.617295 5.382705 53
Segment 3 0 2.874376 1.371880 30
Appendix J Multicollinearity checking: variance inflation factor score Table 23: VIF score after mean centering
Variable VIF scores Variable VIF scores
Print advertising 1.520030 Banner * Print 6.872189
TV advertising 1.385982 Banner * TV 10.648535
Banner advertising 17.563668 Masthead * Print 1.762841 Masthead advertising 2.436638 Masthead * TV 2.068948 Google display network 1.109412 GDN * Print 1.487823 GDN * TV 1.388789 Table 24: VIF scores after mean centering without banner advertising
Variable VIF scores Variable VIF scores
Print advertising 1.520028 Banner * Print 1.036806
TV advertising 1.385918 Banner * TV 1.245817
