Research Master Thesis 2016-2017
Does offline trigger online? – the effects of direct mailing on online
consumer responses along the consumers’ journey to purchase
E.N.M. (Lisan) Lesscher
Student number: 2375044
University of Groningen
Faculty of Economics and Business
Research Master Marketing
e.n.m.lesscher@student.rug.nl
ABSTRACT
Despite the frequent use of traditional marketing media and the fact that firms have to manage their budget across online and offline touchpoints, there is a strong focus on online touchpoints in current research with offline touchpoints mostly being neglected. Moreover, cross-channel effects of customer touchpoints are mostly neglected in current studies. Therefore, this research will study the effects of an offline touchpoint, direct mailing, along the consumers’ journey to purchase. By simultaneously estimating a system of equations based on quasi-experimental data by a large German insurance firm and a well-known search engine operator, this research aims to provide insights in the effectiveness of direct mailing to generate consumer responses in the consumers’ journey to purchase and the relations among the different stages of the consumers’ journey to purchase. Our findings show that direct mailing generates consumer responses in the consideration and purchase stages of the consumers’ journey to purchase and mainly affects consumer responses in the consideration stage of the consumers’ journey to purchase by getting the general topic in the consumers’ mind. Moreover, the different stages of the consumers’ journey to purchase affect each other, causing direct mailing to also affect purchase indirectly via the earlier stages of the consumers’ journey to purchase. Our results are valuable to business as they help explaining the effectiveness of an offline customer touchpoint, which is largely neglected in current research.
Chapter 1 INTRODUCTION
In order to attract consumers and influence conversion and related behaviors, firms allocate significant marketing investment to online and offline media channels. Over the last years, marketing investments have steadily increased—case in point is an increase of €136 million in the net advertising spending realized by the largest advertising markets (online, TV, newspaper, radio, out-of-home and magazine) from 2015 to 2016 and an expected increase of €96 million for 2017 (Deloitte 2016) in the Netherlands. Likewise, in the United States, digital and direct spending have increased by $35 billion over the past four years (Winterberry Group 2016). As a result, the amount of marketing messages is increasing (Hubspot 2017) and multiple customer touchpoints with different types of firms and brands are created. We define customer touchpoints as an episode of (in)direct contact with a firm or brand (Baxendale, MacDonald and Wilson 2015). Understanding the effectiveness of each touchpoint and its role in the journey to purchase of consumers is becoming very important as consumers move through a number of touchpoints across channels, media and devices in their journey to purchase (Kannan, Reinartz and Verhoef 2016). This journey to purchase involves multiple stages, from consideration to visit to purchase (Li and Kannan 2014). Within this journey to purchase, firms face multiple challenges for which research has not provided answers yet. Particularly, understanding the effectiveness of offline touchpoints is important as these touchpoints are used at least as frequently as online touchpoints although largely neglected by research, as we will discuss soon. Therefore, the current research will focus on an offline touchpoint, direct mailing, as this is an important offline touchpoint, which we discuss soon, and the impact of this offline touchpoint is generally ignored (Prins and Verhoef 2007). Overall, one of the unanswered questions, which the current research will focus on is:
How does direct mailing affect consumer responses in the different stages of the consumers’ journey to purchase?
practical relevance, Anderl et al. (2016) suggest the importance of attribution as insights are valuable for firms by helping to explain a channels’ effectiveness and define the interplay of channels. Moreover, guidance is provided to managers on strategic budget allocation across the channels to create a successful consumer journey to purchase (De Haan, Wiesel and Pauwels 2016).
Although important, current studies of attribution face some major issues, which the current research hopes to solve and thereby contribute to both theory and practice. First, there is a strong focus on online channels and touchpoints as most attribution studies typically only consider several forms of online touchpoints and neglect offline touchpoints (e.g. Blake et al. 2015; Li and Kannan 2014). However, traditional channels are used at least as frequently by firms in practice. For example, a global report by McKinsey (2015) shows that the digital share of total spending of 44% is still below 50% although expected to be half of the total spending by 2019. Therefore, firms have to manage and allocate their marketing budget strategically across both online and offline channels (e.g., Dekimpe and Hanssens 2007; Lehmann 2004). The current research will focus on an offline touchpoint and study its effect on consumer responses in the consumers’ journey to purchase.
Table 1: Contribution of this study to existing literature Paper Inclusion of offline touchpoints Cross-channel effects (offline online) Inclusion of multiple stages of consumers’ journey to purchase
Naik and Raman (2003) X X X
Danaher and Dagger (2013) √ X X
Li and Kannan (2014) X X √
Anderl et al. (2016) X X √
Kireyev, Pauwels and Gupta (2016) X X X
De Haan, Wiesel and Pauwels (2016) √ X √
Joo, Wilbur and Zhu (2016) √ √
(offline online) X
Lobschat, Osinga and Reinartz (2017) √ √
(online offline)
√
This research √ √
(offline online) √
Chapter 2 LITERATURE AND BACKGROUND
Figure 1: Conceptual framework
The effect of direct mailing in earlier stages of the consumers’ journey to purchase
Consumers are affected by different touchpoints in different stages of their decision process (Abhishek, Fader and Hosanagar 2012). Regarding the effectiveness of direct mailing, past research has studied the effects of direct mailing on consumer responses in the different stages of the consumers’ journey to purchase. Thereby, its effect on earlier stages of the consumers’ journey to purchase has also been considered. De Haan, Wiesel and Pauwels (2016) take the distinction between firm-initiated and customer-initiated touchpoints into account, which is an important distinction among customer touchpoints in the literature. Firm-initiated touchpoints are (in)direct contact points with the customer initiated by the firm and therefore have the focus of pushing the message to the consumer (Shankar and Malthouse 2007), such as advertising in email or print. Customer-initiated touchpoints are brought about by actions of the (potential) customer (Li and Kannan 2015; Shankar and Malthouse 2007), such as using a research engine website for search or review pages. As indicated by Abhishek, Fader and Hosanagar (2012), these different touchpoints affect consumers in different stages of the consumers’ journey to purchase. With regard to this distinction, it is shown that firm-initiated customer touchpoints seem able to reach consumers unaware of the need for the product (category), whereas customer-initiated touchpoints provide assistance for consumers in need for more information, evaluating alternatives and closer to buying the product (De Haan,
Earlier stages consumers’ journey to
purchase
Direct mailing Purchase behavior
Consideration stage: (organic) search use
Visit stage: clicks on sponsored
Wiesel and Pauwels 2016). In line with this finding, Abhishek, Fader and Hosanagar (2012) show that firm-initiated customer touchpoints usually impact consumers early in the decision process by moving them from a disengaged to an engaged state.
With regard to the effect of direct mailing in earlier stages of consumers’ journey to purchase, direct mailings are shown to trigger interest in a product/service and eventually lead to a final purchase (Roberts and Berger 1999). This interest in the product/service can be best captured by non-branded search terms (i.e., organic search use) according to Ghose and Todri (2016), which implies that direct mailing trigger organic search. Krafft et al. (2007) also state stimulation of interest as one of the advantages of direct mails. Furthermore, Danaher and Dagger (2013) state that direct mail is an effective tool for gaining visibility by being exposed to an advertisement. Related to direct mailing is email marketing as they both deliver a firm-initiated communication with the same speed and control over information transfer (e.g., Dijkstra et al. 2005) although they differ in delivery mechanism (i.e., print vs. Internet). Given this, the effects of email found by research might relate to the effects of direct mailing. Li and Kannan (2014) found significant spillover effects from email marketing to customer-initiated channels (i.e., organic search, sponsored search), which implies that these touchpoints affect consumers to use organic search and click on sponsored search ads. Moreover, they find that these firm-initiated interventions also influence visits through other channels (e.g., direct or referral). Martin et al. (2003) conducted an exploratory study on email marketing and found that email characteristics (i.e., usefulness and amount) negatively influenced website visits via clicking on the email and in general. Rather, the consumers were more likely to visit the physical store when viewing the emails as useful. Anderl et al. (2016) conducted an online attribution study and mapped the customer journey. They found that newsletters in emails might motivate subsequent visits to the firm’s website via clicking on sponsored search ads. Moreover, they find that if consumers respond to the link in the newsletter (i.e., click-through) they are likely to return via (organic) search, which corresponds with findings of Li and Kannan (2014).
search use and clicks on sponsored search ads. Therefore, we predict direct mailing to have an effect on both our earlier stages of the consumers’ journey to purchase – consideration and visit.
The effects among organic search and clicking on sponsored search ads
Within the earlier stages of the consumers’ journey to purchase (i.e., consideration and visit stage), effects might also be observed. Prior research has shown that organic search and clicking sponsored search ads are related. In general, Ansari and Mela (2003) show that organic search use increases the probability of clicking on a sponsored search ad. Organic search might influence clicks on organic and sponsored search ads based on the position of the ad (Agarwal, Hosanagar and Smith 2011; Ghose and Yang 2009) or text characteristics (Rutz and Trusov 2011). Search can be distinguished in organic search and sponsored search. With regard to the position of the ad, organic search results are ordered based on relevance and popularity of a page for a keyword and not influenced by advertiser’s bids, whereas sponsored search results are ordered based on bids. This might cause consumers to trust organic search results more than sponsored search results, which consequently might result in negative effects of search on sponsored search performance, which implies the clicks on sponsored search ads (Jansen and Resnick 2006). Yang and Ghose (2010) studied the interrelationships between organic search and sponsored search and found that the clicking probability on organic search has a positive interdependence with clicking probability on sponsored search, and vice versa. This implies that organic search use influences clicking probability on sponsored search ads as shown in other research, but clicking probability on sponsored search ads also influences organic search use. This positive interdependence is, however, asymmetric with a higher impact of organic search use on the clicking probability of sponsored search than vice versa (i.e., clicking probability of sponsored on organic search use). With regard to the text characteristics, Jerath, Ma and Park (2014) show that less common keywords (i.e., more specific keywords) are associated with more clicks on sponsored search ads per search, whereas for more common keywords (i.e., generic keywords) the probability to click a sponsored seach ad is lower.
sponsored search ads with more specific keywords (i.e., branded keywords) leading to more clicks compared to generic keywords.
The effect of earlier stages of the consumers’ journey to purchase on purchase behavior
As stated, consumers are affected by different touchpoints in different stages of their decision process (Abhishek, Fader and Hosanagar 2012). Regarding the consumers’ journey to purchase, past research has studied the effects of consumer responses in the different stages of the consumers’ journey to purchase. Thereby, the effect of earlier stages of the consumers’ journey to purchase on purchase behavior has also been considered. De Haan, Wiesel and Pauwels (2016) report that earlier stages in the consumers’ journey to purchase affect consumers further down the purchase funnel and are effective in generating sales, in line with findings of previous studies (Li and Kannan 2014; Shankar and Malthouse 2007). An example of an effect of earlier stage in the consumers’ journey to purchase is the effect of traffic to the website, which result in more search for information and a probable purchase (e.g., Wiesel, Pauwels and Arts 2011). Specific to our earlier stages in the consumers’ journey to purchase – consideration and visit stage – prior research shows that organic search and clicking on sponsored search ads both lead to purchase behavior (e.g., Danaher and Dagger 2013; Jansen 2007). Furthermore, Yang and Ghose (2010) show that the positive interdependence between organic and sponsored search causes an increase in purchase, and thereby, profits. Within search, consumers are shown to have a higher probability to purchase when their organic search is directed compared to other consumers (Montgomery et al. 2004). This is supported by Agarwal, Hosanagar and Smith (2012), who show that more generic keywords in search have a negative impact on purchase behavior, whereas search can have a positive effect on purchase for more specific keywords (e.g., branded keywords). A reason for these findings might be that consumers are less informed and search less when using generic keywords, whereas consumers are more informed and search more when using specific keywords (White and Morris 2007; White, Dumais and Teevan 2009).
The effect of direct mailing on purchase behavior
Chapter 3 DATA
We are fortunate to have access to unique data from a German Insurance firm and a well-known search engine operator, which can be used to adequately answer our research question. The insurance firm is a well-known German insurance firm, which belongs to a worldwide insurance group with over 50.000 employees in more than 200 countries. The firm has a well-established multi-channel distribution system, including online presence, own agencies and partners. For reasons of confidentiality, however, we cannot disclose the name of the insurance firm. The data provided by the firm concerns a campaign to promote car insurances. Therefore, the firm sent out direct mailings on the respective campaign to potential customers. Timing wise, the entire campaign ran from September 7, 2015 to October 24, 2015 and the direct mailings were sent out from October 19, 2015 to October 24, 2015.
The provided data is quasi-experimental, which implies that the data contains a treated and a control group and within these groups different observations on consumers’ responses in the different stages of the consumers’ journey to purchase. Hence, we identify the treatment group as having received a direct mailing on the respective campaign, whereas the control group did not receive a direct mailing. The treatment group was selected based on the household potential to converse within a postal code area. Fortunately, the postal code areas of the control group have the same household potential, but contains postal codes that eventually did not receive a direct mailing. In order to validate our control group and check for differences between the groups, GfK data on purchase power of the postal code areas was retrieved. An analysis on the difference in purchase power of the treatment group (M = 111.51) and control group (M = 106.47) show that these do not significantly differ in purchasing power, F(1, 603) = 1.06, p > .10. This provides us with confidence, that our treatment and control group did not differ significantly.
Direct mailing. The direct mailing on the respective campaign was rather
informational as it mainly included information on a car insurance and its relevance. The design for the direct mail was equal for all consumers, with images and a brief description about the insurance highlighted by the campaign and the firm’s logo (see appendix A – logo is nonvisible due to confidentiality). The direct mails were sent based on the household potential of a postal code area. This potential is mainly based on age and income although the exact algorithm of the firm is unknown to us. As stated, the direct mailings were sent out from October 19, 2015 to October 24, 2015.
Search behavior. The search behavior of consumer as a response to the campaign was
Figure 2: Frequency plot generic (black) and branded (red) search query use over time
Clicking behavior. The online clicking behavior on sponsored search ads of consumer
amount of clicks is pretty stable over time with a big spike after the first week of November. Moreover, the plot shows that the amount of impressions is (much) higher over time compared to the number of clicks, but this difference makes sense as impressions occur more frequently compared to consumers actually clicking the sponsored search ad.
Figure 3: Frequency plot of Impressions (black) and Clicks (red) over time
Purchase behavior. The purchase behavior of the consumers was recorded by the
German insurance firm. The data contains the amount of purchases of a postal code area, the contribution amount, the date the purchase was made, the starting date of the contract, through which channel the purchase was made (online/offline) and what insurance type was bought.
When analyzing when the latest policy started, this ranges between 2006 and 2016 with the most policies starting in 2015 (39.56%) and 2016 (59.15%).
Chapter 4 MODEL DEVELOPMENT
Our focus is on studying the effect of an offline touchpoint, direct mailing, on online consumer responses in different stages of the consumers’ journey to purchase. Thereby, we study the influence of direct mail on the probability that a consumer will (1) use (organic) search for information, (2) click on a sponsored search ad and/or (3) purchase a product, using a simultaneous system of equations with an equation for each consumer response.
Simultaneous system of equations: explanation
Simultaneous systems of equations can be used when one equation does not suffice to specify the relations in the market. In a system of equations, multiple endogenous variables are included to be explained based on the equations. The remainder of the variables are predetermined, such as exogenous variables. Exogenous variables are taken for given as they are assumed to be determined outside the system of equations.
The concept of simultaneity suggests that the endogenous variables are explained jointly and simultaneously based on the exogenous variables and disturbance terms. The endogenous variable can be used both to be explained as to explain other variables, in the different equations. Therefore, an endogenous variable cannot be stochastically independent of all disturbance terms.
Consider a system of K equations, where the ith equation is of the form 𝑦𝑖 = 𝑋𝑖𝛽𝑖+ 𝑢𝑖, 𝑖 = 1, 2, … , 𝐾
where 𝑦𝑖 is a vector of the dependent variable, 𝑋𝑖 is a matrix of the exogenous variables, 𝛽𝑖 is the coefficient vector and 𝑢𝑖 is a vector of the disturbance terms of the ith equation.
The ‘stacked’ system can be written as [ 𝑦1 𝑦2 ⋮ 𝑦𝐾 ] = [ 𝑋1 0 ⋯ 0 0 𝑋2 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ 𝑋𝐾 ] [ 𝛽1 𝛽2 ⋮ 𝛽𝐾 ] + [ 𝑢1 𝑢2 ⋮ 𝑢𝐾 ] or more simply as 𝑦 = 𝑋𝛽 + 𝑢
Within the system of equations, it is assumed that there is no correlation of the disturbance terms across observations, so that
where i and j indicate the equation number and, t and s denote the observation number, where the number of observations is the same for all equations. However, we explicitly allow for collective correlation, i.e.,
Ε[𝑢𝑖𝑡 𝑢𝑗𝑡] = 𝜎𝑖𝑗
When the estimation only includes exogenous regressors, the system of equations can be consistently estimated by using ordinary least squares (OLS) or seemingly unrelated regression (SUR) estimation. The OLS estimator is based on the assumptions that the disturbance terms are not contemporaneously correlated (𝜎𝑖𝑗 = 0 ∀ 𝑖 ≠ 𝑗) and have the same variance in each equation (𝜎𝑖2 = 𝜎𝑗2 ∀ 𝑖, 𝑗). The OLS estimator is efficient, as long as the disturbances are not collectively correlated. The disturbance terms of a system of equations are, however, likely to be collectively correlated due to the influence of unconsidered factors that influence the disturbance term in one equation likely also influence the disturbance terms in the other equations. When this collective correlation is ignored and the equations are estimated separately, inefficient estimates of the coefficients are gained. However, efficient estimates can be gained by estimating all equations simultaneously with a generalized least squares (GLS) estimator. The GLS estimator takes the covariance structure of the residuals into account. This estimation method is termed seemingly unrelated regression (SUR) (Zellner 1962). Note that an unbiased OLS estimation requires only that the regressors and the disturbance terms of each single equation are uncorrelated (Ε[𝑢𝑖𝑇𝑋𝑖] = 0 ∀ 𝑖), a consistent SUR estimation requires that all disturbance terms and all regressors are uncorrelated (Ε[𝑢𝑖𝑇𝑋𝑖] = 0 ∀ 𝑖, 𝑗).
In case the regressors of one (or more) equations are correlated with the disturbance terms (Ε[𝑢𝑖𝑇𝑋𝑖] ≠ 0), OLS and SUR estimation provides biased estimates. To correct this, a two-stage least squares (2SLS) estimation or three-stage least squares (3SLS) estimation with instrumental variables (IV) can be used. Herein, the instrumental variables for each equation
Zi can be either identical or different for all equations. Importantly, they may not be
correlated with the disturbance terms of the corresponding equation (Ε[𝑢𝑖𝑇𝑍𝑖] = 0). At the first stage, new (“fitted”) regressors are obtained by
X̂ = 𝑍𝑖(𝑍𝑖𝑇𝑍𝑖)−1𝑍𝑖𝑇𝑋𝑖
The 2SLS estimator is based on the same assumptions about the disturbance terms as the OLS estimator. However, in case the disturbances are collectively correlated, a feasible generalized least squares (FGLS) version of the 2SLS estimation leads to consistent and asymptotically more efficient estimates. This estimation procedure is generally termed 3SLS (Zellner and Theil 1962). While an unbiased 2SLS estimation requires only that the instrumental variables and the disturbance terms of each single equation are uncorrelated (Ε[𝑢𝑖𝑇𝑍𝑖] = 0 ∀ 𝑖), Schmidt (1990) points out that the 3SLS estimator is only consistent if all
disturbance terms and all instrumental variables are uncorrelated (Ε[𝑢𝑖𝑇𝑍𝑖] = 0 ∀ 𝑖, 𝑗).
System of equations: Consumer responses in the consumers’ journey to purchase
We model consumer responses in different stages of the consumers’ journey to purchase – 𝑃𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑖, 𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖, 𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖, and 𝐶𝑙𝑖𝑐𝑘𝑠𝑖 – for postal code i using a
system of equations approach. The system of equations consists of the following equations:
𝑃𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑖 = 𝛼𝑖+ 𝛽0𝐷𝑀𝑖+ 𝛽1𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽2𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽3𝐶𝑙𝑖𝑐𝑘𝑠𝑖+ 𝛽4𝑃𝑃𝑖+ 𝜀𝑖
𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖 = 𝛼𝑖 + 𝛽0𝐷𝑀𝑖 + 𝛽1𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽2𝐶𝑙𝑖𝑐𝑘𝑠𝑖+ 𝛽3𝑃𝑃𝑖+ 𝜀𝑖
𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖 = 𝛼𝑖+ 𝛽0𝐷𝑀𝑖+ 𝛽1𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽2𝐶𝑙𝑖𝑐𝑘𝑠𝑖+ 𝛽3𝑃𝑃𝑖+ 𝜀𝑖
𝐶𝑙𝑖𝑐𝑘𝑠𝑖 = 𝛼𝑖+ 𝛽0𝐷𝑀𝑖+ 𝛽1𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽2𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖+ 𝛽3𝑃𝑃𝑖+ 𝜀𝑖
where i refers to the postal code area, 𝑃𝑢𝑟𝑐ℎ𝑎𝑠𝑒𝑖 is the amount of purchases for a postal code region, 𝐺𝑒𝑛𝑒𝑟𝑖𝑐𝑆𝑒𝑎𝑟𝑐ℎ𝑖 is the average indexed amount of generic search queries for a postal code region, 𝐵𝑟𝑎𝑛𝑑𝑒𝑑𝑆𝑒𝑎𝑟𝑐ℎ𝑖 is the average indexed amount of branded search queries for a postal code region, 𝐶𝑙𝑖𝑐𝑘𝑠𝑖 is the average amount of clicks on sponsored search ads for a
postal code region, 𝐷𝑀𝑖 is whether the postal code is in the treatment group receiving a direct
mail (1) or in the control group not receiving a direct mail (0), and 𝑃𝑃𝑖 is the purchase power per household index of Germany for a postal code region.
For an additional model, we also included timing effects in the model (see appendix B for the system of equations of this additional model). The timing effect is incorporated by including a time difference variable, 𝑇𝑖𝑚𝑒𝑑𝑖𝑓𝑓𝑖, which shows the average amount of days for
account as it is shown that advertising wears out (i.e., decays of time) (East 2003). Moreover, it has a negative effect by decreasing brand awareness (Mahajan, Muller and Sharma 1984). The decay effect can also be included as a decay parameter (e.g., Prins and Verhoef 2007; Risselada et al. 2014), however, we consider decay effects by including the elapsed days since direct mailing was received.
Model estimation
We simultaneously estimate the system of equations on data for all available postal codes areas. Therefore, different estimation methods are checked and compared to find the estimation method to use. We start by checking the OLS estimation of our systems of equations. We use the OLS estimation for our systems and check whether endogeneity is present in one (or more) of the equations. As stated, in case the regressors of one (or more) equations are correlated with the disturbance terms (Ε[𝑢𝑖𝑇𝑋𝑖] ≠ 0), OLS estimation provides biased estimates. The Wu-Hausman test checks the consistency of the OLS estimates under the assumption that the instrumental variable (IV) approach is consistent. When rejecting the null hypothesis, it is suggested that endogeneity is present and OLS is not consistent. Accepting the null hypothesis essentially implies that OLS and IV estimates are similar and endogeneity is not a big problem. The Wu-Hausman test for the second equation is significant (p < .001), which implies that endogeneity is an issue in the system of equations. Therefore, OLS estimation results in biases and cannot be used. When OLS is not used, 2SLS and 3SLS are almost exclusively used for estimation of simultaneous system of equations thanks to their asymptotic efficiency and simplicity (Greene 2002). Given this, we continue with estimating the model with 2SLS and 3SLS estimation.
Model validation
To validate our simultaneous equations models, we have tested multiple assumptions for the separate equations, checked for weak instruments and checked the model fit. First, we tested the assumptions of normality, and homoscedasticity and ‘no perfect collinearity’, which presence might lead to biases. Non-normality is not an issue due to the fact that 3SLS is robust to non-normality (Greene 2002). Homoscedasticity is tested by the Breusch-Pagan test, which tests the null hypothesis that the variances of the standard errors are equal against the alternative hypothesis that the variances of the standard errors are not equal and heteroscedasticity is present. The Breusch-Pagan test for each equation is non-significant except for slight significance for the second equation. However, over the entire system, it is assumed there is no heteroscedasticity present. Lastly and most importantly, multicollinearity does not seem an issue in the models. To check this assumption, the correlation table and VIF values are studied for both models. The results of the ‘no perfect collinearity’ assumption testing are included in appendix C. The correlation table shows that there is no high correlation between the independent variables of each equation except for correlation between the endogenous variables and instrumental variables, which is allowed. The maximum value for the VIF’s is set to 5 based on Leeflang et al. (2015). All equations show VIF values lower than 5. Therefore, we conclude multicollinearity is of no concern in the models.
Second, we checked for weak instruments by conducting an F-test on the instruments in the first stage. The null hypothesis suggests that we have weak instruments and a rejection of the null hypothesis implies that our instruments are not weak, which is good. All F-tests are significant, meaning that our instruments are not weak. Besides the instruments, we checked the model fit for our models. Goodness of fit of a simultaneous system of equations can be measured by the McElroy’s R2 value (McElroy 1977). The McElroy R2 value of our two
Chapter 5 RESULTS
We present the results of the system of equations in tables 2-9. Before discussing our estimation results, we first present initial tests of our expectations.
Model-free evidence. We first explore whether receiving a direct mail or not affects
consumers’ responses in the consumers’ journey to purchase as well as whether the consumer responses before and after the campaign were different for the treatment and control group by conducting multiple one-way ANOVAs. With regard to purchase, the one-way ANOVA of treatment on purchase was not significant, F(1, 591) = .466, p = .495. This implies that the amount of sales does not differ significantly between the treatment group (M = 27.14) and the control group (M = 20.77). The means including their confidence interval are depicted in figure 4.
Figure 4: means of purchase behavior between groups (0 = control, 1 = treatment)
indicate the treatment and control group do not differ significantly in the use of branded and generic search queries. The means including their confidence interval are depicted in figure 5. Although not significant, the Post Hoc tests and plots on the means (see figure 5) do show a lower use of branded search queries and higher use of generic search queries of the treatment group compared to the control group. This may imply that the consumers, who received a direct mailing, use more generic search queries and less branded search queries compared to consumers, who were not confronted with a direct mailing. This might suggest that the campaign induced the generic topic of (car) insurance in consumers’ minds, which translates in the increase in generic search behavior of consumers confronted with direct mailing.
Figure 5: means of organic seach behavior between groups (1 = control, 2 = treatment)
search query use was not significantly different after the campaign (M = .81) for the control group. The one-way ANOVA of before/after campaign on generic search for the control group was significant, F(1, 668) = 9.05, p < .01. After the campaign the control group had a significant higher generic search query use (M = .77) compared to before the campaign (M = .58). Overall, the tests on the difference before and after the campaign of organic search use indicate that branded search use of the treatment group after the campaign was significantly lower compared to before the campaign. This implies that consumers, who received a direct mailing, use less branded search queries after receiving the direct mailing. For the control group, the generic search query use after the campaign was significantly higher compared to before the campaign. This implies that consumers, who were not confronted with a direct mailing, increased their use of generic search queries over time. The means including their confidence interval of all tests are depicted in figure 6.
Figure 6: means of organic search behavior before and after campaign (1 = before, 2 = after)
ANOVA of treatment on CTR was not significant, F(1, 11948) = .22, p = .64. The CTR of the treatment group (M = 11.18) and control group (M = 12.06) does not differ significantly. The one-way ANOVA of treatment on clicks was also not significant, F(1, 11948) = .21, p = .65. The amount of clicks of the treatment group (M = .16) and control group (M = .14) does not differ significantly. The one-way ANOVA of treatment on impressions was significant, F(1, 11948) = 4.25, p < .05. The amount of impressions of the treatment group (M = 1.77) was significantly higher compared to the amount of impressions of the control group (M = 1.27). Overall, the tests on the difference between the treatment and control group indicate that the treatment group has significant higher amount of impressions. However, this does not result in a higher click volume or CTR. The means including their confidence interval of all tests are depicted in figure 7.
Figure 7: means of clicking behavior on sponsored search ads between groups (1 = control, 2 = treatment)
compared to before the campaign (M = 2.59). For the control group, the difference in clicking behavior before and after the campaign was non-significant for all three metrics (i.e. CTR, clicks and impressions). Overall, the tests on the difference before and after the campaign of clicking behavior on sponsored search ads indicate that clicking on sponsored search ads of the treatment group after the campaign was significantly lower. This might suggest that the direct mail reduced the clicking on sponsored search ads of consumers. The means including their confidence interval of all tests are depicted in figure 8.
Figure 8: means of clicking behavior on sponsored search ads before and after campaign (1 = before, 2 = after)
System of equations model. In tables 2-5, we provide the estimation results for the
system of equations, respectively the purchase equation, generic search equation, branded search equation and clicks equation. We will discuss what our results reveal per equation.
.05. This implies that the amount of sales decrease if the generic search use increases. More simplified, the more consumers use generic search queries, the lower the chance of a sale. For branded search query use, this effect is the opposite as the amount of sales increase if the branded search use increases. In other words, the more consumers use branded search queries, the higher the chance of a sale. Also, the direct mailing treatment significantly influences sales. This result implies that the difference in sales when receiving a direct mailing compared to not receiving a direct mailing is significant with the treatment group having a higher chance to purchase. The control variable, purchase power per household index, is insignificant and thereby does not influence the results.
With regard to the significant effect of direct mailing on purchase behavior, we see in terms of elasticity that the effect of direct mailing has an elasticity of .035. This implies that a 1% change in direct mailing increases the amount of sales with .035%. This elasticity is, however, lower compared to previous findings by Danaher and Dagger (2013), who report an elasticity of .095. Moreover, De Haan, Wiesel and Pauwels (2016) report an average elasticity of all of their firm-initiated touchpoints of .003 and Danaher and Dagger (2013) report elasticities of their marketing tools between .011 and .078. Compared to these elasticities of other marketing tools, our elasticity for direct mailing is within this range. With regard to the effectiveness of the direct mailing in terms of return on investment (ROI), a crude calculation provides some insights. When looking at the overall costs of the direct mailing, the amount of purchases by the treatment group and the contribution per sale, the following insights are found. The average investment per sale is €5,31, whereas the average contribution per sale is much higher, €118.97. Moreover, the average contribution per direct mailing is a bit more than €4, which is around 20 times as much as one direct mailing costs the firm. Lastly, the ROI, meaning the benefit per euro spend on direct mailing, is €21,41. Overall, these insights indicate that investing in direct mailing seems efficient and worthwhile as the ROI is positive.
Table 2: results purchase equation
Estimate Std. Error t-value P-value
Second, the results of the equation on the generic search query use (see table 3), which show that the generic search query use is influenced significantly by the branded search query use, = 6.68, p < .05, the amount of clicks on sponsored search ads, = 11.73, p < .01, and the direct mailing, = 2.31, p < .05. This implies that the generic search query use increases if the branded search query use increases. In other words, the more consumers use branded search queries, the higher the chance of them using generic search queries. This makes sense, as using branded or generic search queries is closely related. Also, the amount of clicks on sponsored search ads significantly influence the generic search query use. The result implies that a higher amount of clicks on sponsored search ads also increase the generic search query use and thereby the chance of consumers using generic search queries after clicking on a sponsored search ad. Lastly, the direct mailing treatment significantly influences the generic search query use. The result implies that the difference in generic search query use is significantly higher when receiving a direct mailing compared to not receiving a direct mailing with the treatment group having a higher chance to purchase. This is not in line with the model free results, which suggested that generic search query use of the treatment and control group not significantly differs. Again, the control variable, purchase power per household index, is insignificant and does not influence the results.
Table 3: results generic search equation
Estimate Std. Error t-value P-value
(Intercept) -8.60 2.91 -2.96 .003 ** Branded search 6.68 2.99 2.24 .03 * Clicks 11.73 3.92 2.99 .003 ** Treatment 2.31 1.07 2.15 .03 * Purchase power 0.01 .01 .97 .33 * p < .05 ** p < .01
of branded search queries. Also, the direct mailing treatment significantly influences the branded search query use. The result implies that the difference in branded search query use when receiving a direct mailing compared to not receiving a direct mailing is significant with the treatment group having a higher chance to purchase. This is partly in line with the model free results as the Post Hoc test also suggested that the control group has a higher branded search query ratio compared to the treatment group. Also, the branded search query use is marginal significantly influenced by the purchase power per household index, = -.001, p = .09. Although the effect size is very small, the result implies that a higher purchase power per household index decreases the use of branded search queries.
Table 4: results branded search equation
Estimate Std. Error t-value P-value
(Intercept) .96 .17 5.73 <0.001 *** Generic search .06 .03 2.25 .03 * Clicks .08 .54 .14 .89 Treatment -.25 .10 -2.60 .009 ** Purchase power -.001 .001 -1.69 .09 . . p < .10 * p < .05 ** p < .01 *** p < .001
Lastly, the results of the clicks equation (see table 5), which show that the click volume is only influenced significantly by the generic search query use, = .05, p < .01. This implies that the click volume on sponsored search ads increases if the generic search use increases. In other words, the more consumers use generic search queries, the higher the chance of them clicking a sponsored search ad. This makes sense, given that searching for a product/service in general increases the chance of being confronted with sponsored search ads of different brands and the chance of consumers clicking on one. The control variable, purchase power per household index, is insignificant and does not influence the results.
Table 5: results clicks equation
Estimate Std. Error t-value P-value
An overview of the results of the system of equations (without the timing effect) can be found in figure 9. Based on the findings of the entire system of equations, the net effects of multiple variables in this model can also be reported. With regard to the net effects of direct mailing, these can be reported for the effect of direct mailing on purchase, generic search use and branded search use. The net effect of direct mailing on purchase mediated by generic search use is found to be 72.38 based on the direct and mediated effects. The net effect of direct mailing on purchase mediated by branded search use is found to be 6.64 based on the direct and mediated effects. The net effect of direct mailing on generic search use mediated by branded search use is found to be .64 based on the direct and mediated effects. The net effect of direct mailing on branded search use mediated by generic search use is found to be -.11 based on the direct and mediated effects. With regard to the net effects of search, these can be reported for the effect of generic search on purchase and branded search on purchase. The net effect of generic search use on purchase mediated by branded search use is found to be -3.86 based on the direct and mediated effects. The net effect of branded search use on purchase mediated by generic search use is found to be 186.79 based on the direct and mediated effects.
Figure 9: Overview of results
System of equations mode with timing effect. In tables 6-9, we provide the estimation results
for the system of equations, respectively the purchase equation, generic search equation, branded search equation and clicks equation, together with the results of the model without timing effect to be able to compare the results. We will discuss what our results reveal per equation and how these results differ with the results of the model without timing effect.
When including the time difference variable, the results of the sales equation change (see table 6). Some of the directions of the relationship changed although not significant, some significant effect (i.e., generic search query use and direct mailing treatment) turned insignificant, and the effect sizes changed. The results show that the amount of sales is influenced significantly by the branded search query use, = 261.76, p < .05, and the clicks on sponsored search ads, = -347.52, p < .05. As in the other model, the more consumers use generic search queries, the lower the chance of a sale. However, the effect size is smaller compared to the model without timing effect. For clicks on sponsored search ads, the result implies that the more consumers click on a sponsored search ad, the lower the chance of a sale. This effect was not found in the model without the timing effect and is not entirely in line with the expectation that clicking sponsored search ads leads to more sales. Notable, no significant timing effect is found, which implies that a larger time difference between the campaign and a sale does not significant influence the amount of sales.
Table 6: results of purchase equation and comparison
Estimate Estimates no timing
effect
Std. Error t-value P-value P-value no timing effect (Intercept) -149.06 -296.81 126.96 -1.17 .24 .02 * Generic search 2.51 -25.15 10.01 .25 .80 .03 * Branded search 261.76 354.79 116.82 2.24 .03 * .003 ** Clicks -347.52 85.45 144.66 -2.40 .02 * .67 Treatment 48.60 95.34 39.85 1.22 .22 .02 * Purchase power 0.20 0.30 .28 .72 .47 .29 Time diff .58 x .58 .99 .32 x * p < .05 ** p < .01
mail, = 2.32, p < .05. Most of the effect sizes are a bit larger compared to the model without timing effect, except for the effect size of clicks which turned smaller, and the direction of the effects remain equal. The results imply that a higher use of branded search queries increases the chance of consumers using generic search queries. Moreover, when the amount of clicks on sponsored search ads increase, the chance of consumers using generic search queries also increase. Lastly, the direct mailing treatment significantly influences the generic search query use with the treatment group using significantly more generic search queries compared to the control group. Notable, no significant timing effect is found, which implies that a larger time difference between the campaign and the use of a generic search query does not significant influence the use of generic search queries.
Table 7: results of generic search equation and comparison
Estimate Estimates no timing
effect
Std. Error t-value P-value P-value no timing effect (Intercept) -8.60 -8.60 2.52 -3.41 <.001 *** .003 ** Branded search 6.92 6.68 2.45 2.82 .005 ** .03 * Clicks 10.85 11.73 3.20 3.40 <.001 *** .003 ** Treatment 2.32 2.31 .97 2.38 .02 * .03 * Purchase power .008 0.01 .01 1.06 .29 .33 Time diff -.00 x .003 -.07 .94 x * p < .05 ** p < .01 *** p < .001
Table 8: results of branded search equation and comparison
Estimate Estimates no timing
effect
Std. Error t-value P-value P-value no timing effect (Intercept) .96 .96 .16 6.06 <.001*** <0.001 *** Generic search .06 .06 .02 2.94 .004 ** .03 * Clicks .04 .08 .44 .09 .93 .89 Treatment -.25 -.25 .09 -2.66 .008 ** .009 ** Purchase power -.001 -.001 .001 -1.62 .11 .09 . Time diff -.00 x .0001 -.28 .78 x . p < .10 * p < .05 ** p < .01 *** p < .001
With regard to the results of the final equation for clicks on sponsored search ads (see table 9), the results are also similar to the model without the timing effect included. The results show that clicking on sponsored search ads is only influenced significantly by the generic search query use, = .05, p < .001, which effect size is equivalent to the other model. A higher generic search query use increases the click volume on sponsored search ads to a small extent. As stated, this makes sense as searching for a product/service in general increases the chance of being confronted with sponsored search ads of different brands and the chance of consumers clicking on one.
Table 9: results of clicks equation and comparison
Estimate Estimates no timing
effect
Std. Error t-value P-value P-value no timing effect (Intercept) .16 .13 .23 .68 .50 .60 Generic search .05 .05 .01 3.47 <.001 *** .003 ** Branded search .01 .04 .22 .07 .95 .85 Treatment -.05 -.04 .08 -.62 .54 .62 Purchase power .0001 .00 .001 .09 .93 .88 Time diff .0001 x .00 .38 .71 x ** p < .01 *** p < .001
branded search use is found to be .59 based on the direct and mediated effects, which is similar to the net effect of direct mailing on generic search use of the model without timing effects. The net effect of direct mailing on branded search use mediated by generic search use is found to be -.11 based on the direct and mediated effects, which is similar to the net effect of direct mailing on branded search use of the model without timing effects.
Figure 10: Overview of results
Chapter 6 DISCUSSION
General discussion
With this paper, we model the impact of direct mailing on online consumer responses in different stages of the consumers’ journey to purchase. Thereby, we address the important unanswered research question: How does direct mailing affect consumer responses in the different stages of the consumers’ journey to purchase? In order to answer the question, we
study the influence of direct mailing on consumer responses in different stages of the consumers’ journey to purchases. These stages of the consumers’ journey to purchase are studied by checking the influence of the direct mail on the probability that a consumer will (1) use (organic) search for information, (2) click on a sponsored search ad and/or (3) purchase a product, using a simultaneous system of equations. Moreover, the relationships between the different stages of the consumers’ journey to purchase are included in the model. Our findings are depicted in figure 9 and 10. Overall, we provide evidence that direct mailing affects consumer responses in the different stages of the consumers’ journey to purchase and mainly has an influence on consumer responses in the earlier stages of the consumers’ journey to purchase by getting the general topic of the direct mailing in the consumers’ mind. Our provided insights are valuable to business as it helps explaining the effectiveness of a customer touchpoint (e.g., Anderl et al 2016). Particularly, the effectiveness of offline customer touchpoints, represented by the focal offline customer touchpoint of this paper direct mailing, are valuable to business as these touchpoints still used at least as frequent compared to online customer touchpoints (e.g., McKinsey 2015) although mostly neglected by current research.
Our results provide insights into the consumers’ journey to purchase and the role of direct mailing in this journey. Below, we discuss our findings and the resulting implications for business and research on direct mailing and the consumers’ journey to purchase.
infer a similar process based on their results, namely that firm-initiated touchpoints are able to create awareness and interest by influencing consumers earlier in the consumers’ journey to purchase based on their results. Moreover, the finding is in line with prior research on the effects of direct mailing on earlier stages of the consumers’ journey to purchase. Prior research also shows that direct mail triggers interest in the product/service (Roberts and Berger 1999; Krafft et al. 2007), and this interest in the product/service can be best captured by non-branded search terms (i.e. organic search results) according to Ghose and Todri (2016).
A reason for the suggestion that direct mailing influences the beginning of the consumers’ journey to purchase by getting the general topic in consumers’ mind might be the type of direct mailing, which was rather informational for this campaign. The direct mailing informed the consumers about the type of product/service and not specifically promote the brand. Another explanation for the finding that direct mailing positively influences generic search query use might be based on the dimensions by Dijkstra et al. (2005) and the findings of Joo, Wilbur and Zhu (2016). Dijkstra et al. (2005) suggest two dimensions instrumental in processing information, namely modality and control over information processing. Within the dimension control, delivery and retrieval media can be distinguished based on pacing (Van Raaij 1998). Delivery media use external pacing as the advertiser controls the speed and order of information transfer, whereas retrieval media use internal pacing and allows consumers to process information in their own pace and sequence. Joo, Wilbur and Zhu (2016) also studied the cross-channel effect of offline marketing on online search behavior and found a significant relationship between TV advertising and consumers’ tendency to search for brand-related keywords instead of generic category-brand-related keywords. Their offline customer touchpoint, TV advertising, is a delivery media, whereas our touchpoint is a retrieval media. The opposing media types might explain the opposing findings of the offline touchpoint increasing branded search query use for delivery media and generic search query use for retrieval media.
Dagger 2013). In addition to the direct effect on purchase, direct mailing also indirectly influences purchase through influencing an earlier stage in the consumers’ journey to purchase, which consequently influence purchase.
The results show that the earlier stages in the consumers’ journey to purchase – generic search query use, branded search query use and clicks on sponsored search ads – influence purchase, which is in line with prior research (e.g., Anderl et al. 2016; De Haan, Wiesel and Pauwels 2016). In the model without including the timing effect, generic and branded search query use significantly influence purchase. Generic search query use negatively influences purchase, which implies that a higher generic search query use lowers the chance of a purchase. Branded search query use positively influences purchase, which implies that a higher branded search query use increases the chance of a purchase. This finding is in line with our expectations and the suggestion by Agarwal, Hosanagar and Smith (2012) that the keywords used by consumers can potentially reflect expertise. A provided reason for this effect is that consumers are less informed and tend to search less when using generic keywords, whereas consumers are more informed and tend to search more for specific keywords (White and Morris 2007; White, Dumais and Teevan 2009). In relation to direct mailing, this model shows that direct mailing indirectly leads to less sales by increasing the generic search query use and decreasing branded search query use, which consequently lead to less sales based on the relation among the stages. In the model with timing effect, branded search query use and clicks on sponsored search ads significantly influence purchase. Although the effect size is smaller, branded search query use also positively influences purchase in this model, which implies that a higher branded search query use increases the chance of a purchase. Clicks on sponsored search ads negatively influences purchase, which implies that a higher click volume on sponsored search ads decreases the chance of a purchase. In relation to direct mailing, also this model shows that direct mail indirectly leads to less sales by negatively influencing branded search query use, which consequently positively influences purchase.
search query use. With regard to the consumer responses in the consumers’ journey to purchase, these findings suggest that more use of branded search queries increases sales, but also increases generic search query use, which consequently decreases sales although to a smaller extent. On the other hand, more use of generic search queries decreases sales, but also increases branded search query use, which consequently increases sales.
The generic search query use and clicks on sponsored search ads positively influence each other in both models, which is in line with the results of Anderl et al. (2016). The finding implies that a higher click volume on sponsored search ads positively influence the use of generic search queries and there is a high chance that generic search queries follow up clicks on sponsored search ads. Moreover, a higher generic search use positively influence the click volume on sponsored search ads and there is a high chance that clicks on sponsored search ads follow up generic search queries. Also, click volume on sponsored search ads negatively influences purchase, but only in the model including the timing effect. This result implies that a higher click volume on sponsored search ads decrease the chance of a purchase. With regard to the consumer responses in the consumers’ journey to purchase, these findings suggest that more use of generic search queries increase the click volume on sponsored search ads, which consequently decreases the sales. Furthermore, a higher click volume on sponsored search ads increase the use of generic search queries, which consequently also decrease sales.
Limitations and future research
We acknowledge some limitations of our study. However, these provide interesting future research opportunities.
Although we find that direct mail significantly affects important consumer responses in the different stages of the consumers’ journey to purchase, our control group was relatively small. Despite the fact that Hutchins et al. (2015) showed that the effect sizes were equivalent regardless the size of the control group (for effective public health practice in communities), the results could have been influenced by the control group size as Coe (2002) suggests that the control group size could influence the estimation of effect size. Future research should ideally account for this by using a relatively larger control group and thereby provide more robust results.
mail to explore how the causal relations work and check how the specific channel affects the outcomes.
REFERENCES
Abhishek, Vibhanshu, Peter Fader and Kartik Hosanagar (2012), “Media Exposure through the Funnel: A Model of Multi-Stage Attribution,” Retrieved from
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2158421.
Agarwal, Ashish, Kartik Hosanagar and Michael D. Smith (2011), “Location, location, location: An analysis of profitability of position in online advertising markets,”
Journal of Marketing Research, 48, 1057-1073.
Agarwal, Ashish, Kartik Hosanagar and Michael D. Smith (2012), “Do Organic Results Help or Hurt Sponsored Search Performance,” working paper, University of Texas at Austin.
American Marketing Association (2015), “Dictionary”. (accessed August 2, 2017) [Available at: https://www.ama.org/resources/Pages/Dictionary.aspx].
Anderl, Eva, Ingo Becker, Florian von Wangenheim and Jan Hendrik Schumann (2016), “Mapping the customer journey: Lessons learned from graph-based online attribution modeling,” International Journal of Research in Marketing, 33(3), 457-474.
Ansari, Asim and Carl F. Mela (2003), “E-customization,” Journal of Marketing Research, 40(2), 131-145.
Ansari, Asim, Carl F. Mela and Scott A. Neslin (2008), “Customers Channel Migration,”
Journal of Marketing Research, 45(February), 60-76.
Baxendale, Shane, Emma, K. MacDonald, and Hugh, N. Wilson (2015), “The Impact of Different Touchpoints on Brand Consideration,” Journal of Retailing, 91(2), 235-253. Bawa, Kapil and Robert W. Shoemaker (1987), “The Effects of a Direct Mail Coupon on
Brand Choice Behavior,” Journal of Marketing Research, 24 (November), 370–376. Beirne, Mike (2008), “Direct Mail Still Effective for Marketers,” Brandweek, 20 (June). Blake, Thomas, Chris Nosko and Steven Tadelis (2015), “Consumer heterogeneity and paid
search effectiveness: A large-scale field experiment,” Econometrica, 83(1), 155-174. Coe, Robert (2002), “It’s the effect size stupid: what effect size is and why it is important,”
Paper presented at the Annual Conference of the British Educational Research Association, University of Exeter, England (September 12-14).
Danaher, Peter, J. and Tracey S. Dagger (2013), “Comparing the Relative Effectiveness of Advertising Channels: A Case Study of a Multimedia Blitz Campaign,” Journal of
Marketing Research, 50(4), 517-534.
Dijkstra, Majorie, Heidi E. J. J. M. Buijtels and W. F. van Raaij (2005), “Separate and Joint Effects of Medium Type on Consumer Responses: A Comparison of Television, Print, and the Internet,” Journal of Business Research, 51(10), 527-545.
DMA (2015), Statistical fact book. New York: Direct Marketing Association.
De Haan, Evert, Thorsten Wiesel and Koen Pauwels (2016), “The effectiveness of different forms of online advertising for purchase conversion in a multiple-channel attribution framework,” International Journal of Research in Marketing, 33(3), 491-507.
DeKimpe, Marnik. G., and Dominique M. Hanssens (2007), “Advertising response models,” in Handbook of advertising, G.J. Tellis and T. Ambler (Eds.). Londen: Sage.
East, R. (2003), “Wearout, Carryover Effects and Decay of Advertising,” in The Effect of
Advertising and Display. Boston, MA: Springer.
Feld, Sebastian, Heiko Frenzen, Manfred Krafft, Kay Peters and Peter C. Verhoef (2013), “The effects of mailing design characteristics on direct mail compaign performance,”
International Journal of Research in Marketing, 30(2), 143-159.
Gázquez-Abad, Juan Carlos, Marie Hélène De Canniére, Francisco J. Martínez-López (2011), “Dynamics of Customer Response to Promotional and Relational Direct Mailings from an Apparel Retailer: The Moderating Role of Relationship Strength,” Journal of
Retailing, 87(2), 166-181.
Ghose, Anindya, and Vilma Todri (2016), “Towards a Digital Attribution Model: Measuring the Impact of Display Advertising on Online Consumer Behavior,” MIS Quarterly, 40(4), 889-910.
Ghose, Anindya and Sha Yang (2009), “An Empirical Analysis of Search Engine Advertising: Sponsored Search in Electronic Markets,” Management Science, 55(10), 1605–22. Greene, William H. (2002). Econometrical Analysis. Upper Saddle River, N.J.: Prentice Hall. Hausman, Jerry (1978), “Specification Tests in Econometrics,” Econometrica, 46(6),
1251-1271.
Hubspot (2017), “The ultimate list of marketing statistics”. (accessed February 10, 2017) [Available at https://www.hubspot.com/marketing-statistics].
Hughes, Arthur M. (2012), “Why Email Marketing is King,” Harvard Business Review, (August 21). (accessed January 5, 2017) [Available at https://hbr.org/2012/08/why-email-marketing-is-king].
of immunization effectiveness studies,” Journal of Comparative Effectiveness
Research, 4(3), 1-12.
Jansen, B. J. and M. Resnick (2006), “An Examination of Searchers’ Perceptions of
Nonsponsored and Sponsored Links During Ecommerce Web Searching,” J. Am. Soc.
For Information Science and Technology, 1949-1961.
Jansen, B. J. (2007), “The Comparative Effectiveness of Sponsored and Non-Sponsored Links for Web E-commerce Queries,” ACM Trans. Web, 1(1), article 3.
Jerath, Kinshuk, Liye Ma and Young-Hoon Park (2014), “Consumer Click Behavior at a Search Engine: The Role of Keyword Popularity,” Journal of Marketing Research, LI, 480-486.
Joo, Mingyu, Kenneth C. Wilbur and Yi Zhu (2016), “Effects of TV Advertising on Keyword Search,” International Journal of Research in Marketing, 33(3), 508-523.
Kannan, P. K., Werner Reinartz and Peter C. Verhoef (2016), “The path to purchase and attribution modeling: Introduction to special section,” International Journal of
Research in Marketing, 33(3), 449-456.
Krafft, Manfred, Jürgen Hesse, Jürgen Höfling, Kay Peters and Diane Rinas eds. (2007), Berlin: Springer.
Leeflang, P. S. H., Wieringa, J. E., Bijmolt, T. H. A. and Pauwels, K. H. (2015). Modeling
Markets. New York: Springer
Lehmann, Donald R. (2004), “Metrics for making marketing matter,” Journal of Marketing, 68(4), 73-75.
Lemon, Katherine N., and Peter C. Verhoef (2016), “Understanding Customer Experience and the Customer Journey,” Journal of Marketing, 80(Special Issue), 1-62.
Li, Hongshuang (Alice) and P.K. Kannan (2014), “Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment,”
Journal of Marketing Research, 51(1), 40-56.
Lobschat, Lara, Ernst C. Osinga, and Werner Reinartz (2017), “What Happens Online Stays Online? – Segment-Specific Online and Offline Effects of Banner Advertisements,”
Journal of Marketing Research, forthcoming.
Mahajan, V., E. Muller and S. Sharma (1984), “An empirical comparison of awareness forecasting models for new product introduction,” Marketing Science, 3, 99-118. Marketing Science Institute (2016), “Research Priorities 2016-2018.” Cambridge, Mass.:
Martin, Brett A. S., Joël van Durme, Mika Raulas, and Marko Merisavo (2003), “Email Advertising: Exploratory Insights from Finland,” Journal of Advertising Research, 43(3), 293-300.
Martin, Andrew (2009), “The Long Road to Conversion: The Digital Purchase Funnel,” white paper, Microsoft’s Atlas Institute.
McElroy, M. B. (1977), “Goodness of Fit for Seemingly Unrelated Regressions,” Journal of
Econometrics, 6, 381-387.
McKinsey (2015), “Global Media Report 2015”. (accessed June 20, 2017)[available at http://www.mckinsey.com/industries/media-and-entertainment/our-insights/2015-global-media-report].
Millward Brown (2009), “Using Neuroscience to Understand the Role of Direct Mail”. (accessed May 22, 2017)[available at
http://www.millwardbrown.com/Libraries/MB_Case_Studies_Downloads/MillwardBr own_CaseStudy_Neuroscience.sflb.ashx].
Moffett, Tina, Mary Pilecki, and Rebecca McAdams (2014), “The forrester wave: Cross-channel attribution providers”, Q4 2014. (accessed February 1, 2017) [Available at https://www.forrester.com/report/The+Forrester+Wave+CrossChannel+Attribution+Pr oviders+Q4+2014/-/E-RES115221].
Montgomery, A. L., S. Li, K. Srinivasan and J. C. Liechty (2004), “Modeling Online Browsing and Path Analysis Using Clickstream Data,” Marketing Science, 23(4), 579-595.
Mulpuru, Sucharita, Shar VanBoskirk, Patti Freeman Evans and Doug Roberge (2011), “The Purchase Path of Online Buyers: A Joint Holiday Shopping Study with GSI
Commerce of 15 Retail Websites,” research report, Forrester Research. Naik, Prasad A. and Kalyan Raman (2003), “Understanding the Impact of Synergy in
Multimedia Communications,” Journal of Marketing Research, XL, 375-388.
Peppers, Don and Martha Rogers (1999), Enterprise One-to-One: Tools for Competing in the
Interactive Age. New York: Doubleday.
Prins, Remco and Peter C. Verhoef (2007), “Marketing Communication Drivers of Adoption Timing of a New E-Service Among Existing Customers,” Journal of Marketing, 71(April), 169-183.
