• No results found

CUSTOMER INFORMATION ORIENTATION AND RETENTION

N/A
N/A
Protected

Academic year: 2021

Share "CUSTOMER INFORMATION ORIENTATION AND RETENTION"

Copied!
66
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)
(2)
(3)
(4)

4

Management Summary

Retention is crucial in the survival of firms nowadays (Al-Jader, 2015). As a consequence, researchers’ attention increased towards the retention of customers. In many studies customer satisfaction is the predominant metric for identifying churning or retaining customers (Ahn, Han & Lee, 2006; Capraro, Broniarczyk, Srivastava, 2003). However, can switching behavior be controlled by only focusing on the satisfaction of customers? The digital world we are living in changes perspectives. Following Holtrop et al. (2017) transparency in market information makes customers more aware of switching opportunities. Furthermore, the digitalized market results in more channels available for information obtaining (Neslin et al., 2006). Therefore, the understanding of factors influencing customer retention will be broadened in this research. In contrast to previous research, this research focuses on the information search of consumers on different channels as a predictor for customer retention. Furthermore, the influence of satisfaction on the relationship between the extend of using information channels and retention is investigated.

(5)

5 hours. Looking at the dissatisfied customers, it only resulted in significant results for expert orientation. Indicating that information search is not decisive for dissatisfied customers.

With this study companies can predict the customers that are likely to retain based on the information orientation behavior of customers. The results of this study show that information orientation of customers negatively affects customer retention. In order to reduce the information search of customers on alternative providers, companies should try to keep their customers in their own environment. Therefore, managers are advised to proactively guide customers in their information search on insurance providers. Furthermore, the results of this study show that satisfied customers are sensitive for information about alternative providers, which resulted in an even stronger negative relation with retention in some cases. This indicates for Dutch health insurance providers that satisfaction is not a sufficient condition for customer retention. The findings of this study should be viewed as a first step towards the understanding of the role of different information channels on retention behavior. Therefore, the conceptual framework could be extended. Moreover, this study only investigated the process of consumers in the health insurance industry in the Netherlands. In order to make replication in other contexts possible, this research could be extended to assess generalizability in other industries and countries.

Abstract

In this study the influence of the information search behavior of customers across different information channels on retention in the context of Dutch health insurance providers is examined. Furthermore, the influence of satisfaction on the relationship between the extend of using information channels and retention is investigated. Logistic regression indicated that the search for information makes customers less likely to retain, especially when people search on online and printed information channels. Furthermore, when customers are satisfied it makes the negative relationship between information search and retention stronger. These results indicate that information search behavior in the Dutch Health insurance industry are important drivers of switching behavior and should be considered by firms.

Keywords: customer retention, information search, information channels, logistic regression, health

(6)
(7)
(8)
(9)

9 contact, which might also weaken the loyalty of customers (Neslin et al., 2006). To the best of our knowledge, despite the important consequences that go along with the digitalization, no research is done on the effect of different channels on retention behavior. Customer churning behavior is a big issue in the services industry (Ahn, Han & Lee, 2006). Moreover, following Neslin et al. (2006) the health industry is one of the areas which faces significant problems with churning customers. Therefore, this study will focus on health insurances in the Netherlands. The competition in the health insurance industry in the Netherlands is increasing (Maarse et al., 2016) and switching behavior of customers is increasing from 3,5 % in 2007 to 5,5 % in 2016 (Vektis, 2016). Insurance companies should therefore pay attention to keep their customers and to achieve a long term relationship (JaJaee & Ahmad, 2012).

The aim of this study is to investigate the influence of the searching behavior of customers on retention. More interesting, what influence does the extend of using online, printed, network and expert information channels have on the retention behavior of customers. Furthermore, the influence of satisfaction on the relationship between the extend of using information channels and retention will be investigated. Customers which are satisfied could search for information for self-support and confirmation of their choice (Dholakia et al., 2010), which could weaken the studied relationships. The main question of this study is therefore:

What is the relation between the extend of searching behavior of customers across different information-channels and retention and what influence does satisfaction have on that relationship?

This question is divided in four sub questions, namely:

- What influence does information search of consumers have on the retention behavior of customers?

- What is the effect of the extend of using online, printed, network and expert channels on retention behavior?

- What effect does the use of multiple channels have on the retention behavior of customers? - What influence does satisfaction have on the relationship of information search and

(10)

10 In designing a strategy on the basis of this multichannel environment of information search, consumer heterogeneity should also be taken into account (Konus et al., 2008). Therefore, in addition to the examination of the extend of use of information search on different channels in relation with retention, this study will control for the customer characteristics and their relational variables.

The additional value of this research is the provision of information about the channels that drive retention of customers in the information searching process. A lot of research is done on the effect of channels used for advertising on purchase intentions (Ansari, Mela & Neslin, 2008; Neslin & Shankar, 2009; Payne & Frow, 2004). However, to the best of our knowledge, no research is done on the effect of the use of different information orientation channels on retention behavior. Furthermore, the influence of satisfaction is often used as a direct relationship with retention. However, in this study it is assumed that only managing satisfaction is not enough to retain their customers nowadays, since customers are more empowered and have more options (Simintiras et al., 2015). Therefore, in this research the moderating role of satisfaction and dissatisfaction is examined. In this research it is assumed that satisfied customers could have biased interpretations on the information gathering, which makes them more likely to retain.

The rise of channels which are available for informing and interacting with customers (Neslin et al., 2006) along with the increase in information search, leads to a new challenge that marketers face. Marketers need to understand consumers’ behavior (Konus et al., 2008) and determine the distribution of expenditures across channels (Rangaswamy & Van Bruggen, 2005). In order to get a holistic view on the behavior of the individual customer across the different channels, four different types of research data will be used to investigate the research question, including logfile data, survey data, demographical data and offline orientation data. The combination of attitudinal and behavioral data provides more value to the firm, in the way that it provides a richer context (Breur, 2011). From a managerial perspective, the outcomes of this study will provide valuable insights for health insurance providers in allocating marketing resources across channels. Which information channels should health insurance providers use in order to retain customers? Furthermore, with the insights of the role of satisfaction on this relationship, managers can proactively try to manage satisfaction.

(11)

11 results will be elaborated in chapter 5. Finally, a discussion including managerial implications, limitations and suggestions for further research will be given in the last chapter.

2. Theoretical Framework

(12)
(13)

13

2.3 Retention

Many service providers have shifted their focus from strategic customer acquisition to customer retention (Venkatesan & Kumar, 2004). Following Reinartz & Kumar (2014) it is more profitable to satisfy and retain customers instead of continuously attracting new customers. One reason that retention is more profitable following Reichheld & Sasser (1990) is that retaining customers realize higher revenues and higher margins than new customers. Thus, when a company loses a customer it will affect the revenue (Tamaddoni et al., 2016). Furthermore, the costs of acquiring a new customer, can substantially exceed the costs of retaining a customer (Fornell & Wernerfelt, 1987). Moreover, long term customers accommodate more referrals for the company, more than recently acquired customers (East et al., 2006). Thus, not only the revenue of the churning customer will be lost, also the referrals from that customer will be lost (East et al., 2006; Ganesh, Arnold & Reynolds, 2000).

Gupta, Lehmann & Stuart (2004) found that when a company is able to make a 1 % improvement in retaining customers, this will lead to a 5 % increase in firm value. Therefore, the retention behavior of customers will be investigated in this study. In many studies retention is measured in terms of intention. However, following Bolton (1998) there is a strong difference between intention and the real behavior. Therefore, the real retention behavior of customers is measured in this study.

2.4 Information orientation

The searching behavior of consumers is useful to investigate in order to understand behaviors and choices of customers (Maity, Dass & Malhotra, 2014). For consumers who are looking for a health insurance provider, the information seeking and decision making process is complex (Schram & Sonnemans, 2011). Hence, the channels available for firms for interacting with consumers are increasing (Neslin et al., 2006). A channel can be defined as: ‘a medium through which a firm can

contact and interact with their current and potential customers’ (Neslin et al., 2006). The use of

multiple channels for searching for information has increased by customers and firms (Konuş, Neslin & Verhoef, 2014). The purchasing and switching behaviors have become more complicated through this growing variety of channels (Chiu et al., 2011). For instance, with the introduction of the internet, an important consequence is that firms need to provide consumers with more information about healthcare policies (Wendel & Dellaert, 2005). Consumers have more power through the information availability on the internet (Schoenbachler & Gordon, 2002), because they can more easily compare offerings and prices of service providers.

(14)

14 Because customers do not want to be disappointed by their insurance provider, many consumers will engage in a search for alternative products or services before making a decision about the purchase (Kolesar & Galbraith, 2000). Information search is required in order to make a purchase decision following Choi & Park (2006). Especially in case of the purchase decision of health insurance provider, because it is complex (Hibbard & Peters, 2003). Following Boonen et al. (2016) customers who are actively searching for information are more likely to churn with their current health insurer. Moreover, Boonen & Schut (2011) suggests that consumers are less willing to churn if they are not looking for information or when there is no information available. When no or less information is available on alternative health insurance providers, it makes customers more likely to remain at the current health insurance provider (Jones et al., 2000). Without orientation and information seeking, insurance customers cannot make a choice for switching to another provider (Van der Maat & De Jong, 2008). Thus, the intensity of information search will have a negative effect on retention behavior of customers.

It is important to understand the role of every information channel of the company in order to have an optimal allocation of the expenditures and efforts per channel (Strebel, Erdem & Swait, 2004). This optimal allocation is important, because it is essential to keep the interest of customer by using the right channels and media for communicating with customers (Kumar, 2010). It always has been critical for firms to make decisions in terms of communication media (Ratchford, Talukdar & Lee, 2001). Consumers rely on different information channels, like: online channels, printed channels, experts of health insurance providers and on their network. All these different channels will be addressed in the next sections.

2.5 Online orientation channels

The internet is critical for consumers in order to obtain information about services (Elliott et al., 2012). The online and mobile advertising channels are increasing (Ailawadi et al., 2009). Consequently, many firms are increasing their marketing budget for online channels (Kireyev et al., 2016). Furthermore, the internet is used a lot by consumers in order to search for specific product or service information (Kuruzovich et al., 2008; Ratchford, Talukdar & Lee, 2001), because it allows customers to access information with minimal effort (Park, Chung & Yoo, 2009). The internet has changed decision making processes completely for customers (Goetzinger et al., 2007): customers can exert more control over their orientation process for purchases (Van der Veen & Ossenbruggen, 2015).

(15)

15 in order to compare different products, which in turn has leaded to an increase in information intermediaries (Kuruzovich et al., 2008). Such information intermediaries are also present in the health insurance system in the Netherlands, like: independer.nl and kiesbeter.nl. Furthermore, social media is an increasing popular information source (Xiang & Gretzel, 2010). Social media contains both firm generated and user generated content (Kumar et al, 2016), which increases consumer insights about products and services in terms of perceptions of other consumers and in terms of firm generated information (Mayzlin & Yoganarasimhin, 2012).

Shim et al. (2001) suggest that the online information search of customers can be used as a key predictor of intention to purchase. In this study, it is expected that due the ease of obtaining information and the richness of the information, the intensity of online search will decrease the likelihood of a customer to remain at the current health insurance provider. Intensity of search behavior online is used, since Kolstad & Chernew (2009) suggest that a higher intensity of search will increase the likelihood to churn. The intensity of the online search will be tested with two dimensions: the time spend on searching on websites and number of times the customer used online sources to search for information. In this study online channels consist of website orientation, social media orientation and comparison sites. The following hypothesis is defined: H1: The higher the extend of consumers’ search for information online, the lower the likelihood the customer will retain.

2.6 Printed orientation

Following Hesselink, Henneman & Timmermans (2009) printed media like flyers and newspapers are frequently used in order to acquire information about health insurers. Following Franke, Huhmann & Mothersbaugh (2004) the information of printed media has a positive effect on buying behavior of a customer. However, printed channels are often seen as information rich (Danaher & Rossiter, 2011). Also, Choi & Park (2006) suggest that printed media contain a lot of information. Due to this amount of information it can result in more knowledge of customers, which in turn can lead to a higher likelihood of defection following Capraro, Broniarczyk & Srivastava (2003). However, some studies suggest that due the rise of internet, the consumption of newspapers and other printed media is reduced (Lipowski, 2015). Since printed media has a rich information content, which can result in more knowledgeable customers, it is assumed that the extend of information search on printed channels will have a negative impact on the retention behavior of customers.

H2: The higher the extend of consumers’ search for information on printed channels, the lower the

(16)

16

2.7 Expert orientation

Next to printed media and online media, people may rely on advice of experts or health insurance providers (Kolesar & Galbraith, 2000). When an advisor has more expertise about the insurance providers than the consumer, consumers tend to rely on these experts (Van Swol & Sniezek, 2005). Lee et al. (2017) suggest that face to face contact with experts can have several benefits in the information searching process, like personal focus and immediate feedback. Following Brehmer & Hagafors (1986) expert advice decreases the complexity of decision making. If people are contacting experts and spend time with these experts in order to make decisions, we assume this will have a negative effect on retention. Therefore, the following hypothesis is defined: H3: The higher the extend of consumers’ search for information via experts, the lower the likelihood the customer will retain.

2.8 Network information orientation

Following Van der Veen & Van Ossenbruggen (2015) consumers can obtain information by themselves or by relying on information of other consumers. Consumers which interact with people in their network, like friends and family, can also have an influence on retaining and churning behavior of customers (Nitzan & Libai, 2011). This social influence is defined in this study as the network orientation, which is defined as follows: the variety of interactions a customer has with people in their

network, like friends and family, in order to obtain information about alternative insurance providers.

(17)

17

2.9 Number of channels used

To date, there is an expanding multiplicity of channels through which firms and consumers can communicate and provide information (Valentini, Montaguti & Neslin, 2011). In the context of information search, people often use different media following the channel complementarity theory of Dutta-Bergman (2004). Following this theory, the information of one channel can be filled with the information of another channel, for instance: the information gap of the newspaper can be filled with information of the internet. If consumers use different channels, it will likely induce a greater intensity of information search (Eliott, 2012). Following Ratchford, Lee & Talukdar (2003) the number of channels used for information search can be used as an indicator of intensity of gathering information. With a higher intensity of information search, customers will have more knowledge about the alternative services. Following Carpraro et al. (2003) a customer with more knowledge about the service and alternative service provider, will be less likely to retain with their current provider. Therefore, the following hypothesis is defined: H5: The higher the number of channels consumers’ use for information orientation, the lower the likelihood the customer will retain.

2.10 Satisfaction

Before consumption, customers do have expectations about a certain service and evaluate performance during or after the usage of that specific service (Zhao et al., 2012). Customers are dissatisfied when the service delivery does not match the expectation of the customer (Zeelenberg & Pieters, 2004). However, customers can still be dissatisfied even if the expectations are met. Following Taylor (1997) the fact that another alternative service provider could have offered a better service to the customers can also dissatisfy customers. Research indicates that a customers’ satisfaction level of the current provider can influence the evaluation of an alternative provider in the information orientation process (Dick & Basu, 1994). Satisfied customers are more likely to have biased interpretations about the information obtained about the alternatives (Capraro, Broniaczyk & Srivastave, 2003). Furthermore, Dholakia et al. (2010) suggests that people who are satisfied can search for information in order to confirm and support their previous choice. In turn, this indicates that the level of satisfaction can weaken the assumed negative relationship between searching for information on the different channels and retention.

(18)

18 Determinants for satisfaction are all the encounters and experiences of the customer with the service provider following Sureshchandar et al. (2002). Customer satisfaction is multidimensional in nature (Zhao et al., 2012) and could therefore not be measured on a one item scale. This could be demonstrated with the research of Cronin & Taylor (1992) who used a single item scale in order to measure customer (dis)satisfaction. However, their approach failed in researching, because it did not take into account the richness of the construct (dis)satisfaction (Sureshchandar et al., 2002). Also Oliver (1993) suggests that satisfaction should be attribute based, because features of services like health insurance companies are complex. Performance measures of the service could be used is order to measure satisfaction (Swan & Combs, 1976). After all, if the performance of the health insurance is above the expectation, the customer will be satisfied. Studies have shown that expectations are related to performance of services (Swan & Combs, 1976). Therefore, in this study satisfaction will be considered as a multidimensional construct, where the perceived performance of the health insurer will be used in order to determine satisfaction of the service. In this study three different performance indicators will be measured in order to assess satisfaction, namely: perceived service quality, perceived service costs and perceived choice freedom of healthcare provider.

Perceived service quality. Following Anderson, Fornell & Lehmann (1994) satisfaction and

dissatisfaction are influenced by the perceived quality. Furthermore, in the health insurance context, service quality is an important reason for customers to churn (Thomson et al., 2013). Also following Jajaee & Ahmad (2012) high quality service perceptions lead to satisfied customers. However, following Taylor & Baker (1994) service quality and customer satisfaction should be measured distinct from each other. Hence, Parasuraman, Zeithaml & Berry (1985) argue that service quality is an important performance indicator for customer satisfaction. Therefore, service quality will be used as one of the constructs of customer satisfaction. Service Quality can be defined as: “how well a service provider meets the servive delivery expectations of the customer on a consistent basis” (Parasuraman, Zeithaml & Berry., 1985). Perceived service cost. Moreover, following Boonen et al. (2016) many of the churning customers of

(19)

19

Perceived choice freedom of healthcare provider. Following Duimelinck et al. (2015) the quality of

contracted providers and the freedom to choose a hospital or drug is an important feature that customers use to evaluate the performance of a provider and thus is an important indicator of customer satisfaction of an insurance company. Furthermore, Reitsma- van Rooijen et al. (2015) suggest that freedom of choice of healthcare provider is an important performance measure for customers of health insurance providers in The Netherlands. Therefore, perceived degree of freedom of healthcare provider will be used as a construct of customer satisfaction.

In summary, this study will use satisfaction as a multidimensional construct, where satisfaction can be described as: the difference in expectation and actual experience of the service quality, costs and perceived freedom of healthcare provider. In this study, it is assumed that satisfaction will have a moderating effect on the relationship of information search of different channels and retention, since research suggest that satisfied consumers have biased interpretations of the information obtained (Dick & Basu, 1994; Capraro, Broniarczyk & Stravastava, 2003). Furthermore, the direct effect of satisfaction on retention will be investigated, since research suggest that satisfaction enhances loyalty which indirectly leads to retention (Gustafsson, Johnson & Roos, 2005). Therefore, the following hypotheses are defined: H6: When consumers are satisfied, it will weaken the negative relationship between retention and the extend of searching in online channels (a), printed channels (b), experts (c) &networks (d) and the number of channels used to obtain information (e). H7: When consumers are satisfied, it will have a positive influence on the likelihood of a customer to retain.

2.11 Control Variables

In this study there will be controlled for consumers’ characteristics and relational variables of customers in order to reduce the unexplained variance in the model. Therefore, in this study it is assumed that the retention behavior could be affected due to the heterogeneity of customers.

Age. Hibbard & Peters (2003) found that younger consumers are better in searching information than

(20)

20

Education. Following Boonen et al. (2016) higher educated people are less likely to retain.

Furthermore, according to several researches, consumers’ information searching behavior and decision making behavior differ in educational background. Also Boonen et al. (2016) suggest that people who are higher educated are more involved in information search than lower educated people. This could infer that higher educated people are more likely to search for alternative health insurance providers which in turn could lead to lower retention probabilities. Therefore, there will be controlled for the education level of the consumers. Income. Konuş et al. (2008) suggest that consumers with a higher income will engage in more extensive search behavior. This could infer that people are in turn less likely to retain. Moreover, Peng & Wang (2006) found that consumers with a higher income are more likely to switch. Therefore, income will be taken into account in this research as a control variable. Discount due group contract. In the Netherlands, insurers are allowed to give a discount up to 10% to customers which belong to a specific group, like employees of a certain organization (Duijmelinck et al., 2015). In this study it is expected that consumers will perceive the switching to another company as a loss of discount, which would positively affect customer retention. Therefore, discount due to a group contract will be taken into account in the model as a control variable.

(21)

21 The survey dataset contains the searching and transaction behaviors of 6.445 consumers on health insurance providers, including 63 possible health insurance providers with a time span of measurement from the 1st of October 2013 until 31st of January 2014, resulting in 21.324 data points. This time span is suitable for analyzing switching behavior and information searching behavior, since a consumer in the Netherlands is allowed to switch to another provider of health insurer at the end of every year (Maarse et al., 2016). Furthermore, as indicated in section 2.1, customers are actively searching in the last 3 months of the year for information on health insurance providers.

In addition to the purchase behavior of the respondents, a dataset including all the online log files on comparison websites, insurance companies’ websites and the websites of intermediaries is used. In the observation time of the datasets, all the visits on websites used for orientation of individual respondents are logged. This dataset contains the websites visited for the orientation, the duration of the websites visited and the date of websites visited for each individual respondent. This online orientation data containing 65.535 data points, will be used to investigate the intensity of online searching behavior of the customer. Next to these datasets, a dataset containing the offline searching behavior of individual customers is used. The set contains 16.843 data points, capturing the time and type of offline searching behavior. This dataset will be used to investigate searching behavior on printed, expert and network channels. In order to control for demographics and background variables of the customers, a dataset containing all the demographics and background characteristics, like age, gender and income is used. This dataset contains 6.445 data points.

3.2 Data cleaning

Before merging the datasets, the different datasets are screened on outliers, missing values and oddities. In the log file dataset 13 rows contained a website visit, but did not contain a time duration on the website, which indicates that the customer immediately left the website. Since these URL visits do not contain a time duration, it is assumed that they did not visit the website on purpose. These rows do not contain any information about the online searching behavior of customers and are therefore deleted from the log file dataset.

(22)

22 the predictive mean matching method of hot-deck imputation is applied using a prediction matrix in combination with the package MICE (Van Buuren & Groothuis-Oudshoorn, 2011). With this method the missing values are replaced by predicting the missing value using the data that is available. The available records that are most similar to the missing record is imputed with that particular value. The advantage of this method is that no data will be lost and the accuracy is very high (Winkler & McCarthy, 2005). Following Roth (1994) hot deck imputation can be used for all sorts of missing scenarios, including missing not at random. However, the MNAR should not be greater than 10%. In this case the missing values contain 8.2 % of the households. Therefore, hot deck imputation can be applied.

(23)

23 As shown in this figure, in three of the available datasets there are multiple, but different levels, per individual household level. For instance, for every single information search occasion, a new row is created for the individual household level with the data, time and duration. For this research no time dimension is needed, since the primary focus of this research is to investigate whether different orientation channels have an effect on the retention behavior of customers. In order to make the data sets more suitable for combining, the data is aggregated on household level and the date variables are removed. First, dummy variables per orientation channel are created, in order to make aggregation suitable. The data manipulations of each construct are further explained in paragraph 3.4:

measurement of constructs. Finally, after all the data manipulations the data is aggregated and

merged, resulting in one dataset consisting of 6.445 observations of 22 variables.

3.4 Measurement of constructs

To account for the explanatory variables in this research, some new variables are created in the combined dataset. Furthermore, in the conceptual model there are two constructs that are multidimensional in nature, these constructs will not be measured ex ante. Principal component analysis (PCA) will be used in order to reduce multicollinearity in the subsequent analysis. After the PCA, the variables were formed by summing the items and dividing it by the number of items. For all used measures for the constructs and their scales, a table is represented in Appendix A. The operationalization of all constructs will be discussed in the upcoming paragraphs.

3.4.1 Customer Retention

Customers who stay with their current health insurance provider in 2014 are classified as retaining customers. Customers going to another provider under the same overarching organization, are classified as switchers in this study. A new dummy variable is created for this this variable, where active and passive retainers are indicated as 1 and switching customers are indicated with a 0.

3.4.2 Level of Online Orientation

The intensity of the online channel search of customers will be measured with a multidimensional scale. The extend of online search for information is measured as the time spent gathering information (Ratchford, Lee & Talukdar, 2003) and the number of websites (Klein & Ford, 2003) used to acquire the information. The variable number of websites is created by summing up all the website visits of an individual, resulting in a continuous variable. The variable duration of website orientation is created by summing up the time spent per website visit for each individual and categorizing into the three categories. Since the two variables are not measured on the same scale, the continuous variable

(24)

24 in recoding, the classification of three categories are used. Since this construct exist of two variables, PCA analysis and Crohnbach’s Alpha reliability analysis is performed. The PCA analysis is suitable for level of online orientation with a KMO measure of .5 and a significant Bartlett’s Test of Sericite test, which is considered suitable (William et al., 2010). To preserve the four items as one construct, the items should have a minimum explained variance of .6 (William et al., 2010) and a minimum eigenvalue of 1 (Kaiser, 1960). With .92 explained variance and an eigenvalue of 1.8, which implies that one construct could be used for measuring the level of online orientation. The two items are reliable predictors with a Cronbach’s Coefficient of .769. Therefore, one item will be used for measuring the level of online orientation.

3.4.3 Level of Printed Orientation

A variable for the level of printed orientation is created by summing up the time spent per reading activity, like reading magazines, flyers and quotations. The variable in the original dataset was categorized into eight categories. To keep parsimoniousness and to be able to interpret the results in chapter 4 clearly, the categories are transformed into a three-way categorization, where 1 = 0-1 hours spend, 2 = 1 – 3 hours spend, 3 = more than three hours spend. This three-way categorization will be used for all variables which include categories in order to be consistent.

3.4.4 Level of Expert Orientation

For the level of expert orientation a new variable is created by summing up the time spent per call or visit with different advisors per individual. This variable is also categorized according to the three categories in order to keep consistency, where 1 = 0-1 hours spend, 2 = 1 – 3 hours spend, 3 = more than three hours spend.

3.4.5 Level of Network Orientation

For the level of network orientation, a categorical variable of eight levels is available in the dataset. This variable is also transformed into a variable including three categories. For this variable respondents could indicate if they exchanged information with their family, friends or other people in their network on their health insurance provider.

3.4.6 Number of Orientation Channels Used

Also for the number of orientation channels a new variable is created. First a dummy variable is created for the use of each orientation channel. A new variable is created in the dataset summing all the dummy variables of the online, printed, expert and network orientation channels.

(25)

25

3.4.7 Satisfaction

Since customer satisfaction is multidimensional in nature (Zhao et al., 2012), this construct is measured on a multidimensional scale. Satisfaction is measured using three performance indicators for health insurance in the Netherlands, namely: perceived service quality (Paraguayan, 1985), perceived service costs (Boonen et al., 2016) and perceived choice freedom of healthcare provider (Duijmelinck et al., 2015). For this construct four variables of the dataset are used:

-

I am not satisfied about the service provision of my current health insurer

-

I am not satisfied about the coverage and fees of my health insurer - The premium of my current health insurance provider is too high - I am not satisfied about the fact that my current insurance company does not contract for particular hospitals In order to test the reliability of the measurement of satisfaction, a PCA analysis and the Reliability analysis of Cronbach’s alpha is used. In PCA analysis a single criterion should not be used in order to assess the reliability of the factors (Costello & Osborne, 2005). Therefore, multiple criteria are assessed for reliability. The PCA analysis is considered suitable, since the KMO measure has a value of .790 and a significant Bartlett’s Test of Sericite. With .65 of explained variance and an eigenvalue of 2.6, the four items load on one factor, measuring the underlying construct satisfaction. The four variables are reliable predictors with a Cronbach’s Coefficient of .814. Therefore, one factor of four items will be used to measure satisfaction.

3.4.8 Control variables

(26)

26

3.5 Selective sampling

In the dataset 11.6 % switched to another insurance provider for 2014. Despite that the percentage is high in comparison with the switching rate of the Dutch population of 6.9 % (Vektis, 2014), it is a rather small event taking place for analysis. The 88.4% of respondents which retain, may dominate the statistical analysis, which may bias the predictions. In order to address this issue, a selective balanced sample is created where an equal number of switching respondents and retaining respondents is represented. The sample sizes of the original sample and the balanced sample are shown in table 1. Following Dunkers, Franses & Verhoef (2003) a selective sampling method could lead to more precise parameters. However, the intercept can be affected using selective sampling methods. In this study there is no interest in the constant. Therefore, no additional actions are taken for the possible changes in the constant. SUBGROUP ORIGINAL SAMPLE SIZE % OF ORIGINAL SAMPLE BALANCED SAMPLE SIZE % OF BALANCED SAMPLE RETENTION N = 5700 88.4 % N = 745 50 % CHURN N = 745 11.6 % N = 745 50 % Table 1: Sample size original sample and balanced sample.

3.6 Research Method

In this study the propensity to retain is modelled as a function of customer searching behavior and customer- and relational characteristics. Specifically, the role of the use of different information channels on the retention behavior of customers of insurance companies for their health insurance of 2014 are examined. The central question in this study is whether a respondent does retain, yes or no, based on their searching behavior. Furthermore, a distinction is made between satisfied and unsatisfied customers. A multilevel approach will enable the simultaneous comparison of relationship between variables across satisfied customers and unsatisfied customers (Raudenbush & Bryk, 1992). Since the dependent variable of this research is binary, two types of models can be used: a binary choice model (probit) or a binary logistic regression model (logit). In order to use the most appropriate model for this research, the methods of the models will be compared in paragraph 4.2.1: method

(27)

27 In both models it is assumed that a latent variable !" drives the decision of retention of the individual respondent #, where retention is denoted by $" = 1 and churning by $" = 0. And where (" is a vector of characteristics and * indicates a vector of regression parameters. This latent variable can be defined as follows:

!"= + + (′"* + ." [3.1]

(28)

28

3.7 Model specification

In this section the model specification is presented. The first two lines of the equation are the main model excluding interaction effects. In order to test the moderation effect of satisfaction, an additional interaction effect model will be estimated including the main model and the interactions. After performing the interaction model the model will be divided into two models: one model (1) based on only the satisfied customers and one model (2) based on only the not satisfied customers. The estimation of these different model enables the comparison of the difference between the impact of the behavior of satisfied and not satisfied customers (Van Birgelen et al., 2006).

!"= 6 + *AMM"+ *C3M +"*NOM"+ *PQM"+ *RQS!"+ *TUVW" +*

XVYZ"+ *[O\!"+ *]^QS"+ *A_\^US"

(29)

29

4. Results

In this section validity, reliability and estimates of the models will be addressed. First, the data will be analyzed exploratory. Subsequently, three models will be estimated in this section one main model (1) based on all customers in the balanced sample, one model (2) based on only the satisfied customers and one model (3) based on only the not satisfied customers.

4.1 Exploratory analysis

4.1.1 Descriptive statistics

In this section, the descriptive statistics of the balanced sample set are described containing information of 1.490 households. The average age of this balanced sample set is 50 years, with the lowest age of 18 and the highest age of 86. The income of the households is diffused in low income (39.7 %), medium income (39 %) and high income (21.3 %). Only 7.1 % of the households are part of a discount group. Of all retaining customers in the sample set, 52 % did not spend any time on information search for health insurance providers. Compared to 27 % of the switching customers which did not spend any time on information search, indicating that the customers who stay at their current provider are less inclined to search for additional information. The searching behavior of retaining and switching consumers is shown in figure 4. As can be seen in figure 4, the orientation via an expert is the least used channel. Of the retaining households 93.4 % did not spend any time on this information channel and of the switching consumers, 87.2 % did not spend any time with experts for information. Looking at the differences between satisfied and dissatisfied customers who retain, satisfied customers spend substantial less time on information orientation online and via their network. Furthermore, the dissatisfied customers which retained are more likely to spend time on information orientation on different channels. Looking at the churning customers, there are few differences in the number of channels used for orientation and no time spend for orientation. However, dissatisfied churning customers searched substantially more intense for information on online and expert channels.

(30)

30 The index in the frequency table in appendix B indicates the difference between searching behavior of retaining and churning consumers. Looking at these indexes, there is a substantial difference in searching behavior between the two groups. First, compared to the retaining customers, twice as much churning customers searched for information on at least one channel. Furthermore, there are substantial difference between the searching behavior of retaining and churning customers on all channels. Customers who did churn searched more on online, printed, network and expert channels for information. When looking at the demographics, lower educated consumers with a lower income are more likely to retain at their current health insurance. Vice versa, higher educated consumers and consumers with a higher income are less likely to retain. When looking at the churning customers, there are not much differences between satisfied and not satisfied customers in their searching behavior. In contrast, there are substantial differences between the retaining customers in their searching behavior. Satisfied retaining customers searched substantial less for information online and via their network. The difference between satisfied and non-satisfied customers, and in the retention and switching behavior of customers are presented in a frequency table per group which is presented in Appendix B.

4.1.2 Pearson’s correlation

To obtain some preliminary insights and in order to check correlations between the variables, a Pearson correlation matrix is created. The Pearson correlation coefficients are able to identify correlations between independent and dependent variables which might indicate potential multicollinearity issues. The correlation matrix is shown in table 2. Table 2: Pearson Correlation Matrix PRINT NETW ORK EXPE RT ONLINE CHAN N SATISFA CTION INC OME EDUCA TION

(31)

31 From the correlation matrix significant correlations are found. The variable number of channels used for orientation is significant moderately correlated with the four orientation channels. This is a plausible outcome, since the variable ‘number of channels used’ is based on the variables intensity

print orientation, intensity network orientation, intensity expert orientation and the intensity of online orientation using dummy coding. If these correlations will result in multicollinearity issues is further

(32)

32

The difference between the performance of the models is relatively small. The logit model has the highest score on the hit rate, the Pseudo dC values and the likelihood ratio. Furthermore, the logit model scores better on the AIC and the BIC criteria. Based on these criteria, the logit model performs better than the probit model. Furthermore, the logit model is more convenient for interpretation (Franses & Paap, 2001). Therefore, a logit method will be used for the upcoming analyses.

4.2.2 Model Choice

In order to choose the most appropriate model for this research, a comparison between different models is made. Four different models are compared. A null model including only a constant. Furthermore, the main model including all predictors and the control variables (model 1) is tested. Another model including only the predictors (model 2) is tested. Moreover, a new model will be created. Since the variable number of channels appears to be correlated with four of the predictors of retention, a model excluding this variable is created (model 3). For this comparison the hit rate, Pseudo dC measures, the likelihood ratio test and the information criteria are also used.

MODEL HIT RATE PSEUDO ef LIKELIHOOD

(33)

33 accuracy considerations of parsimoniousness and simplicity, model 3 would fit better. Furthermore, the variable number of channels, which is incorporated in model 1, has potential multicollinearity issues due to the high correlations. Since the two models have slightly different scores, model 3 will be used in the upcoming analysis.

4.3 Model assumptions

Before performing the model, the model needs to be tested on assumptions. However, since logistic regression will be performed, assumptions like homoscedasticity, linearity and normality are violated. The logistic regression makes use of the maximum likelihood estimation, which overcomes these assumptions (Menard, 2002). However, in logistic regression multicollinearity is an assumption which needs to be taken into account. Multicollinearity implies that predictor variables are highly correlated, which could lead to unreliable estimates (Leeflang et al., 2005). The Variance Inflation Factor (VIF) and the tolerance values are used in order to determine if multicollinearity is an issue in this model. The VIF values should not exceed the threshold of 5 (Menard, 2002). The test outcomes shown in Appendix C reveal that the VIF values do not exceed the threshold value of 5 and the tolerance values did exceed 0.2, which indicates that no multicollinearity issues are present in this model.

4.4 Model estimation of the main model

In this paragraph the main model is estimated. First, in order to be able to validate the model, the data will be randomly split into two samples: one estimation sample including 80 % of the data and one validation sample including 20 % of the data. The results of the estimation of the main model are presented in table 5. With an gC statistic of 171.07 (p-value <0.01) the likelihood ratio test indicates that the main model fits the data well. However, the Pseudo dC values are in comparison with similar studies relative low (Kumar & Venkatesan, 2005). After fitting the model, the estimated coefficients and their values can be interpreted.

4.4.1 Interpretation

(34)

34

The intensity of print orientation does have a significant (p value <0.01) and negative effect (odds <1) on the probability that a customer will retain on all three categories of time spend on printed orientation. The odd ratio for searching for information on printed channels for less than one hour is .499 lower compared to the reference level where no orientation on printed channels is done. The marginal effects indicate a 16.7 % decrease in probability of obtaining retention when spending less than 1 hour on printed information search. The odds of 1 till 3 hours spend on orientation via printed channels is .548 lower than the odd ratio of no orientation. The marginal effects indicate that spending 1 till 3 hours on printed orientation, the probability of observing retention decreases with 14.6 %.

COEFFICIENTS DF/DX STD. ERROR Z VALUE P-VALUE ODDS

(35)

35 Furthermore, the odds for orientation time spend longer than 3 hours the odds are .416 lower than no time spend on printed orientation. Moreover, marginal effects indicate a 21 % decrease in the probability of observing retention.

The intensity of network orientation does have a partially significant (p value <0.05) and negative effect (odds <1) on the probability of retention, since only one category is significant. Spending less than one

hour on searching information within the network will lead to a 15.3 % decrease in probability of

(36)

36 negative effect on the retention of customers. The odds ratio for consumers with a group discount is .603 lower than no discount and with a marginal effect indicating that the probability of observing retention decreases with 12.4% for customers with a group discount.

4.4.2 Predictive Validity

In order to assess the robustness of the model, a validation set of 20 % of the data was created. With this part of the dataset, the predictive performance of the main model will be assessed. Commonly used metrics for the predictive performance are: Top Decile Lift (TDL), the Gini coefficient and the cumulative lift curve (Leeflang et al., 2015; Risselada et al., 2010). The top decile lift can be defined as the fraction of customer which retain in the top decile divided by the fraction of customer who retain in the complete dataset (Blattberg, Kim & Neslin, 2008). With this measure, the high retention probability customers can be identified (Risselada et al., 2010). The main model has a Top Decile Lift of 1.49 indicating a 49 % better fit compared to a random selection. The Gini coefficient focuses on the performance of the model across all customers instead of only the highest probability of retentaining customers. The value can range from zero to one, where zero indicates no predictive power and one indicates perfect predictive power (Kamakura et al., 2003). The Gini Coefficient is presented in the area between the lift curve and the random selection shown in figure 5. The Gini coefficient of the main model is .25, which is somewhat higher compared to a comparable study on retention of De Haan et al. (2015). In figure 5 the lift curve is shown, where the grey dotted line indicates the base case, where all the customers have an even switch probability, and the black line indicates the lift curve of the main model. The figure shows that the main model of this research is outperforming the random selection. Furthermore, the validation set classified 60 % correctly compared with 67 % of correct classification of the estimation set, which is low comparing to the benchmark accuracy of 76.5 % (Morrison, 1969). However, comparing the percentages of the estimation and validation set, there can be concluded that no overfitting issues are present.

(37)

37

4.5 Interaction effects

In this study, it is assumed that satisfaction will have a moderating effect on the relationship of information search on different channels and retention, since research suggest that satisfied consumers have biased interpretations of the information obtained (Capraro, Broniarczyk & Stravastava, 2003). Therefore, a model including the interaction terms will be performed. First, the model needs to be tested on the assumption multicollinearity. In this model VIF values are relatively high and exceed the threshold value of 5. Furthermore, some of the tolerance values are lower than 0.2. This indicates that the predictor variables in the interaction model are highly correlated, which could lead to unreliable results (Leeflang, 2015). Since four moderation effects are present in the interaction model, this could lead to higher Variance Inflation Factor values. In order to overcome the multicollinearity issue, the interaction model is divided into four models, where each interaction effect is tested separately in the main model. In all four models the VIF scores still exceed the threshold value, indicating that multicollinearity is still present in the four models. The tables including the VIF scores and Tolerance values of the models are shown in Appendix D: Multicollinearity of the Interaction

Model. Due the multicollinearity issues the models including interactions cannot be tested.

When looking at the descriptive statistics in paragraph 4.1: descriptive statistics, substantial differences can be found between the information orientation behavior and the retention behavior of satisfied and not satisfied customers. Furthermore, looking at figure 6 there are cases where an interaction effect could be expected.

(38)

38 For example, the influence of the orientation behavior on printed channels spending 1 till 3 hours is different for satisfied and not satisfied customers. When satisfied customer search for information on printed channels for 1 till 3 hours, they are more likely to retain than the not satisfied customers. However, for expert orientation no interaction effects are expected. Because there are cases where interaction effects are expected, the main model will be estimated separately on satisfied customers and not satisfied customers. The estimation of these separate models enables the comparison between satisfied and not satisfied customers. In the upcoming paragraphs the estimation, interpretation and validation of these models will be presented.

4.5.1 Satisfied customers model

The results of the main model based on satisfied customers are presented in table 6. With an gC statistic of 152.03 (p-value <0.01) the likelihood ratio test indicates that the model fits the data containing only satisfied customers well. However, the Pseudo dC of this model are in comparison with similar studies relative low (Kumar & Venkatesan, 2005). Since the model fits the data well, the model can be interpreted.

4.5.1.1 Interpretation of the satisfied customers model

In this section only the significant results presented in table 6 will be interpreted. The intensity of print orientation does have a significant (p value <0.05) and negative effect (odds <1) on the probability that a satisfied customer will retain on all three categories of time spend on printed orientation. This result is in line with the results of the model based on all customers of a health insurance provider. The odd ratio for searching for information on printed channels for less than one hour is .546 lower compared to the reference level where no orientation one printed channels is done. The marginal effects indicate a 14.7 % decrease in probability of obtaining retention when spending less than 1 hour on printed information search. The odds of 1 till 3 hours spend on orientation via printed channels is .612 lower than the odd ratio of no orientation. Moreover, a 25.8 % decrease in the probability of observing retention. Furthermore, the odds for orientation time spend longer than 3 hours the odds are .403 lower than no time spend on printed orientation. The intensity of network orientation does have a partially significant (p value <0.05) and negative effect (odds <1) on the probability, since two categories are significant. When customer spend less than one hour of searching for information within the network, will lead a 19.4 % decrease in probability of

(39)

39 satisfied customers spend more than 3 hours it will decrease the probability of observing retention with 21.1 %. The odds of this category are 0.470 lower than no time spend on this type of information search. When looking at the variable intensity of expert orientation, only one category is significant (p value < 0.01) and negative (odds <1). The odd ratio for spending less than one hour is .268 lower compared to the reference level of no time spend on information search with experts. Furthermore, spending less than one hour searching for information by using an expert, will lead a 29,2 % decrease in probability of observing retention. Which is in line with the results of the main model.

COEFFICIENTS DF/DX STD. ERROR Z VALUE P-VALUE ODDS

(40)

40 The results of the intensity of online orientation are also in line with the main model. In the model based on the satisfied customers, two categories (p value <0.01) are significant and negative affecting retention. The odds of low intensity of orientation via online channels is .416 lower compared to the odd ratio of the reference level no orientation. Furthermore, this intensity will lead to a decrease of 18.7 % in the probability of observing retention. Moreover, the odds for high intensity of orientation online are .305 lower than no time spend on printed orientation and causes a 26.7 % decrease in the probability of observing retention. The direction of two categories is the same in this model comparing to the main model based on all customers. Furthermore, the impact of high intensity is almost the same as the main model. The impact of medium intensity online is stronger for satisfied customers. For the control variables, age is significantly (p value < 0.05) positive affecting (odds>1) the retention of customers. With an odd ratio of 1.025 and marginal effects of .6 %, indicating that a one-unit increase in age, will lead to an increase in probability of observing retention of .6 %. Furthermore, customers with a high education level are significant (p value < 0.05) and negative (odds <1) affecting retention. The odds ratio of customers with a higher education level is .624 than customers with a low education level. Furthermore, for higher educated consumers the probability of observing retention decreases 11.7% compared to low educated consumers.

4.5.1.2 Predictive validity of the satisfied customers model

In order to test the robustness of the model, the predictive performance will be assessed using the same criteria as for the main model. The Top Decile Lift of this model is 1.48, indicating a 48 % better fit than the random selection. Looking at the performance of the model across all satisfied customers, the Gini coefficient is .36, which is relatively high compared to the main model and to the comparable study of De Haan et al. (2005).

Looking at the lift curve of figure 7 the model based on satisfied customers is outperforming the base case model with the dotted lines. Furthermore, the validation set classified 69 % correctly compared with 67 % of correct classification of the estimation set, indicating no overfitting. Comparing these values with the benchmark accuracy of 76.5 % (Morrison, 1969) the score of this model is relatively low.

(41)

41

4.5.2 Not Satisfied Customers Model

The results of the main model based on the not satisfied customers are presented in table 7. With an gC statistic of 42.401 (p-value <0.01) the likelihood ratio test indicates that the model fits the data based on only the not satisfied customers well. However, the Pseudo dC of this model is also low in comparison with similar studies.

COEFFICIENTS DF/DX STD.

ERROR Z VALUE P-VALUE ODDS

(42)

42

4.5.2.1 Interpretation of the Not satisfied customers model

The results of the model based on only not satisfied customers are presented in table 7. Looking at the results, there are less significant results than the two other discussed models. Only the significant variables will be interpreted below.

The intensity of print orientation does have a partial significant (p value <0.05) and negative effect (odds <1) on the probability that a satisfied customer will retain on one category of time spend on printed orientation. When a customer spends 1 till 3 hours on orientation via printed channels the odds ratio is 0.612 lower than no orientation. Moreover, a 31.2 % decrease in the probability of observing retention.

For the intensity of expert orientation, two categories are significant (p value <0.01) and negative (odds <1) affecting retention behavior. Compared to the main model and the model with only satisfied customers, expert orientation has more significant variables in the model based on only not satisfied customers. For the category spending less than one hour on information orientation via an expert, will lead a 48.7 % decrease in probability of observing retention. Furthermore, the odd ratio for spending

less than one hour is 0.048 lower compared to the reference level of no time spend on information

(43)

43

4.5.2.2 Predictive Validity of the Not Satisfied Customers Model

For the not satisfied model the robustness is also tested using the same criterial TDL, GINI and the lift curve. The Top Decile Lift of the model based on the not satisfied customer is 1.25, indicating a 25 % better fit than the random selection. However, the TDL is relatively lower than the TDL of the two previous discussed models. The Gini Coefficient is .28, which is higher compared to the comparable study of De Haan et al. (2015). Looking at the lift curve presented in figure 8, the model is outperforming the random selection with the dotted lines. Furthermore, the validation set classified 59 % correctly compared with 66% of correct classification of the estimation set, which is low comparing to the benchmark accuracy of 76.5 % (Morrison, 1969).

(44)

44

5. Conclusion

In this chapter the results of chapter 4 are elaborated. In table 8 an overview of the hypotheses is presented. From the eleven hypotheses, two hypotheses are fully supported, four are partially supported and five are not supported. Furthermore, this chapter will describe the main conclusions and will give an answer on the main question of this study: What is the relation between the extend of

searching behavior of customers across different information-channels and retention and what influence does satisfaction have on that relationship?

HYPOTHESES SUPPORT COMMENT H1 The higher the extend of consumers’ search for information online, the lower the likelihood the customer will retain. Supported H2 The higher the extend of consumers’ search for information on printed channels, the lower the likelihood the customer will retain. Supported H3 The higher the extend of consumers’ search for information via experts, the lower the likelihood the customer will retain. Partially supported One category (0-1 hours spend) is significant. H4 The higher the extend of consumers’ search for information via network orientation channel, the lower the likelihood the customer will retain. Partially

supported Two categories (0-1 hours spend, >3 hours spend) are significant. H5 The higher the number of channels consumers’ use for information orientation, the lower the likelihood the customer will retain. Not supported Could not be tested due to correlation issues. H6 a When consumers are satisfied, it will weaken the negative relationship between retention and the extend of searching in online channels Not supported For satisfied customers, online orientation has a stronger negative effect on retention for two categories. H6 b When consumers are satisfied, it will weaken the negative relationship between retention and the extend of searching in printed channels Partially

(45)

45

5.1 Findings

A key trend in marketing the last years is the increasing interest of marketers on customer retention. In customer retention research, most studies are trying to predict churn, in order to reduce churners and to increase retention. Within these studies, customer satisfaction has been the most used key determinant of retention (Lemon et al., 2002). In this study, retention and switching behavior is investigated in another perspective. We are living in a digital world (Breur, 2011), which results in more familiarity of customers using a variety of channels (Chiu et al., 2011). In this study is anticipated on this new digitalized market by taking into account traditional perspectives of research. Since, customers are more empowered due the growing variety of channels and technologies, this could have an impact on their retention behavior. Therefore, the information searching behavior of customers on different channels are assessed as key predictors of customer retention.

(46)

46 had a significant and negative effect on retention. A more intensive search on online channels results in a stronger negative likelihood of retention. Thus, customers who are more actively searching for information on online channels tend to be less likely to retain. This is in line with the expectations based on previous research, where Shim et al. (2001) suggests that the online information search of customers can be used as a key predictor of intention to purchase. Looking at the satisfied customers, spending more than one hour on online orientation, resulted in a significantly stronger negative relation with retention, which is not in line with the expectations.

5.1.2 Printed information search

Based on the literature, it was expected that orientation of customers on printed channels would negatively affect customer retention. Especially in case of Dutch health insurance providers, where printed media is the most used media for acquiring information on alternatives (Hesselink, Henneman & Timmermans, 2009). This expectation is fully supported by the data. When a consumer is orientating by using printed media channels for information search, customers are less likely to retain at their current health insurer. The likelihood of staying at their current health insurer decreases when the intensity is higher. Spending more than 3 hours on printed orientation resulted in an even stronger negative effect on retention for satisfied customers, which is not in line with the expectations. It is interesting that the negative effect of printed orientation is stronger than the negative effect of online orientation. Contradictory, nowadays, marketers are investing more heavily on online orientation channels and less in printed channels (Graham & Greenhill, 2013).

5.1.3 Expert information search

(47)
(48)

Referenties

GERELATEERDE DOCUMENTEN

hoof van die navorsing, prof. Indien dit suk- sesvol in die toekoms blyk om stccnkool waaraan daar 'n groot tckort is. Hulle is besig om pamflette tc versprci

The study contributed towards the developing and understanding of the academic use of Facebook in order to support the learning experience of Open Distance Learning (ODL) students

Findings from the First Youth Risk Behaviour Survey in South Africa (Reddy et al., 2003), reported that PA levels among South African children have declined over the past decades

[r]

Secondly, this research aimed to explain the interaction effect of sexual orientation and gender on perceived leadership effectiveness through the mediating role of perceived

(2008) empirical research on the IO still is rather scant. Both concepts – EO and IO – seem to be important determinants for the international performance of firms. However, as

More specifically, this research seeks to demonstrate whether the frequency of outbound marketing activities, in the form of commercial emails and mails, have an effect on

• The moderation effects of Relationship Length, NPS promoters, and Number of products on the relationship between Email Marketing/Direct Mail and customer retention are