Segmenting the Dutch financial services industry based on trust and its
determinants using latent class analysis
Dirk Lutke Veldhuis
MSc. Marketing Intelligence Faculty of Economics and Business
s2514672
Supervisor: prof. dr. P.C. Verhoef Co-assessor: dr. A. Bhattacharya
2
Abstract
Trust is vital for the relationship between firms and their customers, as it leads to customer loyalty and satisfaction. Despite its importance, however, trust has barely been used as a segmentation basis in the context of the financial services industry. This study segments the Dutch financial services industry based on trust and its three determinants: competence, customer orientation and transparency. Latent class analysis was used to segment a dataset of 14628 customers from 2015-2018. Outcomes of the analysis yielded three distinct segments, which can be labelled as “low”, “middle” and “high” trust groups. The effect of the three determinants on trust differs across these segments, including opposite directions of the same determinant. These findings result in both theoretical and managerial implications, indicating that different strategies across segments should be used for increasing trust.
Keywords: trust, determinants of trust, segmentation, financial services industry, latent class
3
Content
Introduction ... 5 Theory ... 7 Segmentation ... 7 Trust ... 8Segmentation and trust ... 10
Determinants of trust ... 11
Conceptual model ... 13
Methodology ... 13
Data collection... 14
Data transformation ... 15
Data measurement and scaling ... 17
Trust variables ... 17
Demographics ... 17
Financial characteristics ... 18
Data exploration ... 19
Exploring trust ... 19
Exploring differences between banks ... 20
Exploring customer heterogeneity in trust ... 22
Hierarchical clustering ... 23
Latent Class Analysis ... 24
Model ... 24
Determining the number of segments ... 26
Results ... 29
Hierarchical clustering results ... 29
Latent class analysis results ... 29
Model inspection ... 34
Discussion ... 35
Theoretical implications ... 36
Limitations and future research ... 38
4
Conclusion ... 40
References ... 41
Appendix A - GfK market survey questions... 44
5
Introduction
Segmentation is a vital tool for structuring heterogeneity in a firm’s customer base (Wedel & Kamakura, 1998). In addition, customer segmentation can be used for understanding motivations of customers and is therefore key for successful customer value management (Kumar, 2018; Verhoef & Lemon, 2013). Customer segmentation can be performed with different segmentation bases. Studies which segment the financial services industry use for instance perceived value, behavioral and psychographic bases as segmentation basis (Bijmolt, Paas, & Vermunt, 2004; Floh, Zauner, Koller, & Rusch, 2014; Muhamad, Melewar, & Alwi, 2012).
Surprisingly, little attention is paid to the concept of trust as a segmentation basis in marketing literature. Trust plays a major role in the relationship between firm and customer. This has the following reasons. First of all, trust fosters value to the customer, which results in customer loyalty (Sirdeshmukh, Singh, & Sabol, 2002). Moreover, trust has a positive influence on relationship commitment, also resulting in higher customer loyalty (Morgan & Hunt, 1994). Lastly, trust has a positive effect on customer satisfaction and long-term orientation (Geyskens, Steenkamp, & Kumar, 1998). One may argue that in the relationship between banks and customers, trust may play an even more important role compared to other industries. To illustrate, customers lend out their savings to a bank, and rely on them to properly manage this money. The 2015 salary scandal of ABN AMRO’s executive board evidently illustrates the role that trust plays in this sector. Thousands of customers left the bank because of the (in their eyes) excessive salary increase of the board members, and thus the broken trust between ABN AMRO and its customers (Hensen, 2015).
6
determinants of trust. These determinants are subsequently being used by Moin, Devlin, & McKechnie (2017) to classify UK bank customers into either a low or high trust group. By creating two groups of customers based on a trust score, they are actually the first ones to segment a financial services market based on trust. However, they did not use the determinants directly to form segments, but instead split the customer base in a low and high trust segment.
By applying latent class analysis, a customer base can be segmented on both the determinants of trust as well as trust itself. By using this approach, customer heterogeneity may be better explained than by segmenting based on the degree of trust alone. This deeper customer knowledge can serve bank managers in their decisions about the allocation of their resources across their customer base. Also, the latent class approach in general has some interesting advantages over traditional clustering methods (Vermunt & Magidson, 2002). In summary, the need for research into trust-based segmentation in the financial services industry, which includes the use of the determinants of trust, has been identified. This need is not being met by existing literature, which leads to the following research objective of this study:
Segment customers of the financial services industry on both the determinants of trust and trust itself.
7
Theory
Segmentation
Market segmentation is a disaggregative method of viewing a heterogeneous market, which is characterized by dissimilar demand, as two or more homogeneous markets that respond similar within the submarket but differ in response between submarkets (Smith, 1956). Customer segmentation is a key tool to understand that which drives customers. Furthermore, a proper segmentation yields valuable insights, as it helps to determine the allocation of marketing resources across segments (Verhoef & Lemon, 2013). Lastly, Kumar (2018) argues that segmentation can help firms in their customer relationship management efforts.
Various studies have used segmentation in the context of banking and the financial services industry. Noteworthy is the study of Bijmolt et al. (2004), where they segment the European financial services industry based on ownership of different financial products. Segmenting this market has some practical implications for managers, as targeting different customers and countries may require distinct marketing strategies. For instance, the segment with customers mainly from the Benelux and Germany has low product penetration of credit cards and cheque books. In contrast, the segment with mainly Greek customers has high penetration rates for these products. Also, by using demographic variables, the study shows that customers in the Greek segment tend to be younger. Another relevant segmenation study is that by Floh et al. (2014), in which they segment the customer base of a large European financial firm. By using dimensions of perceived value as a segmentation basis, the researchers are able to segment the customers into three distinct groups. By using latent class analysis as their segmentation technique, the study makes it possible to link the segments with different dimensions of perceived value. The
rationalists, for instance, give higher weight to cognitive dimensions of perceived value, while for
8
and beliefs, and not to product details or attractive pricing. On the contrary, in the segment called
economic rationality group, customers chose for Islamic financial products solely based on
economic rationality and not on any religious values. All three segmentation studies provide examples of the value that segmentation could have for firms in the financial services industry. One can use various segmentation bases to segment a customer base, such as observable (e.g. demographic variables) or unobservable (e.g. perceived value) bases (Wedel & Kamakura, 1998). To illustrate, the Muhamad et al. (2012) study uses both (religious) values as perceived benefits as segmentation bases. Although many bases for segmentation have been used, surprisingly, the use of trust as a segmentation basis remains quite unexplored.
Trust
Customer decision making is believed to be guided by high order mental constructs, including that of trust (Garbarino & Johnson, 1999). The concept of trust can be defined in various ways. An early definition is presented by Rotter (1967), who defines it as “an expectancy held by an individual or a group that the word, promise, verbal or written statement of another individual or group can be relied upon”. More related to a marketing context, Singh & Sirdeshmukh (2000) describe customer trust as “the expectations held by the customer that the service provider is dependable and can be relied on to deliver on its promises”. The importance of trust in the relationship between firm and customer has been identified in marketing literature. A noteworthy study is that of Morgan & Hunt (1994), in which they underline the importance of trust in a relational context. The study identifies trust as a key mediating variable in relationship commitment, that results in customer loyalty. The generalizability of this finding is demonstrated by Garbarino & Johnson (1999), who add that trust plays a role in strong relationships between firm and customer. In addition, Sirdeshmukh et al. (2002) found that trust has an effect on customer loyalty, although partially mediated by value. Moreover, a meta-analysis of multiple countries and industries of Geyskens et al. (1998) found that building trust is effective for increasing customer loyalty and satisfaction.
9
more extensive knowledge of the (often complex) financial products they sell, compared to the customers who buy them. Diacon & Ennew (1996) state that this information asymmetry may lead to a problematic development of trust in this relationship. Furthermore, banks have a duty of stewardship to properly manage customers’ savings, and these customers therefore have to trust banks that they will fulfill this duty with integrity. The existence of this trust is also beneficial from a societal perspective on a macro-level, because when a large part of a country’s population uses banks for savings, it will result in a national banking system which is more stable (Han & Melecky, 2017). In 2008, Dutch banks launched a code of conduct, in order to restore customer trust in the financial services industry after the economic crisis (Dutch Association of Banking, 2014). This illustrates that trust plays an important role for the Dutch financial services industry too. Two types of trust in relation to the financial services industry can be identified. System trust is defined as customers’ trust in the financial services industry in general, and also includes trust in the (international) banking system and the transfer of money. Institution trust is defined as a customer’s trust in their own bank (Van Esterik-Plasmeijer & Van Raaij, 2017). In this study, the concept of institution trust will be used when mentioning trust.
10
customers with a bachelor’s degree care more about large bonuses than government support compared to customers without such a degree, when asked why they lose trust in banks.
Although customer characteristics are not being used as segmentation bases, they do provide insights into the differences between the customer segments. Previous paragraphs illustrate the influence of demographic factors on customer trust in a financial services setting. Therefore, demographic variables will be included. The dataset used in this study provides details about a customer’s age, gender, education and place of residence. Furthermore, the dataset provides information about a customer’s main bank and financial product ownership. It is expected that these variables are useful in explaining differences between the segments.
Segmentation and trust
Previous sections illustrated the advantages that customer segmentation has for firms, and the importance of trust in relation to the financial services industry. Also, previous sections have shown that trust in banks differs among sociodemographic characteristics of customers. It is therefore remarkable that segmentation studies with trust as a segmentation basis barely exist. A study of Dimitriadis, Kouremenos, & Kyrezis (2011) segments users of self-service technology bank channels (e.g. mobile banking applications) based on either high or low trust in these channels. By using discriminant analysis, they test whether these groups are statistically different. As the two trust segments differ sufficient from each other, the study is first in successfully using trust as a segmentation basis in the financial services context. Moin et al. (2017) go even further by segmenting customers based on their trust in banks in general (i.e. institution trust). Like the previous study, the customer base is split into a high and low trust group, and differences between these two segments are investigated. By conducting a cluster analysis and an analysis of variance subsequently, they are able to identify sociodemographic differences between high and low trustees. Age, marital status, and ethnicity, for instance, differ significantly between the two trust segments. Furthermore, a higher income is related to higher trust. However, the study states that there are no significant differences in trusting belief in terms of gender.
11
determinants integrity, benevolence, competence and predictability to determine the customers’ trust in the channels. However, both studies do not use these determinants directly for segmenting their customer base, instead using them to measure trust and to segment the customer base into high and low trust groups. For banks, knowing the extent of trust that different customers (e.g. males versus females) have might be interesting, but deeper knowledge is also available. When segmenting customers directly based on both trust and its determinants, banks are able to decompose how trust is related to its determinants across different customers. To illustrate, it would be possible to investigate if some determinants of trust matter more for one specific customer segment. This would expand the current field of study relating to trust-based segmentation. In the next section, the determinants of trust and their applicability to segmentation of the financial services industry are being discussed.
Determinants of trust
Much research has been done into the determinants of trust in relation to the financial services industry. Nienaber, Hofeditz, & Searle (2014) conducted a meta-analysis of organizational trust within the financial services industry. Their analysis covers twenty empirical studies in the financial services industry from multiple countries, in the period of 2002-2011. Although most studies focused on retail banking, both internet banking and financial services were also included. The meta-analysis resulted in a wide range of determinants of trust, such as competence,
cooperation and performance. However, only four constructs appear frequently in the studies,
namely: shared values, satisfaction, communication and reputation. What becomes clear from the study is that the determinants of trust are not always the same across different settings. For instance, the determinant reputation is a significant determinant of trust in a Malaysian financial services study, but not significant in a Taiwanese one. Therefore, it could be argued that identifying the determinants of trust is context specific. To illustrate, a study of Van Esterik-Plasmeijer & Van Raaij (2017) specifically focused on the Dutch financial services industry and found the following determinants of bank trust: competence, integrity, customer orientation and transparency. In contrast, for the UK financial services market, Ennew et al. (2011) found the determinants to be:
integrity and consistency, expertise and consistency, communications, concern and benevolence
12
determinants of trust to use. This study uses the dataset from the Banking Confidence Monitor of the Dutch Association of Banking (2018), which measures the following three determinants of trust: competence, customer orientation and transparency. Therefore, these three determinants will be used in this study. The following sections will explain each of them in more detail.
Sirdeshmukh et al. (2002) describe competence as the “expectation of consistently competent performance from an exchange partner”, and name it a dimension of trustworthy behavior. Customers rely on a wide range of services from a bank (e.g. credit card payments, bank transfers). They have to trust the banks in that these services will work and are safe to use. Also, as customers lend out their savings, they have to lay trust in the competence of a bank that it will not go bankrupt and thus not lose their savings.
Customer orientation is described as “the belief that prescribes the unit of analysis of every
marketing action and reaction to be the individual customer” (Hoekstra, Leeflang, & Wittink, 1999; Ramani & Kumar, 2008). In other words, it is the extent to which banks are focused on the customer perspective and their relationship with the customers. An example of customer-oriented behavior could be when banks advise products to the customers that are in the best interest of the customer, but less so in the bank’s interest. When customers have the feeling that the bank cares for them, they might be more likely to place trust in the institution. This is closely related to operational benevolence, a concept also considered as an determinant of trust (Sirdeshmukh et al., 2002).
Transparency is described as the extent to which information is shared in a business relationship,
13
Conceptual model
In figure 1, the conceptual framework of this study is presented. Institution trust (i.e. the trust customers have in their main bank) is determined by the perceived competence, customer orientation and transparency of the bank. Based on this relationship, the customer base is segmented using latent class analysis. In other words, the way how a customer’s institution trust is formed by competence, customer orientation and transparency determines the customer’s segment membership. This results in an unknown number of segments, which are different from each other in terms of model parameters. More precisely, the effects of the three determinants on institution trust are different over the various segments.
Furthermore, the study examines whether the segments differ in terms of demographic and financial characteristics. Demographics included in this study are the customer’s age, gender, education and the place of residence. In addition, financial information, including a customer’s main bank and their ownership of financial products is being used to describe the customer segments. Note that both demographics and financial characteristics are not being used for segmenting the customer base, but solely for profiling the segments.
This framework makes it possible to investigate differences in trust between customers, and relating this to the determinants of trust. This is the first study so far applying this approach, and therefore may result in deep customer knowledge which has not been obtained before. This could therefore have valuable implications from a managerial point of view.
14
Methodology
This section starts with an overview of the collection, transformation and measurement of the dataset used in this study, after which, descriptive statistics are provided. This is followed by a data exploration part, where a deep dive into the relationship between the various trust variables will be conducted. The segmentation part of this methodology section consists of two approaches: hierarchical clustering and latent class analysis. Although the main focus of this study is using latent class analysis, the hierarchical clustering approach is included to illustrate differences between the two. Both will be discussed comprehensively in this section.
Data collection
This study uses information from a dataset of the Dutch Association of Banking, which annually investigates customer trust into the Dutch financial services industry. These results are subsequently presented in the Banking Confidence Montior, and is publicly accessible (Dutch Association of Banking, 2018). The data itself is collected by research institute GfK, from the period 2015-Q1 until 2018-Q2. Surveys are conducted on customers of all Dutch banks, namely ABN AMRO, ING, Rabobank, Volksbank (consisting of SNS, Regio Bank and ASN Bank) and other banks representing Argenta, BinckBank, Triodos Bank, Centraal Beheer, Knab, LeasePlan Bank, NN Bank, NIBC and Woonfonds. The larger banks (i.e. ABN AMRO, ING, Rabobank) are underrepresented in the dataset. Therefore, a weighting factor is added which can be used for the calculation of a representative trust average for the Dutch financial services industry.
15
may lead to biased results. As there are no outliers, the final dataset contains 14628 customers of the Dutch financial services industry.
Data transformation
Customers can agree or disagree with fourteen different statements about their main bank. This is measured on a 5-point Likert scale, ranging from “completely disagree” to “completely agree”. The survey containing these statements can be found in Appendix A.
Each of the determinants (i.e. competence, customer orientation and transparency) is measured by a predetermined combination of these statements. To check internal consistency, a reliability analysis has been performed. In addition, a confirmatory factor analysis (CFA) has been conducted (table 2). Both tests offer satisfying results. The three Cronbach’s alpha values are well above the threshold value of 0.6, and the standardized loadings are also sufficient. The CFA yields satisfactory global fit indices (CFI = .960; TLI = .951; RMSEA = .081; SRMR = .035). In addition, the CFA model with three components (competence, customer orientation and transparency) has as lower AIC and BIC scores than the baseline model with only one component (table 1). A likelihood ratio test between these two models is highly significant (p<0.001), thus resulting in a preference for the three-construct model.
So, the concepts of competence, customer orientation and transparency are internally consistent and can be used for subsequent analysis. The constructs are created by averaging the scores of the relevant items, as no distinction in importance between the statement has been made.
Table 1
Confirmatory factor analysis: differences in information criteria between CFA models
Model AIC BIC
One construct 382230 382443
16
Table 2
Results of confirmatory factor analysis (loadings) and reliability analysis (Cronbach’s α)
Transparency Customer Orientation Competence Reliability (α)
Bank is open 0.95
Bank is honest 0.95 0.98
Bank is communicating clear 0.78
Bank is informing proactive
about changes in policy 0.72
Bank listens to customers 0.88
Bank advises products which are
in interest of customer 0.85
Bank supports in making
financial choices 0.83
Bank searches for financial solutions
together with customer 0.80 0.92
Bank has financial knowledge 0.81
Bank has knowledgeable staff 0.84
Bank offers insights 0.80
Bank is punctual 0.83
Bank is easy to contact 0.72
17
Data measurement and scaling
Trust variables
Institution trust, the trust a customer has in his or her main bank, is measured on a 5-point scale by asking customers about the degree of trust they have in their main bank. This scale ranges from “very low trust” to “very high trust” in the bank. This results in a mean of 3.283. However, this value is not representative for the Dutch financial services industry in general, as the use of the weighting factor is not applicable anymore due to the deletion of missing data. The remainder of this paper therefore does not use the weighting factor.
An overview of the means and standard deviations of the trust variables, including the three determinants of trust can be found in table 3. Just like the variable institution trust, the three determinants range from 1 to 5. Furthermore, visual inspection of the variables’ distribution learns that they are normally distributed (figure 2).
Table 3
Descriptive statistics of trust variables (N=14628)
Mean Standard Deviation
Institution Trust 3.283 0.817
Competence 3.683 0.705
Customer Orientation 3.383 0.791
Transparency 3.332 0.928
Demographics
18 Table 4 Demographic characteristics (N=14628) Gender % Age % Female 50.8 18-34 17.1 Male 49.2 35-49 27.5 50-64 31.9 65+ 23.5 Education % District % Low 26.6 North 11.9
Middle 39.9 West (three big cities) 13.9
High 33.5 West (other) 27.0
East 22.1
South 25.1
Financial characteristics
Besides demographic details, customers are also asked to state their main bank (i.e. the bank which feels the most important for the customer). In addition, they are asked which financial products they currently have with their main bank. Table 5 gives an overview of this information, where the percentages at the financial product column are penetration rates. The customer’s main bank is anonymized in the data, therefore Bank A to E is used in the remainder of this thesis instead of the real bank names, where Bank E represents the other banks.
Table 5
Financial characteristics (N=14628)
Main bank % Financial product %
Bank A 25.3 Checking account 97.6
Bank B 24.1 Savings account 81.9
Bank C 27.7 Investments 11.0
Bank D 12.3 Mortgage 23.2
Bank E 10.6 Credit account 6.9
19
Data exploration
Exploring trust
Institution trust is determined by the degree of competence, customer orientation and transparency of the financial institution. When looking at the Pearson correlation coefficients in figure 2, all values are positive and highly significant (p<0.001), indicating that there is a positive association between the determinants of trust and institution trust itself. This also becomes clear when looking at the line charts in the same figure. There is a clear positive slope with institution trust on the vertical axis and three the determinants on the horizontal axis. Furthermore, it is clear that the three determinants reinforce each other, as evidenced by the visible positive slope. And, when looking at the distribution of each variable, they move in the same direction.
Figure 2
20
To formally test whether the three determinants influence institution trust, a linear regression is conducted with:
𝑌𝑖 = 𝛽0+ 𝛽1𝐶𝑖+ 𝛽2𝐶𝑂𝑖 + 𝛽3𝑇𝑖+ 𝜖𝑖
Where 𝑌𝑖 denotes the institution trust score, 𝐶𝑖 the competence score, 𝐶𝑂𝑖 the customer orientation score and 𝑇𝑖 the transparency score for customer i. The estimates of parameters 𝛽 are displayed in
table 6. Table 6
Linear regression results (institution trust as dependent variable)
Estimate Standard error t-value p-value
Intercept 0.990 0.026 38.247 0.000***
Competence -0.015 0.013 -1.202 0.229
Customer Orientation 0.334 0.014 24.336 0.000***
Transparency 0.366 0.010 38.037 0.000***
sig. levels: *p<.05. **p<0.01. ***p<0.001
A positive relationship can be found between both customer orientation and transparency on institution trust, with estimates of 0.334 and 0.366 respectively. The competence variable has a p-value of 0.229, and has therefore no significant relationship on institution trust. This is in line with literature, given that competence is not a determinant of trust in every setting.
Exploring differences between banks
22 Exploring customer heterogeneity in trust
From current literature, the question arises whether there are differences in institution trust between the various levels of the demographic variables. This is investigated by an analysis of variance, resulting in significant outcomes for every demographic variable. So, there are significant differences in means of institution trust between the various levels of the demographic variables. In table 7, the p-values from the ANOVA are presented together with the group means of institution trust of the various variables.
There is a significant difference in institution trust between males and females, as females have higher scores. Looking at age, the youngest age group scores the highest value in trust, whereas the other age groups do not seem to differ a lot from each other. When examining a customer’s education, the trust score tends to decrease as the education level increases. Finally, also the district variable has a significant effect on institution trust. The highest institution trust can be found in the East, whereas customers in the three biggest cities have the lowest trust.
Table 7
Results of analysis of variance with demographics (dependent variable: institution trust)
Variable Group Mean Significance
Gender Male 3.231 0.000*** Female 3.336 Age 18-34 3.422 0.000*** 35-49 3.257 50-64 3.257 65+ 3.271 Education Low 3.342 0.000*** Middle 3.286 High 3.232 District North 3.292 0.000***
West (three big cities) 3.236
West (other) 3.249
East 3.322
South 3.305
23
Hierarchical clustering
Hierarchical clustering is a method of structuring customer heterogeneity. With this approach, the number and structure of the segments are not known a priori. Clustering is done based on (dis)similarity across a set of segmentation variables, which are in this case institution trust, competence, customer orientation and transparency. The hierarchical clustering approach is agglomerative, meaning that it starts with very small segments and keeps adding them up into larger segments. Distance is measured based on the Euclidean scale, and the Ward’s method is used as a cluster agglomeration algorithm, in which the total within-cluster variance is minimized. The agglomeration algorithm continues merging segments based on similarity until only one segment, consisting out of all observations, remains. The optimal number of segments is reached when the difference (and thus information loss) between two merged segments is too large. One can detect this visually, by examining a scree plot and dendrogram. The scree plot (figure 6) shows a clear “elbow” at k=2, thus indicating a large information loss when going from two segments to one. This is confirmed by the dendrogram (figure 7), which also clearly favors the two-segment solution. In short, the hierarchical clustering procedure yields two segments, which will be discussed more in-depth in the results section.
Figure 6
24
Figure 7
Dendrogram of hierarchical clustering outcomes
Latent Class Analysis
Model
25
(Vermunt & Magidson, 2002). A hierarchical clustering procedure is included in this study to illustrate the differences between the two segmentation techniques.
The latent class regression model can be stated as (Vermunt & Magidson, 2002):
𝑓(𝒚𝑖|𝜃) = ∑ 𝜋𝑘𝑓𝑘(𝒚𝑖|𝜃𝑘) 𝐾
𝑘=1
Where 𝑦𝑖 denotes a customer’s i institution trust score, K is the number of segments, and 𝜋𝑘 denotes the prior probability of belonging to latent segment k. The distribution of 𝑦𝑖 given the model parameters 𝜃, 𝑓(𝑦𝑖|𝜃), is assumed to be a mixture of class-specific densities, 𝑓𝑘(𝑦𝑖|𝜃𝑘). The model can be extended by adding covariates. Covariates, or concomitant variables, are not used as segmentation variables but variables used for profiling the segments. But, extending the model with concomitant variables yields output which is very hard to interpret, due to the categorical nature of the concomitant variables (e.g. education) which results in a double base level interpretation.
Estimating the parameters of the latent class regression model is done by the maximum likelihood (ML) method. Subsequently, the iterative expectation-maximization (EM) algorithm is used to find the ML estimates. Although estimates are useful for the purpose of this study, also important is the classification of customers into segments. Classification can be done based on posterior class membership probabilities, in which each customer i is assigned to segment k with the highest posterior probability. Posterior class membership probabilities are given by:
𝜋𝑘|𝑌𝑖 =
𝜋𝑘𝑓𝑘(𝑦𝑖, 𝜃𝑘) ∑ 𝜋𝑘 𝑘𝑓𝑘(𝑦𝑖, 𝜃𝑘)
26 Determining the number of segments
The use of information criteria is preferred when choosing the optimal number of segments (K) in a latent class analysis (DeSarbo et al., 2006; Vermunt & Magidson, 2002; Wedel & Kamakura, 1998). The standard likelihood ratio test, however, cannot be used as the asymptotic chi-square distribution is not valid. Three types of information criteria for determining K are commonly mentioned in the literature, namely the Akaike information criteria (AIC), Bayesian information criteria (BIC) and the consistent Akaike information criteria (CAIC). All three criteria make use of the likelihood function to determine which model is the best, where the model with the lowest value indicates a best fit with the “true model”. The difference in the three criteria is mainly concerned with the penalties for sample size and number of parameters, where BIC and CAIC penalize more heavily than AIC. As the current field of study does not offer a univocal recommendation about which information criteria to use, this study uses all three for comparison. In addition, the entropy statistic (𝐸𝑘) can be used for determining the number of clusters. 𝐸𝑘 measures how well segments are separated, and ranges between zero and one, where one
indicates a perfect separation. The entropy statistic is calculated by the following equation (Wedel & Kamakura, 1998): 𝐸𝑘 = 1 − ∑ ∑ − 𝐾 𝑘=1 𝜋𝑘|𝑌𝑖ln 𝜋𝑘|𝑌𝑖/𝐼 𝐼 𝑖=1
The FlexMix model is running six times with K ranging from one to six. The scores of the information criteria and entropy statistics of the different models are presented in table 8, together with the log-likelihood values (LL). In addition, figure 8 shows a line graph of the BIC scores over the different segment sizes. This is visually the same as the AIC/CAIC scores, as these values lay very close to each other.
27
should favor the K=6 model as it has the best separation between segments. However, both the K=3 and K=4 model do a fair job in segment separation, as their entropy scores are still quite high. This leaves room for favoring the K=3 and K=4 models too.
Table 8
Information statistics of different number of segments
Number of segments (K) LL BIC AIC CAIC 𝐸𝐾
1 -12933 25915 25877 25896 - 2 -12085 24275 24192 24234 0.66 3 222608 -445053 -445182 -445117 0.76 4 244054 -487888 -488062 -487975 0.78 5 260289 -520299 -520519 -520409 0.68 6 355039 -709743 -710008 -709876 0.86 Figure 8
Graph of BIC scores of different number of segments (K)
Next to the use of information criteria, also managerial and theoretical perspectives should be taken into account when choosing the optimal number of segments. Important are the five criteria of effective segmentation, stating that segments should be measurable, substantial, accesible,
28
homogeneous but externally heterogeneous, so that every segment can be targeted with its own marketing strategy. This is not the case for the model with K=5, as the means of institution trust are not significantly different between two segments1. In addition, the model has two segments with almost identical parameters. This is also true for K=6, in which the response parameters are identical for three segments. Furthermore, both the K=5 as the K=6 models result in segments which are very small (5.3% and 7.2% respectively, as can be observed in table 9), which does not meet the substantiality criteria very well. When comparing the resulting models K=3 and K=4, the first one is favoured as the segments are more distinct from each other. The reason for this is that the third and fourth segment of the K=4 model are very similar to the second segment of the K=3 model, when looking at the intercept and competence coefficients. Furthermore, the segment sizes are more substantional at K=3, as K=4 has a fairly small segment only covering 8.1% of the market. So, the K=3 model yields segments which are both differentiable and substantial. The measurability criterium is met too, as the number and characteristics of customers within the segment can be determined with accuracy. And as the three segments are all substantial, it meets the accessibility and actionability criteria.
Table 9
Segment sizes over number of segments (N=14628)
Segment K=2 K=3 K=4 K=5 K=6 1 657 2963 5510 7957 1469 2 13971 3682 7934 1909 1135 3 7983 1184 2575 7894 4 3379 781 1490 5 1406 1064 6 1576
29
Results
The results section starts with the outcomes of the two-segment hierarchical clustering procedure, which will be briefly discussed. Afterwards, the outcomes of the latent class analysis are presented. The analysis resulted in three distinct segments, which will be extensively described and named in this section. The final part is dedicated to model inspection, in which a comparison with the hierarchical clustering results will also be made.
Hierarchical clustering results
The hierarchical clustering procedure yields two segments, which are presented in table 10. Both segments are almost equal in size, covering around 47% and 52% of the market respectively. Noteworthy is the large difference in institution trust between the two. The mean of institution trust of segment 1 is around 36% higher than that of segment 2. The difference in means of the trust determinants between the two segments is of similar size. To formally test the differences in means between the two segments, a Kruskal-Wallis analysis of variance is conducted. All p-values are highly significant (p<0.001), indicating that there are significant differences in means between the two segments. Based on this, segment 1 can be labeled as a “high” trust segment, while segment 2 can be named a “low” trust segment.
Table 10
Hierarchical clustering results: segment description – differences are investigated using Kruskal-Wallis tests
Segment 1 Segment 2 p-value
Size (%) 6914 (47.3%) 7714 (52.3%) Institution trust 3.82 2.80 0.000*** Competence 4.17 3.25 0.000*** Customer Orientation 3.98 2.85 0.000*** Transparency 4.10 2.64 0.000*** sig. levels: *p<.05. **p<0.01. ***p<0.001
Latent class analysis results
30
meaning that it relates to the degree of institution trust within the segment. All three intercepts are highly significant (p<0.001) and correspond to the segment’s mean of institution trust which can be found in table 12. Segment 1, for instance, has the highest mean of institution trust and also the highest intercept. Roughly speaking, the segments can be classified as a “low”, “middle” and “high” trust segments for segments 1, 2 and 3 respectively. Similar to the interpretation of the intercept, coefficients of the independent variables can be interpreted in the same way as a standard regression. A positive coefficient value indicates a positive relation of the independent variable on institution trust and vice versa for negative values. Besides, the size of the coefficient indicates the strength of the relationship. This strength can subsequently be used to compare importance of the determinant on trust across the three segments.
Table 11
Latent class model results: coefficients
Segment 1 Segment 2 Segment 3
Size (%) (20.3%) (25.2%) (54.5%) Intercept 2.730*** -0.686*** 0.000*** Competence -0.565*** 0.995*** 0.000 Customer Orientation 0.680*** 0.067** 0.000 Transparency 0.137*** -0.089*** 1.000*** sig. levels: *p<.05. **p<0.01. ***p<0.001
31
Table 12
Latent class model results: means – differences are investigated using Kruskal-Wallis tests
Segment 1 Segment 2 Segment 3 p-value
Size (%) 2963 (20.3%) 3682 (25.2%) 7983 (54.5%) Institution trust 3.43 3.05 3.33 0.000*** Competence 3.00 3.59 3.33 0.000*** Customer Orientation 3.12 3.59 3.39 0.000*** Transparency 3.46 3.91 3.66 0.000*** sig. levels: *p<.05. **p<0.01. ***p<0.001 Table 13
Latent class model results: demographic differences – differences are investigated using Chi-square tests
Variable Segment 1 Segment 2 Segment 3 p-value (𝑋2)
Age 18-34 111 93 100 0.000*** 35-49 103 95 100 50-64 89 101 100 65+ 103 109 100 Gender Male 110 105 100 0.000*** Female 90 95 100 Education Low 90 101 100 0.005*** Middle 99 102 100 High 109 97 100 District North 99 110 100 0.632
West (three cities) 100 96 100
West (other) 103 102 100
East 100 97 100
South 98 98 100
32
Table 14
Latent class model results: differences in main bank– differences are investigated using Chi-square tests
Bank Segment 1 Segment 2 Segment 3 p-value (𝑋2)
Bank A 105 99 100 0.003*** Bank B 101 95 100 Bank C 106 107 100 Bank D 78 92 100 Bank E 100 107 100 sig. levels: *p<.05. **p<0.01. ***p<0.001 Table 15
LCA results: differences in financial product ownership– differences are investigated using Chi-square tests
Financial product Segment 1 Segment 2 Segment 3 p-value (𝑋2)
33
The latent class model yields three segments of customers, whose degree of institution trust is influenced through different ways by the determinants of this trust. This difference of influence makes up the main difference between the three segments, which will be discussed more comprehensively below.
Segment 1 – The heart followers
The first segment is the smallest (20.3%) and has the highest intercept and thus the highest institution trust of all segments. Across all segments, customer orientation is the most important determinant for trust in segment 1. In addition, it is also the most important determinant within the segment, with a strong positive relationship of 0.680. Also, transparency is positively influencing trust in this segment, although weaker than customer orientation. Competence of a customer’s main bank decreases the institution trust a customer has, which can be derived from the strong negative relationship. When examining the demographic variables, it can be observed that customers in this segment are the youngest and most highly educated of all three segments, because of the highest indexes for the 18-34 age group and high education group (table 13). Furthermore, segment 1 has the highest concentration males of all segments. The highest density of Bank A and Bank B customers can be found in segment 1 (table 14). Although segment 1 has the highest index for credit accounts, it cannot be stated that this segment holds more credit accounts than other segments due to the insignificant result of the Chi-square test (table 15).
In short, for customers in segment 1, trust increases when a bank is open and honest. Furthermore, these customers appreciate it when the bank listens to them, and supports them in making financial decisions. Customers are not impressed by punctual behavior and superior knowledge of the bank, as it will decrease the customers’ trust. As the customers build trust based on “soft” and relational aspects, and not with more rational and functional ones, it indicates that these customers make decisions with their heart and less with their head. Therefore, these customers are called the heart
followers.
Segment 2 – The rationalists
34
2 is also the only segment in which transparency negatively influences trust. The demographic characteristics are also quite different, as segment 2 has the highest density of customers of 65+ and the lowest density of customers of 18-34 and 35-49. In addition, the customers in this segment have had the lowest education across all segments.
To conclude, customers in segment 2 tend to trust their bank when it shows that possesses financial knowledge, and is able to share insights with their customers. Moreover, trust increases when the bank has knowledgeable staff who are punctual and easy to contact. Thus, the “soft” and relational aspects of the bank are not important for customers in this segment, in which it differs from the first segment. Customers in the second segment tend to build trust mainly based on aspects that are the most functional for them as a customer. Considering these customers make decisions based on reason and knowledge, they are called the rationalists.
Segment 3 – The transparency fundamentalists
The third segment is the largest, covering 54.5% of the population. It has an intercept of 0.000, which is in between the intercepts of the two other segments. The mean of institution trust (3.33) is also the closest to the overall mean of the population (3.28). In this segment, the only significant variable is that of transparency, with an estimate of 1.000. This means that institution trust is fully positively influenced by transparency for customers in this segment. The other determinants of trust have no influence on trust here. Segment 3 is also the only segment with a majority of females in it, given their highest index. Next to this, segment 3 has the most customers of Bank D, and the least Bank C customers compared to other segments.
In short, every change in a bank’s transparency score is fully incorporated in the bank’s trust score. Customers in this segment do not care about the extent to which the bank is oriented towards them, neither do they care about the competence of the bank. As transparency is the only determinant which matters for building trust, customers in this segment are called the transparency
fundamentalists.
Model inspection
35
the Kruskal-Wallis analyses. This means that the classification of segments into a “low”, “middle” and “high” trust segment is based on significant differences and therefore valid. The model also yielded segments which are significantly different based on four of the six concomitant variables, namely age, gender, education and main bank. Comparing the results of the latent class analysis to the results of the hierarchical clustering procedure, the performance is even more satisfying for two important reasons. First, the hierarchical clustering approach resulted in two segments, with one “low” trust segment and one “high” trust segment. This is, however, less accurate than the three-segment solution of the latent class analysis, in which next to a “low” and “high” trust segment, a “middle” trust segment is also identified. Second, in the hierarchical clustering procedure, it is only possible to compare the two segments in terms of their (trust) means. The relationship of the determinants on institution trust is assumed to be the same for both segments. However, with the latent class analysis, one is able to compare segments in terms of this relationship, and find out that this relationship is not the same for the three segments. This, too, is the result of a more accurate segmentation approach. Finally, the entropy statistic can be examined to assess model quality. The statistic equals 0.76, which is fairly close to the maximum of 1. Ultimately, the statistic indicates a fair degree of separation.
Discussion
36 Theoretical implications
Results of this study show that the effect of trust determinants on trust can differ across segments. Transparency, for instance, has a positive effect on trust for high- and middle trustees, but a negative effect for low trustees. This difference in effects has two implications for theory.
First, it expands the possibilities for segmentation based on trust. Current literature relating to the segmentation of the financial services industry has only used trust itself as segmentation base (Dimitriadis et al., 2011; Moin et al., 2017). By including the determinants of trust as segmentation bases, as done in this study, one can create more accurate segments. This is because segmenting on trust alone does not yield information about how this trust is formed, let alone that it can explain differences in this trust formation between segment. The approach of including the determinants as segmentation base can tell how trust is formed (e.g. the different directions of transparency on trust), and therefore shows superior performance in structuring heterogeneity of a customer base. Second, these findings have implications for research relating the effects of trust determinants on the formation of trust. The current field of study has identified a wide range of determinants, which all show to have a positive effect on trust (Ennew et al., 2011; Nienaber et al., 2014). However, this study shows that this positive effect can vary between segments and that the effect can even be negative. When again examining transparency, it shows a substantial difference in positive effect between segment 1 and segment 3. And for segment 2, this effect even is negative. The difference in effects between this study and the current literature may be explained due to the fact that this study accounts for customer heterogeneity, by analyzing the effects of trust determinants on trust across segments.
37
Relating the influence on satisfaction, three types of incluence factors can be identified: hygiene (dissatisfiers), motivators (satisfiers) and dual factors (both dissatisfier and satisfier) (Herzberg, Mausner, & Bloch Snyderman, 1959). In the case of competence as a determinant of institution trust, it could be labelled as a dual factor. In a slightly different context, research has already identified competence as a dual factor. Johnston (1997) researches the determinants of service quality in a retail banking setting, and found that the determinant competence is both a satisfier as a dissatisfier. In other words, a change of competence leads to both an increase as well as a decrease in satisfation. The only study researching the relationship between competence and institution trust in the Dutch financial services industry is that of Van Esterik-Plasmeijer & Van Raaij (2017). They found a significant parameter estimate of 0.100, indicating a postive relationship. However, they also state that determinants of trust might know dissatisfiers and satisfiers, and provide competence as an example. According to them, it is not possible for a customer to generate trust in a bank if competence falls under a certain level. When the bank’s competence is above a certain level, it positively influences the degree of trust a customer has in the bank. The Van Esterik-Plasmeijer & Van Raaij (2017) study does not tell from which level of competence the deterimant changes from being a dissatisfier to a satisfier. However, it should be noted that the study calculates the coefficient of competence over the entire dataset. It might very well be possible that when accounting for customer heterogeneity relating the influence of competence on institution trust, the effect has an oppostive direction in two segments. As this study accounts for customer heterogeneity based on the three segments, it can offer a method of finding this dissatisfier-satisfier level of competence. Literature relating to this topic calls for more research into the (dis)satisfier function of the determinants of trust (Van Esterik-Plasmeijer & Van Raaij, 2017). This study meets this research objective in that it offers insights in the (dis)satisfier function in one of the determinants of trust, namely competence.
38
In addition, it has the highest density of young customers. This is in line with research, as trust tends to be higher for young customers. Moreover, research also states that trust tends to be higher for females, which is also the case in this study. Segment 1 holds the highest density of males across all segments, while segment 2 with lower trust has a lower percentage of females. Education differs significantly across segment, meaning that it is related to segment membership. This is not in line with a current literature (Fungáčová et al., 2017; Järvinen, 2014). Therefore, this is the first study that finds an effect of education level on trust. An explanation might be that segmenting based on trust and its determinants yields more accurate results than aggregated data, as done in previous research. This might explain the differences in favor of this study. Therefore, final contribution to the literature can be made that using trust and its determinants for segmentation sheds new lights on the role of demographic factors, with especially education, on institution trust.
Limitations and future research
39
approach, one should consult Wedel & Kamakura (1998). The final limitation relates to the fact that this study only used three determinants of trust in its segmentation analysis, while much more of them can be found in the literature (Nienaber et al., 2014). Using more and other determinants might result in better performance in structuring customer heterogeneity. Furthermore, it offers insights in the importance and significance between the various determinants, and might even find negative effects. Future segmentation studies which include more determinants could therefore be of deep interest, and can contribute to the understanding in how trust between customer and firm is determined.
Managerial implications
40
Figure 9
Effect of trust determinants on the three trust segments
Conclusion
41
References
Bijmolt, T. H. A., Paas, L. J., & Vermunt, J. K. (2004). Country and consumer segmentation: Multi-level latent class analysis of financial product ownership. International Journal of
Research in Marketing, 21(4), 323–340. https://doi.org/10.1016/j.ijresmar.2004.06.002
DeSarbo, W. S., Kamakura, W. A., & Wedel, M. (2006). Latent Structure Regression. In
Handbook of Marketing Research, 394–417.
Diacon, S. R., & Ennew, C. T. (1996). Ethical issues in insurance marketing in the UK.
European Journal of Marketing, 30(5), 67–80.
Dimitriadis, S., Kouremenos, A., & Kyrezis, N. (2011). Trust-based segmentation Preliminary evidence from technology-enabled bank channels. International Journal of Bank Marketing,
29(1), 5–31.
Dutch Association of Banking. (2014). Toekomstgericht Bankieren: Maatschappelijk Statuut Code Banken Gedragsregels.
Dutch Association of Banking. (2018). Banking Confidence Monitor 2018.
Ennew, C., Kharouf, H., & Sekhon, H. (2011). Trust in UK financial services: A longitudinal analysis. Journal of Financial Services Marketing, 16(1), 65–75.
https://doi.org/10.1057/fsm.2011.8
EY. (2016). World Islamic Banking Competitiveness Report 2016. EY. Retrieved from
http://www.ey.com/Publication/vwLUAssets/EY-world-islamic-banking-competitiveness-report-2014-15/$FILE/EY-world-islamic-banking-competitiveness-report-2014-15.pdf Floh, A., Zauner, A., Koller, M., & Rusch, T. (2014). Customer segmentation using unobserved
heterogeneity in the perceived-value-loyalty-intentions link. Journal of Business Research,
67(5), 974–982. https://doi.org/10.1016/j.jbusres.2013.08.003
Fungáčová, Z., Hasan, I., & Weill, L. (2017). Trust in banks. Journal of Economic Behavior and
Organization. https://doi.org/10.1016/j.jebo.2017.08.014
Garbarino, E., & Johnson, M. S. (1999). The Different Roles of Satisfaction, Trust, and Commitment in Customer Relationships. Journal of Marketing, 63(2), 70.
https://doi.org/10.2307/1251946
Geyskens, I., Steenkamp, J.-B. E. ., & Kumar, N. (1998). Generalizations about trust in
42
Marketing, 15(3), 223–248. https://doi.org/10.1016/S0167-8116(98)00002-0
Han, R., & Melecky, M. (2017). Broader use of saving products among people can make deposit funding of the banking system more resilient. Journal of International Financial Markets,
Institutions and Money, 47, 89–102. https://doi.org/10.1016/j.intfin.2016.11.005
Hensen, C. (2015). ASN trekt meer klanten, maar de exodus blijft uit. NRC Handelsblad. Herzberg, F., Mausner, B., & Bloch Snyderman, B. (1959). Motivation to Work.
Hoekstra, J. C., Leeflang, P. S. H., & Wittink, D. R. (1999). The Customer Concept : The Basis for a New Marketing Paradigm. Journal of Market Focused Management, 4(1), 43–76. https://doi.org/10.1023/A:1009856028285
Hultman, J., & Axelsson, B. (2007). Towards a typology of transparency for marketing management research. Industrial Marketing Management, 36(5), 627–635.
https://doi.org/10.1016/j.indmarman.2006.04.001
Jansen, D. J., Mosch, R. H. J., & van der Cruijsen, C. A. B. (2015). When Does the General Public Lose Trust in Banks? Journal of Financial Services Research, 48(2), 127–141. https://doi.org/10.1007/s10693-014-0201-y
Järvinen, R. A. (2014). Consumer trust in banking relationships in europe. Marketing
Intelligence and Planning, 32(6), 551–566. https://doi.org/10.1108/IJBM-08-2013-0086
Johnston, R. (1997). Identifying the critical determinants of service quality in retail banking: importance and effect. International Journal of Bank Marketing, 15(4), 111–116. Knell, M., & Stix, H. (2009). Trust in Banks? Evidence from normal times and from times of
crises. In Oesterreichische Nationalbank, 1–29.
Kumar, V. (2018). A Theory of Customer Valuation: Concepts, Metrics, Strategy and
Implementation. Journal of Marketing, 82(1), 1–19. https://doi.org/10.1509/jm.17.0208 Leisch, F. (2004). FlexMix: A general framework for finite mixture models and latent class
regression in R. Journal of Statistical Software, 11, 1–18. https://doi.org/10.18637/jss.v011.i08
Moin, S. M. A., Devlin, J. F., & McKechnie, S. (2017). Trust in financial services: The influence of demographics and dispositional characteristics. Journal of Financial Services Marketing,
22(2), 64–76. https://doi.org/10.1057/s41264-017-0023-8
43
Muhamad, R., Melewar, T. C., & Alwi, S. F. S. (2012). Segmentation and brand positioning for Islamic financial services. European Journal of Marketing, 46(7), 900–921.
https://doi.org/10.1108/03090561211230061
Nienaber, A.-M., Hofeditz, M., & Searle, R. H. (2014). Do we bank on regulation or reputation? A meta-analysis and meta-regression of organizational trust in the financial services sector.
International Journal of Bank Marketing, 32(5), 367–407.
https://doi.org/10.1108/IJBM-12-2013-0146
Ramani, G., & Kumar, V. (2008). Interaction Orientation and Firm Performance. Journal of
Marketing, 72(1), 27–45. https://doi.org/10.1509/jmkg.72.1.27
Rotter, J. B. (1967). A new scale for the measurement of interpersonal trust. Journal of
Personality, 35(4), 651–665.
Singh, J., & Sirdeshmukh, D. (2000). Agency and trust mechanisms in consumer satisfaction and loyalty judgments. Journal of the Academy of Marketing Science, 28(1), 150–167.
https://doi.org/10.1177/0092070300281014
Sirdeshmukh, D., Singh, J., & Sabol, B. (2002). Consumer Trust, Value, and Loyalty in Relational Exchanges. Journal of Marketing, 66(1), 15–37.
https://doi.org/10.1509/jmkg.66.1.15.18449
Smith, W. R. (1956). Product differentiation and market segmentation as alternative marketing strategies. Journal of Marketing, 21(1), 3–8.
Van Esterik-Plasmeijer, P. W. J., & Van Raaij, F. W. (2017). Banking system trust, bank trust, and bank loyalty. International Journal of Bank Marketing, 35(1), 97–111.
https://doi.org/10.1108/IJSE-09-2015-0234
Verhoef, P. C., & Lemon, K. N. (2013). Successful Customer Value Management: Key Lessons and Emerging Trends. European Management Journal, 31(1), 1–15.
Vermunt, J. K., & Magidson, J. (2002). Latent class cluster analysis. In Applied Latent Class
Analyis, 89–106. https://doi.org/10.1017/CBO9780511499531.004
Wälti, S. (2011). Trust no more ? The impact of the crisis on citizens ’ trust in central banks.
Journal of International Money and Finance, 31(3), 593–605.
Wedel, M., & Kamakura, W. A. (1998). Market Segmentation: Conceptual and Methodological
44
Appendix A - GfK market survey questions
1. How much confidence do you have in banks?
2. Can you explain why you have <answer to Question 1> in banks? 3. At which bank or banks do you bank?
4. Which bank do you regard as your main bank?
5. How much confidence do you have in your main bank? [5-point scale] 6. Can you explain why you have <answer to Question 5> in your main bank?
7. To what extent do you agree or disagree with the following statements with regard to your main bank? [5-point Likert scale]
... is open ... is honest
... communicates in a language I understand
... actively informs me of changes in products and services ... listens to customers
... advises on products that are in the interest of customers ... supports me in making financial choices
... searches for solutions with me in the case of financial setbacks ... has knowledge of banking affairs
... has expert personnel
... makes my banking affairs transparent ... meets agreements reached
... is easily accessible (online, by telephone, in branch) … is a solid bank financially
46
Appendix B – Results of the other latent class models
K=2 model results Segment 1 Segment 2 Size (%) (4.5%) (95.5%) Intercept 3.301*** 0.308*** Transparency 0.214*** 0.379*** Customer Orientation 0.652*** 0.259*** Competence -0.790*** 0.220*** sig. levels: *p<.05. **p<0.01. ***p<0.001 K=4 model resultsSegment 1 Segment 2 Segment 3 Segment 4
Size (%) (14.6%) (54.2%) (8.1%) (23.1%) Intercept 3.148*** 0.000*** -1.000*** -0.522*** Transparency 0.170*** 1.000*** 0.250*** -0.043* Customer Orientation 0.680*** 0.000 0.000 0.302*** Competence -0.709*** 0.000 0.750*** 0.754*** sig. levels: *p<.05. **p<0.01. ***p<0.001 K=5 model results
Segment 1 Segment 2 Segment 3 Segment 4 Segment 5
47
K=6 model results
Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6
