The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction

(1)

The Heterogeneity, Nonlinearity and

Asymmetry of the Relationship between

Attribute Performance and Customer

Satisfaction

Consequences for Importance-Performance Analysis

Marnix M. van Loenen

Faculty of Economics and Business, University of Groningen,

Master Thesis Marketing Research, December 2009

Zuiderweg 367, 9744AH Groningen, The Netherlands, +31612925996,

m.m.van.loenen@student.rug.nl, s1500309

(2)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂

Management Summary

The marketing concept is accepted widely nowadays and is based on the identification and satisfaction of customer needs (Kotler 2002). The goal of this concept is to achieve sustainable success, but standard short term measures like sales, profit or even economic expressions such as GNP and utility are unable to reveal the required information. Customer satisfaction research is the obvious solution and is able to improve both the identification and understanding of the satisfaction of customer needs if performed properly and extensively. Furthermore, satisfying customers is the primary obligation of a company and a defensible and appropriate company objective (Peterson and Wilson 1992). In this paper, two studies have been used of organizations that have performed customer satisfaction surveys. The studies were performed in various settings and contained distinct measurements and attributes.

(3)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃

which provides different importance scores for different attribute performance rates. Finally, a link with customer behavior is created. As respondents indicated their intention to recommend the organization, the relationship between customer satisfaction and recommendation intention is investigated as well. Recommendation intention reveals information with relation to the consequences of customer satisfaction for the different segments and is therefore also a valuable tool in developing marketing and resource allocation strategies.

(4)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₄

Preface

Customer satisfaction is a well researched topic in literature. Besides, in today’s financial crisis and economic recession, companies can distinguish themselves by delivering improved service and focusing on customer satisfaction. Potential customers are hard to persuade in these times, but providing additional service may convince them. However, customers should not be treated as homogenous entities and differentiation strategies for improving customer satisfaction have to be developed.

While I was working in a Vodafone store, I realized that delivering improved service from a customer’s perspective was by far the most effective method for a great deal of the customers. Other customers, however, did prefer or simply needed a different approach, which was, for example, more focused on sales. In turn, dissatisfied customers were returning to the store and did not cause frustration for the unsatisfied customer only, but also for the store personnel. By the time I contacted Store Support for this master thesis project, I was already very interested in service quality and customer satisfaction within firms. Additionally, the market research that has to be done before improvement strategies can be developed and implemented seemed to be of great importance and an interesting challenge to me. For this reason, I contacted Store Support to do this research on their behalf. Store Support is a company that uses mystery shopping and customer satisfaction surveys as a tool to support organizations, firms and (semi)governments to achieve their goals and objectives with regard to service quality, customer contact and customer satisfaction. The company is a professional organization for five years and has been growing very rapidly during its existence.

(5)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₅

1 Introduction

Standard (short term) measures, like sales, profit or even economic expressions such as GNP and utility fall short in providing information about the satisfaction of needs of the customer and they do not facilitate the identification of these customer needs. Since the marketing concept is founded on the believe that firms should pursuit the identification and satisfaction of customer needs more effectively than their competitors, in order to achieve sustained success (Day 1994; Kotler 2002), it can be concluded that customer satisfaction is an important construct in marketing. Furthermore, a fundamental finding among a great deal of research supporting the concept, is that satisfying customers is determined to drive higher firm profitability through higher customer retention rates and increased repurchase behavior (Mittal and Kamakura 2001). Following this view, numerous firms include customer satisfaction (CS) in their mission statement or business slogan. The measure provides information about the level of both identification and satisfaction of customer needs and is therefore considered to be important in the marketing concept. The act of measuring is even a proper market connection and in that way a part of the market orientedness of the firm.

These arguments have led to a focus on customer satisfaction, complemented with the statement of Peterson and Wilson (1992) that satisfying customers is the primary obligation of a company and a defensible and appropriate company objective. Consequently, customer satisfaction has become a major cornerstone for customer-oriented business practices and managers become more and more aware of the essence of customer satisfaction, obviously affected by the abundant literature written on this topic (Peterson and Wilson 1992; Szymanski and Henard 2001). As a result, the collection of customer satisfaction data typically is the largest item of a firm’s annual expenditure on market intelligence and it is often one of the few systematic customer intelligence firms generate (Wilson 2002). Direct survey methods are the most widely used means of measuring customer satisfaction (Yi 1990; Peterson and Wilson 1992) and these customer satisfaction surveys (CSS) are conducted mostly on a yearly or ad hoc basis.

(7)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₇

Given that the customer satisfaction construct is found to have great influence on financial and other firm performance measures, researchers as well as managers aim to discover what determines the satisfaction of customers. According to the conclusions of Mittal and Kamakura (2001), however, the weights of the elements for two sets of respondents with different characteristics are found to be statistically different. The elements determining customer satisfaction (attributes) may therefore change among different customer segments. Moreover, the role of each of the attributes in determining customer satisfaction may also differ, since this relationship is often perceived to be asymmetric and nonlinear (Kano 1984; Mittal, Ross and Baldasare 1998; Matzler et al 2004). A proper determination and combination of the element weights, customer segments and nonlinearity will result in improved decision making on resource allocation strategies. To date, however, these factors are not combined into one study. In order to make this combination, the research question of this thesis is as follows: What are characteristics of the relationship between the attribute performance and customer satisfaction? This research consists of three main elements.

First, a method will be presented to estimate what elements are important in determining customer satisfaction for various customer segments. The weights of each of the attributes that determine customer satisfaction will be estimated using latent class regression analysis with overall satisfaction scores as dependent variable and attribute performance scores as independent variables. Furthermore customer characteristics will be used as covariates or concomitant variables in the latent class regression. This methodology of using statistical approaches that include latent variables to model customer satisfaction, has proven to outperform direct ratings of respondents on weights or attribute importance (Gustafsson and Johnson 2004). Moreover, self-stated importance ratings often display a lack of discriminating power between customer satisfaction attributes, since everything is rated as being very important (Myers 2001). Garver (2003) mentions latent class regression as an important tool for further research to calculate statistically inferred importance and performance ratings. In essence, this tool develops segments of customers, based on similar regression coefficients and results in customer segmentation driven by attribute and overall performance scores. Wu et al (2006) have already applied such a methodology, but assuming only linear relationships and without using customer characteristics as covariates.

(8)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₈

revised several times (Crompton and Duray 1985; Matzler, Sauerwein and Heischmidt 2003; Matzler et al 2004; Eskildsen and Kristensen 2006; Deng, Kuo and Cheng 2008). Depending on the interplay of the two dimensions (importance and performance), strategies supporting the marketing concept can be derived (Matzler et al. 2004). The identification of needs is being estimated on the importance scale and the satisfaction of needs on the performance scale. Thus, presenting the results of a customer satisfaction survey on an action grid gives the managers new insights in how to reach persistent success.

Third, a dummy variable regression analysis will be performed, because the asymmetric impact of each attribute’s performance on overall satisfaction has to be assessed before managerial implications can be derived. Since the action grid of the importance-performance analysis is only able to display the linear relationship between customer satisfaction and attribute performance, extensive resource allocation strategies can just be derived after considering nonlinearity and asymmetry. This analysis is included because Mittal, Ross and Baldasare (1998) have investigated the relationship between the attribute-level performance and overall satisfaction and concluded that this relationship is asymmetric and nonlinear. Furthermore, Matzler et al (2004) have applied importance-performance analysis in combination with nonlinearity, resulting in similar conclusions. The dummy variable regression analysis contains an investigation of the attribute weights when attribute performance is low or high compared with average attribute performance.

(9)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₉

2 Literature Review

This chapter discusses the developments in customer satisfaction research over time and the relevance of this construct in today’s literature and business. Furthermore, this chapter deals with the measurement of customer satisfaction by discussing the construct in more detail, as well as how it is defined, what its components are and the link between attribute performance and overall satisfaction.

2.1 Developments in Customer Satisfaction Research

(10)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₀

2007) and the inclusion of customer characteristicshave further enriched the quality of the existing literature (Athanassopoulos 2000; Mittal and Kamakura 2001; Wu et al. 2006; Cooil et al. 2007). In short, customer satisfaction is related to all business elements and contains a great deal of information. Antecedents, consequences and moderators of customer satisfaction have been researched in a growing number of academic studies as well, but the development in measuring customer satisfaction is still making progress (Szymanski and Henard 2001).

2.2 Measuring Customer Satisfaction

Although the antecedents and consequences of customer satisfaction have been subject to research frequently, Dholakia and Morwitz (2002) have discovered that the actual measurement of customer satisfaction alone already changes one-time purchase behavior and relational behaviors (likelihood of defection, aggregate product use, and profitability) of the appraised customer. Moreover, when it is measured, customer satisfaction will significantly contribute to decision-making in the quality/operations area and in overall strategic management control within the firm (Piercy and Morgan 1995). The essence of gathering information from customers and establishing market connections is investigated by Gebhardt, Carpenter and Sherry (2006). They note that market connections are crucial for creating a market-oriented culture and enabling the identification of needs and the degree of satisfaction with those needs. It may consist of, for example, surveys, focus groups and other research activities, but also cross-functional field studies. Customer satisfaction information usage, in turn, lies at the heart of a firm’s market orientation (Morgan, Anderson and Mittal 2005). Furthermore, market orientation has a positive impact on organizational performance (Kirca, Jayachandran and Bearden 2005; Verhoef and Leeflang 2009). Thus, actually measuring customer satisfaction is relevant in addition to only considering the satisfaction of customers as a construct. Moreover, offering opportunities to the customers to reveal their opinions and maintaining interaction also improves the financial results of companies through the (improved) market orientation of the organization.

(11)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₁

economy. In order to measure this quality, the American Customer Satisfaction Index (ACSI) was introduced (Fornell et al. 1996). The ACSI actually is a measure of customer satisfaction which is uniform and comparable among firms, industries and national economies and in this way is trying to capture the quality of economic output as a supplement of the quantitative measures like GNP. Throughout the years, its usefulness is proved by the great deal of literature referring to this measure. It has been used to demonstrate that a firm’s financial performance is related to the level of satisfaction, as reported by its customers (see for example Anderson and Fornell 2000; Athanassopoulos 2000; Cronin, Brady and Hult 2000; Homburg and Giering 2001; Mittal and Kamakura 2001; Gustafsson and Johnson 2004). The ACSI score is based on perceived quality and customer expectations as determinants of the customer satisfaction. It gives a clear understanding of the consequences of customer satisfaction on customer loyalty and customer complaint. Moreover, it enables a comparison of perceived performance between competitors (Varki en Rust 1997). Comparing customer satisfaction with competitors is also a major element of the marketing concept, since the sustained success can only be achieved when satisfying and identifying customer needs more effectively than competitors.

(12)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₂

3 Methodology

Besides the aforementioned literature, several other theoretical findings support the attribute-based approach in which the strong relationship between perceived attribute performance and customer satisfaction are emphasized even further. Firstly, the methodology is based on the belief that customers are more likely to review the attributes rather than the entire product (Gardial et al. 1994). Secondly, customers may have mixed feelings toward a product or service, because they may be satisfied with one attribute but dissatisfied with another. By letting the customers judge several attributes simultaneously, the performance and importance of each single attribute can be displayed in a relative relationship which enables a comparison between attributes within the firm and between competitors (Oliva, Oliver, and Bearden 1995). Furthermore, practical or managerial implications resulting from the analysis of attributes and overall CS are very straightforward and easy to understand. Finally, Mittal, Ross and Baldasare (1998) believe that an attribute-level approach is justified since it provides researchers (and managers) a higher level of specificity and diagnostic usefulness as compared to a general satisfaction approach. So, in this paper one-time attribute performance ratings and overall satisfaction with the organization will be used to identify their relationship, both linear and asymmetric or nonlinear. The attribute weights determined over a single cross-sectional analysis, however, should not be generalized over the entire life of a customer relationship (Mittal, Katrichis and Kumar 2001) and key drivers of overall satisfaction may vary over time (Mittal, Kumar and Tsiros 1999). Thus the existence of a relationship between perceived attribute performance and overall satisfaction scores is built up around several service quality elements as defined by Parasuraman et al. (1988), in this chapter the characteristics of this relationship will be further elaborated on.

(13)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₃

3.1 Customer Heterogeneity

“The estimation of a single aggregate regression equation across all consumers in a sample may be inadequate and potentially misleading if the consumers belong to a number of unknown classes (segments) in which the regression parameters differ.” (Leeflang et al. 2000, pp 454)

As stated before, the amount of customer satisfaction research is quite extensive. Nevertheless, studies handling this topic vastly assume consumer homogeneity (Garry 2007). Positive exceptions are for instance Athanassopoulos (2000), Mittal and Kamakura (2001), Wu et al. (2006) and Cooil et al. (2007), who consider the different behaviors of distinct consumer groups with respect to customer satisfaction. These different behaviors may initiate firms to focus on different attributes for customers with different levels of intrinsic retainability, i.e. since the importance and performance of attribute differs for each group of customers, managers should set up segment-specific improvement activities. Segmentation of customers has already revealed information with respect to behavioral patterns that otherwise would have been masked under a universal assessment of satisfaction scores (Athanassopoulos 2000) and therefore is considered to be useful. To emphasize this finding, Mittal and Kamakura (2001) computed attribute weights for two sets of respondents with different characteristics and were found to be statistically different.

(14)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₄

employed, to uncover the nature of heterogeneity (Wu et al 2006). All these findings show a wide variety of support for the fact that satisfaction ratings are being dependent on customer characteristics as well as attribute performance, and it motivates a fully segmented approach (Wu et al. 2006).

3.2 Latent Class Regression Analysis

To optimize overall satisfaction, firms maximize performance on attributes that have the largest weight in determining overall satisfaction (Mittal and Kamakura 2001). This statement implies that the characteristics of the relationship between customer satisfaction and perceived attribute performance include that performance on attributes determines overall satisfaction, and that different attributes have different weights. The magnitude of these weights can be determined by performing a regression of the attribute performance on overall satisfaction ratings. Such a statistical method derives the importance of the attributes implicitly from the data (Crompton and Duray 1985; Jaccard, Brinberg and Ackerman 1986). Gustafsson and Johnson (2004) have compared direct importance ratings with a variety of methods for statistically deriving attribute importance. The authors conclude that direct ratings are outperformed by statistical approaches that include latent variables to model satisfaction. Moreover, stated importance ratings often display a lack of discriminating power between customer satisfaction attributes, everything is rated as being very important (Myers 2001).

Wu et al. (2006) propose a maximum likelihood based latent structure factor analytic methodology, based on expectancy-disconfirmation theory in customer satisfaction. This is a method to calculate statistically inferred importance ratings for each customer group through creating dimensions from attributes. It develops segments of customers based on similarity in regression coefficients and results in customer segmentation driven by attribute and overall performance scores. The superiority of the Wu et al. (2006) model over a number of various other model specifications is empirically demonstrated using customer satisfaction survey data. Segments were created and graphically displayed from customer data on performance, expectation and disconfirmation of attributes only, while no customer characteristics were taken into account. Furthermore, linear and symmetric relationships between the attributes and overall satisfaction were assumed. This research, however, shows that creating segments is valuable in analyzing customer satisfaction survey data.

(15)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₅

segments of customers based on similarity in regression coefficients and results in customer segmentation driven by attribute and overall performance scores. Simultaneously, through the concomitant variable mixture regression the derived segments are profiled with demographic and customer characteristic variables (Leeflang et al. 2000). Subsequently, a regression analysis with dummy variables is used to assess the asymmetric relationship between attribute-level performance and overall satisfaction. The indication by Mittal, Ross and Baldasare (1998) that nonlinearity and asymmetry exists requires such an additional analysis.

The results of customer satisfaction surveys should be presented clearly, so that understanding the findings requires no sophisticated statistical knowledge and implementation of improvements is straightforward. Also the comparability of results of several surveys and scores of competitors require a proper presentation method. The most suitable tool for evaluating the attributes and translating the results of a customer satisfaction survey into action is the action grid resulting from an importance-performance analysis, since it provides a clear overview of attributes that require additional (less) attention or improvement efforts (Martilla and James 1977, Sampson and Showalter 1999; Garver 2003). Since the introduction of the Importance–Performance analysis (IPA) by Martilla and James (1977), the technique has been applied in numerous papers and research in various industries like tourism (Wade and Eagles 2003) and other service providers (Ford, Joseph and Joseph 1999). It is an easy-to-use and easy-to-understand technique that can yield important insights into the attributes that require more versus less attention, it yields prescriptions for the management of customer satisfaction and is presented on a two-dimensional grid, which causes quite straightforward interpretable results (Matzler, Sauerwein and Heischmidt 2003). Typically, data from customer satisfaction surveys are used to

construct a two-dimensional action grid matrix where importance is depicted along the x-axis and performance (satisfaction) along the y-axis (Matzler et al. 2004). In figure 1, the recommendations for customer satisfaction management emerge and it offers a visual display of analysis results, which is easily interpreted by practitioners who may not be very knowledgeable of statistical procedures (Crompton and Duray 1985). In this way, it is a potentially valuable tool for translating market research results into action.

(16)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₆

3.3 Nonlinearity and Asymmetry

There is, however, a remark since using a statistical method implying linear relationships is insufficient, because several researchers have found nonlinearity and asymmetric properties of the relationship between attribute performance and overall satisfaction (Mittal, Ross and Baldasare 1998; Matzler et al. 2004). Since quality attributes can be divided into three categories (basic, performance and excitement [see Kano et al. 1984]), with distinct characteristics, asymmetries should be considered, in order to assess the impact of the different attributes on overall satisfaction correctly (Matzler et al. 2004). Basic factors are entirely expected or basic necessities that cause dissatisfaction if not fulfilled, but do not lead to customer satisfaction if fulfilled or exceeded; negative performance on these attributes has a greater impact on overall satisfaction than positive performance has. Excitement factors, conversely, surprise the customer when performance is high and are in essence the opposites of basic factors. These factors increase customer satisfaction if delivered well, but do not cause dissatisfaction if they are not delivered; in other words, positive performance on these attributes has a greater impact on overall satisfaction than negative performance. Performance factors lead to satisfaction if performance is high and to dissatisfaction if performance is low. In this case the attribute performance–overall satisfaction relationship can be considered as linear and symmetric.

(17)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₇

3.4 Recommendation Intention and Customer Satisfaction

(18)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₈

4 Data collection and analysis

The main purpose of this section is to give a proper indication of the method of data collection, adjustment and data structure, as well as method of data analysis. The models that will be presented are based on attribute performance scores and overall satisfaction ratings from respondents of customer satisfaction surveys. This data is supplemented with customer characteristics and background information of the respondents and together, these are the covariates or concomitant variables. Since transaction-specific satisfaction is a less fundamental indicator of the firm's past, current, and future performance (Anderson, Fornell, and Lehmann 1994), the respondents judge the organizations in a complete perspective.

Two studies will be used to test the relationship between the attribute performance scores and overall satisfaction as well as the role of customer heterogeneity in this relationship. The data of both studies are collected through customer satisfaction surveys performed by Store Support, a commercial external party specialized in mystery shopping and customer satisfaction research. The research process has been performed in consultation with the sponsoring organizations, which are non-profit organizations in both cases. Findings of the surveys have already been communicated to the clients, latent class analysis, however, has not been part of these reports. A wide variety of customer characteristics was measured during both surveys, but not all of them were usable in this paper. Moreover, effects of customer characteristics are most likely to be industry- and/or category-specific, so in the view of generalizability, these may have little value (Mittal and Kamakura 2001). The usability of the customer characteristics, however, will be shown in general. Due to confidentiality, research outcomes are limited and the names of the organizations are hidden.

(19)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₁₉

standards and proper briefings to the interviewers. Since the surveys were conducted by professionals, this should not be a problem in the present data. The problem of reactivity is overcome by using the latent class regression approach, which distills the underlying (true) performance scores from the data. Furthermore, Mittal and Kamakura (2001) mention that observed satisfaction ratings are error-prone measures of true (latent) satisfaction, which means conclusions cannot be derived from individual, but only from (partly) aggregated scores. Personal characteristics and background information of the respondents were also collected, since the customers of firms are not homogenous, and the satisfaction ratings are likely to be influenced by a person’s vision and/or environment. Mittal and Kamakura (2001) also found that a particular indicated satisfaction score does not necessarily have the same implication for the one respondent as compared to another, which is, in fact, heterogeneity. Therefore, latent classes will be created on the basis of differences in needs and satisfaction with those needs and personal characteristics of respondents, i.e. the methodology is based on differences in the relationships between overall satisfaction and attribute perceived performance.

For the models, several assumptions are made. The first premise underlying this modeling approach is that the attribute scores measured are determinants of the overall satisfaction. Secondly, the relationship between the determinants and overall satisfaction is different for distinct groups of customers. Furthermore, other elements, which are not being measured, could also possibly influence the overall satisfaction score. These elements are represented by the error terms in the equations. These errors, or residuals, are assumed to be normally distributed and independent, which will be tested after modeling is executed.

Study 1 (Festival)

(20)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₀

As mentioned before, the methodology is based on the relationship between attribute performance and overall satisfaction. Heterogeneity causes differences in this relationship among customer groups, which will be accounted for in the model. Using the latent class regression method, differences in this relationship will be detected and composed into segments. Equation (1) displays the model which will be used to distill the importance of each attribute per customer segment.

i m mi m|s 0|s i|s

=

+

∑

+

ε

M

Α

β

OS

(1) Let: i = 1, . . . , I customers; s = 1, . . . , S latent/unknown segments; Ami = 1, . . . , M attribute scores for customer i;

β0|s = the intercept or basic overall satisfaction score of segment s;

βm|s = the importance of attribute m in segment s;

OSi|s = observed overall satisfaction rating for customer i, given segment s;

εi = the error term of the equation for customer i.

The dependent variable – overall satisfaction – is modeled as a function of the explanatory variables – the attributes – as in a standard multiple regression model. The differences in coefficients for each independent variable and attribute scores between the segments will be presented as respectively importance and performance of attributes on the importance-performance action grids. In order to make the scores comparable among customer segments, the standardized coefficients will be used in the action grid as indicators of importance. Customer characteristics will be used as covariates or concomitant variables in this model, which means that the segments will be simultaneously profiled with the customer characteristics. This is not a two-step procedure, since such models have the disadvantage that the membership probabilities of the covariates are performed independently from the model estimation and optimizes a different criterion (Leeflang et al 2000). Therefore, equation (1) is expanded with the core of the concomitant variable approach: the prior probabilities are explicitly reparameterized as functions of these concomitant variables.

It is assumed that the respondents arise from the population of the visitors of the festival which is a mixture of S unobserved segments, with unknown proportions π1, ..., πs. It is unknown in advance to

which segment a particular respondent belongs. The properties of the probabilities that a respondent belong to segment s are as follows:

(21)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₁

Given that respondent i comes from segment s, it follows from equation (1) that the distribution function for the overall satisfaction score is represented by the general form fs (Ami|βm|s). Here βm|s

denotes all unknown parameters for segment s. To develop a latent class model with covariates, let c = 1,…, C concomitant variables, and zci the value of the c-th concomitant variable for respondent i.

The unconditional distribution for the regression is formulated as follows:

)

C

,

β

A

(

|

)

β

,

(A

, S s ms mi s s m mi

z

f

s

∑

=

π

(3)

In equation (3), the core of the concomitant variable approach, πs|Z, is based on the prior

probabilities of segment membership that are explicitly reparameterized as functions of the concomitant variables (Leeflang et al. 2000). In order to determine πs|Z, a logistic formulation is used

with parameters denoting the impact of each covariate on the prior probability for segment s. Since the segments are based on a-priori probabilities, there is not a single solution. The results, however, can be compared using the BIC and CAIC information criteria, which are based on the log-likelihood of the model solution. For this type of information criteria, the lowest value means a best fitting solution. The covariates or concomitant variables used are transportation method, living region, party (the composition of the visiting group), reason for coming, money spent, age, and on which part of the day the interview was held.

(22)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₂

the asymmetric impact of attribute-level performance on overall performance (Matzler et al 2004). Table 1 provides an overview of how the attribute scores are divided into the three categories. For example, the Parking attribute has an average of 7.4 and standard deviation is 1.5, then respondents consider this attribute to have low performance when rating it ≤ 5.9 and high performance when rating the attribute ≥ 8.9. The Parking attribute may now have a larger impact on customer satisfaction for respondents who rated the attribute as performing low versus respondents who rated the attributes as performing high. This difference is revealed by regressing both the low performance dummy variable and the high performance dummy variable on customer satisfaction simultaneously.

Attribute Score Low Performance Dummy High Performance Dummy

≤ Average – Standard Deviation 1 0 > Average – Standard Deviation and

< Average + Standard Deviation

0 0

≥ Average + Standard Deviation 0 1

Table 1, The division of attribute performance scores for dummy variable regression analysis

Study 2 (Business Organization)

This study is based on a customer satisfaction survey which was held among Dutch businesses with a membership in a business organization. This organization represents on both national and international level the common interests of those businesses and industries. The organization also provides various services to its members and organizes several network meetings. The objectives of the organization include representation of the companies on national and international scale and delivering service to companies such as network meetings.

(23)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₃

The methodology is largely similar to the methodology of study 1. The determinants of overall satisfaction in this study, however, are composed dimensions following the five subcategories of the questionnaire. Using the dimension scores – which are the average scores on the attribute questions in the particular subcategory – is justified since the 16 original attributes are highly mutually correlated and descriptive validity will be improved using this methodology. Second, the dimensions are input for the latent class analysis, with the dimensions as independent and overall satisfaction as the dependent variable in the same way it is done in the previous study. The differences in coefficients and attribute scores for each independent variable are again presented as respectively importance and performance of attributes on the importance-performance action grids. In order to make the scores comparable among the customer segments, the standardized coefficients will be used in the action grid as indicators of importance. For this model, recommendation rates are used as a control variable to check the internal consistency of each segment and relating the satisfaction scores to customer behavior. Another regression analysis is included in order to check whether overall satisfaction is related to intention to recommend and whether this relationship is different among the segments.

Thus, attributes are summarized in dimensions in this study which, in turn, will be regressed against the overall satisfaction scores. Heterogeneity causes differences in this relationship among customer groups, which will be accounted for as well. Using the latent class regression method, differences in this relation will be detected and composed into segments. Equation (4) displays the model which will be used to distill the importance of each dimension per customer segment.

i ti s t 0|s i|s

=

+

∑

+

ε

T t |

D

β

OS

(4) Let: i = 1, . . . , I customers; s = 1, . . . , S latent/unknown segments;

Dti = 1, . . . , T average dimension scores for customer i;

β0|s = the intercept or basic overall satisfaction score in segment s;

βt|s = the importance of dimension t in segment s;

OSi|s = observed overall satisfaction rating for customer i, given segment s;

Equation (4) is the mathematical representation of the latent class regression analysis. It measures the overall satisfaction of consumer i, given latent/unknown segment s. For every dimension t in segment s, parameters (βt|s) are measured. Equation (2) and (3) further complement this model,

(24)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₄

overall satisfaction, but which are not included in the model. It also includes measurement or methodological errors. The covariates or concomitant variables used in this model are membership activity; location of the head office (province); the age of the relationship between the organization and the company; number of meeting visits; what the respondent expects from meetings; whether new customers were contracted through meetings; whether respondents miss topics at meetings and method of information acquisition.

The results can be compared using the BIC and CAIC information criteria, which are based on the log-likelihood of the model solution. For this type of information criteria, the lowest value means a best fitting solution. Subsequently, the importance and performance scores will be displayed on the action grids. Furthermore, the asymmetric impact of each attribute’s performance on overall satisfaction has to be assessed before managerial implications can be derived in this study as well. To this end, a regression analysis with dummy variables will be performed similar to the methodology of the first study in this report.

Ultimately, the internal consistency in combination with the intention to recommend is tested. The latent class regression results and the distinctiveness of the segments will be tested with a binary logistic regression analysis with the recommendation intention (yes/no) as dependent and overall satisfaction as independent variable.

(5)

(6)

Let:

i = 1, . . . , I customers;

s = 1, . . . , S latent/unknown segments; η0|s = the constant or intercept of segment s;

η1|s = the weight of OSi given segment s;

RIi|s = recommendation intention probability of customer i, in segment s;

OSi = observed overall satisfaction rating for customer i;

This analysis results in different interpretations of the impact of overall satisfaction on the intention to recommend. The logistic regression is displayed in equations (5) and (6). It will result in recommendation intention probabilities between 0 and 1, with probabilities <0.5 means no intention to recommend and scores >0.5 indicate a positive intention to recommend.

(25)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₅

5 Results

Study 1 (Festival)

The latent class regression analysis resulted in several possible solutions, ranging from a homogeneous population to be considered to segmentation in five segments. Table 2 displays the model heuristics, which indicate that the three-segment solution is the best solution according to the BIC and CAIC scores. These criteria are based on the log-likelihood of the solutions and include a penalty for the number of parameters included in the model. For this type of information criteria a lower value indicates a better model fit, which leads to a three-segment solution.

Segments (S) BIC CAIC

1 2350.22 2357.22

2 469.64 513.64

3 267.41 186.41

4 1144.21 1026.21

5 1155.62 1000.62

Table 2, Model 1 selection diagnostics

(26)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₆

the activities, while the scores on catering facilities are higher in segment 3, compared to segment 1 and scores on signing are significantly higher in segment 2 compared to segment 1.

The results have face validity, since the model provides no unreasonable results following from the scores on the attributes for all segments. The predicted values following from the model range between 6.4 and 9.9, while the range of the residuals is largest in segment 2. Residual diagnostics results in satisfied assumptions in all cases. The disturbances are normally distributed for each of the separate segments and there is no correlation between the independent variables of the model and the residuals. Moreover, there are no outliers visible on a scatter plot, indicating that the model is not suffering high leverage from a small part of the data. In addition, according to Variance Inflation Factors, the attributes are not mutually correlated, since VIF score did not exceed 1.5.

Model 1, N = 928 Average Scores Parameter Values Standardized Coefficients

Aggregate model (Constant) 3.355***

N = 928 Activities 7.7 .381*** .393

Catering facilities 7.5 .140*** .179

Signage 7.2 .034 .048

Adj. R2 = 0.279 Public toilets 6.5 .031* .051

Mean OS = 8.1 Parking 7.0 .046** .069

Segment 1 (Constant) 3.800***

N = 560 Activities 7.6 .361*** .393

(60%) Catering facilities 7.4 .087*** .121

Signage 7.2 .026 .040

Adj. R2 = 0.252 Public toilets 6.5 .067*** .121

Mean OS = 8.1 Parking 7.0 .033 .052

N = 246 Activities 7.7 .370*** .339

Signage 7.4 .097** .120

Adj. R2 = 0.274 Public toilets 6.5 -.027 -.039

Mean OS = 8.1 Parking 7.1 .039 .057

N = 122 Activities 8.1 .533*** .540

Signage 7.2 -.096 -.126

Adj. R2 = 0.422 Public toilets 6.6 -.021 -.034

Mean OS = 8.3 Parking 6.9 .119** .171

Dependent Variable: Overall satisfaction with the event *= p-value <0.1, **= p <0.05, ***= p <0.01

Table 3, Model 1 latent class regression results

(27)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₇

horizontal axis with axis scales equal among segments. The figures display the importance and performance scores of each of the independent variables. For each of the segments, the activities are most important in determining the overall satisfaction, but performance is good as well although it is both more important and performed better for customers in segment 3. The catering facilities are less important and have lower performance for each segment compared to the activities. In segments 2 and 3, this attribute is part of Quadrant I (“Keep up the good work”), while in segment 1 it is part of quadrant IV (“Possible overkill”). There are no attributes located in the high importance/low performance quadrant, which in itself is positive. Public toilets (segment 1) and parking (segment 3), however, are very close to the border of this quadrant and therefore need some attention.

(28)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₈

Before managerial implications can be derived, the asymmetry of the relationship between the attribute score and overall satisfaction has to be measured. Table 4 shows the impact of each of the variables when performance is low and also the importance of the variables when attribute performance is high for each of the segments. The parameter scores in table 4 are based on regression analyses where attribute scores are recoded into low performance when below mean minus standard deviation. High performance is based on attribute scores above mean plus standard deviation. Table 4 shows therefore the relationship between attribute performance and overall satisfaction when respondents are dissatisfied with a particular attribute (column 1) and when respondents are highly satisfied with the attribute (column 2). Strikingly, the significance of the attributes is particularly present when performance is high, while the same attributes that were significant before, again have the highest importance.

Segment 1 Dummy-variable regression coefficients

Adj. R2 = 0.209 Low Performance High Performance

Activities -0.485*** 0.553***

Catering facilities -0.076 0.356***

Signage -0.110 0.090

Public toilets -0.090 0.281***

Parking 0.084 0.234*

Activities -0.538*** 0.648***

Signage -0.162 0.258

Public toilets 0.023 0.122

Parking -0.106 -0.107

Activities -0.290 0.617***

Signage -0.004 -0.081

Public toilets -0.068 0.023

Parking -0.326 0.477*

Table 4, Dummy variable regression results, model 1

(29)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₂₉

for all segments, because its importance is high with high performance, but is not affecting the overall satisfaction with low performance. Signage is a significant variable for segment 2 only and is a performance factor as well. The public toilets and parking are excitement factors for respondents in segment 1, it really helps improve their overall satisfaction when performance is high, but low performance is not influencing the overall satisfaction score. The same is true for the parking attribute for segment 3 although this could also be marked as performance factor. Surprisingly, basic factors have not been found in this study.

Based on the importance/performance scores of the different segments and the customer characteristics of respondents in these segments, profiles can be created to indicate what distinguishes the segments. In table 5, the within class percentages of each segment are displayed. A description of each of the segments is given in tables 6-8.

(30)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₀

Customer characteristics per segment Segment 1 (60%) Segment 2 (27%) Segment 3 (13%)

Gender Male 53% 62% 62% (p-value = 0.023) Female 47% 38% 39%

Total 100% 100% 101%

Transportation Car 65% 86% 19% (p-value = 0.004) Public Transportation 25% 4% 11% Bicycle 7% 6% 34% Walking 1% 0% 26% Boat 2% 0% 9% Other 0% 4% 1% Total 100% 100% 100% Region Groningen 53% 88% 62% (p-value = 0.000) Other northern provinces 32% 2% 14% Eastern provinces 11% 4% 1% Other Dutch provinces 2% 6% 20% Other countries 3% 0% 3% Total 100% 100% 100% Company Single 10% 20% 15% (p-value = 0.000) Family 17% 33% 34% Group 37% 17% 14% Partner 36% 31% 37% Total 100% 100% 100%

Reason Boats and ships 65% 44% 36% (p-value = 0.000) Activities 3% 10% 6% Artists 1% 5% 14% Sociability 27% 28% 44% Other 4% 13% 0% Total 100% 100% 100% Spending €0 to €10 17% 19% 9% (p-value = 0.000) €11 to €25 28% 27% 23% €26 to €50 34% 31% 35% €51 to >€150 20% 24% 33% Total 100% 100% 100% Age in years <15 to 35 19% 28% 15% (p-value = 0.000) 36 to 45 21% 26% 22% 46 to 55 24% 25% 25% 56 to 65 20% 15% 22% 66 to >86 15% 6% 17% Total 100% 100% 100%

Part of the day Early afternoon (12 to 3 p.m.) 28% 67% 48% (p-value = 0.000) Late afternoon (3 to 6 p.m.) 47% 8% 28% Evening (6 to 9 p.m.) 26% 26% 23%

Total 100% 100% 100%

Day of interview Day 1 37% 26% 51% (p-value = 0.000) Day 2 16% 24% 19% Day 3 25% 26% 18% Day 4 22% 24% 12%

Total 100% 100% 100%

Number of visiting days 1 Day 70% 63% 34% (p-value = 0.000) 2 Days 17% 18% 19%

3 Days 6% 7% 16%

4 Days 7% 13% 31%

Total 100% 101% 100%

Returns next time Yes 95% 96% 99% (p-value = 0.119) No 5% 5% 1%

_Total _100% _101% _100%

(31)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₁

Segment 1: Self-supplying spectators of ships

This group contains relatively many women. Visitors in this segment have come mainly by public transportation, even 85% of all public transportation visitors are in this group. Slightly more than half of the respondents live in Groningen; the rest is from surrounding provinces in the northern or eastern area. Visitors in this segment mainly come with a partner or a group. The most important reason for coming are the boats and ships and about one third of this group spends between €26 and €50, while overall spending is on average, as well as the age of the visitors in this group. Segment 1 contains almost all visitors who are interviewed in the late afternoon (83% of all respondents between 15 and 18h are in this group). The spreading around the days is on average, while visitors in this segment come for just 1 day. Most of these visitors will return the next time.

Respondents in this segment like the activities, which are quite important, while they are dissatisfied with the toilets, which is a significant variable either. Since these visitors are by car or public transport and do not live very close to the event, they appreciate clean toilets, but those should be improved as well. The public toilets and parking are excitement factors, they really help improve overall satisfaction when performance is high, but low performance is not considerably influencing the overall satisfaction. The boats and ships are the most important reason for coming and although they visit the event with a group or partner, spending is not quite high and catering is relatively unimportant (approx. equal to the toilets). Therefore this segment is called the self-supplying spectators of ships segment. The organization should focus on improving the catering facilities and toilets in order to make these visitors more satisfied, because these attributes have high importance when performance is also high. In this way, the position of the attributes on the action grid will move towards the “Keep up the good work” quadrant as is displayed in appendix I.

Table 6, Description of model 1, segment 1

Segment 2: People just driving by

The second segment consists mainly of men, almost all visitors are by car and coming from

(32)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₂

this segment, while toilets have a low performance, but also low importance. Signing is a significant variable for these people and they are quite satisfied with it as well. Visitors in this segment are by car, live in the area (within 40 kilometers) and have come relatively often alone. They spend very little, which is most likely caused by the fact that they visit early in the afternoon and have to drive a car, so catering facilities will not be used intensively. In order to make these visitors more satisfied, the organization should focus on the signage, since low attribute performance has a significant negative effect, while high attribute performance has significant positive influence on satisfaction. A further improvement of the attribute will make it move towards the first quadrant as is shown in appendix I. Furthermore, by offering improved catering facilities, visitors in this segment may become more satisfied, spend more and might even visitors by car to make use of more environmentally responsible transportation.

Segment 3: Entertainment seeking locals

In this segment, many visitors are male, many have come walking or by bicycle, which includes 60% of visitors. These people most likely live close to the event terrain. This corresponds to the percentage of 62% of respondents from the Groningen province in this group. The relationship between region and transportation method is significant for this cluster. Most people are with their partner and relatively many visitors are with the family, while a small percentage is with a group. The most important reason for people in this segment is the sociability and 50% of the respondents who said to come for the artists are in this cluster. Only 9% of respondents in this cluster spend between €0 and €10. One third of the respondents in this segment spend more than €50 which results in above average spending. The distribution of the age categories of respondents in this group is equal to segment 1. About 50% of the interviews are held early in the afternoon. The visitors in this group are the first day visitors, but come to the event on more than one day as well. Everyone in this segment will return the next time the festival will be held.

(33)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₃

performance on parking facilities, since this is important when performance is low as well as in cases where performance on this attribute is high. Although this attribute is presented in the “Low priority” quadrant at this moment, improvements on performance are necessary and will make it move towards the “Keep up the good work” quadrant. Appendix I contains also an improvement path on the action grid for this segment.

Study 2 (Business Organization)

For this model, the latent class regression analysis resulted in several possible solutions, ranging from a single up to five segments to be considered. As table 9 shows, the three-segment solution is the best solution according to the BIC and CAIC scores. These criteria are based on the log-likelihood of the solutions and include a penalty for the number of parameters included in the model. For this type of information criteria a lower value indicates a better model fit, which leads to a three-segment solution.

Segments (S) BIC CAIC

1 591.94 598.94

2 679.57 716.57

3 260.57 327.57

4 878.03 975.03

5 378.68 505.68

Table 9, Model 2 selection diagnostics

(34)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₄

have a significantly negative effect on overall satisfaction for members in this segment. This suggests that members in this segment dislike member meetings or at least are discontented with the large quantity of meetings that is being organized. Finally, the results clearly indicate that the differences in parameter values would have been remained unmasked when considering only the aggregate model.

Model 2, N = 327 Average Scores Parameter Values Standardized Coefficients

Aggregate model (Constant) -0.461

N = 327 Advocacy 3.7 0.271*** 0.181

Quality of Network 3.8 0.691*** 0.583

Adj. R2 = 0.535 Member Meetings 3.9 -0.027 -0.020

Information Provision 3.8 0.035 0.025

Mean OS = 3.7 Attainability 4.3 0.116** 0.089

Segment 1 (Constant) 0.047

N = 187 Advocacy 3.8 0.327*** 0.252

(57%) Quality of Network 4.0 0.564*** 0.476

Adj. R2 = 0.449 Member Meetings 4.0 -0.110 -0.101

Information Provision 3.9 0.147* 0.125

Mean OS = 4.0 Attainability 4.5 0.059 0.048

Segment 2 (Constant) -0.590

N = 99 Advocacy 3.6 0.236** 0.149

(30%) Quality of Network 3.5 0.794*** 0.648

Adj. R2 = 0.607 Member Meetings 3.8 0.222** 0.161

Information Provision 3.8 -0.134 -0.093

Mean OS = 3.5 Attainability 4.2 0.017 0.013

Segment 3 (Constant) 1.563*

N = 41 Advocacy 3.5 -0.182 -0.164

(13%) Quality of Network 3.3 0.339** 0.357

Adj. R2 = 0.394 Member Meetings 3.8 -0.488*** -0.450

Information Provision 3.6 0.186 0.153

Mean OS = 3.0 Attainability 3.9 0.535*** 0.576

Dependent Variable: Overall satisfaction with the organization *= p-value <0.1, **= p <0.05, ***= p <0.01

Table 10, Model 2 latent class regression results

(35)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₅

segment 2 and 3, while the scores in segment 2 are again significantly higher than scores in segment 3. All scores for the dimensions in segment 1 are significantly superior to the scores in both segment 2 and segment 3 as well. In turn, only performance on attainability are superior in segment 2 compared to segment 3, the other are not significantly different between segments 2 and 3.

The results have face validity, since the model provides no unreasonable results following from the average dimension scores for all segments. The predicted values following from the model range between 1.0 and 4.9, while the range of the residuals is largest in segment 1. Residual diagnostics results in satisfied assumptions in all cases. The disturbances are normally distributed for each of the separate segments. There is no correlation between the independent variables of the model and the residuals. Furthermore, according to the Variation Inflation Factor, the dimension are not mutually correlated, since this value did not exceed 1.8 so multicollinearity is successfully suppressed.

(36)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₆

In order to make a proper comparison between the segments, the standardized coefficients are used as input for the importance-performance analysis. The action grids are created with the absolute standardized coefficients as input for importance and mean attribute score as input for performance. The results are shown in figure 4 above, with importance on the vertical and performance on the horizontal axis, with axis scales equal among segments. The figures display the importance and performance scores of each of the independent variables per segment. For both segments 1 and 2, the quality of network of the organization is the most important dimension in determining the overall satisfaction, but performance scores are remarkably different. For segment 1, the performance is relatively high, while the performance is extremely low for segment 2. The attainability of the organization is the most important dimension for segment 3, while member meetings are quite important also, but as table 10 shows, its relationship with customer satisfaction is negative. Furthermore, the quality of network is next most important dimension of segment 3. For this first and second dimension, performance is on average, while for the third performance scores are even lower than performance in segment 2 on quality of network. Furthermore, attainability is performing well in cases where its importance is relatively low, while advocacy has relatively higher performance ratings in segments with relatively higher importance, although it is not performing above average in any segment. Note that both segment 2 and 3 have no dimensions located in Quadrant I (“Keep up the good work”), which implies insufficient satisfaction of needs for those customers. This also corresponds to the average overall satisfaction scores per segment which is highest in segment 1 and lowest in segment 3. The organization should, based on the action grids, concentrate on advocacy in order to improve customer satisfaction for segment 1 customers and little on information provision. Improvements in segment 2 can be made by concentrating on the quality of network with little attention to member meetings and advocacy. For segment 3, quality of network and information provision should be improved and also attainability in order to improve satisfaction.

(37)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₇

the performance of this dimension is low, and advocacy is only significantly important when performance is high. Segment 3 contains besides the quality of network dimension also member meetings as significant variable for both low and high performance. Furthermore, the attainability dimension is only significant when performance is low. Again, the member meetings dimension has an opposite effect in segment 3; low performance of this dimension is improving customer satisfaction, while high performance lowers the overall satisfaction significantly.

Advocacy -0.332*** -0.045

Quality of Network -0.564*** 0.485***

Member Meetings -0.019 -0.045

Information Provision -0.374*** 0.105

Attainability 0.096 0.152*

Advocacy -0.234 0.394**

Quality of Network -1.099*** 0.721***

Member Meetings -0.337* 0.277

Information Provision 0.105 -0.291

Attainability -0.187 -0.134

Advocacy -0.036 -0.229

Quality of Network -0.679*** 0.464**

Member Meetings 0.320* -0.658***

Information Provision 0.046 0.104

Attainability -0.295* 0.292

Table 11, Dummy variable regression results, model 2

(38)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₈

customer satisfaction for segment 1, but is significantly important in both segment 2 and 3. This dimension is a performance factor in segment 2 and an excitement factor in segment 3. Although advocacy is a basic factor in segment 1, it is an excitement factor for segments 2 and 3, while its overall importance is higher in segment 2. The provision of information can be seen as excitement factor in segments 2 and 3, but according to table 11 it is still not significantly related to overall satisfaction when performance is high in both cases. The importance of the attainability is quite low for all segments, but has the largest role as performance factor in segment 3.

(39)

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction ₃₉

Constant η1|s Odds ratio [EXP(η1|s)]

Aggregate -5.478*** 2.369*** 10.687

Segment 1 -6.270*** 3.055*** 21.221

Segment 2 -4.957*** 2.074*** 7.957

Segment 3 -4.147** 1.716** 5.562

Dependent Variable: Recommendation intention of the event *= p-value <0.1, **= p <0.05, ***= p <0.01

Table 12, Logistic regression results, model 2

The recommendation intention of members depends on the overall satisfaction. For the aggregate model as well as for the three segments, the results of the logistic regression are displayed in table 12 and graphically in figure 6 for the three segments only. The Nagelkerke R-square of each of the segments were respectively 0.543, 0.436 and 0.216, which indicates the predictive power of overall satisfaction on the recommendation intention of members in each of the segments. The members in segment 1 have the highest recommendation probability, while segment 2 and 3 members are almost equally likely to recommend the organization, even though overall satisfaction is lowest for segment 3 members. The aggregate result is the average of the results from the segmented logistic regression and is for that reason not graphically represented. The odds ratios give a clear insight in the velocity with which the recommendation intention probability inclines when the customer satisfaction improves.

Figure 6, Probability of recommendation for every overall satisfaction score, model 2

The Heterogeneity, Nonlinearity and Asymmetry of the Relationship between Attribute Performance and Customer Satisfaction