Master Thesis Marketing Intelligence Annemiek de Graaf

(1)

Master Thesis Marketing Intelligence

Annemiek de Graaf

Linking Performance to Satisfaction

Differentiating the core service attributes

from the side aspects.

(2)

Linking performance to satisfaction:

Differentiating the core service attributes from the side aspects.

Master Thesis Marketing Intelligence

16th of January, 2017

Annemiek de Graaf

S2114097 Roelantstraat 2 1055 LP Amsterdam annemiek_degraaf@live.nl +31655884123 1st supervisor:

dr. ir. Maarten J. Gijsenberg

m.j.gijsenberg@rug.nl 2nd supervisor dr. Jelle T. Bouma j.t.bouma@rug.nl External supervisor Raphael Saidof raphael-van.saidof@klm.com University of Groningen Faculty of Economics and Business

Department of Marketing

(3)

MANAGEMENT SUMMARY

The importance of performance in service environments has been widely recognized, but the link is difficult to evaluate given the intangible nature of services, difficulties in standardization and the simultaneous production, delivery and consumption of services. Furthermore, one has to take into account that several nonlinearities and asymmetries in the relation have been identified. In order to understand the relationship between objective service performance and customer satisfaction, in this research I will combine several theories to test the existence of a hierarchy in the impact of different service aspects on satisfaction.

Identifying service attributes that are important to customers has been a major focus of customer and market research for decades. In order to understand what drives customer satisfaction, the total service has to be analyzed at a smaller, more manageable level. Three different levels of service attributes have been identified, namely core, semi-peripheral and peripheral service attributes. Based on previous literature it is expected that some attributes will become more important once other attributes have reached a certain performance threshold. In order answer the question ‘What

is the link between objective service performance and customer satisfaction and is there a hierarchy in the impact of different service aspects on satisfaction?’ I will estimate a multiple

threshold regression model, to compare the impact of the service attributes in three different situations.

(4)

Lastly, the peripheral service attributes, which were considered to be ‘bonus’ attributes, should not be neglected in earlier stages of the hierarchy. The results show that even if the core attributes are not performing satisfactory, the peripheral attributes do have a positive impact on satisfaction.

(5)

PREFACE

This thesis is the last part of my Master Marketing Intelligence at the University of Groningen. I got the opportunity to finalize my study by doing a graduate internship at Air France-KLM. I would like to thank Jelle Bouma and Hans Zijlstra for making this possible. Being able to use data from the largest airline of the Netherlands provided me with many opportunities and an interesting and meaningful topic for my thesis.

I would like to thank my supervisor Maarten Gijsenberg for guiding me through the process, providing me with feedback and support to continuously improve my thesis. Furthermore I would like to thank my external supervisor Raphael Saidof for his support in collecting and preparing the data and his help with the statistical analyses. Lastly I would like to thank all the colleagues at Air France-KLM who contributed by sharing their knowledge and expertise.

(6)

CHAPTER 1: INTRODUCTION

The relationship between customer satisfaction and service performance has been extensively researched (Yueng et al., 2013). Since the early 1970s, the volume of customer satisfaction research has been impressive. Numerous studies have been devoted to identify the antecedents of satisfaction and develop meaningful measures of the construct (Churchill and Suprenant, 1982). In the early nineties, Anderson and Sullivan (1993) noticed the growing managerial interest in customer satisfaction as a means to evaluate performance. One year later, Taylor and Baker (1994) identified the critical roles of service performance and satisfaction in the formation of customer’s purchase intentions. Researchers and practitioners are very much interested in understanding what drives customer satisfaction because it has been identified as an antecedent of increased market share and profitability (Anderson, Pearo and Widener, 2008).

In the service industry, identifying the determinants of satisfaction is complicated by the fact that a service encompasses many attributes. An understanding of these attributes is required and managers often focus on enhancing particular attributes of the service (Bacon, 2012). According to Danaher (1997), determining the relative impact of service attributes is one of the most important objectives of customer satisfaction measurement. Effective service management requires an understanding of the relationship between performance of one aspect of a service, and the overall evaluation of the service (Bacon, 2012).

(9)

2 failure on the perceived service performance is larger than the impact of an improvement. Likewise, Falk, Hammerschmidt and Schepers (2010) found an asymmetry in the delighting capacities of utilitarian versus hedonic quality attributes on customers, as their relationship with a firm matures. This illustrates the complexity of the relation between performance and satisfaction.

Another complicating element is that a service is comprised of several attributes which can be classified into different bundles of attributes. In one study the researchers make a distinction between core and peripheral attributes of a service, in order to understand which components of the service concept are most important to customers (Anderson, Pearo and Widener, 2008). Another classification is the distinction between attributes that contribute to satisfaction and attributes that contribute to dissatisfaction (Kano et al., 1984). The importance that customers place on service performance attributes as drivers of satisfaction and loyalty, is a critical input to a firm’s resource allocation strategy and quality improvement efforts (Gustafsson and Johnson, 2004). In this research I will focus on examining the relationship between objective service performance (OSP) and customer satisfaction and, in doing so, investigate whether there is a hierarchy in importance. Several theories will be combined in order to explore a new type of categorization.

The main question that will be addressed in this research is formulated as follows:

‘What is the link between objective service performance and customer satisfaction and is there a hierarchy in the impact of different service aspects on satisfaction?’

(10)

3 satisfaction with services. While there are more antecedents of satisfaction, the data that is available provides an opportunity to test this link.

(11)

4

CHAPTER 2: THEORY

The services literature has been dominated by the study of service performance, service value and satisfaction issues, with particular attention given to identifying the relationships between these constructs (Cronin, Brady and Hult, 2000). This literature serves as a base to define the concepts, identify the relations and suggest a categorization of service attributes.

2.1 Service performance and customer satisfaction

Kano et al. (1984) considered two aspects of a service, namely the objective aspect involving whether or not the quality is met and the subjective aspect involving the customers’ satisfaction. This research is focused at examining the link between measures of objective service performance (OSP) and customer satisfaction.

2.1.1 Service performance

The importance of performance in service industries has accelerated dramatically. Much of this importance has been driven by the realization that high service performance results in higher customer loyalty, higher NPS scores and increased market share (Danaher, 1997). Since service performance is fundamental to the satisfaction profit-chain, firms have to strengthen their understanding of this construct (Falk, Hammerschmidt and Schepers, 2010). According to this chain, customer satisfaction should increase by improving product and service attributes. This should lead to greater customer retention and thus greater profitability (Anderson and Mittal, 2000). Given the need to maintain a high service performance, practitioners and academics have devoted increasing effort to measure service performance (Danaher, 1997). This led to several ways to conceptualize and measure service performance. One measure was introduced by Parasuraman et al. (1988). They propose the SERVQUAL scale which is a gap-based comparison of expectations and performance perceptions of customers. This is in line with the definition used by Dagger, Sweeney and Johnson (2007). They define service quality as a customers’ judgment about an entity’s overall superiority, which is often described in terms of discrepancy between customer’s expectations of, and actual, service performance.

(12)

5 (2015) use the term objective service performance (OSP), which is the extent to which a company succeeds in delivering the service it is promising. OSP is about the actual quality of the service, for example the punctuality of trains, which is an important determinant of customer satisfaction (Gijsenberg, van Heerde and Verhoef, 2015). A study by Burton, Sheather and Roberts (2003) suggests that OSP is a significant predictor of customer satisfaction, apart from the indirect influence via the perceptions of customers. Furthermore, measures of actual performance are often available to managers on a continuous basis, unlike perceptions of performance (Burton, Sheather and Roberts, 2003). Therefore, in this research I will focus on OSP as a measure of service performance, which is defined as the conformity of a service aspect to the standards set by the company.

2.1.2 Customer satisfaction

Customer satisfaction has been studied extensively by researchers and used by firms because the construct is generic, clearly understood by respondents, easy to communicate to managers and it can be universally applied (Gupta and Zeithaml, 2006). Falk, Hammerschmidt and Schepers (2010) define satisfaction as the discrepancy between expectations and perceived quality (i.e., performance). Likewise, Gupta and Zeithaml (2016) state that customer satisfaction is a customer’s judgment that a product or service meets or falls short of expectations. High customer satisfaction ratings are widely believed to be the best indicator of a company’s future profits (Anderson and Sullivan, 1993).

(13)

6

2.2 The relation between service performance and satisfaction

Service performance and customer satisfaction are widely recognized as key influences in the formation of customers’ purchase intentions in service environments. However the specific nature of the relationship between these constructs remains unclear (Taylor and Baker, 1994). To encourage performance improvements that will lead to an optimal level of satisfaction, it is necessary to understand what drives customer satisfaction (Anderson and Sullivan, 1993). A considerable amount of research has examined the cause-effect relationship between OSP and customer satisfaction (Larivière, 2008). Although the existence of the effect of performance on satisfaction has been well established, the way in which it influences satisfaction is not clear (Burton, Sheather and Roberts, 2003).

As was mentioned before, several asymmetries have been found in the relation between OSP and customer satisfaction. Anderson and Mittal (2000) argue that an important step towards successful implementation of the satisfaction-profit chain is to account for asymmetry and nonlinearity. Asymmetry occurs when the impact of an increase is different from the impact of a decrease (Falk, Hammerschmidt and Scheper, 2010). An example of such a link comes from Gijsenberg, van Heerde and Verhoef (2015) who demonstrate that the impact of a negative OSP shock is larger than the impact of positive OSP shock, and that it also takes longer to recover from negative versus positive performance changes. Nonlinearity on the other hand implies that the effect of an increase or decrease in OSP can have a diminishing or exponential impact on satisfaction over time (Anderson and Mittal, 2000). This shows that the relation between the constructs is not straightforward. Therefore, in this research I will focus on examining the link between OSP and customer satisfaction whilst accounting for the different types of relationships. In the next section I will propose a categorization of the service attributes.

2.3 Categorization of service attributes

(14)

7 1997). Identifying product and service attributes that are important to customers has been a major focus of customer and market research for decades (Gustafsson and Johnson, 2004). This indicates that, in order to understand what drives satisfaction, the total service has to be analyzed at attribute level. By separating the service into smaller, more manageable design components and understanding the service elements, a firm can increase the knowledge of their entire service (Hume, 2008). Mittal, Ross and Baldasare (1998) advocate the use of such a multi-attribute model. They argue that: 1) customers are more likely to evaluate a product at the attribute level, 2) managers usually work at the attribute level rather than the overall service level, and 3) it enables researchers to conceptualize phenomena better. To improve the overall performance of a service experience, managers often focus on enhancing particular attributes of the service. But identifying which attributes to improve is often challenging (Bacon, 2012). Therefore, the current research will consider separate service aspects and examine the impact of each group of aspects on customer satisfaction. This will provide practitioners with insights into which aspects should be improved, and in what order, to achieve the greatest gains in satisfaction.

In this research, I will focus on examining whether a hierarchy exists in the relationship between OSP and satisfaction, identifying aspects that are the core of the service and aspects that are peripheral. Anderson, Pearo and Widener (2008) define core attributes as ‘what’ is delivered and peripheral being ‘how’ it is delivered. They refer to Iacobucci and Ostrom (1993) who state that the core is the part of the service we think of when we name the service, for example the dinner served at a restaurant. Peripheral attributes are relational aspects such as the friendliness of the waiter (Iacobucci and Ostrom, 1993), or physical aspects such as the design of the restaurant (Anderson, Pearo and Widener, 2008). Peripheral attributes can be seen as those facilitating the core offering, whilst the core is the basic reason for the business to be in the market (Hume, 2008). In this research, core aspects are the essential service aspects that need to be in place whilst the peripheral aspects are seen as ‘bonus’ aspects.

(15)

8 survive, such as food and water, a house to live in and clothing. Higher in the hierarchy, aspects such as friendship, love, and education become important and at the top is self-actualization (Maslow, 1943). Linking this to service encounters, it can be expected that several aspects, the crucial needs, are supposed to be in place, otherwise people will never be satisfied. For example, if the waiters in the restaurant will not take your order, you do not get anything to eat or drink and thus you will never be satisfied, not matter how friendly they are. In this research, I will test a hierarchy that draws upon this theory for categorizing attributes according to their impact on satisfaction.

Another prominent categorization of attributes is introduced by the Japanese researcher Noriaki Kano (1984) who proposed that there is a difference in attributes that have an impact on customer satisfaction, and attributes that have an impact on dissatisfaction. In this theory, a distinction is made between five categories of service attributes that differ in their relation with satisfaction, which are as follows (Kano et al., 1984): 1) Attractive quality attribute, this attribute gives satisfaction if present but no dissatisfaction if absent. 2) One-dimensional quality attribute, attributes that fall within this category are positively and linearly related to satisfaction, and thus the greater the degree of fulfilment, the more satisfied customers are. 3) Must-be quality attribute, attributes that result in customer dissatisfaction if absent but do not contribute to satisfaction if present. 4) Indifferent quality attribute, attributes that do not cause satisfaction nor dissatisfaction. 5) Reverse quality attribute, attributes that cause dissatisfaction if present and satisfaction if absent. This theory shows that some attributes only impact satisfaction, some only impact dissatisfaction, some attributes have no impact at all and some can have impact on both.

A third categorization comes from the researchers Iacobucci and Ostrom (1993). They differentiate in their paper between core and peripheral service attributes. Core attributes are the core of a service, these are the aspects we think of when we name the service, and peripheral service attributes are all the other aspects (Anderson, Pearo and Widener, 2008). This theory strengthens the idea that some aspects are vital to the existence of the service and some can be considered a bonus.

(16)

9 model where insights from these models are combined to define three categories of service aspects, namely: core, semi-peripheral and peripheral service attributes. In the next sections I will explain the rationale behind the categories and introduce the hypotheses.

2.3.1 Core service attributes

Identifying performance attributes that drive satisfaction is critical to a firm’s resource allocation strategy and quality improvement efforts (Gustafsson and Johnson, 2004). Therefore, the goal is to distinguish attributes that companies should focus on from attributes that are not essential, in order to maximize customer satisfaction. Those attributes can be seen as the basic needs in the hierarchy of Maslow (1943) and it can be related to the must-be quality attributes from the Kano model. These attributes contribute to customer dissatisfaction when absent but do not contribute to customer satisfaction if present (Kano, 1984). Therefore, the first category of attributes that can be defined will be termed the core service attributes. Those attributes are expected to have an effect on satisfaction as long as they are not satisfactory, but the effect will wear off once a satisfactory level of quality has been reached. This leads to the first hypothesis:

H1: The core service attributes will have a larger effect on customer satisfaction as long as the level of performance is not satisfactory yet. The effect will wear off once the performance is

above this level.

2.3.2 Semi-peripheral service attributes

After the core aspects are in place, a second category of attributes becomes important: the

semi-peripheral service attributes. From Maslow (1943), we gain the insight that some aspects should

(17)

10 This leads to the second hypothesis:

H2: Once the core service attributes have reached a satisfactory level, the semi-peripheral service attributes will have a larger (positive or negative) impact on customer satisfaction.

2.3.3 Peripheral service attributes

Besides the core- and semi-peripheral attributes, a service encompasses many aspects that are not vital to the service but do influence the total experience. Those are termed the peripheral service

attributes. Referring to the theory of Maslow (1943), those attributes can be compared with higher

order needs that become important only when the lower order needs are fulfilled. For example, no matter how friendly the courier is that brings your package, if the package is three days overdue (and you do not need the product anymore) the friendliness will probably not relieve your dissatisfaction with the service. In Kano’s hierarchy, attractive-quality attributes are identified. Those are the aspects that will not contribute to dissatisfaction if absent, but will contribute to satisfaction when present (Kano, 1984). It is expected that if the core- and semi-peripheral of the service is not right, the peripheral service attributes will not be able to restore satisfaction levels. Therefore the third hypothesis is formulated as follows:

(18)

11 A visualization of the relations between the constructs is shown in the conceptual model below.

Figure 1. Conceptual model

Customer satisfaction Semi-peripheral

service attributes

Core service attributes

Peripheral service attributes

H3

(19)

12

CHAPTER 3: DESIGN

3.1 Data collection and description

In order to verify the existence of a hierarchy in service attributes, Air France-KLM has provided data on both the performance of the service attributes and customer satisfaction with the service. At the moment, Air France-KLM uses two types of measurements. The first measure is objective, the Quality Observer (QO) program, in which passengers serve as ‘mystery shoppers’ to judge the aspect of the flight. During their travel they answer questions such as ‘Did the plane leave on

time?’ ‘Did the stewardess smile when she greeted you?’ and ‘Did you have the choice between 2 meals?’ Participants can answer these questions usually with yes or no (100% and 0%), which

leads to a percentage of conformity with the standards, indicating the OSP on those aspects. In some cases, the questions have four answer options. Such as the question ‘Was the agent in the

lounge polite and considerate?’ In this case the answer options are: 1) Yes, both polite and

considerate, 2) Polite but not considerate, 3) Considerate but not polite and 4) No, not polite nor considerate. With these options, yes and no are still recoded into 100% and 0% but option 2 and 3, where only half of the service attribute is as it is supposed to be, are recoded into 50%.

With this program, Air France-KLM aims to measure the performance of their service in order to optimize it and offer the right service level all around the globe. For this program, frequent flyers of Air France and KLM are invited to become a member of a pool. Based on the need for QOs per route, members of this pool are invited to be a Quality Observer on their flight. The goal of Air France-KLM is to have at least one QO per route per week. The research company running the program keeps track of the flights of QOs and the routes that still need to be covered for that week. This leads to an average of 3,500 respondents per month. Within the period under investigation (October 2015 till September 2016), 41,780 respondents used the QO app to evaluate the conformity of the service aspects.

(20)

13 The second measurement is a subjective measure of satisfaction, called eSCORE. It is a survey for which selected passengers receive an email after a flight, in which they are invited to answer questions about their experience. Those questions are more general and subjective compared to QO, to assess the satisfaction of passengers with their travel. Passengers are asked to rate several aspects of the flight on a 5-point Likert scale (excellent, very good, good, fair and poor). The answers of respondents have been recoded to a 10-point interval scale by Air France-KLM, in order to make the satisfaction numbers easier to interpret for the business. The aim for this program is to obtain a representative sample of the total passengers in nationality, travel reason, business class/economy class etcetera. Based on this, selected passengers will get an invitation. The response rate of about 10% is taken into account whilst sending out invitations. eSCORE has on average 75,000 respondents per month. Within the twelve months under investigation, 919,925 respondents filled in the satisfaction survey. The eSCORE data contains more general information about the passenger and the flight. For the dataset under consideration the respondents have an average age of 48.5 and 87% of the respondents are members of the Flying Blue program (53% ivory, 17% gold and 17% platinum). In appendix I an overview of all the variables present in both data files is given.

3.2 Aggregating and merging the data

In order to be able to compare the two measurements, I had to aggregate and combine the data into one file. Several choices were made with regards to the level of aggregation. First, I decided to aggregate the data per month because Air France-KLM has approximately one Quality Observer per week per airport. In order to be able to see some variation in the data (i.e. not only performance levels of 0% and 100%), the average performance of one month is used to compare with the average satisfaction scores of that month.

(21)

14 The last variable that I used to aggregate the data is the priority status of the passenger. Passengers can have Sky Priority status, which gives them several benefits on the ground, such as priority at the check-in desks, access to the lounge, being the first passengers to be called on board and the luggage of Sky Priority is placed on luggage belts at the arrival airport first. I chose to use this variable as an aggregation variable because there are several questions dedicated to this status, both in QO and eSCORE.

After aggregating the data on these break variables, the files were merged into one file of 4,647 cases. These cases represent the average level of performance for the ground services per station per month, with a distinction between priority and non-priority. Because of the decision to aggregate the data on airport level, in-flight variables were not taken into consideration in this analysis.

3.3 Data reduction

After aggregating the data, I took a few steps to come to a manageable dataset suited to test the hypotheses. In order to select variables to use in further analyses, I performed several regression analyses to select the variables per touch point that significantly contribute to the satisfaction of that touch point. The output of the regression analyses is shown in appendix II.

3.3.1 Check-in

(22)

15 In order to reduce the amount of variables used for further analysis, I conducted a multiple regression analysis. In this analysis, the variables indicating the performance of the check-in aspects were used as independent variables, and the satisfaction score with the overall check-in procedure was used as dependent variable. This analysis shows that the variable concerning whether the agent is well-dressed has a significant impact on satisfaction (β=.006, p=.02), therefore this variable will represent check-in. Furthermore, by excluding cases pairwise instead of listwise the variable clean and organized shows to have a significant influence on satisfaction of check-in (β=.002, p=.03). I applied the same method to the luggage drop off variables, which led to the variable agent polite as a representative (β=.007, p=.01).

3.3.2 Lounge

Likewise, I performed a stepwise regression analysis on the lounge variables in order to identify representatives for this aspect of the journey. Two variables show to have a significant influence on the overall satisfaction with the lounge, namely whether the agent was polite and considerate (β=.044, p<.01) and whether the agent was visible and approachable (β=.018, p<.05).

3.3.3 Boarding

The regression concerning the performance of the different aspects of the boarding process and the satisfaction with this process, provided three variables that will be used in further analysis. These are the variables indicating whether Sky Priority passengers were the first to go on board (β=.008, p=.00), whether the boarding process started on time (β=.010, p<.01) and whether the boarding staff greeted the passengers with a smile and eye contact (β=.004, p<.01).

3.3.4 Arrival

(23)

16 As an overview, the variables that will be used in further analysis are presented in the table below:

Table 1. Variables used in further analysis

3.4 Classification of service attributes

After selecting the variables that will be used in further analysis, the next step is to group the variables into the three categories: core service aspects, semi-peripheral service aspects and peripheral service aspects. In order to categorize the attributes, I consulted several experts within Air France-KLM. The knowledge that is available within the organization is very valuable for grouping the variables into the three categories. In order to bundle this knowledge, I sent a questionnaire to employees of the Customer Insights department. The questionnaire asked respondents to place the variables mentioned above into the three categories. The survey that has been used, and the results, are shown in Appendix III. According to the experts, the division of variables into the three categories is as follows:

Table 2. Division of variables into categories

Core service attributes Semi-peripheral service attributes Peripheral service attributes

Checkin_clean_org Dropoff_agent_polite Checkin_agent_welldressed

Arrival_clean_organized Lounge_agent_polite_considerate Boarding_greeted_smile_eye

Boarding_skyp_first Lounge_agent_visible_approachable

Boarding_timeliness Arrival_Skyp_luggage_first

Category Representative variables

(24)

17

3.5 The model

In the eSCORE satisfaction questionnaire, people are asked to rate their satisfaction with all aspects of their travel. Since I aggregate on departure station, the in-flight aspects cannot be taken into account. To eliminate the effect of omitted variables as much as possible, I chose to discard the overall satisfaction score as dependent variable. This variable represents the satisfaction of customers with the entire service, but this research only considers the ground service aspects. Based on previous research within Air France-KLM it is known that the in-flight service aspects have a large impact on satisfaction (Customer Experience Drivers, 2016). In order to exclude the effect of omitting the in-flight variables, the dependent variable is composed of customer satisfaction with regards to overall check-in, overall boarding, overall lounge and overall post-flight. The dependent variable used for analysis, overall satisfaction ground, is an average of these four variables.

In order to be able to test whether a hierarchy exists, the level at which the performance of the service aspects can be considered satisfactory has to be assessed. The first step in determining whether the performance of the elements is satisfactory or not, is to calculate the average performance level per group. Therefore, I created two new variables, the first consisting of the mean of the core variables (Mean_Core) and the second representing the mean of the semi-peripheral variables (Mean_Semi). In order to determine the level of performance that can be considered satisfactory, I consulted the expert within Air France-KLM with regards to the Quality Observer program. In line with what is used within Air France-KLM as a target for the performance of the service, a level of 90% conformity is used as a satisfactory level of performance.

The mean variables and the threshold performance level are used to calculate three indicator variables, one per level in the hierarchy:

I1 Indicator variable 1 represents the situation in which the core variables did not reach a

satisfactory performance level.

I2 Indicator variable 2 represents the situation in which the core variables did reach a satisfactory

(25)

18 I3 Indicator variable 3 represents the situation in which the core- and the semi-peripheral

variables have reached a satisfactory performance level and the peripheral variables will become of importance. 𝐼1[1,0] = {1 = 𝑀𝑒𝑎𝑛_{0 = 𝑀𝑒𝑎𝑛}𝐶𝑜𝑟𝑒 < 90 𝐶𝑜𝑟𝑒 ≥ 90 } 𝐼2[1,0] = {1 = 𝑀𝑒𝑎𝑛𝐶𝑜𝑟𝑒 ≥ 90 𝐴𝑁𝐷 𝑀𝑒𝑎𝑛_{0 = 𝐸𝑙𝑠𝑒} 𝑆𝑒𝑚𝑖 < 90} 𝐼₃[1,0] = {1 = 𝑀𝑒𝑎𝑛𝐶𝑜𝑟𝑒 ≥ 90 𝐴𝑁𝐷 𝑀𝑒𝑎𝑛𝑆𝑒𝑚𝑖 ≥ 90 0 = 𝐸𝑙𝑠𝑒 }

In order to test the hypotheses, I perform a multiple regression analysis. The indicator variables that were introduced in the previous section are used to make a distinction between three different situations. The model that will be tested can be formulated as follows:

𝑌_𝑖 = 𝛼_0𝑖+ 𝛼_1𝑖𝐼_2𝑖 + 𝛼_2𝑖𝐼_3𝑖+ (𝛽_𝐼1𝑖… 𝛽_𝐼10𝑖) 𝐼1𝑖 [ 𝑋_1𝑖 . . . 𝑋_10𝑖] + (𝛽_{𝐼𝐼1𝑖}… 𝛽_{𝐼𝐼10𝑖}) 𝐼2𝑖 [ 𝑋_1𝑖 . . . 𝑋_10𝑖] + (𝛽𝐼𝐼𝐼1𝑖… 𝛽𝐼𝐼𝐼10𝑖) 𝐼3𝑖 [ 𝑋_1𝑖 . . . 𝑋10𝑖] + 𝜀

This model is used to estimate a threshold regression model. The 10 variables, indicated by X1 till

X10, are multiplied by each indicator variable. This results in 30 independent variables representing

each variable in the three situations. I will compare the betas and significance levels of the variables to infer whether the hierarchy exists. The table below gives an overview of the Xn terms

(26)

19

Table 3. Overview of X-terms and corresponding variables

Xn and variable names

X1 CORE_Checkin_Clean_org X6 SEMI_Lounge_polite_considerate

X2 CORE_Boarding_skypfirst X7 SEMI_Lounge_agents_visible

X3 CORE_Boarding_timeliness X8 SEMI_Arrival_skyprio_luggage

X4 CORE_Arrival_cleanorg X9 PERI_Checkin_well_dressed

(27)

20

CHAPTER 4: RESULTS

In this chapter I will present the results of the analysis. First, I perform several preliminary checks after which the frequencies and descriptives of the data are given. Subsection 4.3 is devoted to the regression analysis and next I conduct several robustness checks of which the results are presented in section 4.4. The chapter will conclude with the support for the hypothesis, which I will discuss in chapter 5.

4.1 Preliminary checks

The original data contains several missing values due to the structure of the questionnaires. The eSCORE questionnaire has several so called skeleton questions, which are being asked to every respondent, but the remainder of the questions is divided in three sets of questions. Each respondent only answers one set of questions, thus missing values exist for the questions that were not answered. By aggregating the data, this problem is largely solved. Furthermore, the decision to take the average of the four ground satisfaction scores as dependent variable reduces the amount of missing values. The remainder is filtered out during the analysis, by excluding cases listwise. The Quality Observer data also copes with the problem of missing values, because some questions are specific for situations that do not apply to all passengers. For example the questions about the lounge are only asked to passengers eligible to visit the lounge. Again by aggregating the data this problem is largely solved. The remainder of the missing values were filtered out during the analysis. I checked the data on outliers but none were present in the dataset.

4.1.1 Initial analyses

The first part of the research question stated in chapter 1 is to examine the link between service performance and customer satisfaction. In order to test whether this link exists in itself, regardless of the possible hierarchy, I performed a multiple regression analysis. The analysis is performed with the 10 variables indicating the performance of the service aspects as independent variables, and the satisfaction with the ground services as dependent variable. The results of the regression are presented in appendix IV.

(28)

21 of the arrival area (β=.008, p<.01), the politeness and considerateness of agents in the lounge (β=.005, p<.01) and the friendliness of greeting at the boarding area (β=.004, p<.01) prove to be significant predictors of overall ground satisfaction. An increase in the performance on these attributes has a positive impact on the satisfaction with the ground service. One variable shows a negative relation with satisfaction. If the variable indicating whether the luggage of sky priority passengers was the first on the luggage belt at the arrival airport improves, the satisfaction decreases (β=-.002, p<.01). I will discuss this unexpected result in further detail in chapter 5.

As was mentioned before, the in-flight variables are not included in this research because of the decision to aggregate on airport level. The decision to use the mean of the ground satisfaction variables as dependent variable is expected to account for the effect of omitted variables. In order to test whether the effect of in-flight satisfaction is indeed as large as is expected, I performed a regression analysis with overall satisfaction as dependent variable and the satisfaction of check-in, boarding, lounge, in-flight and arrival as independent variables. The results, presented in appendix IV, show that in-flight satisfaction has the largest effect on overall satisfaction. Since the in-flight satisfaction contributes a lot more to overall satisfaction compared to the ground satisfaction, the possibility of a halo effect has to be taken into account. It might be that the in-flight experience has such an impact on passengers, that the satisfaction with this item also influences the ratings of the other items. In order to test this, I performed a robustness check with overall satisfaction instead of overall ground satisfaction as dependent variable. The results of this analysis are presented and discussed in section 4.4.

4.1.2 Testing assumptions

(29)

22 error terms in situation 1, the variables multiplied by indicator 2 are divided by the standard deviation in situation 2 and indicator 3 variables by the standard deviation in situation 3. E.g., the variable Core_boarding_skypfirst*Indicator 1 is now divided by .71219, Core_boarding_Skypfirst *Indicator 2 is divided by .56861 and Core_boarding_skypfirst*Indicator 3 is divided by .65575. Since heteroscedasticity is a problem with the variance in the data, the parameter estimates are not affected by the remedy.

Secondly, I tested whether the normality assumption holds. In order to detect whether the error terms are normally distributed, a histogram is presented in Appendix V. This histogram gives no direct reason to believe that the assumption has been violated, but to be certain I performed the Kolmogorov-Smirnov test, of which the results can also be viewed in Appendix V. The Kolmogorov-Smirnov test is significant (p=.000, statistic=.057), which means that the model is significantly different from a normal distribution. This means that the p-values cannot be trusted, because these are based on normal distribution. This might lead to wrong acceptance or rejection of hypotheses. In order to overcome the problem of nonnormality, bootstrapping can be performed. Bootstrapping is a procedure in which results of multiple models is averaged. The models are estimated on a bootstrap sample drawn from the original sample. Each sample leads to predictions and these predictions are averaged, which will result in more stable predictors (Leeflang et al., 2015). To be certain that I can trust the significance of the variables, I will perform a bootstrap regression.

4.2 Characteristics

(30)

23

Table 4. Mean and standard deviations

Variables Mean St. dev.

CORE_Checkin_Clean_org 90.7 19.7 CORE_Boarding_skypfirst 79.5 30.9 CORE_Boarding_timeliness 66.7 34.4 CORE_Arrival_cleanorg 89.3 21.8 SEMI_Dropoff_polite_considerate 94.3 16.2 SEMI_Lounge_agents_visible 83.5 25.0 SEMI_Lounge_polite_considerate 83.6 23.9 SEMI_Arrival_skyprio_luggage 51.1 41.0 PERI_Checkin_well_dressed 97.5 13.3 PERI_Boarding_greeting 77.9 24.3

Graph 1 presents the average performance of the core, semi-peripheral and peripheral items in relation with the satisfaction. This graph shows that for some periods the satisfaction and performance move parallel (e.g., the increase between Dec’15 and Feb’16 and from August to September) but for other periods the lines move contradictory (e.g., the performance of

semi-peripheral items and

satisfaction between

March and April). The line that seems to move

most similar to

satisfaction is the

performance of the core

items, which would

support the notion that those items are most impactful in influencing satisfaction. 5,5 5,6 5,7 5,8 5,9 6,0 6,1 6,2 6,3 6,4 6,5 65,0 70,0 75,0 80,0 85,0 90,0

Oct'15 Nov'15 Dec'15 Jan'16 Feb'16 Mar'16 Apr'16 May'16 Jun'16 Jul'16 Aug'16 Sep'16

Relation between performance and satisfaction

Core average Semi-peripheral average

Peripheral average Satisfaction

(31)

24 In order to give an idea of the occurrence of the three situations in the data, table 5 presents the division of the cases between the three situations. This shows that the majority of the cases falls within situation 1: the core variables did not reach a satisfactory level, which means that there is much to gain with regards to performance of the core items. The second largest category is the third situation in which the core and semi variables have reached the threshold of 90% performance.

Table 5. Division of cases among situations

4.3 Regression analysis

In order to estimate the model, I performed a multiple regression with bootstrapping. The bootstrap was performed with 4,964 samples and a confidence interval of 95%. Since the p-values cannot be trusted because the data is not normally distributed, the significance of the variables will be derived from the confidence interval. If the interval contains zero, the variable does not have a significant effect on ground satisfaction. But if zero is not included in the interval, the variable can be considered to be significant. I performed a second bootstrap regression analysis with the confidence level set at 90%. This regression shows the variables that have a marginally significant effect on overall ground satisfaction, comparable with the .10 significance level. The model fit is assessed by examining the adjusted R2, which is a measure of the amount of fluctuation in the dependent variable that is explained by the model, corrected for the amount of variables (Leeflang et al., 2015). The adjusted R2 for this regression model is .156 which is quite small, but the model is significant with an F-value of 4.551 (p=.000). It is no surprise that the adjusted R2_{is not too}

high since only a selection of the ground variables are taken into account.

The results of the bootstrap regression analysis are presented in Appendix VI and in table 6 an overview of the results is presented.

Situation Count Percentage

Situation 1 2,758 74%

Situation 2 466 12%

(32)

25

Table 6. Overview of the results of the multiple regression.

Variable B 95% Confidence interval

Lower Upper I1_CORE_Checkin_Clean_org I2_CORE_Checkin_Clean_org I3_CORE_Checkin_Clean_org .003 -.013 .019 -.001 -.037 -.040 .007 .007 .072 I1_CORE_Boarding_skypfirst* I2_CORE_Boarding_skypfirst** I3_CORE_Boarding_skypfirst .004 .011 -.008 .000 -.001 -.036 .007 .025 .017 I1_CORE_Boarding_timeliness I2_CORE_Boarding_timeliness I3_CORE_Boarding_timeliness .001 .001 -.011 -.002 -.007 -.031 .004 .011 .006 I1_CORE_Arrival_cleanorg* I2_CORE_Arrival_cleanorg I3_CORE_Arrival_cleanorg* .008 .020 -.041 .005 -.012 -.070 .011 .041 -.011 I1_SEMI_Dropoff_polite_considerate I2_SEMI_Dropoff_polite_considerate I3_SEMI_Dropoff_polite_considerate .001 .000 -.026 -.004 -.011 -.066 .007 .006 .029 I1_SEMI_Lounge_polite_considerate** I2_SEMI_Lounge_polite_considerate* I3_SEMI_Lounge_polite_considerate .004 .012 -.008 -.001 .004 -.040 .008 .018 .024 I1_SEMI_Lounge_agents_visible I2_SEMI_Lounge_agents_visible* I3_SEMI_Lounge_agents_visible .003 -.008 .004 -.001 -.014 -.025 .007 -.002 .042 I1_SEMI_Arrival_skyprio_luggage* I2_SEMI_Arrival_skyprio_luggage I3_SEMI_Arrival_skyprio_luggage -.003 -.002 -.002 -.005 -.005 -.016 -.001 .001 .012 I1_PERI_Checkin_well_dressed I2_PERI_Checkin_well_dressed I3_PERI_Checkin_well_dressed .001 .012 -.010 -.005 -.005 -.024 .006 .060 .005 I1_PERI_Boarding_greeting* I2_PERI_Boarding_greeting I3_PERI_Boarding_greeting .005 .002 .002 .000 -.006 -.012 .009 .011 .020 Indicator_2 Indicator_3** -.974 10.390 -7.323 -.599 4.563 20.787 *= Significant predictors of satisfaction at .05

(33)

26

4.4 Robustness checks

In order to check the robustness of the results, I performed the regression analysis again with performance levels of 85% and 95% as threshold level. In appendix VII the results of the robustness checks are presented. The results are not really different compared to the performance level of 90%. The most important difference is that the variable SEMI_Dropoff_polite_ considerate becomes significant with the 85% threshold in situation 3 (β=.015, 95% CI[0.29; -.006]). This still does not provide support for hypothesis 2 as the effect of semi variables was expected to become larger in situation 2, and this is not the case. What is striking is that the effect in situation 3 is negative, which would imply that the more polite and considerate the agents in the drop-off area are, the less satisfied the passengers are.

Furthermore, the effect of SEMI_Arrival_Skyprio_luggage becomes significant in situation 2 with the 85% threshold (β=-.004, 95% CI[-.007; -.001]). The effect is slightly larger than in situation 1 (.001 difference). A final notable difference is that the effect of CORE_Arrival_cleanorg becomes larger in situation 2 with an 80% performance threshold (β=.017, 95% CI[.007; .030]). The beta value increases from .006 to .017 from situation 1 to 2. This effect is not visible with a performance threshold of 85%. This provides an indication that the level of 80% is still too low to be considered satisfactory. In situation 2, where the core aspects perform above 80%, improvements of the performance above the level of 80% still have a significant positive influence. This provides support for the threshold level higher than 80%. Since there are no large differences between 85% and 90%, I will stick to the target level of 90% used within Air France-KLM to provide management with more meaningful results.

(34)

27 the regression are presented in appendix VII. The results show that only two variables are significant. The SEMI variable indicating whether the lounge agents are visible is significant in situation 2 and the PERI variable concerning the clothing of check-in agents, also in situation 2. The fact that most of the ground variables do not significantly influence overall satisfaction provides support for the notion that in-flight variables are important predictors of overall satisfaction. The fact that some of those variables do significantly contribute to overall ground satisfaction shows that deciding to use this variable as dependent variable is an effective choice in overcoming the effect of omitted variables.

4.5 Hypothesis table

Following the results presented in section 4.3 the support for the hypotheses is stated in table 7. The results will be discussed in the next chapter.

Table 7. Support for the hypotheses

Hypothesis Supported

H1 (Effect of the core wears off once satisfactory) Partly

H2 (Effect of semi becomes larger once core is satisfied) Partly

H3 (Effect of the peripheral becomes larger once core and semi are

satisfied)

(35)

28

CHAPTER 5: DISCUSSION AND CONCLUSION

In this chapter I will discuss the results presented previously, answer the main- and subquestions followed by a conclusion.

5.1 Discussion

The goal of this research is to provide insight into the relation between objective service performance and customer satisfaction, with particular interest in the hierarchy in which performance of service attributes influences satisfaction. The main research question is therefore formulated as follows:

‘What is the link between objective service performance and customer satisfaction and is there a hierarchy in the impact of different service aspects on satisfaction?’

The first part of the question can be addressed using the results of the multiple regression analysis performed in section 4.1. The aim of the analysis is to show that the performance of some service aspects has an impact on satisfaction. The results show that 4 out of 10 variables show the expected positive correlation. The higher the performance on the service attributes, the higher the satisfaction. The fact that not all service attributes contribute significantly to the overall ground satisfaction is also as expected. The service of Air France-KLM encompasses so many aspects and thus the satisfaction with this service has many influencers. Besides this, there are other aspects influencing the satisfaction of passengers. For example their reason for traveling, whether they travel for business, holidays, family reunions or maybe a funeral has a large impact on their state of mind which is likely to impact satisfaction.

(36)

29 luggage first better spend somewhere else. Since the main goal of this thesis is not to identify the individual relations of the service attributes and satisfaction, I did not perform any further analyses to examine this unexpected relation. The issue will be discussed in limitations of this research and suggestions for further research.

In order to answer the second part of the question, I formulated three hypotheses. In the next section, the support for these hypotheses is the topic of discussion.

5.1.1 Hypothesis 1

The first hypothesis states that the effect of the core attributes would have a larger effect on customer satisfaction if the performance is not satisfactory, and the effect will wear off once a satisfactory level of performance has been reached. The results of the regression show that in situation 1, when the core variables did not reach a satisfactory level of 90%, two core variables significantly explain part of the variance in customer satisfaction with the ground service. The cleanliness of the arrival area (β=.008, 95% CI[.005; .011]) and whether the sky priority passengers were allowed to board first (β=.004, 95% CI[.000; .007]) prove to be significant predictors of satisfaction. In situation 2 however, once the core has reached a satisfactory level of 90%, further increase of the performance of those items does not significantly contribute to customer satisfaction any longer. The bootstrap regression with the confidence interval set at 90% shows that there is a marginally significant effect of sky priority boarding in situation 2 (β=.011, 90% CI[.001; .022]), which weakens the support for the first hypothesis slightly.

(37)

30 The other two core variables, clean and organized check-in area and timeliness of boarding, are not significant in any of the three situations. The fact that two variables are significant in situation 1, and not anymore or only marginally in situation 2 and 3, provides support for the notion that the effect of improving the performance of core attributes increases satisfaction up until the level of 90%. After the threshold has been reached, extra improvements do not significantly increase satisfaction anymore and might even lower satisfaction, according to the negative correlation in situation 3 for the state of the arrival area. Hypothesis 1 is therefore considered to be partially supported by this research.

The second hypothesis that I address is whether some service aspects should be fulfilled before other aspects will have an influence on satisfaction. Hypothesis 2 therefore states that once the core service attributes have reached a satisfactory level, the semi-peripheral service attributes will have a larger impact on customer satisfaction. This hypothesis is partly supported by the results of the analysis. Two of the four semi-peripheral variables show significant relations with the dependent variable in situation 2, meaning that the influence of the semi-variable becomes significant once the core variables have reached the 90% performance threshold. The variable concerning how polite and considerate the lounge staff is has a significant effect in situation 2 (β=.012, 95% CI[.004; .018]). The effect of this service aspect is marginally significant in situation 1 (β=.004, 90% CI[.000; .007]), providing support for the hypothesis that the effect on satisfaction becomes larger, once the core variables have reached the threshold level of 90%. As soon as the core items perform well enough, the focus should be on also getting the semi-peripheral items at this satisfactory performance level. As expected, the effect wears off in situation 3 when the core as well as the semi-peripheral items have reached a satisfactory level. So increasing the performance of semi-peripheral items will have an effect up until the performance is satisfactory and any further increases do not significantly increase satisfaction.

(38)

31 CI[-.005; -.001]). The effect is no longer significant in situation 2 and 3. Also the variable indicating the visibility of lounge agents is negatively correlated with satisfaction in situation 2 (β=-.008, 95% CI[-.014; .002]). This does prove that the effect of semi-peripheral variables becomes larger once the core has been fulfilled, and wears of once the semi-peripheral attributes also perform above the 90% threshold. But the negative correlation with satisfaction is unexpected. It might be that people in the lounge are mainly there to relax and rest, and do not want to be disturbed by too many agents. If the agent is overly present, it might be disturbing. In situation 2 the semi-peripheral attributes do not perform at the 90% threshold, which means that the variable indicating whether the agents at the lounge are polite and considerate is also underperforming. Taking this into account, it might be that the agents in the lounge were not polite and considerate and thus people will become less satisfied if they are too present. In situation 3, when the semi-peripheral variables also perform above the threshold, this effect is not visible anymore. Because of the mixed results, the hypothesis that the effect of semi-peripheral variables becomes larger in situation 2 is only partly supported.

The third hypothesis under consideration for this research is whether the effect of the peripheral service attributes becomes larger once lower-order attributes (core and semi-peripheral) perform satisfactory. As is shown in table 6, the effect of the peripheral service attributes is not significant in situation 3. The only significant effect of the peripheral service attributes is the greeting in the boarding area in situation 1 (β=.005, 95% CI[.000; .009]), but this effect is no longer significant within the 95% confidence interval in situation 2 and 3. The variable PERI_Checkin_well_dressed is not significant in any of the situations. Therefore, no support was found for hypothesis 3. The lack of support for this hypothesis might indicate that the service attributes considered a ‘bonus’ are important to customers, even if the core and semi-peripheral attributes are underperforming. This is an important finding because this indicates that even if the core attributes are not in place, a friendly greeting whilst boarding the plane can impact satisfaction positively.

(39)

32 18.824]). This indicates that overall ground satisfaction is a lot higher in situation 3, compared to situation 1 and 2. It might be that being in a situation where the core and semi-peripheral aspects perform above the 90% threshold increases satisfaction that much, that the individual effect of the peripheral items does not significantly contribute anymore. This is an important finding as it indicates the positive effect on satisfaction of performing above 90% for the core and semi-peripheral items.

5.2 Conclusion

Following this discussion, I conclude that there definitely is a link between the performance of service attributes and satisfaction. Several items show to significantly influence the satisfaction in (mostly) a positive way. Thus the higher the performance of the ground service attributes the higher the satisfaction of customers with the ground service, with a few exceptions.

The second part of the main question is more complex to answer as the hypotheses received mixed support. The focus of this research is to test whether a hierarchy exists in service attributes in order to provide managers with practical advice on how to manage the performance of different service attributes. The partial support for hypothesis 1 means that we can assume that there are core aspects to a service. Those attributes have to be at a satisfactory performance level, otherwise satisfaction will decrease. But once the threshold level has been reached, extra efforts in improving the performance will have less impact on satisfaction and might even backfire. This is in line with what Kano (1984) defines as must-be quality attributes. Those attributes have to be in place (i.e. at a satisfactory level) in order for customers to be satisfied with the service, but any extra effort will not necessarily result in extra satisfaction.

(40)

33 can be identified, the relation with satisfaction is not straightforward and should be examined with great care.

I did not find support for a third level of service attributes, the peripheral attributes, that become important after the core and semi-peripheral attributes perform satisfactory. No significant effects of these attributes were found in situation 3. There was however a marginally significant, but very large, effect of the variable indicating situation 3. This means that the satisfaction of passengers is significantly higher in the situation where the core and semi-peripheral items perform satisfactory. Thus, even though partial or no support was found for hypothesis 2 and 3, performance of the core and semi-peripheral attributes should be at 90% to increase customer satisfaction.

(41)

34

CHAPTER 6: RECOMMENDATIONS

In the last chapter of this thesis I will provide managers with practical insights and implications based on the research. Furthermore, I will discuss the limitations of the study and give directions for future research.

6.1 Managerial implications

The goal of the research is to provide mangers with practical implications with regards to the decision on which attributes to focus, in order to improve satisfaction as much as possible. As was mentioned in chapter 2, performance management of services is more complex compared to products because of the many attributes they are composed of, the simultaneous production and consumption and most important, the unclear relation between performance and satisfaction.

First, this research provides proof for the existence of a relationship between objective service performance and customer satisfaction. Therefore, companies should monitor both constructs closely and identify the role of each service attribute. This research made use of expert knowledge to categorize the service attributes, which can also be applied within other companies. Dividing the attributes in categories can aid in monitoring and researching the impact on satisfaction.

The second implication that can be drawn from this research is that managers should focus on identifying the core attributes of their service, those that would impact satisfaction negatively if they perform below a threshold level. Once these attributes are identified, effort should be taken to get the performance of these attributes at this threshold level. Once the threshold level is reached, any extra investments to increase the performance even further can be considered a waste of time and money, as the improvements will not, or only marginally, significantly contribute to satisfaction.

(42)

35 above the threshold level will result in higher satisfaction, which can be implied by the positive impact of situation 3 in the regression model.

The last implication that follows the results of this research is that companies should put effort into the attributes that can be considered ‘bonus’, especially when the core attributes perform below threshold. The results show that the peripheral attributes are able to increase customer satisfaction only when the core elements were underperforming. This implicates that the peripheral items are able to relieve some of the negative impact of the underperforming core elements on satisfaction. For example, if the order of sky priority boarding is not maintained properly, a friendly greeting by the staff can counter the negative impact on satisfaction somewhat.

6.2 Limitations

I conducted this research with great care and several limitations were accounted for along the way. Still there are some limitations that have to be taken into account whilst reading and interpreting the results of the study.

One limitation of this study is a limitation of the Quality Observer data. The QO program is very useful in tracking the performance of the service but it has some limitations when comparing the performance figures with satisfaction. The QO program has a limited number of observations. The performance of one station in one month is usually based on the answers of 4 to 10 respondents, whilst the eSCORE satisfaction scores are based on a much larger number of questionnaires. Within Air France-KLM it is expected that the QO program is useful in tracking the overall performance and this performance does not fluctuate dramatically, but for the comparison with the satisfaction scores it would have been desirable to have more observations. This would also have made it possible to aggregate on a lower level than month and station.

(43)

36 What is known about the QOs, is that they are all Flying Blue members in the higher levels (Gold or Platinum). This is important to note because this means that people filling in the QO survey are frequent flyers, loyal Air France-KLM customers, and this leads to a relatively high numbers of Sky Priority passengers. The analyses will not be affected by this overrepresentation of Sky Priority passengers because the QO data is matched with satisfaction scores of passengers who also had this Sky Priority status. But for managerial implications, it is important to take this into consideration. It can be expected that aspects such as Sky Priority signage, the order of boarding and the luggage of Sky Priority passengers being the first on the luggage belt will have greater influence on the satisfaction of Sky Priority passengers compared to non-Sky Priority passengers.

Another limitation of the study is that it was not possible to include the in-flight service attributes. As was mentioned before, these attributes prove to have a large impact on the overall satisfaction of passengers. This has been taken into account by taking the satisfaction with ground services as dependent variable instead of the overall satisfaction, but it does mean that the analysis does not provide a complete picture of the service of the company. The main goal of this research iss to test whether the proposed hierarchy exists, and for this goal the data proved to be useful. But for managerial implications, gaining insights into which attributes to improve first and to what level, the in-flight variables are important.

Finally, this research has been performed in a specific industry with one specific company, and therefore the generalizability of the research can be questioned. Flights are quite complex services, with many service attributes that might impact satisfaction. Besides that, flying can be considered an experience for some people and the emotions accompanying the experience affect the satisfaction of passengers as well. For simpler services, for example a taxi ride or getting a hair cut, the analysis might not be as complex and the need for categorizing service attributes might be less.

6.3 Suggestions for further research

(44)

37 the model in a different environment. As was mentioned before, the airline industry is very complex, encompasses many service attributes and satisfaction with the service has many contributors. It is interesting to see whether the hierarchical model, estimated in a different setting, would show similar results.

Furthermore, I would suggest to (temporarily) collect Quality Observer data on a daily basis, covering every route every day, to be able to compare the performance to satisfaction on a daily basis. It is expected that this would increase the correlation between the constructs and it would lead to more reliable results. Because of the higher number of QO cases, this would also provide the opportunity to aggregate on a lower level, for example flight number, and include in-flight variables in the regression to provide the company with a complete picture of their service. Next to collecting data more frequently, I would also suggest to include some personal questions with regards to nationality, age and travel reason. This would provide the opportunity to test whether differences among people influence the satisfaction with performance levels, which might lead to interesting insights.

(45)

38

REFERENCES

Anderson, Eugene W. and Vikas Mittal (2000) “Strengthening the Satisfaction-Profit Chain.”

Journal of Service Research, 3(2), 107-120.

−−− and Mary W. Sullivan (1993) “The Antecedents and Consequences of Customer Satisfaction for Firms.” Marketing Science, 12(2), 125-143.

Anderson, Shannon, Lisa Klein Pearo and Sally K. Widener (2008) “Drivers of Service

Satisfaction. Linking Customer Satisfaction to the Service Concept and Customer Characteristics.” Journal of Service Research, 10(4), 365-381.

Bacon, Donald R. (2012) “Understanding Priorities for Service Attribute Improvement.” Journal

of Service Research, 15(2), 199-214.

Brady, Michael K., J. Joseph Cronin and Richard R. Brand (2002) “Performance-only

measurement of service quality: a replication and extension.” Journal of Business

Research, 55, 17-31.

Burton, Suzan, Simon Sheather and John Roberts (2003) “Reality or Perception? The Effect of Actual and Perceived Performance on Satisfaction and Behavioral Intention.” Journal of

Service Research, 4(4), 292-302.

Churchill, Gilbert A. JR. and Carol Suprenant (1982) “An Investigation into the Determinants of Customer Satisfaction.” Journal of Marketing Research, 19, 491-504.

Cronin, Joseph J., Michael K. Brady, and Thomas M. Hult (2000) “Assessing the Effects of Quality, Value, and Customer Satisfaction on Customer Behavioral Intentions in Service Environments.” Journal of Retailing, 72(6), 193-218.

Customer Experience Drivers (2016) Air France-KLM document

Dagger, Tracey S., Jillian C. Sweeney and Lester W. Johnson (2007) “A Hierarchical Model of Health Service Quality. Scale Development and Investigation of an Integrated Model.”

Journal of Service Research, 10(2), 123-142.

Master Thesis Marketing Intelligence Annemiek de Graaf