How Service Performance Affects Customer Satisfaction and Product Usage

(1)

How Service Performance Affects

Customer Satisfaction and Product Usage

In the public transport industry

(2)

How Service Performance Affects

Customer Satisfaction and Product Usage

In the public transport industry

University of Groningen Faculty of Economics and Business

MSc Marketing

Marketing Management & Marketing Intelligence MASTER THESIS June 20, 2016 F.V.N. Wolterink Grevingaheerd 122 9737 SR Groningen Telephone: +31 (0)647181988 fwolterink@gmail.com Studentnumber: 1781200

Supervisor (first): Dr. Ir. M.J. Gijsenberg

(3)

Management summary

Service performance is an important topic in marketing literature. However, variance in service performance is relatively new and therewith the effects are underestimated. This study focuses on this topic by answering: Does variability in service performance matter more than average service

performance in the effects on customer satisfaction and product usage? In order to answer this

question, a VARX model is used to analyze the effects of service performance and its variance on output measures as customer satisfaction and product usage. This analysis is done with a dataset from a Dutch public transport provider, which contains about 57 observations. Findings show that average service performance is more influential for customer satisfaction and variability in service performance does affect product usage. The current study provides managerial relevance because it shows that providing average service performance is not enough for firms. Variability in service performance needs to be managed. Furthermore, the study contributes to existing literature as variability in service performance is a rather new concept in marketing literature and therefore relatively few research has been done on the topic.

(4)

Preface

This master thesis announces the end of my study period at the University of Groningen. After a bachelor in International Business and a master in Marketing, my time as student has come to an end. Being the last hurdle between me and my degree, this thesis has been quite challenging. I applied for the topic of marketing effectiveness because of the availability of a public transport provider dataset. Next to my study, I work as a management-assistant in retail at central station in Groningen. There, I experience all kind of troubles that customers have with public transport. Therefore, I was extremely motivated to work with a dataset from a Dutch public transport provider. However, the dataset asked for an analysis method I had never used before which made it extremely challenging for me. I had to discover everything with regard to the method, the results, the interpretation, and so on.

Finally, I succeeded. For that, I would like to thank my first supervisor Maarten Gijsenberg as he was the one who always motivated me to continue and discover the analysis method. I really appreciate his valuable guidance during the process of discovering the analysis method and writing this thesis. Furthermore, I would like to thank my friends, who always supported me in what I was doing and made it possible for me to write this thesis in a way that people without much knowledge of the marketing concepts could understand.

(5)

1. Introduction

Service performance has been topic of research for decades. Most research is about service performance in a single setting: from employee to customer, for example in the hospitality industry (Lewis and McCann (2004), Akbaba (2006)). The most important aspect of a service is the lack of tangible cues which can be used to evaluate the service (Parasuraman, Zeithaml, Berry, 1985). Therefore, service evaluation is highly dependent on the perception of customers. Consequently, this intangibility of services, makes it hard for firms to evaluate the service performance. The inseparability of production and consumption of the service adds to the difficulty of evaluating service performance. It indicates that performance is dependent on how the employees and equipment deliver the service and thus service performance cannot be guaranteed as constant (Parasuraman et al., 1985; Rust, Inman, Jia, Zahorik, 1999). Researchers have been developing models to evaluate service performance and how failures in service performance can be recovered (Parasuraman et al. (1985), Cronin and Taylor (1994), Lewis and McCann (2004)). These studies mainly conclude that there are gaps between service performance, from the providers’ perspective, and service quality, as perceived from the receiver’s perspective (Parasuraman et al., 1985). All of them acknowledge the importance of service quality for the customer to come back and keep spending money at the same firm. After a service performance failure, it is hard to restore the satisfaction of customers (Lewis and McCann, 2004). Therefore, measuring service performance and securing a determined level of performance is necessary. However, all this is about service performance and consequences in a one-to-one setting.

(8)

2

cargo is perishable (Van Doorn and Verhoef, 2008). Moreover, if travelers experience delays in their journey, it may lead to missed connections or appointments which is a big inconvenience for the traveler. Therefore, it seems logical that for example firms in the Dutch public transport industry take service performance, in their case punctuality, as their main goal (NS, Prorail, 2016).

But does this actually make sense? Does higher service performance pay off? Does higher punctuality make travelers happier? Does higher punctuality lead to more travels? Is it a must that punctuality is constant or can day-to-day variability in punctuality be forgiven by the traveler? That is what the current research aims to investigate with data from a Dutch public transport provider.

Data available for this research is over five years of monthly data from a public transport provider. This data allows service performance to be measured as punctuality of vehicles. Moreover, the variance within the monthly punctuality of vehicles is available which allows this study to focus on the (un)predictability of the punctuality defined as day-to-day variability in service performance. As measure of product usage, total travel distance is available. Two measures of customer satisfaction are available.

This study will be answering the following research question:

Does variability in service performance matter more than average service performance in the effects on customer satisfaction and product usage?

(9)

3

(10)

4 H1 + H2 + H3 -H4 -H5 + H6 +

2. Theoretical Framework

The focus of this study is on the dynamic effects of service performance on customer satisfaction and product usage. However, as there are many similarities between service performance and product quality, literature on product quality is used as well to motivate this study.

2.1 Conceptual model

Before relevant literature on the topic will be discussed, a conceptual model will help to visualize the objective of this study. Figure 1 shows the conceptual model of this study.

Figure 1: Conceptual model

Note: The effects will be estimated on short-term and long-term.

As figure 1 shows, the objective is to estimate the influence of service performance on customer satisfaction and product usage. There is a difference between the average service performance per month and its variance from day-to-day. This variance implicates an upward but also downward movement of service performance. As the data available for this study has several severe drops in performance, it is interesting to investigate whether these drops have an influence on both dependent variables: customer satisfaction and product usage. The remainder of this chapter will theoretically explain the foundation of this conceptual model.

2.2 Product quality

A customer can easily determine the quality of a product by its tangible cues like style, label, feel, package, color (Parasuraman et al., 1985). However, rating product quality is also partly determined by expectations prior to usage (Olshavsky and Miller, 1972). These expectations are based on brand names, price, physical appearance, and retailer reputation (Dawar and Parker,

(11)

5

1994). Rust, Inman, Jia, and Zahorik (1999) argue that these expectations are cumulatively effective after usage. Moreover, a certain level of expected quality is formed (Rust et al., 1999). Olshavsky and Miller (1972) found that higher expectations beforehand, lead to better product quality ratings after usage. This indicates that the stronger the brand equity before usage, the better the product quality rating after usage. Hence, it is clear that product quality is based on cumulative perceptions and experiences of customers. However, what happens when product quality is not constant over time? How do customers respond to fluctuations in product quality?

2.2.1 Variability in product quality

There are several forms of fluctuations in product quality. It can be that there is a severe quality issue with a specific product as studied by Van Heerde, Helsen and Dekimpe (2007), which they defined as a product-harm crisis. They argue that product-harm crises cause severe damage to the company’s revenues and brand equity. Moreover, Van Heerde et al. (2007) find that the effectiveness of marketing variables as advertising is not of the same magnitude after a product-harm crisis as it was before the crisis which emphasizes a long-term effect of the crises. Dawar and Pillutla (2000) find that customer expectations prior to the product-harm crises are important for the effects of the crisis on brand equity. As customer expectations are high before the crisis, they are more likely to get a better understanding of the product-harm crisis through firm responses which eventually leads to less harm to their brand equity (Dawar and Pillutla, 2000). In cases of extremely strong and high expectations, firms can even block the negative effect of product-harm crises on brand equity and turn it into a positive effect (Dawar and Pillutla, 2000). However, in all cases, product-harm crises have a negative effect on brand equity and sales (Van Heerde et al. (2007), Dawar and Pillutla (2000)). Consequently, it is important to prevent product-harm crises from happening. Therefore, it is suggested that the best solution is a highly controlled production process in which product failures are eliminated (Van Heerde et al., 2007). In case of a real product-harm crisis, a remedy seems rather easy. Once the specific product line and series is defined, a recall on those products can be done to provide the customers a solution. However, this is not restoring brand equity and sales to pre-crisis levels.

(12)

6

implicates that there are no crisis situations (as mentioned in Van Heerde et al., 2007), but just little deviations from the normal quality level. Rust et al. (1999) studied the influence of customer expectations on perceived quality. They found that expected variance from the expected quality, leads to choices for lower quality with less expected variability. Explanations for that can be found in the risk aversive nature of customers (Rust et al., 1999). It is assumed that high quality goes along with high variability which is unsafe and risky in the eyes of customers (Rust et al., 1999). This indicates that customers are looking for stable and constant product quality. Moreover, quality must meet expectations in order to be preferred (Rust et al., 1999).

2.3 Service performance

As services are intangible, heterogeneous and inseparable, the quality, or rather performance, is hard to measure (Parasuraman, Zeithaml, Berry, 1985). The intangible characteristic entails that it is not easy to understand how the customer will evaluate the service performance. The main difference between consuming a product versus consuming a service is that production and consumption are separated when consuming a product while they cannot be separated when consuming a service (Parasuraman et al., 1985). This leads to problems with securing a certain level of service performance because the performance is directly dependent on the staff providing the service. Moreover, as with product quality, service performance is subjective to the interpretation of customers (Boulding, Kalra, Staelin, 1999). Customers use prior beliefs about service performance to interpret new information, which in the long run leads to less importance of new information (Boulding et al., 1999). This indicates that customers will use prior beliefs and experiences to forecast what will happen during a next service delivery in order to make an overall service performance assessment (Boulding et al., 1999).

2.3.1 Average service performance’ influence on customer satisfaction

(13)

7

attribute levels is determinant in the effect on customer satisfaction. This indicates that customers are not only influenced by the average service performance, but by the performance per attribute of the service. However, for the current study, the relationship of average service performance is investigated. Therefore, the following hypothesis is formulated:

H1: Higher average service performance leads to higher customer satisfaction. 2.3.2 Average service performance’ influence on product usage

Rust et al. (1999) found that higher expected quality is not always a straightforward relationship to a higher preference. In their experiments, nearly half of their respondents chose for a lower quality option while a high quality option was available. As said before, quality must meet expectations in order to be preferred (Rust et al., 1999). However, in line with previous research (Anderson and Sullivan, 1993), Boulding et al. (1999) found that customer make repurchase decisions on the basis of overall quality assessment. Overall quality assessment includes attribute level evaluation of performance, prior beliefs, new information about the service, but also perceptions of competitors’ service performance (Mittal et al. (1998), Boulding et al. 1999)). Therefore, it is necessary to constantly provide a higher or equal level of service performance in order to stay competitive and be the preferred option for the customer. Previous research concluded that customers adjust usage levels to changes made by the firm, like changes in perceived service quality (Bolton and Lemon (1999), Tam (2004)). Moreover, Sriram, Chintagunta and Manchanda (2015) confirm this by concluding that customers who experience low levels of quality, are more likely to terminate service and thus stop product usage. Therefore, it can be argued that average service performance will influence product usage. Consequently, following hypothesis can be formulated:

H2: Higher average service performance leads to higher product usage.

2.4 Variability in service performance

Parasuraman et al. (1985) argue that services are heterogeneous which means that the service is not a constant. It can fluctuate day by day. This means that there is always a degree of variability within service performance. As with product quality, there are several types of fluctuations: crisis situations and little deviations from standard performance.

(14)

8

perceived service quality, which they approach as customer satisfaction. These failures, which they name losses, have a bigger effect in comparison to effects from the same magnitude in the opposite direction: gains. Therewith they find evidence for asymmetric effects between losses and gains. Moreover, they let their effects depend on a trend in service performance and find that a failure after an upwards trend hurts more in comparison to another failure in a row of failures. Moreover, this view is confirmed by Bolton (1998), as those findings show that customers are highly sensitive to losses during service encounters. Boulding et al. (1999) researched the effect of large positive shifts in service performance which is approachable as the opposite of Gijsenberg et al. (2015). They found that service performance is influenced by the company itself in terms of service delivery (Boulding et al., 1999). When the company has large positive shifts in service performance with average performance held constant, customers will adapt their perceived standard level of quality and will be disappointed when a next service delivery is not of the previous high quality (Boulding et al., 1999). Indirectly, they all argue that customers are sensitive to variability in service performance, downwards (Bolton, (1998), Gijsenberg et al., (2015)) and upwards (Boulding et al., 1999).

(15)

9

Sriram, Chintagunta and Manchanda (2015) recently researched the influence of service quality on customer retention rates, with the use of data on video-on-demand services. Moreover, Sriram et al. (2015) conclude that higher variability of service quality leads to less appreciation of improvements in average quality. This is in line with the findings of previous research (Boulding et al., (1999) Rust et al. (1999)), therefore it is likely that customers are sensitive to variability of service performance. Sriram et al. (2015) conclude that variability of service quality is highly underestimated. Therefore, they propose further research on variability of service quality in multiple settings

2.4.1 Variability in service performance’ influence on customer satisfaction

Already in 1993, Anderson and Sullivan indicate that there are negative effects of disconfirmation on customer satisfaction. Disconfirmation occurs when delivered service is not meeting expectations. When service performance shows variability, it leads to situations of disconfirmation which hurt customer satisfaction. Reimann, Lünemann and Chase (2008) provide reasons for this damage to customer satisfaction. They argue that uncertainty avoidance is the reason why customers are less satisfied after a service defect (Reimann et al., 2008). This means that when customers experience variability in service performance, they feel the service is becoming unsecure which causes uncertain situations for them. If customers cannot rely on a service providing a constant level of performance, they tend to avoid the uncertainty that comes along (Reimann et al., 2008). In other words, if service performance shows high variability, the service performance becomes more unpredictable which causes uncertainty for customers which makes them less satisfied. However, Reimann et al. (2008) have done their research in a B2B market which assumes that there are huge differences across countries with respect to levels of uncertainty avoidance (Hofstede, 1993). Moreover, within the respondents in their research, there is a wide diversity of origins, from Spain to Sweden, which causes huge differences on the uncertainty avoidance index as measured by Hofstede (Reimann et al., 2008). For the current study, data used consists of a Dutch public transport provider which means that there is a single level of uncertainty avoidance as indicated by Hofstede (1993). However, in line with Reimann et al. (2008), following hypothesis is formulated:

(16)

10

2.4.2 Variability of service performance’ influence on product usage

Rust et al. (1999) concluded that consumers are more likely to choose a brand, when the brand performs as expected. Assuming that expected performance is constant, fluctuations would not lead to repeated usage. Boulding et al. (1999) confirm this, by arguing that accidental large positive shifts in service performance lead to an increased standard for the customer. When service performance is as normal with the next service delivery, it will be perceived as lower overall assessment of service performance (Boulding et al., 1999), and low service performance leads to termination behavior (Sriram et al., 2015).

As stated before, Sriram et al. (2015) found evidence for the positive relationship between variability of service quality and termination rates. When variability of service quality increases, termination rates increase as well. A reason for this relationship can be found in the customers’ uncertainty avoidance. Variability lowers expectations of utility of the service, which activates the customers risk-aversion which leads in turn to termination (Sriram et al., 2015). This indicates that customers are not willing to take the risk of less utility. Therefore, customers are avoiding risky services and thus are likely to stop using services when they have a certain level of uncertainty. Customers want a predictable service performance before they rely on it. Bolton, Lemon and Bramlett (2006) confirm this view by concluding that increased variability of service performance leads to a decreased value of the service contract and lower likelihood of renewal in B2B settings. Variability in service performance indicates the level of uncertainty that customers have to face (Bolton et al. (2006), Sriram et al. (2015)). Consequently, following hypothesis is formulated:

H4: Variability in service performance lead to lower product usage.

2.5 Customer satisfaction and product usage

(17)

11

H5: Higher customer satisfaction leads to higher product usage.

The relationship between product usage and customer satisfaction seems less logical as the other way around. However, considering that this study will be in the public transport industry, it can be a reasonable relationship. In the Netherlands, it is common to use public transport a lot so it might not always be the customers’ choice. It can be that using public transport is the only option to arrive at a destination. Think of customers who are not in possession of a car and the distance from A to B is too far for using a bicycle. These customers are somewhat ‘forced’ into public transport when they need to travel long distances. Nevertheless, also these customers have formed expectations which are nurtured by experiences (Boulding et al., 1999). As they discover that public transport serves their needs, it can be that this positive experience will cause a higher perceived performance and thus a higher customer satisfaction (Tam, 2004). However, they might not like every attribute of the service, it still can be that more product usage leads to higher customer satisfaction because determination of customer satisfaction can be on attribute-level (Mittal et al., 1998). Consequently, the following hypothesis is formulated:

(18)

12

3. Research design and methodology

3.1 Data structure

Data from a major Dutch public transport provider is used for the analysis. This dataset contains variables on service performance, customer satisfaction, and product usage who provide monthly data from January 2007 until September 2011. In order to control for the effects mentioned before, control variables are added to this dataset. Because the data is from a Dutch public transport provider, holidays are based on the average Dutch schedule of school holidays as implicated by the Ministry of Education, Culture and Science. To control for temperature effects, monthly data of average temperature in the Netherlands is added from the Royal Netherlands Meteorological Institute (KNMI).

3.2 Description of variables

The variables included in the analysis will be explained according to their construction which will be helpful in interpreting the results.

3.2.1 Dependent variable: Customer Satisfaction

Customer satisfaction is available in two variables. The first shows the percentage of respondents who evaluated the public transport provider with a 7 or higher (on a scale of 10). This group of respondents can be described as customers that are really satisfied with the company. The second variable shows the absolute average customer satisfaction score which can be seen as the more appropriate measure as all respondents are included. In order to perform the analysis, one of the two variables should be chosen as dependent variable because of multicollinearity issues. However, it might be interesting to see whether there is a difference in dynamic effects of the overall average customers in comparison to the more satisfied customers. Therefore, the analysis will be performed twice with both customer satisfaction variables for the purpose of a robustness check.

3.2.2 Dependent variable: Product Usage

(19)

13 3.2.3 Service Performance

Service performance is measured in two punctuality variables. The first variable is the punctuality of vehicles in a 3-minute window. It measures whether the vehicles arrive within three minutes from the predetermined arrival time. The second variable is the punctuality of vehicles in terms of made connections with other vehicles. In order to perform the analysis, one of the two punctuality variables should be chosen because of multicollinearity issues. The consequences of a missed connection are worse, in comparison to a vehicle that arrives only 5 minutes late at its destination, because a missed connection may lead to a delay of on average more than 30 minutes. Therefore, the punctuality variable that measures the punctuality in terms of made connections, performance connections, will be used for the analysis.

3.2.4 Variance in Service Performance

From both variables mentioned in section 3.2.3, the variance is available. This variance should be interpreted as the day-to-day fluctuations in punctuality of the vehicles. When variance is high, the punctuality shows a lot of difference across days within the month. When variance is low, the punctuality shows less difference across days within the month. As explained in section 3.2.3. the variable variance performance connections will be used for the analysis.

3.2.5 Descriptive statistics

In order to get an insight in the data, table 1 provides descriptive statistics for the endogenous variables. As table 1 shows, customer satisfaction %7 shows more variance as compared to customer satisfaction average score. This could indicate that this measure is more influential. There is also a big variance in variance performance connections.

Table 1: Descriptive statistics endogenous variables

Mean Minimum Maximum

Customer satisfaction %7 74.8% 65.76% 80.41%

Customer satisfaction average score 6.97 6.75 7.14

Performance connections 92.34% 86.65% 95.28%

Variance performance connections 8.84 1.54 62.83

(20)

14 3.2.6 Exogenous variables

The control variables included in this analysis are holidays, temperature and trend which are constructed as follows.

As the data is on a monthly base, the holidays variable is constructed as a dummy variable being 1 during a holiday month and 0 during the other months. There are no full holiday months except for summer holidays, therefore, only summer holidays are accounted for.

Temperature is another control variable constructed as an interval variable with the average temperature in the Netherlands during that month in degrees Celsius. The KNMI uses a single meteorological station to determine the average temperature known as ‘De Bilt’, which is located in the center of the Netherlands. As the public transport covers the whole country, this central location seems to be an appropriate measure for temperature.

The last control variable checks whether there is a deterministic trend in the data.

3.3 Methodology

The data used for this analysis is time series data. There are 57 months from January 2007 until September 2011. In order to analyze time series data, a vector autoregressive (VAR) model is used.

3.3.1 VARX model

The regular VAR model only allows endogenous variables to be included. A VARX model is able to show, next to endogenous effects, exogenous effects that are affecting the dependent variable over time (Srinivasan, Vanhuele, and Pauwels, 2010). This means that a VAR model will for example show bad service performance in a given week will lead to less product usage in a few months. When including an exogenous variable as holidays, the VARX model will for example capture the effect of holidays leading to less travelers and so less product usage. Consequently, VARX is able to show a complete picture of the dynamics in time series. VARX is commonly used in marketing literature (e.g. Srinivasan et al. (2010), Nijs, Dekimpe, Steenkamp and Hanssens (2001)). The VARX equation tested in this study will be:

𝑌

_𝑡

= 𝛼 + ∑

𝑃_𝑝=1

𝛷

𝑌

_𝑡−𝑝

+ 𝛹𝑋

_𝑡

+ 𝜀

_𝑡

(1)

(21)

15

Where

α

is the vector of intercepts,

Φ

is the matrix of endogenous variables at time t – p where p are lags,

Ψ

is the matrix of exogenous variables at time t (holidays, temperature and trend) and

ε

is the matrix of error terms at time t.

The complete notation of the model is presented after unit root testing and when the number of lags is determined.

3.3.2 Unit root tests

The VARX model requires endogenous variables to be stationary instead of evolving over time (Srinivasan, Vanhuele, Pauwels, 2010). There are two commonly used tests to check for the presence of unit roots in the endogenous variables. One is the augmented Dickey-Fuller test, which tests a null hypothesis that the variable has a unit root (Dickey and Fuller, 1979). When the test shows that the endogenous variable has a unit root, even after a constant and trend are added, the variable needs to be transformed into its first or second difference. The other test for stationarity is the Phillips-Perron test, which runs from the same null hypothesis as the augmented Dickey-Fuller test (Phillips and Perron, 1988).

As stated before, it is important that the endogenous variables are stationary. Therefore, the augmented Dickey-Fuller test is performed. Table 2 shows the results from the test. These are the values for the T statistic which need to be above the critical values in order to be significant. These critical values are mentioned in table 2 below the variables. The variables who fail to reject the null hypothesis, because the T statistic is not above the critical value, are both customer satisfaction variables. The variables show non-significant results with trend and intercept test which means that those two variables contain a unit root and are not stationary over time. The Phillip-Perron test shows the same results.

(22)

16

rejected. This means that the KPSS test is showing the same results as the augmented Dickey-Fuller and Phillips-Perron test: both customer satisfaction variables are not stationary.

In order to overcome the problem of evolving variables, both customer satisfaction variables should be transformed to their first difference. When transformed to the first difference, the variables show a significant result on the augmented Dickey-Fuller and Phillip-Perron unit root tests, and a non-significant result on the KPSS unit root test (table 2). This means that both variables are transformed from non-stationary to stationary. In the rest of the analysis, customer satisfaction %7 and customer satisfaction average score will be used in first difference: Δ customer satisfaction %7 and Δ customer satisfaction average score.

Table 2: Unit root tests

Note: values displayed in green are at 1% significant, in yellow are at 5% significant, and in red are at 10% significant or non-significant. 10% significance is considered to be non-significant.

Augmented Dickey-Fuller test

Phillip-Perron KPSS

Intercept (I) and Trend (T) I + T I + T I + T

Customer satisfaction %7 -3.22 -2.72 0.19

Customer satisfaction average score -2.59 -2.59 0.19

Performance connections -4.36 -4.41 -

Variance performance connections -5.69 -5.67 -

Traveled KMs -5.32 -4.57 -

Δ Customer satisfaction %7 -6.22 -6.12 0.06

Δ Customer satisfaction average score -6.55 -6.66 0.06

Test critical values 1%

-4.14 -4.14 0.21

5% -3.50 -3.50 0.15

(23)

17 3.3.4 Lags

VARX models are estimating the endogenous variables by the means of their past (Srinivasan et al., 2010). This is represented by the lags of the endogenous variables. However, when including too much lags, the model will suffer easily from too much parameters (Horváth, Leeflang, Wieringa and Wittink, 2005). There are four endogenous variables per model, which indicates that there are 16 additional parameters per lag to estimate. Therefore, it is important to systematically choose the number of lags.

The number of lags is dependent on information criteria. Two available criteria are the Akaike Information Criterion (AIC) and the Schwartz Information Criterion (SIC). However, SIC has a harder punishment for additional parameters in comparison to AIC so therefore SIC is considered to be the determinant information criterion. The models are estimated from 0 up to 10 lags which provides a list of values for SIC. The absolute lowest values for SIC indicates the optimal number of lags. Table 3 shows the list of SIC values per lag length. As the table indicates, there are only values for lags 0 to 8. This is because with 9 or 10 lags, the different lags are too similar which causes multicollinearity. SIC shows an optimal lag length of 1. Therefore, the models will be estimated with only 1 lag.

Table 3: Lag length selection

Δ customer satisfaction %7 Δ customer satisfaction average score

Lag length SIC SIC

0 53.937 46.681 1 53.724 46.385 2 53.921 46.576 3 54.358 47.017 4 55.076 47.663 5 55.486 48.034 6 55.848 48.419 7 55.744 48.376 8 56.054 48.501

(24)

18 3.3.5 VARX detailed specification

As stated before, the VARX model will be estimated with only 1 lag. The full notation is presented in equation 2.

[

𝑉𝑆𝑃𝐶

_𝑡−𝑝

𝑇𝐾

_𝑡−𝑝

_]

+

[

ASPC = average service performance connections VSPC = variance service performance connections TK = traveled KMs

HOLIDAY = holiday dummy

OLD = measurement dummy (old vs new measurement) TEMP = average temperature

TREND = deterministic trend p = number of lags

ε

= error term

(25)

19 3.3.6 Granger-causality

As first indication of causal relationships between variables, the Granger-causality test can be done (Granger, 1969). The null hypothesis in this test is: X1 does not cause changes in X2. Table 4 shows the results from the Granger-causality test for the endogenous variables. Beware that the significant results do not imply causal relations but only correlations. The results show that there are correlations between both performance variables and both customer satisfaction variables. Also interesting to see is that traveled KMs correlates with one performance variable and both satisfaction variables. However, conclusions cannot be drawn from these correlations.

Table 4: Granger-causality test

Variables 1 2 3 4 5

1. Δ Customer satisfaction %7 - .363 .923 .900 .125

2. Δ Customer satisfaction average score .998 - .891 .980 .063

3. Performance connections .000 .001 - .050 .022

4. Variance performance connections .026 .052 .112 - .084

5. Traveled KMs .034 .015 .022 .159 - Note: values are p-values for granger causality.

3.3.7 Impulse response functions

(26)

20

impulse response functions on short-term, long-term and the dust-settling time which they define as medium-term. It is interesting to investigate on short-term, medium-term and long-term because there might be permanent effect on the dependent variables. As this study works with monthly data, short-term is defined as 1 month, medium-term is defined as 3 months (1 quarter) and long-term is defined as 12 months (1 year).

3.3.8 GFEVD

(27)

21

4. Results

The data runs from January 2007 until September 2011 and thus consists of 57 monthly observations per variable. However, traveled KMs has three unknown values in the first three months of the observations. Therefore, these first three months are excluded from the analysis. This changes the total of observations in 54. Because of the transformation to the first difference, the first observation of both ΔCS7 and ΔACS is unknown. However, as there are three missing values for traveled KMs, the first difference transformation has no influence on the total number of observations.

4.1 VARX

The variables used for VARX are all used in their original form, except for the customer satisfaction variables who are transformed to their first difference. There is no log transformation applied because interpretation as elasticities will be lost due to the use of generalized impulse response functions.

4.1.1 Model fit

One thing that can be found in general output of VARX is the model fit. This is based on the R2 and the adjusted R2 per model. Table 5 shows the results for the model fit. For the first model, the fit for Δ customer satisfaction %7 is approximately 18%. This means that 18% of the variance is explained by the endogenous and exogenous variables in the model. This is not very high as compared to traveled KMs. For traveled KMs the fit is approximately 76%. In the second model, the fit is slightly better. Approximately 20% for Δ customer satisfaction average score and 76% for traveled KMs. This indicates that Δ customer satisfaction average score has slightly more explanatory power than Δ customer satisfaction %7.

Table 5: Model fit

Model Δ Customer satisfaction %7 Model Δ Customer satisfaction average score

Δ Customer satisfaction %7

Traveled KMs Δ Customer satisfaction average score

Traveled KMs

R2 _.18 _.76 _.20 _.76

(28)

22 4.1.2 Exogenous variables

As stated before, the interpretation of VARX estimates for endogenous variables is not possible. The only estimates that can be interpreted directly from VARX output are the estimates of the exogenous variables. These variables are not included in GFEVD and impulse response functions so therefore it is necessary to check them with VARX output for face validity.

The table in appendix 1 shows the results of VARX analysis for both models with the different customer satisfaction measurements. Significant results at 5% are highlighted in green and significance results at 10% are highlighted in yellow (T-statistics). As expected, there is a negative effect of holidays on traveled KMs in both models. An explanation for that can be that there are less customers commuting between home and work during holiday months. The old KM-measurement captures the KM-measurement error for traveled KMs in both models. Furthermore, in both models, traveled KMs is following a trend upwards as the exogenous variable trend is significant and has a positive coefficient for traveled KMs. There is no significant trend in both measures of customer satisfaction. Unfortunately, there is no effect of temperature in both models. An explanation for that might be that the effect during summer is captured in the holiday variable and the effects during winter are not strong enough.

4.2 Impulse Response Functions

As previously mentioned, impulse response functions show the results of a shock in one endogenous variable relative to their expected values without a shock (Srinivasan et al. (2010), Pauwels, Hanssens and Siddarth (2002)). Generalized impulse response functions (GIRFs) are derived from the VARX estimates using the Monte Carlo simulation approach with 1000 obtained standard errors which is more as mentioned in Srinivasan et al. (2010). However, more repetitions will lead to more stable and fitted results. First, the model with Δ customer satisfaction %7 (ΔCS7) and traveled KMs (TK) will be analyzed on short-term (1 months), medium-term (3 months) and long-term (12 months) which will be referred to as model 1. Second, the model with Δ customer satisfaction average score (ΔACS) and traveled KMs (TK) will be analyzed in the same way and will be referred to as model 2. Average service performance connections will be abbreviated as ASPC and variance of service performance connections as VSPC.

(29)

23

actual value, the redlines are two standard errors. However, when zero is between the red lines, it means that the effect can be zero. Therefore, when zero is between the red lines, the effect is not significant. However, this might be unclear as the red line may fluctuate around zero. Fortunately, there is another method to check significance. This is to calculate the T-statistic. This statistic is calculated by dividing the coefficient by the standard deviation. The absolute value of the

T-statistic is subjective to T-distribution tables. Values above 1.96 are at 5% significant and values

above 1.645 are at 10% significant which is commonly used for interpretation of impulse response functions (Gijsenberg et al., 2015).

4.2.1 Model 1: Δ Customer satisfaction %7 and Traveled KMs

In figure 2, the cumulative impulse response functions of the first dependent variable of model 1 (ΔCS7) are presented.

Figure 2: Accumulated GIRFs with responses of ΔCS7 to the other endogenous variables

(30)

24

The T-statistics that belong to these figures are presented in appendix 1, table 1. Note that ΔCS7 is a first differenced variable, which indicates that significant effects are permanent (Gijsenberg et al., 2015).

A 1 SD shock in ASPC leads to a positive response of ΔCS7 (figure 2, graph 2.1). However, the effect smoothens in the second month already. Notably, the effects are at 10% significant in month 1 and 2. Moreover, since the first differenced character of the responding variable, the effect is permanent. This means that hypothesis 1 can be confirmed at short-term, medium-term and long-term level. It is in line with expected results and a logical result because when service performance is higher, customers are more satisfied.

ΔCS7 shows a little negative response to a 1 SD shock in VSPC (figure 2, graph 2.2) which seems to be partly in line with expected results, however, there is no significant effect observed in the response of ΔCS7 to VSPC (appendix 2, table 1). This means that hypothesis 3 has to be rejected for model 1. Research on this aspect was exploratory, therefore it is hard to explain why changes in VSPC do not lead to a severe lower ΔCS7. However, a partly explanation might be that ΔCS7 represents customers who are relative highly satisfied as they scored the public transport company a 7 or higher. Therefore, it might be that these customers are not so influenced by VSPC.

ΔCS7 responds negatively to a 1 SD shock in TK (figure 2, graph 2.3). This effect appears to be significant at 5% in month 2 and at 10% in month 3. Moreover, this implies again a permanent effect on ΔCS7. Therefore, hypothesis 6 can be confirmed on all levels but in the opposite direction. It was expected that there would be a positive relationship between product usage and customer satisfaction. However, results show that there is a negative relationship between them. It can be that the public transport provider is not able to handle more TK which leads to crowdedness in vehicles when TK increases. Crowdedness may lead to too few seats for the amount of travelers which in turn leads to lower ΔCS7.

(31)

25

Figure 3: Accumulated GIRFs with responses of TK to the other endogenous variables.

Surprisingly, TK responds in the first two months negatively to a 1 SD shock in ΔCS7 (figure 3, graph 3.1). However, none of the periods show significant effects (appendix 2, table 2) and therefore, hypothesis 5 is rejected for model 1.

Similar to ΔCS7, TK responds positively to a 1 SD shock in ASPC (figure 3, graph 3.2). It indicates that a higher average service performance leads to higher product usage. This effect smoothens in the second month. However, also none of the periods show a significant effect (appendix 2, table 2) and therefore hypothesis 2 is rejected for model 1.

(32)

26

TK responds negatively to a 1 SD shock in VSPC which is in line with expected results (figure 3, graph 3.3). However, the effect is already gone after the second month. Interestingly, this effect appears to be significant at 10% in the second period (appendix 2, table 2). This means that a 1 SD shock in VSPC leads to negative responses of TK in the second month. Therefore, hypothesis 4 can be confirmed on medium-term level. This confirms that customers who experience higher variability in service performance, are more likely to lower their product usage.

4.2.2 Model 2: Δ Customer satisfaction average score and Traveled KMs

In figure 4, the cumulative impulse response functions of the first dependent variable of model 2 (ΔACS) are presented.

Figure 4: Accumulated GIRFs with responses of ΔACS to the other endogenous variables

-0,02 -0,01 0 0,01 0,02 0,03 0,04 0,05 1 2 3 4 5 6 7 8 9 10 11 12

4.1 Response of ΔACS to ASPC

(33)

27

Similar to ΔCS7, ΔACS responds positively to a 1 SD shock in ASPC (figure 4, graph 4.1). When service performance increases, customers are more satisfied. Although, the effect is smoothened in the second month already. The effect appears to be significant at 10% for the first two months (appendix 3, table 1). However, as with ΔCS7, ΔACS is a first differenced variable. Therefore, every significant effect is interpreted as permanent effect and thus hypothesis 1 can be confirmed for model 2. This indicates that shock in average service performance has a permanent positive effect on average customer satisfaction.

Also similar to ΔCS7 is the response of ΔACS to a 1 SD shock in VSPC (figure 4, graph 4.2). This response is negative which is in line with expected results. However, there are no significant effects observed which indicates that hypothesis 3 has to be rejected for model 2 as well (appendix 3, table 1). As said before, research on the variability in service performance was explorative, therefore it is hard to explain why results show no significant effects.

The response of ΔACS to TK shows a negative line (figure 4, graph 4.3). This indicates that a 1

SD shock in TK causes a negative movement of ΔACS. In other words, when product usage

increases by 1 SD, average satisfaction decreases. Moreover, this effect is in every period significant, at 5% on medium-term and at 10% on short- and long-term (appendix 3, table 1). However, because of the first differenced nature of ΔACS, effects are in any case permanent. This indicates that hypothesis 6 is confirmed, but in the opposite direction from what was expected. The cumulative impulse response functions of other dependent variable in model 2, TK, are displayed in figure 5. Corresponding T-statistics are presented in appendix 3, table 2.

(34)

28

Figure 5: Accumulated GIRFs with responses of TK to the other endogenous variables

The response of TK to a 1 SD shock in ASPC is positive (figure 5, graph 5.2) which would indicate that increased average service performance leads to more product usage. However, there are no significant effects (appendix 3, table 2). Therefore, hypothesis 2 is rejected for model 2 as well. On contrary, the response of TK to a 1 SD shock in VSPC is negative (figure 5, graph 5.3). Moreover, this effect appears to be significant at 10% in month 2 and month 4, so at medium-term. This indicates that higher variability in service performance leads to lower product usage. Therefore, hypothesis 4 is confirmed on medium-term level.

(35)

29

4.3 GFEVD

As mentioned before, GFEVD indicates the dynamic exploratory value on the dependent variables from each endogenous variable (Srinivasan et al. (2010), Pesaran and Shin (1998)). In order to answer the research question: Does variability in service performance matter more than average

service performance in the effects on customer satisfaction and product usage, it is necessary to

decompose the variance of customer satisfaction and product usage. In terms, GFEVD is done for ΔCS7, and TK in model 1, and ΔASC, and TK in model 2.

Table 6: GFEVD results model 1

Model 1: ΔCS7 Model 1: TK

Month ΔCS7 ASPC VSPC TK ΔCS7 ASPC VSPC TK

1 100.00% 0.00% 0.00% 0.00% 3.79% 0.11% 1.28% 94.82%

3 92.87% 3.13% 1.11% 2.89% 7.25% 18.79% 1.14% 72.82%

12 92.85% 3.14% 1.12% 2.89% 7.25% 18.81% 1.16% 72.78% Table 6 shows the GFEVD results for model 1. The results show the percentage of variance that is explained by the endogenous variables to ΔCS7 in the left panel and TK in the right panel. Logically, both dependent variables do for a large part explain themselves. However, relevant for the research question of this study are the effects of ASPC and VSPC. In the left panel, the explaining power of ASPC is slightly higher than the explaining power of VSPC. However, as the results in chapters 4.2.1 show, only ASPC is significant for both dependent variables. Moreover, for TK, VSPC is significant on medium-term. When comparing percentages in the third month (medium-term), it can be seen that ASPC explains more variance of TK as compared to VSPC (18.79% versus 1.14%). This indicates that the research question is not confirmed for model 1.

Table 7: GFEVD results model 2

Model 2: ΔACS Model 2: TK

Month ΔACS ASPC VSPC TK ΔACS ASPC VSPC TK

1 100,00% 0,00% 0,00% 0,00% 5,92% 0,05% 0,91% 93,12%

3 93,71% 2,68% 0,31% 3,30% 9,13% 18,36% 0,77% 71,74%

(36)

30

however, the effect is incredibly small. Therefore, model 2 does not confirm the research question either.

4.4 Hypotheses testing

Model 1 and 2 tested both the six hypotheses with variables ΔCS7, TK, and ΔACS, TK respectively as dependent variables. An overview is presented in table 8. The results show permanent significant effects for hypothesis 1 and 3, although effects measured for hypothesis 3 were in the opposite direction. Moreover, hypothesis 4 is confirmed on medium-term for both models. Lastly, there is a strange short-term effect measured that confirms hypothesis 5 in the opposite direction.

Table 8: Results on hypotheses

s*

- -

H6: Higher product usage leads to higher

customer satisfaction.

s* s* s*

(37)

31

5. Discussion and Conclusions

5.1 Discussion

Findings show that average service performance is an important permanent driver for customer satisfaction which is in line with previous research (Anderson and Sullivan, 1993). Model 1 and model 2 found both evidence for this which indicates that it is a robust result. Previous research also suggested that average service performance would be a driver for product usage (Rust et al. (1999), Boulding et al. (1999), Sriram et al. (2015). However, current research did not find evidence for that.

Initially, theoretical reasoning indicated that customers are sensitive for variability in service performance in both aspects, customer satisfaction and product usage (Anderson and Sullivan (1993), Boulding et al. (1999), Rust et al. (1999), Sriram et al. (2015)). However, findings only show evidence for the influence of variability in service performance on product usage. Both models showed the same medium-term effect which indicate that the result is robust. Moreover, the fact that variability has a small but significant influence indicates that it should not be underestimated and be more investigated.

The relationship between customer satisfaction and product usage is quite the opposite from what theory would expect. Surprising is the short-term negative influence of customer satisfaction on product usage. It makes no sense that more satisfied customers will lower or terminate their product usage. This matter should be subject in further research to make sure if it is a consistent result. Also, only one model (with average customer satisfaction) showed this evidence which already indicates that the effect is not robust.

(38)

32

in both models, it is a robust result. Therefore, more research needs to be done to be able to generalize results about the relationship between product usage and customer satisfaction. The main objective of this study was to answer the research question:

Does variability in service performance matter more than average service performance in the effects on customer satisfaction and product usage?

As results in chapter 4.3 show, the research question has to be answered with a no. Results show that variability in service performance does not matter more than average service performance does in the effects on customer satisfaction and product usage. There is not much previous research that has been investigating the effects of variability in service performance which makes it difficult to find an explanation for this result. However, there are also a few theoretical explanations for positive effects of variability in service performance, which are also not observed but may explain why negative relations are not observed either. Variability may lead to positive results when the firm responds adequately to critical incidents and with that intensifies the relationship with the customer (Van Doorn and Verhoef, 2008). Bolton et al. (2006) confirm this view by arguing that extreme positive outcomes lead to managers attributing more value to contracts which leads to renewal. This could indicate that the positive effects of variability outweigh the negative effects.

5.2 Managerial implications

(39)

33

part of customers of the public transport network in the Netherlands travel by a special public transport card. This makes it easy for public transport firms to follow their customers’ path and thus also if their customers experience variability in performance. Especially missed connections can be detected because these will cause a detour or longer traveling times. Furthermore, this information should be used to monitor the degree of experienced variability and in case of high variability, the public transport firm can try to prevent the customer from leaving by specific responses which intensify the relationship between the customer and the firm (e.g. Van Doorn and Verhoef, 2008). An example of firm responses might be that in case of delays and missed connections, the firm acts proactively (e.g. by means of an app) in order to compensate for the damage done, instead of waiting until the customer filed a complaint or in the worst case does nothing. Moreover, a proactive attitude of the firm can be implemented to prevent variability in service performance (e.g. by means of notifications) in case of delays or problems on the route which is usually used by the customer. However, this requires an individual approach to customers which might be costly.

Disturbing are the permanent negative effects from product usage on customer satisfaction. It indicates that customers are less satisfied when product usage increases. As mentioned before, it might be that increased product usage causes crowdedness in vehicles which in turn decreases customer satisfaction. Managers should overcome this problem by analyzing which services are frequently used and create the right balance between demand and supply. In the current industry setting, the data gathered with the public transport card can be used to analyze frequently used routes. Models can forecast the amount of travelers on which managers can adjust the amount and length of vehicles needed. This could overcome the negative side effects that are now arising from an increase in product usage.

5.3 Limitations and further research suggestions

(40)

34 5.3.1 Limitations

The first limitation lies in the length of the data. There is only four and a half years of monthly observations available. Most variability occurs during winter, which indicates that a longer period would contain more winters and thus more variability. Maybe effects of variability in service performance will be measured when more winters are included and thus a longer period of time is covered. Moreover, the data comes from only one public transport provider. Data from several public transport providers would make the analysis more reliable and generalizable.

Considering the methodology, there are also some limitations. As said before, VARX allows to show a complete picture of dynamics in time series. However, adding lags of all endogenous variables means a strong increasing number of parameters per added lag. This lead to a model with only 1 lag while more information might be hidden in other lags. So the strong increasing number of parameters is a limitation of using VARX models. Furthermore, the customer satisfaction variables have both low R2 scores. Only about 20% of the variance is explained by the variables included in the VARX, which is relatively low. Therefore, it might be that there are some critical factors missing. Although, from a theoretical perspective, there is no indication from other potential variables.

(41)

35

performance by the customers (Parasuraman et al., 1985). When individual data would be available, measures of perceived average service performance and perceived variability of service performance could be used for a repetition of this analysis. If obtaining those data is possible, the performance measurements would include more dynamics of service performance. So the limitation can be found in the data being on monthly aggregate level.

5.3.2 Further research suggestions

Although evidence for the influence of variability in service performance on customer satisfaction is not found, it can be interesting to investigate the same matter with data that covers a longer period of time. Gijsenberg et al. (2015), found significant effects for average service performance over time with a dataset that is about six years of observations. Moreover, the asymmetric effects that have been discovered, can also be present in variability of service performance influences. Also Rust et al. (1999) gave leads for this by concluding that worse-than-expected quality hurts more than better-than-expected quality helps. Additionally, Anderson and Sullivan (1993) already found more negative effects of negative disconfirmation in comparison to positive effects of positive disconfirmation. Therefore, further research should focus on finding more evidence of the influences of variability in service performance on both customer satisfaction and product usage, including asymmetric effects for positive shifts versus negative shifts in variance in service performance.

Moreover, as the concepts of customer satisfaction and product usage are closely related to each other, there might be reason to believe that there is a mediating effect of customer satisfaction between service performance (average as well as variance in) and product usage. Anderson and Sullivan (1993) gave indications for an effect between perceived quality and repurchase intentions through customer satisfaction. Therefore, it could be of managerial interest to investigate whether this mediating effect exists especially for variance in service performance.

(42)

36

6. References

Anderson, Eugene W., and Sullivan, Mary W. (1993) “The Antecedents and Consequences of Customer Satisfaction for Firms”, Marketing Science, Vol. 12, No. 2, pp. 125 – 143

Akbaba, Atilla (2006), “Measuring Service Quality in the Hotel Industry: A Study in a Business Hotel in Turkey”, Hospitality Management, Vol. 25, pp. 170 – 192

Bolton, Ruth N., (1998) “A Dynamic Model of the Duration of the Customer’s Relationship with a Continuous Service Provider: The Role of Satisfaction”, Marketing Science, Vol. 17, No. 1, pp. 45 – 65

Bolton, Ruth N., and Lemon, Katherine N. (1999) “A Dynamic Model of Customers’ Usage of Services: Usage as an Antecedent and Consequence of Satisfaction”, Journal of Marketing

Research, Vol. 36, pp. 171 – 186

Bolton, Ruth N., Lemon, Katherine N., Bramlett, Matthew D. (2006) “The Effect of Service Experiences over Time on a Supplier’s Retention of Business Customers”, Management Science, Vol. 52, No. 12, pp. 1811 – 1823

Boulding, William, Kalra, Ajay, and Staelin, Richard (1999) “The Quality Double Whammy”,

Marketing Science, Vol. 18, No. 4, pp. 463 – 484

Cronin, J. Joseph, and Taylor, Steven A. (1994), “SERVPERF Versus SERVQUAL: Reconciling Performance-Based and Perceptions-Minus-Expectations Measurement of Service Quality”,

Journal of Marketing, Vol. 58, pp. 125 – 131

Dawar, Niraj, and Parker, Philip (1994) “Marketing Universals: Consumers’ Use of Brand Name, Price, Physical Appearance, and Retailer Reputation as Signals of Product Quality”, Journal of

Marketing, Vol. 58, pp. 81 – 95

(43)

37

Dickey, David A., and Fuller, Wayne A. (1979) “Distribution of the Estimators for Autoregressive Time Series With a Unit Root”, Journal of the American Statistical Association, Vol. 74, No. 366, pp. 427 – 431

Gijsenberg, Maarten J., Van Heerde, Harald J., Verhoef, Peter C. (2015) “Losses Loom Longer Than Gains: Modeling the Impact of Service Crises on Perceived Service Quality over Time”,

Journal of Marketing Research, Vol. 52, pp. 642 – 656

Granger, C.W.J. (1969) “Investigating Causal Relations by Econometric Models and Cross-spectral Methods”, Econometrica, Vol. 37, No. 3, pp. 424 – 438

Hofstede, Geert (1993) “Cultural Constraints in Management Theories”, Academy of Management

Executive, Vol. 7, No. 1, pp 81 – 94

Horváth, Csilla, Leeflang, Peter S.H., Wieringa, Jaap E., Wittink, Dick R. (2005) “Competitive reaction- and feedback effects based on VARX models of pooled store data”, International Journal

of Research in Marketing, Vol. 22, pp. 415 – 426

Lewis, Barbara R. and McCann, Pamela (2004), “Service Failure and Recovery: Evidence from the Hotel Industry”, International Journal of Contemporary Hospitality Management, Vol. 16, Iss 1, pp. 6 – 17

Mittal, Vikas, Ross, William T., Baldasare, Jr. and Patrick M. (1998) “The Asymmetric Impact of Negative and Positive Attribute-level Performance on Overall Satisfaction and Repurchase Intentions”, Journal of Marketing, Vol. 62, No. 1, pp. 33 – 47

Nijs, Vincent R., Dekimpe, Marnik G., Steenkamp, Jan-Benedict E.M., and Hanssens, Dominique M. (2001) “The Category-Demand Effects of Price Promotions”, Marketing Science, Vol. 20, No. 1, pp. 1 – 22

How Service Performance Affects Customer Satisfaction and Product Usage

How Service Performance Affects

Customer Satisfaction and Product Usage

How Service Performance Affects

Customer Satisfaction and Product Usage

Management summary

Preface

Table of contents

1. Introduction

2. Theoretical Framework

3. Research design and methodology

𝑌

= 𝛼 + ∑

𝛷

𝑌

+ 𝛹𝑋

+ 𝜀

α

Φ

Ψ

ε

[

𝛥𝐶𝑆7

𝐴𝑆𝑃𝐶

𝑉𝑆𝑃𝐶

𝑇𝐾

] = [

] + ∑

[

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

𝛷

]

[

𝛥𝐶𝑆7

𝐴𝑆𝑃𝐶

𝑉𝑆𝑃𝐶

𝑇𝐾

]

+

[

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

𝛹

]

+ [

𝜀

𝜀

𝜀

𝜀

]

ε

4. Results

s s s

_]

_]