The impact of network service performance on customer satisfaction and loyalty: High-speed internet service case in Korea

Expert Systems with Applications Expert Systems with Applications 32 (2007) 822–831 www.elsevier.com/locate/eswa

The impact of network service performance on customer satisfaction and loyalty: High-speed internet service case in Korea Kwang-Jae Kim a

a,*

, In-Jun Jeong b, Jeong-Cheol Park c, Young-Jun Park b, Chan-Gyu Kim d, Tae-Ho Kim e

Division of Mechanical and Industrial Engineering, Pohang University of Science and Technology, Pohang, Kyungbuk 790-784, Republic of Korea b IT Services Research Division, Electronics and Telecommunications Research Institute, Daejeon 305-700, Republic of Korea c Consulting Division, LG CNS, Seoul 100-768, Republic of Korea d Customer Satisfaction Management Group, KMAC, Seoul 150-869, Republic of Korea e Quality Management Office, KT Corp., Seongnam, Kyunggi 463-711, Republic of Korea

Abstract The high-speed internet service has achieved a remarkable increase in penetration in recent years. In order to survive in this competitive market, companies should continue to improve their service performance. The high level of service performance is believed to be an effective way to improve customer satisfaction and loyalty. This paper aims to identify the causal relationship among network performance, customer satisfaction, and customer loyalty in the high-speed internet service context. Using the data collected from 51 current users of a VDSL service in Korea, this paper derives two types of the causal relationship models, namely, cross-sectional model and longitudinal model. The modeling results are discussed from both descriptive and prescriptive perspectives.  2006 Elsevier Ltd. All rights reserved. Keywords: Network performance; Customer satisfaction; Customer loyalty; High-speed internet service

1. Introduction The high-speed internet service, based on the ADSL (asymmetric digital subscriber lines) or VDSL (very highspeed digital subscriber lines) technology, has achieved a remarkable increase in penetration in recent years. For example, in Korea, the number of subscribers in the service increased from 3 million (households) in October 2000 to 4 million in December 2000, and to 7.8 million in December 2001, and to 10 million in November 2002. However, after this point of time, the rate of increase has been significantly lowered. There was an increase of only one million subscribers in 2003 (Korea National Statistical Office, 2004).

*

Corresponding author. Tel.: +82 54 279 2208; fax: +82 54 279 2870. E-mail addresses: [email protected] (K.-J. Kim), [email protected] (I.-J. Jeong), [email protected] (J.-C. Park), [email protected] (Y.-J. Park), [email protected] (C.-G. Kim), [email protected] (T.-H. Kim). 0957-4174/$ - see front matter  2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2006.01.022

The number of subscribers at the end of 2004 is estimated to be 12 million (Ministry of Information & Communication, 2004). This figure indicates that almost eight in every ten households are now subscribed to the service, and that this market has already entered a near-saturation stage. Until recently (say, until 2002), the companies in this market believed their competitiveness comes from a fast acquisition of new customers, a typical characteristic of new telecommunication services. However, as this market gets saturated, customer retention has become more critical than new customer acquisition. Consequently, the service performance level that was considered good enough in the past is no longer adequate. In order to survive in this competitive market, companies should continue to improve their service performance effectively. In the high-speed internet service industry, as in any other service industry, the high level of service performance is a differentiator in competition, and in fact, an effective way to improve customer satisfaction and loyalty.

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

The service performance in high-speed internet service consists of two dimensions – network performance and customer-service performance. According to a recent study (Kim et al., in press), network performance is considered about four times more important than customer-service performance. The objective of this paper is to identify the causal relationship among network performance, customer satisfaction, and customer loyalty in the high-speed internet service context. It is generally accepted that customer satisfaction is a transaction-specific measure, while customer loyalty is formed by a customer’s cumulative experience with the service over time. In this view, two types of the causal relationship models are derived in this paper, namely, cross-sectional model and longitudinal model. The cross-sectional model focuses on the interrelationship at a specific time point, while the longitudinal model explicitly considers the cumulative effects over time. Section 2 constructs the research hypotheses after a brief review of the related literature. Section 3 describes the research methodology with a focus on the data collection issues. Section 4 presents the results of the cross-sectional model. Then, Section 5 extends the cross-sectional model to the longitudinal model. The interpretation and utilization of the results, from both the cross-sectional and longitudinal models, are discussed in Section 6. Finally, concluding remarks are given in Section 7. 2. Theoretical background and research hypotheses 2.1. Service performance, customer satisfaction, and customer loyalty Service performance is defined as the level of a service, and can be categorized into two critical aspects: operational and relational performance (Stank, Goldsby, & Vickery, 1999). Operational performance is related with the physical features of the service, while relational performance is concerned with the service delivery process. In the high-speed internet service case, network performance and customerservice performance correspond to operational performance and relational performance, respectively. Customer satisfaction is defined as a customer’s overall judgment on disconfirmation between the expected and perceived service performances (Anderson & Sullivan, 1993; Ramaswamy, 1996). If the perceived performance meets or exceeds the expectation, the customer is satisfied; otherwise, dissatisfied. Customer satisfaction is a transaction-specific measure. This means that a customer evaluates his/her perception of performance relative to expectation in each service encounter, independently of the other occasions (Bitner, 1990; Bolton & Drew, 1991; Parasuraman, Zeithaml, & Berry, 1988; Ramaswamy, 1996). Customer loyalty is defined as a customer’s attitude to the service (Ramaswamy, 1996; Stank et al., 1999). It is formed by a customer’s cumulative experience with the service over time, not by a specific service encounter. It is

823

widely accepted that customer loyalty has a strong relationship with customer satisfaction, and that it is an antecedent of financial outcome (Anderson, Fornell, & Lehman, 1994; Anderson & Sullivan, 1993; Fornell, 1992; Innis & La Londe, 1994; Reichheld & Sasser, 1990). To date, many studies have been conducted to investigate and understand the relationship among service performance, customer satisfaction, and customer loyalty in various service sectors, including hotel service (Choi & Chu, 2001; Gundersen, Heide, & Olsson, 1996; Kim & Cha, 2002; Lemmink, Ruyter, & Wetzels, 1998), tourism service (Baker & Crompton, 2000; Vogt & Fesenmaier, 1995), medical service (Raju, Lonial, Gupta, & Ziegler, 2000; Raju & Lonial, 2002; Thomas, Wan, Lin, & Ma, 2002), and telecommunications service (Gerpott, Rams, & Schindler, 2001; Kim et al., in press; Kim, Park, & Jeong, 2004). Most of these studies have shown that there exist significant causal relationships in a sequential linkage from service performance to customer satisfaction, and then to customer loyalty. 2.2. Research hypotheses This paper aims to identify the causal relationship among network performance, customer satisfaction, and customer loyalty in the high-speed internet service context. As mentioned earlier, service performance in high-speed internet services consists of network performance and customer-service performance. In this service sector, network performance is considered about four times more important than customer-service performance (Kim et al., in press). That is, customer satisfaction (or dissatisfaction) is determined primarily by network performance. Hence, this paper focuses on network performance. As the measures of network performance, ‘‘download speed’’, ‘‘upload speed’’, ‘‘packet transfer delay’’, and ‘‘packet loss rate’’ are employed in this paper. (For a detailed description on these measures, see Table 1 in Section 3.1). Based on a review of the existing studies mentioned in Section 2.1, the following hypotheses are constructed: (H1) The download speed positively affects satisfaction; (H2) The upload speed positively affects satisfaction; (H3) The packet transfer delay negatively affects satisfaction; (H4) The packet loss rate negatively affects satisfaction; (H5) Customer satisfaction positively affects loyalty.

customer customer customer customer customer

Note that the download speed and the upload speed are larger-the-better-type measures, while the packet transfer delay and the packet loss rate are smaller-the-better-type measures. The structural equation model constructed based

824

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

Table 1 Description and measurement unit of the observed variables Construct

Variable

Description

Measurement unit

Network performance

Download speed Upload speed Packet transfer delay Packet loss rate

Network transmission speed in downloading Network transmission speed in uploading Round trip time of a packet in a ping test Percentage of packets lost in a ping test

Megabits per second (MBPS) Megabits per second (MBPS) Milliseconds (MS) Percentage (%)

Customer satisfaction

Customer satisfaction

Overall degree of satisfaction with the high-speed internet service provided in a specific transaction

7-point scalea

Customer loyalty

Recommendation intention Resubscription intention

Intention of recommending the current high-speed internet service to others Intention of resubscribing to the current high-speed internet service Intention of subscribing to new services of the company

7-point scale

New service subscription intention a

7-point scale 7-point scale

1 = strongly disagreeable; 2 = disagreeable; 3 = slightly disagreeable; 4 = neutral; 5 = slightly agreeable; 6 = agreeable; 7 = strongly agreeable.

Network performance Download speed Recommendation intention

e3

Upload speed Customer satisfaction

Customer loyalty

Resubscription intention

e4

e1

e2

New service subscription intention

e5

Packet transfer delay

Packet loss rate

Fig. 1. The hypothesized structural equation model.

on these research hypotheses is presented in Fig. 1. Here, ei (i = 1, . . . , 5) denotes the error associated with the endogenous variables. The covariances among the network performance measures are also included in the model, because they are likely to move together. 3. Research methodology 3.1. Definition and measurement of variables Network performance is directly observed by the four measures mentioned in Section 2.2. Customer satisfaction is directly observed using a 7-point scale, while customer loyalty is not considered directly observable. Instead, customer loyalty is indirectly observed by the following three indicators: ‘‘recommendation intention’’, ‘‘resubscription intention’’, and ‘‘new service subscription intention’’. In summary, this study has one latent variable (i.e., customer loyalty) and eight observed variables (i.e., four network performance measures, customer satisfaction, and three indicators of customer loyalty). The description and measurement unit of the observed variables are given in Table 1.

3.2. Data collection The data for the customer satisfaction and the three indicators of customer loyalty were collected through a survey from a panel of high-speed internet service users. The panel consisted of 51 users of the VDSL service provided by a major telecommunication company in Korea. The survey was conducted using a website designed for benchmarking tests on high-speed internet services. The customer satisfaction and the indicators of customer loyalty were evaluated by the panel members visiting the website. The network performance measures were automatically measured during the visit so that the subjective evaluation on customer satisfaction and customer loyalty and the network performance measurement were synchronized. The survey was carried out sixteen times – twice a month (every two weeks) over an 8-month period, from March 2003 to October 2003. The panel members were asked to participate in all of the 16 surveys. In total, 816 responses were collected. Table 2(a) shows a portion of the collected data along with some summary statistics.

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

825

Table 2 The collected data Panel member

Survey time point

Network performance

Customer satisfaction

Download speed

Upload speed

Packet transfer delay

Packet loss rate

(a) Unstandardized data P1 1 2 .. . 16 Mean Std. dev. ... ... P51 1 2 .. . 16 Mean Std. dev. Gross mean Gross std. dev.

10.72 10.59 .. . 10.83 10.66 0.10 .. . 10.58 10.02 .. . 1.40 9.26 2.81 10.00 3.03

10.81 10.82 .. . 10.82 10.33 1.84 .. . 10.75 9.76 .. . 2.91 8.79 2.71 8.79 2.72

7.80 8.00 .. . 7.50 8.11 0.64 .. . 14.30 13.50 .. . 0.00 10.00 4.20 12.61 17.38

0.00 0.27 .. . 0.00 0.09 0.14 .. . 0.13 0.19 .. . 0.00 0.07 0.10 0.35 4.26

(b) Standardized data P1 1 ..2 . 16 Mean Std. dev. .. .. . . P51 1 2 .. . 16 Mean Std. dev. Gross mean Gross std. dev.

0.6a 0.7 .. . 1.7 0.0 1.0 .. . 0.5 0.3 .. . 2.8 0.0 1.0 0.0 1.0

0.3 ..0.3 . 0.3 0.0 1.0 .. . 0.7 0.4 .. . 2.2 0.0 1.0 0.0 1.0

0.5 ..0.2 . 1.0 0.0 1.0 .. . 1.0 0.8 .. . 2.4 0.0 1.0 0.0 1.0

0.6 ..1.3 . 0.6 0.0 1.0 .. . 0.6 1.2 .. . 0.7 0.0 1.0 0.0 1.0

a

Customer loyalty Recommendation intention

Resubscription intention

New service subscription intention

4.00 4.00 .. . 4.00 3.69 0.95 .. . 7.00 7.00 .. . 6.00 5.94 1.00 4.75 1.54

1.00 3.00 .. . 4.00 3.25 1.00 .. . 7.00 7.00 .. . 6.00 6.04 0.58 4.50 1.56

4.00 3.00 .. . 3.00 3.31 0.95 .. . 7.00 6.50 .. . 6.00 6.01 0.40 4.44 1.62

3.00 3.00 .. . 3.00 3.19 0.98 .. . 6.00 6.50 .. . 6.00 6.01 0.40 4.30 1.60

0.3 ..0.3 . 0.3 0.0 1.0 .. . 1.1 1.1 .. . 0.1 0.0 1.0 0.0 1.0

2.3 ..0.3 . 0.8 0.0 1.0 .. . 1.7 1.7 .. . 0.1 0.0 1.0 0.0 1.0

0.7 ..0.3 . 0.3 0.0 1.0 .. . 2.5 1.2 .. . 0.0 0.0 1.0 0.0 1.0

0.2 ..0.2 . 0.2 0.0 1.0 .. . 0.0 1.2 .. . 0.0 0.0 1.0 0.0 1.0

(10.72–10.66)/0.10 = 0.6.

3.3. Data standardization A preliminary analysis examining the collected data was conducted to check the homogeneity of the panel members. It was found that, for the customer satisfaction and the indicators of customer loyalty, the response pattern (i.e., mean and standard deviation) shows an important difference among the panel members, even for the same network performance. Such a difference was not unexpected because it is well known that the same range (out of a 7-point scale) is not exploited by all survey respondents (Barone & Lombardo, 2004). For a relationship modeling purpose, it is desired to remove the noise effect due to such a difference in response patterns. Hence, a data standardization was performed. The individual data for each panel member were standardized so that their mean and standard deviation became zero and one, respectively. As an example, in Table 2(a), for panel member P1 the download speed in the first survey (i.e., 10.72 MBPS) was standardized to be (10.72–10.66)/

0.10 = 0.6. The standardized data, given in Table 2(b), were used in subsequent analyses. 4. Results The hypothesized structural equation model, shown in Fig. 1, was fitted using the standardized data. The generalized least square (GLS) method was employed to estimate the model, because a multivariate normal distribution of the observed variables is not assumed and some of the observed variables are ordinal-scaled (Jo¨reskog & So¨rbom, 1996, p. 240). AMOS 4.0 software was used for the model estimation (Arbuckle & Wothke, 1999). The resulting model fit indices were as follows: p-value = 0.046, GFI = 0.993, AGFI = 0.981, NFI = 0.958, CFI = 0.982, RMR = 0.028. Compared with the standards in the literature (Kline, 1998), the fit indices are acceptable in general, although the p-value is slightly low. The estimated path coefficients and covariances are presented in Table 3. The download speed and the upload

826

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

Table 3 The estimated path coefficients and covariances of the structural equation model Path/covariance

Estimated value (p-value)

(From)

(To)

(Path) Download speed Upload speed Packet transfer delay Packet loss rate Customer satisfaction Customer loyalty Customer loyalty Customer loyalty

! ! ! ! ! ! ! !

Customer satisfaction Customer satisfaction Customer satisfaction Customer satisfaction Customer loyalty Recommendation intention Resubscription intention New service subscription intention

0.136 0.280 0.050 0.043 0.517 1 1.046 0.914

(0.000) (0.000)

(Covariance) Download speed Download speed Download speed Upload speed Upload speed Packet transfer delay

M M M M M M

Upload speed Packet transfer delay Packet loss rate Packet transfer delay Packet loss rate Packet loss rate

0.546 0.193 0.165 0.282 0.114 0.174

(0.000) (0.000) (0.000) (0.000) (0.001) (0.000)

speed have highly significant and large effects on the customer satisfaction, while the packet transfer delay and the packet loss rate have weakly significant effects. Note that the path coefficients from the packet transfer delay and the packet loss rate to the customer satisfaction are negative because they are smaller-the-better type measures. In high-speed internet service, the speed is a very important attribute because the download/upload of multimedia files and network games has become a major need of the internet users in Korea (Kim et al., in press). Another interesting finding is that the effect of the upload speed on the customer satisfaction is about twice larger than that of the download speed. In the past, downloading was the main use for the internet service. Nowadays, uploading is more prevailing due to new-fashioned applications such as instant messengers (e.g., AOL, 2004; ICQ, 2004; MSN, 2004) and personal websites. For example, in Korea, more than ten million people, almost 25% of the national population, maintain their personal websites. A personal website, called ‘‘Blog’’, necessitates frequent uploading over the internet (Koreans, 2004). As a result, the upload speed has become more influential on customer satisfaction. The customer satisfaction also shows a highly significant (positive) effect on the customer loyalty. This indicates that a strong positive relationship between customer satisfaction and customer loyalty exists in high-speed internet service. The covariances among the network performance measures as well as the effects of the customer loyalty on its three indicators are also highly significant, as expected. In summary, hypotheses H1, H2, and H5 are strongly supported, and H3 and H4 are weakly supported. 5. Longitudinal structural equation model Customer loyalty is a consequence of a customer’s cumulative satisfaction over time as mentioned in Section

(0.001) (0.000) (0.142) (0.197) (0.000)

2.1. This means that customer loyalty at time T is affected not only by customer satisfaction at T but also by customer satisfaction of previous time points. In this section, a new structural equation model considering such longitudinal effects is constructed and analyzed. Hereafter, this model will be referred to as the longitudinal structural equation model, as opposed to the cross-sectional structural equation model presented in Section 4. 5.1. Research hypotheses For the longitudinal structural equation model, the hypotheses, originally presented in Section 2.2, are modified as follows: (H 01 ) The download speed at time point i positively affects customer satisfaction at i (i = T  k, . . . , T); (H 02 ) The upload speed at i positively affects customer satisfaction at i (i = T  k, . . . , T); (H 03 ) The packet transfer at i delay negatively affects customer satisfaction at i (i = T  k, . . . , T); (H 04 ) The packet loss rate at i negatively affects customer satisfaction at i (i = T  k, . . . , T); (H 05 ) Customer satisfaction at i positively affects customer loyalty at T (i = T  k, . . . , T). Note that one time unit is two-weeks long because the survey was conducted every two weeks. The effect of network performance on customer satisfaction is expected to be stationary over time. Therefore, it would be sufficient to include the relationship only at T in our longitudinal model. (The network performance-to-customer satisfaction paths in Fig. 2 show this point.) The constant k represents the width of the time-window for the hypothesized longitudinal effect. To date, the identification of the proper value of k has not been addressed in the literature, particularly in the high-speed internet service

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

827

Network performance Download speed (T) Recommendation intention (T )

e3

Resubscription intention (T )

e4

New service subscription intention ( T )

e5

Upload speed (T ) Customer satisfaction (T )

Customer loyalty (T )

Packet transfer delay (T )

e1 Packet loss rate (T )

e2

Customer satisfaction (T–1) Customer satisfaction (T–2) Customer satisfaction (T–3) Customer satisfaction (T– 4) Customer satisfaction (T–5) Original path/covariance

Customer satisfaction (T– 6)

Added path/covariance

Fig. 2. The hypothesized longitudinal structural equation model.

context. Based on a statistical correlation analysis between customer satisfaction and customer loyalty in the collected data, the value of k was set at 6, a conservative estimate. The longitudinal model constructed based on the modified research hypotheses is presented in Fig. 2. 5.2. Data transformation The data for the longitudinal model fitting were obtained by transforming the original (standardized) data based on the value of k = 6. The data transformation scheme is shown in Fig. 3. The customer loyalty data at T, the customer satisfaction data from (T  6) to T, and the network performance data at T should be aligned in a single record. As a result, the effective number of records decreased from 816 (=16 time points · 51 panel members) to 510 (=10 time points · 51 panel members). 5.3. Model fitting results The longitudinal model was fitted using the transformed data. The resulting model fit indices were unacceptable. In order to improve the model fit, some paths were added. More specifically, the covariances between two successive customer satisfactions for [T  6, T  1], between customer satisfaction at (T  1) and e1, and between e3 and e5 were

added. The added paths are plausible intuitively as well as theoretically. The revised model turned out to be acceptable: p-value = 0.253, GFI = 0.980, AGFI = 0.967, NFI = 0.845, CFI = 0.981, RMR = 0.046. The estimated path coefficients and covariances are presented in Table 4. The results for the network performanceto-customer satisfaction paths are almost the same as those in the cross-sectional model in Section 4. The only noticeable difference is that the effect of the packet transfer delay on customer satisfaction is now larger and more significant, while the effect of the packet loss rate is not significant enough. The customer satisfactions at time points T through (T  4) have highly or moderately significant (positive) effects on the customer loyalty at T. This means that customer satisfaction levels at the four recent past time points (i.e., two-month period) as well as at the current time point collectively determine the current level of customer loyalty. The effect at T (=0.565), which is almost the same as its counterpart (=0.518) in the cross-sectional model, is overwhelmingly larger than the others. The magnitude of the effect gradually drops in general as the time point moves farther from T. If the longitudinal effect of customer satisfaction on customer loyalty had not been considered, the effects at or before (T  1) would have been pooled as part of the modeling error. The covariances among the network

828

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

Original data format NP 1*

CS 1** CL 1***

NP 7

CS 1

CS 2

CS 3

CS 4

CS 5

CS 6

CS 7

CL 7

NP 8

CS 2

CS 3

CS 4

CS 5

CS 6

CS 7

CS 8

CL 8

CL 3

NP 9

CS 3

CS 4

CS 5

CS 6

CS 7

CS 8

CS 9

CL 9

CL 4

NP 10

CS 4

CS 5

CS 6

CS 7

CS 8

CS 9

CS 10

CL 10

CS 5

CL 5

NP 11

CS 5

CS 6

CS 7

CS 8

CS 9

CS 10

CS 11

CL 11

NP 6

CS 6

CL 6

NP 12

CS 6

CS 7

CS 8

CS 9

CS 10

CS 11

CS 12

CL 12

NP 7

CS 7

CL 7

NP 13

CS 7

CS 8

CS 9

CS 10

CS 11

CS 12

CS 13

CL 13

NP 8

CS 8

CL 8

NP 14

CS 8

CS 9

CS 10

CS 11

CS 12

CS 13

CS 14

CL 14

NP 9

CS 9

CL 9

NP 15

CS 9

CS 10

CS 11

CS 12

CS 13

CS 14

CS 15

CL 15

NP 10

CS 10

CL 10

NP 16

CS 10

CS 11

CS 12

CS 13

CS 14

CS 15

CS 16

CL 16

NP 11

CS 11

CL 11

NP 12

CS 12

CL 12

NP 13

CS 13

CL 13

NP 14

CS 14

CL 14

NP 15

CS 15

CL 15

NP 2

CS 2

CL 2

NP 3

CS 3

NP 4

CS 4

NP 5

NP 16

CS 16

…

16 time points × 51 panel members = 816 records

Transformed data format

* Network **

performance data in the first survey

Customer satisfaction data in the first survey

***

CL 16

10 time points × 51 panel members =510 records

Customer loyalty data in the first survey

Fig. 3. The data transformation scheme.

Table 4 The estimated path coefficients and covariances of the longitudinal structural equation model Path/covariance

Estimated value (p-value)

(From)

(To)

(Path) Download speed (T) Upload speed (T) Packet transfer delay (T) Packet loss rate (T) Customer satisfaction (T) Customer satisfaction (T  1) Customer satisfaction (T  2) Customer satisfaction (T  3) Customer satisfaction (T  4) Customer satisfaction (T  5) Customer satisfaction (T  6) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T)

! ! ! ! ! ! ! ! ! ! ! ! ! !

Customer satisfaction (T) Customer satisfaction (T) Customer satisfaction (T) Customer satisfaction (T) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T) Customer loyalty (T) Recommendation intention (T) Resubscription intention (T) New service subscription intention (T)

0.167 0.299 0.146 0.043 0.565 0.085 0.116 0.051 0.075 0.024 0.039 1 0.976 0.972

(0.000) (0.000)

(Covariance) Download speed (T) Download speed (T) Download speed (T) Upload speed (T) Upload speed (T) Packet transfer delay (T) e1 Customer satisfaction (T  1) Customer satisfaction (T  2) Customer satisfaction (T  3) Customer satisfaction (T  4) Customer satisfaction (T  5) e3

M M M M M M M M M M M M M

Upload speed (T) Packet transfer delay (T) Packet loss rate (T) Packet transfer delay (T) Packet loss rate (T) Packet loss rate (T) Customer satisfaction (T  1) Customer satisfaction (T  2) Customer satisfaction (T  3) Customer satisfaction (T  4) Customer satisfaction (T  5) Customer satisfaction (T  6) e5

0.456 0.122 0.099 0.235 0.096 0.166 0.115 0.176 0.143 0.134 0.101 0.112 0.080

(0.000) (0.002) (0.017) (0.000) (0.022) (0.000) (0.001)a (0.000)a (0.000)a (0.000)a (0.005)a (0.003)a (0.001)a

a

Added covariances.

(0.000) (0.000) (0.001) (0.259) (0.000) (0.006) (0.000) (0.102) (0.019) (0.441) (0.221)

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

performance measures and the effects of the customer loyalty on its three indicators are also highly significant as in the cross-sectional model. In summary, hypotheses H 01 , H 02 , and H 03 are strongly supported, while H 04 fails to get supported. Hypothesis H 05 is strongly supported for i = T, (T  1), (T  2), and (T  4); moderately supported for i = (T  3); and unsupported for i = (T  5) and (T  6). 6. Interpretation and utilization of the path coefficients The obtained coefficients can be utilized for both descriptive and prescriptive analysis purposes. The descriptive analysis is to understand the impact of a change in network performance on the improvement of customer satisfaction and loyalty. In contrast, the prescriptive analysis is to form alternative implementation strategies to achieve an improvement target of customer satisfaction and loyalty. The fitted cross-sectional models are: Customer loyalty ¼ 0:517  Customer satisfaction, Customer satisfaction

ð1Þ

¼ ð0:136  Download speedÞ þ ð0:280  Upload speedÞ  ð0:050  Packet transfer delayÞ  ð0:043  Packet loss rateÞ

ð2Þ

and the fitted longitudinal models are: Customer loyalty ðT Þ ¼ ð0:565  Customer satisfaction ðT ÞÞ þ ð0:085  Customer satisfaction ðT  1ÞÞ þ ð0:116  Customer satisfaction ðT  2ÞÞ þ ð0:051  Customer satisfaction ðT  3ÞÞ þ ð0:075  Customer satisfaction ðT  4ÞÞ;

ð3Þ

Customer satisfaction ðT Þ ¼ ð0:167  Download speed ðT ÞÞ

pose the download speed is increased at each time point in [T  4, T] by one unit. This would lead to an increase in the customer satisfaction at each time point in [T  4, T] by 0.167, and subsequently an increase in the customer loyalty level at T by 0.149 (=0.167 · 0.565 + 0.167 · 0.085 + 0.167 · 0.116 + 0.167 · 0.051 + 0.167 · 0.075). The impact of the other network performance measures can be interpreted in a similar manner. The descriptive analysis results can be validated by checking how accurately the models predict the customer satisfaction and customer loyalty levels, given the network performance measures. The fitted cross-sectional and longitudinal models were compared in terms of their prediction accuracy. A popular measure of prediction accuracy is the prediction error, which is the absolute difference between the predicted value and the actual value. The prediction error is comparable to the training error in model assessment and selection (Hastie, Tibshirani, & Friedman, 2001). A statistical test comparing the average prediction errors of the two models was conducted. In predicting the customer satisfaction level, the two models did not show a significant difference (pvalue = 0.504). This was expected because customer satisfaction is considered transaction-specific in both models. However, in predicting the customer loyalty level, the longitudinal model outperformed the cross-sectional model. More specifically, the difference in prediction accuracy is highly significant with respect to recommendation intention (p-value = 0.007), and weakly significant with respect to resubscription intention (p-value = 0.158) and new service subscription intention (p-value = 0.155). This indicates that the longitudinal model has a better capability in describing the causal relationship among network performance, customer satisfaction, and customer loyalty in the high-speed internet service case of this paper. 6.2. Prescriptive analysis

þ ð0:299  Upload speed ðT ÞÞ  ð0:146  Packet transfer delay ðT ÞÞ.

829

ð4Þ

Note that the models in (1)–(4) include only the significant path coefficients (i.e., p-value < 0.2). The coefficients should be interpreted in terms of the standard deviation unit since the standardized data were used for the model fitting. The descriptive and prescriptive analysis results for these fitted models are presented below. 6.1. Descriptive analysis In the cross-sectional model, an increase in the download speed at time T by one unit leads to an increase in the customer satisfaction level at T by 0.136, and subsequently an increase in the customer loyalty level at T by 0.070 ( = 0.136 · 0.517). In the longitudinal model, an increase in the download speed at T by one unit leads to an increase in the customer satisfaction level at T by 0.167. Now, sup-

Suppose an increase in the customer loyalty level at T by one unit is desired. In the cross-sectional model, such a target has to be achieved by improving the customer satisfaction at T, or equivalently, by improving the network performance measures at T. As shown in (1) and (2), the increase of the download speed, upload speed, packet transfer delay, and packet loss rate at T by one unit each improves the customer satisfaction level at T by 0.136, 0.280, 0.050, and 0.043, respectively, and then the customer loyalty level at T by 0.070 (=0.136 · 0.517), 0.145 (=0.280 · 0.517), 0.026 (=0.050 · 0.517), and 0.022 (=0.043 · 0.517). Based on these marginal contribution rates, several alternative strategies can be formed: (i) increasing the download speed at T by 14.286; or (ii) increasing the upload speed at T by 6.897; or (iii) increasing the download speed at T by 7.143 and the upload speed at T by 3.448; or (iv) increasing the download speed at T by 5.714 and the upload speed at T by 2.759, but decreasing the packet transfer delay at T by 7.692; and the like.

830

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831

In the longitudinal model, the target can be achieved by improving the network performance measures not just at T, but in the range of [T  4, T]. As shown in (3) and (4), the increase of the download speed, upload speed, and packet transfer delay at each time point in [T  4, T] by one unit each improves the customer satisfaction level at the same time point by 0.167, 0.299, and 0.146, respectively, and then the customer loyalty level at T by 0.149 (=0.167 · 0.565 + 0.167 · 0.085 + 0.167 · 0.116 + 0.167 · 0.051 + 0.167 · 0.075), 0.267 (=0.299 · 0.565 + 0.299 · 0.085 + 0.299 · 0.116 + 0.299 · 0.051 + 0.299 · 0.075), and 0.130 (=0.146 · 0.565  0.146 · 0.085  0.146 · 0.116  0.146 · 0.051  0.146 · 0.075). Based on these marginal contribution rates, as in the cross-sectional model case, several alternative strategies can be formed: (i) increasing the download speed by 6.711 each in [T  4,T]; or (ii) increasing the upload speed by 3.745 each in [T  4, T]; or (iii) increasing the download speed by 3.356 each in [T  4,T] and the upload speed by 1.872 each in [T  4, T]; or (iv) increasing the download speed by 2.685 each in [T  4, T] and the upload speed by 1.498 each in [T  4, T], but decreasing the packet transfer delay by 1.538 each in [T  4,T]; and the like. In both the cross-sectional and the longitudinal cases, one can devise infinitely many alternatives, at least mathematically, to achieve the given target. Among such alternatives, the analyst needs to select the ‘‘optimal’’ strategy considering other factors, such as technical and financial feasibility. As described above, for a given target of improvement, the cross-sectional model requires a drastic change in a single time period, while the longitudinal model has the flexibility of distributing the required efforts over several time periods. In this view, the longitudinal model is more likely to form an implementable and feasible strategy. The longitudinal model can serve as a basis for developing more practical and effective improvement plans. 7. Concluding remarks This paper has investigated the causal relationship among network performance, customer satisfaction, and customer loyalty in the high-speed internet service in Korea. Two types of the relationship models have been derived, namely, cross-sectional model and longitudinal model. The major findings from the models are as follows. First, the speed-related network performance measures (i.e., download speed and upload speed) have highly significant and large effects on customer satisfaction. Interestingly, the upload speed has the largest effect among the network performance measures. Second, the customer satisfaction levels for the recent two-month period have significant, cumulative effects on the current level of customer loyalty. The derived cross-sectional and longitudinal models have been interpreted from the descriptive and prescriptive perspectives. The descriptive analysis allows one to understand the impact of network performance on customer satisfaction and loyalty. In contrast, the prescriptive analysis

provides insights as to forming effective improvement strategies. The longitudinal model has been shown to have advantages over the cross-sectional model in both descriptive and prescriptive perspectives. References Anderson, E. W., Fornell, C., & Lehman, D. R. (1994). Customer satisfaction, market share, and profitability: findings from Sweden. Journal of Marketing, 58(3), 53–66. Anderson, E. W., & Sullivan, M. (1993). The antecedents and consequences of customer-satisfaction for firms. Marketing Science, 12(2), 125–143. AOL Instant Messenger. (2004). Available from http://www.aim.com. Arbuckle, J. L., & Wothke, W. (1999). AMOS 4.0 user’s guide. Chicago: SmallWaters Corp. Baker, D. A., & Crompton, J. L. (2000). Quality, satisfaction and behavioral intentions. Annals of Tourism Research, 27(3), 785–804. Barone, S., & Lombardo, A. (2004). Service quality design through a smart use of conjoint analysis. The Asian Journal on Quality, 5(1), 34–42. Bitner, M. J. (1990). Evaluating service encounters: the effects of physical surroundings and employee responses. Journal of Marketing, 54(April), 69–82. Bolton, R. N., & Drew, J. H. (1991). A longitudinal analysis of the impact of service changes on customer attitudes. Journal of Marketing, 55(January), 1–9. Choi, T. Y., & Chu, R. (2001). Determinants of hotel guests’ satisfaction and repeat patronage in the Hong Kong hotel industry. Hospitality management, 20, 277–297. Fornell, C. (1992). A national customer satisfaction barometer: the Swedish experience. Journal of Marketing, 55(1), 1–21. Gerpott, T. J., Rams, W., & Schindler, A. (2001). Customer retention, loyalty and satisfaction in the German mobile cellular telecommunication market. Telecommunication Policy, 25, 249–269. Gundersen, M. G., Heide, M., & Olsson, U. H. (1996). Hotel guest satisfaction among business travelers. Cornell Hotel and Restaurant Administration Quarterly(April), 72–81. Hastie, T., Tibshirani, R., & Friedman, J. (2001). The elements of statistical learning: Data mining, inference, and prediction (pp. 193– 224). New York: Springer-Verlag. ICQ. (2004). Available from http://www.icq.com. Innis, D. E., & La Londe, B. J. (1994). Customer service: the key to customer satisfaction, customer loyalty, and market share. Journal of Business Logistics, 15(1), 1–27. Jo¨reskog, K. G., & So¨rbom, D. (1996). LISREL 8: user’s reference guide. Chicago: Scientific Software International. Kim, W. G., & Cha, Y. (2002). Antecedents and consequence of relationship quality in hotel industry. Hospitality management, 21, 321–338. Kim, K., Cho, H., Jeong, I., Park, J., Park, Y., Kim, C. & Kim, T. (in press). Service quality analysis and improvement: Development of a systematic framework. International Journal of Industrial Engineering. Kim, M., Park, M., & Jeong, D. (2004). The effects of customer satisfaction and switching barrier on customer loyalty in Korean mobile telecommunication services. Telecommunications Policy, 28, 145–159. Kline, R. B. (1998). Principles and practice of structural equation modeling. New York: Guilford Press. Korea National Statistical Office. (2004). Available from http:// www.nso.go.kr. Koreans nuts over Cyworld. (2004). JoonAng Ilbo, September 30. Available from http://joongangdaily.joins.com/200409/30/200409302150515239900090609061.html. Lemmink, J., Ruyter, K. D., & Wetzels, M. (1998). The role of value in the delivery process of hospitality services. Journal of Economic Psychology, 19, 159–177.

K.-J. Kim et al. / Expert Systems with Applications 32 (2007) 822–831 Ministry of Information and Communication, Republic of Korea. (2004). Available from http://www.mic.go.kr. MSN Messenger. (2004). Available from http://www.msn.com. Parasuraman, A., Zeithaml, V., & Berry, L. (1988). SERVQUAL: a multiple-item scale for measuring customer perceptions of service quality. Journal of Retailing, 64(Spring), 12–40. Raju, P. S., & Lonial, S. C. (2002). The impact of service quality and marketing on financial, performance in the hospital industry: an empirical examination. Journal of Retailing and Consumer Services, 9, 335–348. Raju, P. S., Lonial, S. C., Gupta, Y. P., & Ziegler, C. (2000). The relationship between market orientation and performance in the hospital industry: a structural equations modeling approach. Health Care Management Science, 3, 237–247.

831

Ramaswamy, R. (1996). Design and management of service processes: Keeping customers for life (pp. 330–360). Massachusetts: AddisonWesley. Reichheld, F. F., & Sasser, W. E. (1990). Zero defections: quality comes to services. Harvard Business Review, 68(5), 105–111. Stank, T. P., Goldsby, T. J., & Vickery, S. K. (1999). Effect of service supplier performance on satisfaction and loyalty of store managers in the fast food industry. Journal of Operations Management, 17, 429–447. Thomas, T. H., Wan, B., Lin, Y. J., & Ma, A. (2002). Integration mechanisms and hospital efficiency in integrated health care delivery systems. Journal of Medical Systems, 26(2), 127–143. Vogt, C. A., & Fesenmaier, D. R. (1995). Tourists and retailers’ perceptions of services. Annals of Tourism Research, 22(4), 763–780.