• 沒有找到結果。

Revised planning matrix of quality function deployment

N/A
N/A
Protected

Academic year: 2021

Share "Revised planning matrix of quality function deployment"

Copied!
24
0
0

加載中.... (立即查看全文)

全文

(1)

Revised Planning Matrix of Quality Function Deployment

WEI-JAW DENG and YING-FENG KUO

Quality function deployment (QFD) has been adopted to improve product quality and development in many fields. Numerous studies have demonstrated that attribute importance and attribute performance have a causal relationship and the customer self-stated raw importance is not the actual importance of customer attribute. These findings generate questions regarding the applicability of the conventional planning matrix (PM) of QFD. This study presents a revised PM that integrates a back-propagation neural network (BPNN) and three-factor theory to assist practitioners in determining actual importance of customer attributes. An illustrative case demonstrates the effectiveness of the revised PM and identifies shortcomings generated when applying the conventional PM.

Keywords: QFD, VOC, planning matrix, BPNN, three-factor theory, customer attribute importance

_____________________

Wei-Jaw Deng (corresponding author) is an associate professor at Graduate School of Business Administration, Chung Hua University, 707, Sec. 2, Wu Fu Road, Hsinchu, Taiwan. E-mail: simond@chu.edu.tw. Ying-Feng Kuo is a professor at Department of Information Management, National University of Kaohsiung, 700, Kaohsiung University Road, Kaohsiung, Taiwan.

(2)

INTRODUCTION

An intensely competitive and consumer-oriented market mandates that companies

transform their operations into ‘customer-focused’ organizations. Global leadership

belongs to those organizations that satisfy or exceed customer requirements [Kathawala and Jaideep, 1994]. In the context of product or service improvements, quality function deployment (QFD) had been applied as a highly effective means of acquiring strong influential product or service technical characteristics that achieve that satisfy customer requirements and improve product quality or customer satisfaction [Trappey and Trappey, 1996; Miyoung and Haemoon, 1998; Pun et al., 2000; Gonzalez et al., 2003; Gonzalez et al., 2004]. Notably, QFD, a concept Akao [1990] first introduced in Japan

in 1966, was first utilized atMitsubishi’sKobeshipyard in 1972 and wasintroduced in

the United States in 1983. QFD has subsequently been adopted as a product development and quality improvement tool worldwide [Tan and Shen, 2000]. As a cross-functional planning tool, QFD ensures that the voice of the customer (VOC) is systematically deployed throughout product planning and design stages [Shin et al., 2002].

House of quality (HOQ), which is occasionally called the A-1 matrix, is the first and primary building block of QFD [Martins and Aspinwall, 2001]. Using HOQ chart can assist QFD practitioners in identifying principal customer requirements (what) and determine which product or service technical characteristics influence customer requirements (how). The planning matrix (PM), one part of HOQ, is a tool that helps QFD practitioners re-prioritize customer attributes. The PM comprises two VOC types: qualitative VOC (what customers want) and quantitative VOC (priority of customer wants). Customers wants can be identified by listening to the VOC; however, how customers prioritize their wants is cannot be directly identified based on the raw importance stated by customers [Tan and Shen, 2000]. Additionally, numerous studies indicated that attribute performance and importance share a causal relationship [Sampson and Showalter, 1999; Oh, 2001; Ryan and Huyton, 2002; Matzler et al.,

2004a]. Restated, when attribute performance changes, relative attribute importance

does as well [Matzler and Sauerwein, 2002]. Consequently, self-stated raw importance of customer attributes does not sufficiently reflect the actual importance of customer attributes. This finding engenders questions regarding the applicability of the conventional PM of QFD and indicates the need to modify PM.

In previous studies, researchers frequently replaced the customer self-stated raw attribute importance with statistically inferred attribute importance in their studies

[O’Leary and Adams 1982; Dolinsky and Caputo, 1991; Matzler and Sauerwein, 2002;

Simpson et al., 2002; Matzler et al., 2003; Matzler et al., 2004a; Matzler et al., 2004b].

(3)

conventional statistical methods for obtaining the implicitly derived importance of attributes [Garver, 2003]. However, these statistical methods assume that (1) data are relatively normal, (2) the relationships between independent and dependent variables are linear and (3) multicollinearity between independent variables is relatively low [Taylor, 1997]. Such assumptions are almost always violated in customer satisfaction research [Garver, 2002]. Therefore, the implicitly derived importance of attributes via statistical methods can be biased and misleading.

Artificial neural networks (ANNs) are analytical techniques modeled on learning processes of the human cognitive system and brain neurological functions. Numerous researchers applied ANNs for their prediction or classification researches in the science and social science field. Garver [2002] argued that ANNs overcome the limitations of conventional statistical applications. Bishop [1994] indicated that in situations in which non-normal data, multicollinearity, and non-linear relationships exist, neural networks outperform multiple regression models. Burger et al. [2001] indicated that ANNs generally have high degrees of freedom, and, thus, can model the non-linearity of a process significantly better than regression techniques. Furthermore, West et al. [1997] showed that ANNs have significant advantages over conventional statistical methods for assessing customer attributes, preference, judgment and choice behavior. Therefore, PM practitioners can employ ANNs to measure actual importance of customer attributes that are then used to acquire actual quantitative VOC. Only when actual quantitative VOC has been acquired can product/service technical characteristics that influence customer satisfaction be identified.

This study presents a revised PM approach that comprises a back-propagation neural network (BPNN), three-factor theory and natural logarithmic transformation. The proposed PM approach considers the causal relationship between attribute performance and attribute importance and overcomes the limitations associated with conventional statistical methods for assessing implicitly derived actual importance of customer attributes. Therefore, QFD practitioners can acquire the final importance of customer attributes and then apply these final importance to subsequent analysis of HOQ. Finally, QFD practitioners can acquire the principal product/service technical characteristics that affect customer satisfaction most and then apply these technical characteristics to enhance product/service design, product/service quality and competitive advantage.

The rest of this paper is organized as follows. Section 2 reviews relevant literature, particularly that concerning QFD, three-factor theory of customer satisfaction and BPNN. To elucidate the final importance of customer attributes, Section 3 introduces a revised PM approach. Next, Section 4 demonstrates how to implement the proposed revised PM approach and notes any shortcomings produced by applying the conventional PM approach. Conclusions are finally drawn in Section 5.

(4)

LITERATURE REVIEW Quality function deployment

Sullivan [1986] defined QFD as an effective means of translating customer requirements into appropriate technical requirements for each stage of product deployment and production. Martins and Aspinwall [2001] defined QFD as a pro-active technique designed to identify, prioritize and deploy the VOC into every level of an organization. Ozgener [2003] defined QFD as an effective means of identifying critical customer attributes and create a specific relationship between customer attributes and product design parameters of a product or service. The overall objective of QFD is to decrease the duration of product development and reduce start-up problems while simultaneously improving product or service quality, increasing customer satisfaction and marketing a product or service at a low cost. Furthermore, the primary contribution of applying QFD is the acquirement of the VOC and the identification of principal product/service technical characteristics that affect customer satisfaction most.

The HOQ is the first and primary building block of QFD. Notably, HOQ has a set of standard components that comprise customer attributes, customer importance ratings of attributes, engineering characteristics, relationship matrix between customer attributes and product/service technical characteristics, roof matrix among product/service technical characteristics, competitive analysis and final relative importance ratings of product/service technical characteristics. The HOQ foundation is the belief that products should be reflect customer desires and tastes—marketing people, design engineers and manufacturing personnel must work together as a team from time a product is first conceived [Hauser and Clausing, 1988]. The planning matrix is a part of HOQ and a tool that assists QFD practitioners in re-prioritizing customer attributes. A typical planning matrix comprises raw importance, competitive analysis, a target, improvement ratio, sales point and final importance (see Fig. 1).

(5)

Numerous studies have revised the conventional QFD in an effort to improve the effectiveness of QFD. For instance, Vanegas and Labib [2001] developed a fuzzy QFD (FQFD) model for identifying optimum priority of engineering characteristics. They utilized a Fuzzy Analytical Hierarchy Process (FAHP) to address customer attribute importance and integrated fuzzy set theory into relationship matrix analysis. Ramasamy and Selladurai [2004] proposed a fuzzy logic QFD (FL-QFD) that prioritizes engineering characteristics. This approach differs from the FQFD approach in that it

utilizes a knowledge-based fuzzy rule. Park et al. [2005] employed Taguchi’s robust

design method in obtaining the HOQ top matrix weights and proposed a signal-to-noise ratio QFD (S/N-QFD) approach. However, these studies did not address the possible problem in explicitly self-sated importance of customer attributes.

Matzler and Hinterhuber [1998] first integrated Kano’s model of customer

satisfaction into QFD. They proposed a quality improvement index (QI) that prioritizes customer attributes.

QI= (raw relative importance)

×(evaluation of own product-evaluation of competitor’s product) (1)

However, this method retained the problem as in the studies that just mentioned in previous paragraph. Later, Tan and Shen [2000] generated an approximate transformation function for applying the PM in QFD to adjust the importance ratio for each customer attribute based on the Kano model analysis. The approximate transformation function is as follows:

 

k

adj IR

IR0 1 (2)

where IRadj is the adjusted improvement ratio, IR is the original improvement ratio,0

and k is the Kano parameter for different attribute categories. However k is subjectively determined by QFD practitioners. For instance, QFD practitioners can arrange a possible set of k values as 1/2, 1 and 2 for must-be, one-dimensional and attractive attributes, respectively. As discussion had addressed in their own study, choosing an

appropriate numerical value for k is critical and dependent on QFD practitioners’

experience. Therefore, this proposed approximate transformation function still cannot effectively help QFD practitioners acquire actual quantitative VOC.

Three-factor theory of customer satisfaction

Kano et al. [1984] developed a model that distinguishes between different quality

attribute types. Kano’s model divides product or service quality attributes into five

(6)

of which influences customer satisfaction differently. Other studies of customer satisfaction, however, suggest that quality attributes can be understood using three categories: basic factors, performance factors, and excitement factors [Brandt, 1988; Johnston, 1995; Matzler et al., 1996; Oliver, 1997; Matzler and Hinterhuber, 1998; Anderson and Mittal, 2000; Matzler and Sauerwein, 2002]. The basic factors are similar to must-be quality elements. The performance factors are similar to one-dimension quality elements. The excitement factors are similar to attractive quality elements.

Matzler et al. [2004a] elucidate these three factors. Basic factors (dissatisfiers) are

minimum requirements that produce consumer dissatisfaction when not fulfilled, but do not result in customer satisfaction when fulfilled or exceeded; that is, negative performance for these attributes has a greater impact on overall satisfaction than positive performance. Excitement factors (satisfiers) are attributes that increase customer satisfaction when delivered, but cause no dissatisfaction when not delivered. That is, positive performance for these attributes has a stronger influence on overall consumer satisfaction than negative performance. Performance factors produce satisfaction when performance is high and dissatisfaction when performance is low. The relationship between customer attribute performance and overall customer satisfaction is nonlinear and asymmetrical for basic and excitement attributes. For performance attributes, the relationship between customer attribute performance and overall

satisfaction is linear and symmetrical [Ting and Chen, 2002; Matzler et al., 2004a].

Consequently, customer attributes have two key characteristics in three-factor theory. (1) Importance of a basic or excitement attribute is based on its performance. Basic attributes are crucial when performance is low and are unimportant when performance is high. Excitement factors are critical when performance is high and are irrelevant when performance is low [Sampson and Showalter, 1999; Ting and Chen, 2002;

Matzler et al., 2004a]. (2) The relationship between attribute performance and overall

customer satisfaction is asymmetrical. Consequently, the applicability of the conventional PM approach that utilized explicit customer self-stated raw importance of attributes requires modification.

Back-propagation neural network

Artificial neural networks are information-processing systems that have specific performance characteristics common to biological neural networks. A standard ANN comprises numerous simple processing elements called neurons or nodes. Each neuron is connected to other neurons through directed communication links, each with an associated weight. These weights represent information utilized by the net to solve a given problem. Typically, ANNs are modeled into one input layer, one or several hidden layers, and one output layer [Fausett, 1994]. Each layer has neurons that are

(7)

connected to other neurons in adjacent layer(s). Each neuron in a neural network is a processing unit containing a summation function and an activation function. A weight (W) returns a mathematical value for the relative strength of connections for transferring data from one layer to another; a summation function (s) computes the weighted sum of all input elements entering a neuron. An activation function, f (s), is utilized for transforming the output such that it falls within an acceptable range (usually 0–1). This transformation is conducted before the output reaches the next level. The input layer can be considered the model stimuli and the output layer is the input stimuli outcome. The hidden layer determines the mapping relationships between input and output layers, whereas the relationships between neurons are stored as weights of connecting links. The input layer is analogous to independent variables, and the output layer is analogous to dependent variables.

Artificial neural networks can be classified into several categories based on supervised and unsupervised learning methods and feed-forward and feedback recall architectures. A BPNN is a neural network that uses a supervised learning method and feed-forward architecture. A BPNN is one of the most frequently utilized neural network techniques for classification and prediction [Wu et al., 2006] and is considered advanced multiple regression analysis that can accommodate complex and non-linear data relationships [Jost, 1993]. The learning algorithm in a BPNN differs from traditional feed-forward neural networks: first, a BPNN uses an activation function for

the hidden unit sjand not the input value; and, second, the gradient of the activation

function is contained [Law, 2000]. The output of a BPNN is compared with the target output and an error is calculated for each training iteration. This error is then back propagated to the neural network and utilized to adjust the weights, thereby minimizing

the mean squared error between the network’s prediction output and the target output.

Consequently, the BPNN model yields predictive output that is similar to the target output. The details for the back-propagation learning algorithm can be found in Medsker and Liebowitz [1994], Russell and Norvig [1995], or Law [2000].

Artificial neural networks are applied in numerous fields, such as pattern recognition, financial management, medical diagnosis, and forecasting for tourism demand, sales, innovation performance, and stock market returns [Tu, 1996; Vach et al., 1996; Schumacher et al., 1996; Burger et al., 2001; Enke and Thawornwong, 2005; Chang and Wang, 2006; Wang and Chien, 2006]. Burger et al. [2001] indicated that ANNs generally have high degrees of freedom; thus, they can accommodate the non-linearity of a process under study significantly better than regression approaches. Wang and Chien [2006] demonstrated that the BPNN approach outperforms statistical regression methods when forecasting performance. West et al. [1997] addressed that

(8)

ANNs outperform statistical procedures in explaining variance and out-of-sample predictive accuracy. Garver [2002] noted that ANNs overcome the limitations of conventional statistical models that are typically visible in customer satisfaction research, and that ANNs are well-suited to evaluating the relative importance of customer satisfaction attributes. Tsaur et al. [2002] measured the importance scores for service aspects at nine international hotels and compared these importance scores with those derived using a logistic regression model. They concluded that the capability of ANNs in modeling non-linear interactions improve model prediction performance and variance interpretation.

METHODOLOGY OF THE REVISED PLANNING MATRIX APPROACH The method for designing a BPNN model

A BPNN model is one of the most widely used ANN models. Therefore, general commercial ANN software packages (e.g., NeuroShell 2; NeuroSolutions 5; NeuFrame; etc.) can be utilized by practitioners when building a BPNN model. The BPNN architecture comprises one input layer, hidden layers and one output layer. The BPNN parameters include a number of hidden layers, a number of hidden neurons, an activation function, learning rate, momentum, etc. All of these parameters have significant impact on neural network performance.

In most cases, one hidden layer is sufficient for computing arbitrary decision boundaries for outputs [Khaw et al., 1995]. Kaastra and Boyd [1996] indicated that a neural network with one hidden layer with a sufficient number of neurons can approximate any continuous function. Therefore, the number of hidden layers is usually set at 1. For the number of neurons, BPNN practitioners can refer the selection principle

of neuron number introduced in Haykin’s book [1999] to determine the number of

neurons in the hidden layer. After performing some trials or applying the Taguchi method that revealed by Khaw et al. [1995], the final number of neurons in the hidden layer can be determined.

The activation function is mathematical formula that determines the output of a processing neuron. In a standard BPNN, input layer neurons typically use linear

activation functions, whereas all other neurons use a sigmoid function (



x

e x f   1 1 ;

output value in the interval [0, 1]) or hyperbolic tangent (



x x

x x e e e e x f     ; output value

in the interval [-1, 1]) [Kaastra and Boyd, 1996]. For setting the learning rate and momentum, most ANN software packages provide default values for both parameters that typically work well. The common practice is to start training with a high learning

(9)

rate such as 0.7 and decrease the rate as training proceeds. Therefore, BPNN practitioners can use the default values provided by the ANN software packages.

Hush and Horne [1993] demonstrated there are three conditions for terminating network learning: (1) when the root mean square error (RMSE) between the expected value and network output value has reduced to a preset value; (2) when the preset number of learning iterations has been reached; or, (3) when the RMSE of a validation sample has begun to increase. The first two conditions are based on the preset values. However, these two conditions can cause an over-fitting problem in the BPNN model. Smith [1996] suggested utilizing a cross-validation procedure that prevents over-fitting in the BPNN model. Practitioners of BPNN can select a cross-validation procedure as the rule to terminate learning in most ANN software packages.

Acquiring the implicitly derived importance of attributes

Many researchers and practitioners use statistical techniques (e.g., correlation analysis, multiple regression, and structural equation models) to acquire the implicitly derived

importance of attributes. For example, O’Leary and Adams[1982] utilized Pearson’s

correlation analysis to obtain the implicitly derived importance of attributes; Dolinsky

and Caputo [1991], Matzler and Sauerwein [2002], and Matzler et al. [2004a] implicitly

derived attribute importance utilizing a multiple regression analysis with overall customer satisfaction as a dependent variable and attributes of performance as

independent variables; Matzler et al. [2003] and Matzler et al. [2004b] applied partial

correlation analysis to derive the implicitly derived importance of attributes. Simpson et

al. [2002] employed structural equation models to yield cause-effect relationships

among variables. The total effect (direct effect plus indirect effect) coefficient is the importance of antecedent variables on the dependent variable. Although statistically inferred attribute importance is superior to self-stated attribute importance, these

statistical approaches still have three assumptions that are always violated in the real world and generate misleading results.

As previous paragraph’s descriptions and the work by Garver [2002] and Tsaur et

al., [2002], this study presents a BPNN approach that comprises the natural logarithmic transformation for measuring the implicitly derived importance of attributes. The natural logarithmic transformation of attributes’performance (independent variables) is performed prior to the execution of BPNN to capture additional diminishing return or sensitivity for independent variables [Anderson and Sullivan, 1993]. The proposed approach for acquiring the implicitly derived importance of attributes has four steps:

Step 1: Transform all attributes’performance (AP) into a natural logarithmic form

 

AP i 1,2,...n ln

APii

(10)

where nis the total number of attributes.

Step 2: Natural logarithmic attributes’performance (ln

 

APi ) and overall customer

satisfaction (OCS) are included in a BPNN as input variables in the input layer and as output variables in the output layer.

Step 3: Train and test the objective BPNN.

After practitioners have built the objective BPNN by using the method

described in subsection “The method for designing a BPNN model”,

practitioners then train and test the objective BPNN. The mean absolute

percentage error (MAPE) and goodness-of-fit (R ) are the indicators for2

evaluating the overall objective BPNN model performance. Both formulas are as follows: MAPE= i n a a y n n i i i i *100% 1,..., 1 1  

 (4)

where y is the predicted output,i a is the actual output andi n is the

number of samples. 2 2 1 R RMSE   (5) where RMSE=

n a y n i i i

  1 2

and  is the variance of all actual output.2

After training is complete, the R of the test data set, which is very2

close to 1, indicates that BPNN model has excellent goodness-of-fit. The MAPE of test data set, which is very close to 0, indicates that BPNN model has precise prediction ability. Practitioners can then use this BPNN model for further analysis.

Step 4: Acquire the implicitly derived importance (IDI ) of the input variablesi

(attributes) of the BPNN model.

The measure of IDI for the BPNN is calculated as follows. After thei

BPNN model is trained, the total weight of a connecting path is calculated by taking the sum of the weights for the two links in the path from the input to output layer. Since the objective BPNN has only one hidden layer, the number of connecting paths for an input neuron is equal to the number of neurons in the hidden layer. The average weight for all connecting paths is then calculated as a total weight for an input aspect [Tsaur et al., 2002].

I i H w w IDI h H ih ho i   

(6)

(11)

where IDI is the weight of the input variablei i (implicitly derived

importance of the customer attribute i), iI, w is the weight of the linkih

between input variable i and the hidden-layer unit h, hH, who is the

weight of the link between the hidden-layer unit h and output-layer unit o ,

O

o , H is the set of hidden-layer units, I is the set of input-layer units,

O is the set of output-layer units, and H is the cardinality of the hidden

layer.

Revised planning matrix approach

The implicitly derived importance, derived using the approach described in subsection “Acquiring the implicitly derived importance of attributes,”is then input into the PM approach. The quantitative VOC generated using this revised PM approach is extremely effective for QFD practitioners attempting to identify principal product/service technical characteristics that most affect customer satisfaction. The revised PM approach comprises the following eight steps:

Step 1: Identify customer wants (qualitative VOC). Personal interviews and/or focus

groups are typically utilized during this step. Discovery of articulated wants and exciting wants that when fulfilled delight the customer is an important goal in this step.

Step 2: Collect customer perceptions of attributes. A questionnaire-based survey is

typically utilized during this step. The questionnaire assesses attribute

performance and overall customer satisfaction perception for own company’s

product/service and the benchmarking competitor’s product/service.

Step 3: Construct the BPNN prediction model for overall customer satisfaction via

practitioner programming or a specific ANN software package. Train the BPNN prediction model and test its performance. For details of the development of the BPNN prediction model, refer to subsection “The method for designing a BPNN model”and steps 1–2 in the subsectionAcquiring the implicitly derived importance of attributes.”For details regarding the testing and training of the BPNN model, refer to step 3 in the

subsection “Acquiring the implicitly derived importance of attributes.”

Step 4: Acquire the implicitly derived importance of each attribute (IDI ) viai

procedures in step 4 of subsection “Acquiring the implicitly derived

importance of attributes”when the BPNN prediction model has been trained completely.

Step 5: Execute competitive analysis and set the target. After analyzing the collected

(12)

competitive strategy of own company, QFD practitioners can establish an appropriate target level for each customer attribute.

Step 6: Calculate the improvement ratio for each customer attribute using Eq. (7).

i i i P T IR  , where (7) i

IR : improvement ratio of the customer attribute i

i

T : target performance level of the customer attribute i

i

P : current performance level of the customer attribute i

Step 7: Calculate the final importance of each customer attribute using Eq. (8).

i i i IR IDI I*   , where (8) * i

I : final importance of the customer attribute i

i

IR : improvement ratio of the customer attribute i

i

IDI : implicitly derived importance of the customer attribute i

Step 8: Calculate the final relative importance of each customer attribute using Eq.

(9).

 n i i i i I I RI 1 * * * , where (9) * i

RI : implicitly derived final relative importance of the customer attribute i

n : the total number of customer attributes

Using this revised PM approach, QFD practitioners can acquire the actual final relative importance of customer attributes and then apply these importance into subsequent analysis of HOQ. Consequently, the principal product/service technical characteristics that affect customer satisfaction can be acquired and utilized to improve product/service design and/or quality.

A CASE STUDY

Preparation and execution of the questionnaire survey

In this section, an example case is presented to demonstrate the application of the revised PM approach for product/service quality improvement stage. The example case is a case study of customer satisfaction improvement for hot spring hotel in Taiwan. For this purpose, a standardized questionnaire with closed-response questions using five-point Likert scale (1 = completely dissatisfy to 5 = completely satisfy) was developed based on the literature of hotel customer satisfaction and practical hot spring

(13)

hotel circumstance [Miyoung & Haemoon, 1998; Tsang & Qu, 2000; Antony et al., 2004; Lau et al., 2005; Mohsin & Ryan, 2005]. The questionnaire included questions concerning customer satisfaction within the five different customer satisfaction dimensions. These five customer satisfaction dimensions are adopted from SERVQUAL [Parasuraman et al., 1988] and are tangible, reliability, responsiveness,

assurance and empathy. Focus group’s members (hotel managers, scholars and

customers) decided twenty customer satisfaction statements (customer attributes) after they referred with references and had sufficient discuss. Therefore, twenty customer satisfaction statements of hot spring hotel and one overall customer satisfaction statement were listed in questionnaire. Customers with who had consumed services from the focal hot spring hotel and had the consumption experience of benchmarking

competitor’s services were asked to rate their degree of satisfaction for each attribute

and overall customer satisfaction for hotels. A total 412 valid questionnaires were collected for analysis.

To verify reliability and construct validity of the formal questionnaire, factor

analysis was conducted to verify the construct validity and Cronbach’s value for

each dimension was computed to verify the reliability. The factor analysis was based on the principal component analysis with varimax rotation, eigenvalue exceeding 1 and factor loadings exceeding 0.4. The test value of Kaiser-Meyer-Olkin (KMO) was 0.914.

The p value of the Bartlett’ssphericity testwasalmost zero. Moreover, the cumulative

variance explained is 58.140 %. Consequently, the construct validity of the questionnaire was quite good [Kaiser, 1974]. The 20 customer satisfaction statements regarding hot spring hotel service were classified into three dimensions, namely

“empathy and assurance”,“responsibility and reliability”and “tangibility”. Cronbach’s

 values for each dimension of hot spring hotel service satisfaction ranged from 0.8239 to 0.8915 (see Table 1). This demonstrates that the scales of the formal

questionnaire have considerable reliability (Cronbach’s values for each dimension

greater than 0.7) [Nunnally, 1978].

Building a BPNN model and acquiring the implicitly derived importance of attribute This study uses NeuroSolutions 5 software to build the BPNN prediction model for overall customer satisfaction at a hotel. The BPNN model is modeled as one input layer, one hidden layer, and one output layer. The 20 attributes are the neurons in input layer and overall customer satisfaction is the only neuron in the output layer. The activation function used in hidden layer and output layer is a hyperbolic tangent function. The values for learning rate and momentum are both 0.7 and decrease as training proceeds. The learning terminative rule is a cross-validation procedure suggested by Smith [1996]. The data from 412 valid questionnaires is assigned 70% into the train case, 20% into the

(14)

test case and 10 % into the validation case which just like Bloom’s research did [Bloom, 2004].

Table 1 Results of factor analysis Extracted dimension Statement Factor loading Eigenv-alue Variance explained (%) Cronba ch’s 18 Have customers’bestinterestatheart 0.709

11 Prompt reply to customers 0.702

06 Reasonable price 0.685

08 Dependability in handling customers’ service problem

0.612

19 Understand the specific needs of customers

0.598 4.277 21.384 0.8915

09 Perform service right at the first time 0.561 07 Provision of services as promised 0.550 12 Willingness to help customers 0.538 Reliability

and Assurance

04 Convenient hotel location 0.468 20 Personal warm care given by staff 0.755 17 Easy to getstaff’sattention & help 0.677 14 Courtesy and friendliness of staff 0.646 15 Knowledgeableto answercustomers’

request

0.603 3.951 19.754 0.8542

13 Provision of safe environment and equipment

0.535

16 Individual attention for customer 0.516 Responsibility and Empathy 10 Readinessto respond to customer’s requests 0.465

01 The physical facilities are visually appealing

0.829

02 Multiple hot spring facilities 0.728 05 Availability of adequate fire &first aids

facilities and instructions

0.703 3.400 17.002 0.8239 Tangibility

03 Cleanness of hot spring facilities 0.659 Cumulative variance explained 58.140%

(15)

The number of neurons in the hidden layer is crucial to BPNN model performance. No precise formula exists for determining the number of neurons in the hidden layer. Maren et al. [1990] demonstrated that the bound of neurons in first hidden layer was between 2N+1 and OP(N+1), where N is the number of input variables and OP is the number of output variables. Since N is 20 and OP is 1 in this case, the bound of neurons in the hidden layer is [21, 41]. Subsequently, some experimental networks (20-21-1, 20-26-1, 20-31-1, 20-36-1 and 20-41-1) were performed where the 20-21-1 network has 20 neurons in the input layer, 21 neurons in the hidden layer and 1 neuron in the output

layer. The RMSE, R2 and MAPE results indicate that the 20-31-1 network

outperforms other networks. Therefore, 31 neurons are finally assigned into hidden layer (see Table 2).

Table 2 Results of Experimental BPNN models

Model RMSE Goodness-of-fit

(R )2 MAPE (%) 20-21-1 0.106a (0.269)b 0.894 (0.800) 3.996 (11.157) 20-26-1 0.070 (0.250) 0.930 (0.830) 2.433 (10.363) 20-31-1 0.050 (0.237) 0.950 (0.847) 1.696 (10.511) 20-36-1 0.081 (0.251) 0.919 (0.825) 3.059 (11.713) 20-41-1 0.105 (0.267) 0.895 (0.798) 3.956 (11.537) Linear Regression 0.311 0.689 7.041a (46.372)b

aValue for training case

bValue for testing case

To identify the advantage of BPNN application, this study created a linear regression prediction model (20 attributes as independent variables and overall customer satisfaction as the dependent variable) such as that has been applied popularly when deriving the importance of attributes. Of the data from the 412 valid questionnaires, 80% data was used for building regression model and 20% data was used for testing.

Table 2 also lists the RMSE, R and MAPE for this regression model. Clearly, the2

BPNN model is superior to the linear regression model for generalizing the learning pattern of data.

Based on the training result of the 20-31-1 BPNN model, the final weights between input neurons and hidden neurons and the final weights between hidden neurons and output neurons can be acquired. Thus, the implicitly derived importance of attributes is

(16)

acquired via Equation. (6). The second column in Table 4 lists the implicitly derived importance of attributes.

Analytical results of the revised PM and conventional PM

In this example case, other raw importance adjustment methods (e.g., sales point) are not considered as the goal of this case is to illustrate clearly the proposed revised PM approach. Additionally, to present the different analytical results between the conventional PM and revised PM, the self-stated raw importance of customer attributes were also measured. A Likert five-point scale was utilized to measure the importance of each customer attribute. Table 3 lists the analytical results for the example case obtained via the conventional PM approach. Table 4 lists the analytical results for the example case acquired with the revised PM approach. The values of raw importance and current performance (Tables 3 and 4) are the average rating of all respondents.

According to the final relative importance (Tables 3 and 4), the importance of customer attributes derived by the revised PM approach differ significantly from those

obtained using the conventional PM approach. Taking the customerattribute“Perform

service right at the first time”asan example, its importance percentage is 4.50% in the conventional PM. However, the importance percentage is 13.9% using the revised PM due to the influence of the attribute structure characteristic of customer satisfaction. Therefore, the degree of importance increased markedly. Taking the customer attribute “Easyto getstaff’sattention andhelp”as another example, its importance is 4.59% with the conventional PM. However, the importance increased to 15.5% with the revised PM. That is, the priority consequences of customer attributes (quantitative VOC) derived using the conventional PM approach will be inappropriate. Thus, QFD practitioners must be aware that the conventional PM does not consider three-factor theory of customer satisfaction. The referential information acquired using the conventional PM can cause QFD practitioners to misidentify the principal product/service technical characteristics that affect customer satisfaction most and then derive inappropriate management actions.

(17)

Table 3 The conventional planning matrix Customer Attribute Raw Im po rt an ce C ur re nt P er fo rm an ce fo r ow n pr od uc t C ur re nt P er fo rm an ce fo r C om pe ti to r’ s Ta rg et Im pr ov em en tR at io F in al Im po rt an ce F in al R el at iv e Im po rt an ce (% )

01. The physical facilities are visually appealing own

4.12 3.69 4.13 5 1.355 5.583 5.59

02. Multiple hot spring facilities 4.17 3.62 4.48 5 1.381 5.760 5.76 03. Cleanness of hot spring facilities 4.31 3.77 3.97 4 1.061 4.573 4.58 04. Convenient hotel location 4.07 3.69 3.44 4 1.084 4.412 4.42 05. Availability of adequate fire &first aids

facilities and instructions

4.11 3.71 3.82 4 1.078 4.431 4.43

06. Reasonable price 3.94 3.52 3.34 4 1.136 4.477 4.48

07. Provision of services as promised 4.18 3.58 4.33 5 1.397 5.838 5.84 08. Dependability in handling customers’

service problem

4.26 3.64 3.94 4 1.099 4.681 4.68

09. Perform service right at the first time 4.10 3.65 3.88 4 1.096 4.493 4.50 10. Readinessto respond to customer’srequests 4.07 3.60 3.79 4 1.111 4.522 4.53 11. Prompt reply to customers 4.06 3.35 3.85 4 1.194 4.848 4.85 12. Willingness to help customers 4.16 3.68 4.47 5 1.359 5.652 5.66 13. Provision of safe environment and

equipment

4.11 3.61 4.06 5 1.385 5.693 5.70

14. Courtesy and friendliness of staff 4.33 3.84 4.11 5 1.302 5.638 5.64 15. Knowledgeableto answercustomers’

request

4.21 3.55 3.92 4 1.127 4.744 4.75

16. Individual attention for customer 3.90 3.37 3.67 4 1.187 4.629 4.63 17. Easy to getstaff’sattention and help 4.00 3.49 3.84 4 1.146 4.585 4.59 18. Havecustomers’bestinterestatheart 4.23 3.50 4.17 5 1.429 6.043 6.05 19. Understand the specific needs of customers 4.06 3.53 3.93 4 1.133 4.601 4.60 20. Personal warm care given by staff 4.29 3.63 3.71 4 1.102 4.727 4.73

(18)

Table 4 The revised planning matrix Customer Attribute Im pl ic it ly de ri ve d Im po rt an ce C ur re nt P er fo rm an ce fo r ow n pr od uc t C ur re nt P er fo rm an ce fo r C om pe ti to r’ s Ta rg et Im pr ov em en tR at io F in al Im po rt an ce F in al R el at iv e Im po rt an ce (% )

01. The physical facilities are visually appealing own

0.015 3.69 4.13 5 1.355 0.020 1.12

02. Multiple hot spring facilities 0.081 3.62 4.48 5 1.381 0.112 6.67 03. Cleanness of hot spring facilities 0.011 3.77 3.97 4 1.061 0.012 0.71 04. Convenient hotel location 0.053 3.69 3.44 4 1.084 0.057 3.39 05. Availability of adequate fire &first aids

facilities and instructions

0.043 3.71 3.82 4 1.078 0.046 2.74

06. Reasonable price 0.014 3.52 3.34 4 1.136 0.016 0.95

07. Provision of services as promised 0.056 3.58 4.33 5 1.397 0.078 4.64 08. Dependability in handling customers’

service problem

0.031 3.64 3.94 4 1.099 0.034 2.02

09. Perform service right at the first time 0.213 3.65 3.88 4 1.096 0.233 13.9 10. Readinessto respond to customer’srequests 0.021 3.60 3.79 4 1.111 0.023 1.37 11. Prompt reply to customers 0.077 3.35 3.85 4 1.194 0.092 5.47 12. Willingness to help customers 0.046 3.68 4.47 5 1.359 0.063 3.75 13. Provision of safe environment and

equipment

0.118 3.61 4.06 5 1.385 0.163 9.70

14. Courtesy and friendliness of staff 0.022 3.84 4.11 5 1.302 0.029 1.73 15. Knowledgeableto answercustomers’

request

0.244 3.55 3.92 4 1.127 0.275 16.4

16. Individual attention for customer 0.010 3.37 3.67 4 1.187 0.012 0.71 17. Easy to getstaff’sattention and help 0.227 3.49 3.84 4 1.146 0.260 15.5 18. Havecustomers’bestinterestatheart 0.056 3.50 4.17 5 1.429 0.080 4.76 19. Understand the specific needs of customers 0.041 3.53 3.93 4 1.133 0.047 2.79 20. Personal warm care given by staff 0.025 3.63 3.71 4 1.102 0.028 1.67

(19)

CONCLUSION

As overall customer satisfaction is the ultimate goal when applying QFD in product/service design and/or improvement stages, QFD practitioners must be mindful that customer self-stated raw importance of customer attributes do not represent the actual attribute importance because the three-factor theory of customer satisfaction has not been considered.

In contrast to methodologies proposed by in previous studies, this study presented a novel approach integrating a BPNN, three-factor theory and natural logarithmic transformation for acquiring implicitly derived customer attribute importance. The weights between input and output neurons in the BPNN are the actual importance of customer attributes considered the attribute category characteristic in three-factor theory of customer satisfaction. Applying the BPNN allows practitioners to accommodate non-normal, multicollinearity, and non-linear phenomena in practical systems. Furthermore, from the perspective of workload in conducting a questionnaire-based survey, the proposed approach eliminates the need to measure the raw importance of customer attributes. This unnecessary process is time-consuming for both analysts and respondents. Subsequently, the revised PM approach proposed in this study effectively assists QFD practitioners in identifying the actual final relative importance of customer attributes, applying these final importance to subsequent analysis of HOQ and identifying the correct principal technical characteristics of a product/service that affect customer satisfaction most. Consequently, business managers can utilize this information derived from QFD analysis to make more appropriate product/service design and/or quality improvement decisions and achieve competitive edge.

REFERENCES

Akao, Y. (1990) Quality Function Deployment: Integrating Customer Requirements into Product Design, Cambridge, MA: Productivity Press.

Anderson, E.W. and Mittal, V. (2000) Strengthening the satisfaction-profit chain, Journal of Service Research, 3(2), pp. 107-120.

Anderson, E.W. and Sullivan, M.W. (1993) The antecedents and consequences of customer satisfaction for firms, Marketing Science, 12 (Spring), pp. 125-143. Antony, J., Antony, F.J. and Ghosh, S. (2004) Evaluating service quality in a UK hotel

chain: a case study, International Journal of Contemporary Hospitality Management, 16(6), pp. 380-384.

(20)

Physics, 65(6), pp. 1803-1832.

Bloom, J.Z. (2004) Tourist market segmentation with linear and non-linear techniques, Tourism Management, 25(6), pp. 723-733.

Brandt, R.D. (1988) How service marketers can identify value-enhancing service elements, The Journal of Service Marketing, 2(3), pp. 35-41.

Burger, C.J.S.C., Dohnal, M., Kathrada, M., and Law, R. (2001) A practitioners guide to time-series methods for tourism demand forecasting-a case study of Durban, South Africa, Tourism Management, 22(4), pp. 403-409.

Chang, Pei-Chann and Wang, Yen-Wen (2006) Fuzzy Delphi and back-propagation model for sales forecasting in PCB industry, Expert Systems with Applications, 30(4), pp. 715-726.

Dolinsky, A.L. and Caputo, R.K. (1991) Adding a competitive dimension on importance-performance analysis: an application to traditional health care systems, Health Marketing Quarterly, 8(3/4), pp. 61-79.

Enke, D. and Thawornwong, S. (2005) The use of data mining and neural networks for forecasting stock market returns, Expert Systems with Applications, 29(4), pp. 927-940.

Fausett, L. (1994) Fundamentals of neural networks, Englewood Cliffs, NJ: Prentice-Hall.

Garver, M.S. (2002) Using data mining for customer satisfaction research, Marketing Research, 14(1), pp. 8-12.

Garver, M.S. (2003) Best practices in identifying customer-driven improvement opportunities, Industrial Marketing Management, 32(6), pp. 455-466.

Gonzalez, M.E., Quesada, G. and Bahill, T. (2003) Improving product design using quality function deployment: the school furniture case in developing countries, Quality Engineering Journal, 16(1), pp. 47-58.

Gonzalez, M.E., Quesada, G., Picado, F. and Eckelman, C.A. (2004) Customer satisfaction using QFD: an e-banking case, Managing Service Quality, 14(4), pp. 317-330.

Hauser, J.R. and Clausing, D. (1988) The house of quality, Harvard Business Review, 66, pp. 63- 73.

Haykin, S. (1999) Neural networks: A comprehensive foundation, Upper Saddle River, NJ: Prentice-Hall.

Hush, D. R. and Horne, B.G. (1993) Progress in supervised neural networks, IEEE Signal Processing Magazine, 10(1), pp. 8-39.

Johnston, R. (1995) The determinants of service quality: satisfiers and dissatisfiers, International Journal of Service Industry Management, 6(5), pp. 53-71.

(21)

system, Credit World, 81(4), pp. 26-33.

Kaastra, I. and Boyd, M. (1996) Designing a neural network for forecasting financial and economic time series, Neurocomputing, 10(3), pp. 215-236.

Kaiser, H.F. (1974) An index of factorial simplicity, Psychometrika, 39, pp. 31-36. Kano, N., Seraku, N., Takahashi, F. and Tsuji, S. (1984) Attractive quality and must-be

quality, Hinshitsu (Quality, The Journal of Japanese Society for Quality Control), 14, pp. 39-48.

Kathawala, Y. and Jaideep, M. (1994) Implementing QFD: a systems approach, The TQM Magazine, 6(6), pp. 31-37.

Khaw, J.F.C., Lim, B.S., and Lim, L.E.N. (1995) Optimal design of neural network using the Taguchi method, Neurocomputing, 7(3), pp. 225-245.

Lau, P.M., Akbar, A.K. and Fie, Y.G. (2005) Service quality: a study of the luxury hotels in Malaysia, Journal of American Academy of Business, 7(2), pp. 46-55. Law, R. (2000) Back-propagation learning in improving the accuracy of neural

network-based tourism demand forecasting, Tourism Management, 21(4), pp. 331-340.

Maren, A.J., Harston, C.T., and Pap, R.M. (1990) Handbook of Neural Computing Applications, San Diego: Academic Press.

Martins, A. and Aspinwall, E.M. (2001) Quality function deployment: an empirical study in the UK, Total Quality Management, 12(5), pp. 575-588.

Matzler, K. and Hinterhuber, H.H. (1998) How to make product development projects

moresuccessfulby integrating Kano’smodelofcustomersatisfaction into quality

function deployment, Technovation, 18(1), pp. 25-38.

Matzler, K. and Sauerwein, E. (2002) The factor structure of customer satisfaction: An empirical test of the importance grid and the penalty-reward-contrast analysis, International Journal of Service Industry Management, 13(4), pp. 314-332.

Matzler, K., Bailom, F., Hinterhuber, H.H., Renzl, B. and Pichler, J. (2004a) The

asymmetric relationship between attribute-level performance and overall customer satisfaction: a reconsideration of the importance–performance analysis, Industrial Marketing Management, 33, pp. 271-277.

Matzler, K., Fuchs, M. and Schubert, A.K. (2004b) Employee satisfaction: doesKano’s

model apply?, Total Quality Management & Business Excellence, 15(9/10), pp. 1179-1198.

Matzler, K., Hinterhuber, H.H., Bailom, F. and Sauerwein, E. (1996) How to delight your customers, Journal of Product and Band Management, 5(2), pp. 6-18.

Matzler, K., Sauerwein, E. and Heischmidt, K.A. (2003) Importance–performance analysis revisited: the role of the factor structure of customer satisfaction, The Service Industries Journal, 23(2), pp. 112-129.

(22)

Medsker, L. and Liebowitz, J. (1994) Design and development of expert systems and neural networks, New York: Macmillan.

Miyoung, J. and Haemoon, O. (1998) Quality function deployment: an extended framework for service quality and customer satisfaction in the hospitality industry, International Journal of Hospitality Management, 17(4), pp. 375-390.

Mohsin, A. and Ryan, C. (2005) Service quality assessment of 4-star hotels in Darwin, Northern Territory, Australia, Journal of Hospitality & Tourism Management, 12(1), pp. 25-36.

Nunnally, J.C. (1978) Psychometric Theory, New York: McGraw-Hill.

Oh, H. (2001) Revisiting importance-performance analysis, Tourism Management, 22(6), pp. 617-627.

O'Leary, J.F., and Adams, M.B. (1982) Community Views Concerning Urban Forest Recreation Resources, Facilities and Services, Chicago: North Central Forest Experiment Station.

Oliver, R.L. (1997) Customer satisfaction: A behavioral perspective on the consumer, New York, NY: McGraw-Hill.

Ozgener, S. (2003) Quality function deployment: a teamwork approach, Total Quality Management & Business Excellence, 14(9), pp. 969-979.

Parasuraman, A., Zeithaml, V.A. and Berry, L.L. (1988) SERVQUAL: a multiple-item scale for measuring consumer perceptions of service quality. Journal of Retailing, 64(1), pp. 12-40.

Park, J.H., Yang, K.M. and Kang, K.S. (2005) A quality function deployment

methodology with signal and noise ratio for improvement of Wasserman’s

weights, International journal of Advanced Manufacturing Technology, 26, pp. 631-637.

Pun, K.F., Chin, K.S. and Lau, H. (2000) A QFD/hoshin approach for service quality deployment: a case study, Managing Service Quality, 10(3), pp. 156-170.

Ramasamy, N.R. and Selladurai, V. (2004) Fuzzy logic approach to prioritize engineering characteristics in quality function deployment (FL-QFD), International Journal of Quality & Reliability Management, 21(9), pp. 1012-1023.

Russell, S.J. and Norvig, P. (1995) Artificial intelligence: a modern approach, Englewood Cliffs, NJ: Prentice-Hall.

Ryan, C. and Huyton, J. (2002) Tourists and aboriginal people, Annals of Tourism Research, 29(3), pp. 631-647.

Sampson, S.E. and Showalter, M.J. (1999) The performance-importance response function: Observations and implications, The Service Industries Journal, 19(3), pp. 1-25.

(23)

Schumacher, M. Robner, R. and Vach, W. (1996) Neural networks and logistics regression: Part I, Computational Statistics & Data Analysis, 21, pp. 661-682. Shin, J.S., Kim, K.J. and Chandra, M.J. (2002) Consistency check of a house of quality

chart, International Journal of Quality & Reliability Management, 19(4), pp. 471-484.

Simpson, J.T., Kollmannsberger, C., Schmalen, H. and Berkowitz, D. (2002) New product development in German and US technology firms, European Journal of Innovation Management, 5(4), pp. 194-207.

Smith, M. (1996) Neural Networks for Statistical Modeling, Boston, MA: International Thomson Computer Press.

Sullivan, L.P. (1986) Quality function deployment, Quality Progress, 19, pp. 39-50.

Tan, K.C. and Shen, X.X. (2000) Integrating Kano’smodelin the planning matrix of

quality function deployment, Total Quality Management, 11(8), pp. 1141-1151. Taylor, S.A. (1997) Assessing regression-based importance weights for quality

perceptions and satisfaction judgments in the presence of higher order and/or interaction effects, Journal of Retailing, 73(1), pp. 135-159.

Ting, S.C. and Chen, C.N. (2002) The asymmetrical and non-linear effects of store quality attributes on customer satisfaction, Total Quality Management, 13(4), pp. 547-569.

Trappey, C. and Trappey, A. (1996) A computerized quality function deployment approach for retail services, Computers and Industrial Engineering, 30(4), pp. 611-622.

Tsang, N. and Qu, H. (2000)Servicequality in China’shotelindustry:aperspective

from tourists and hotel managers, International Journal of Contemporary Hospitality Management, 12(5), pp. 316-326.

Tsaur, Sheng-Hshiung, Chiu, Yi-Chang and Huang, Chung-Huei (2002) Determinants of guest loyalty to international tourist hotels-a neural network approach, Tourism Management, 23, pp. 397-405.

Tu, J.V. (1996) Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes, Journal of Clinical Epidemiology, 49(11), pp. 1225-1231.

Vach, W., Robner, R. and Schumacher, M. (1996) Neural networks and logistics regression: Part II, Computational Statistics and Data Analysis, 21, 683-701. Vanegas, L.V. and Labib, A.W. (2001) A fuzzy quality function deployment (FQFD)

model for deriving optimum targets, International journal of production research, 39(1), pp. 99-120.

Wang, Tai-Yue and Chien, Shih-Chien (2006) Forecasting innovation performance via neural networks-a case of Taiwanese manufacturing industry, Technovation, 26,

(24)

pp. 635-643.

West, P.M., Brockett, P.L. and Golden, L.L. (1997) A comparative analysis of neural networks and statistical methods for predicting consumer choice, Marketing Science, 16(4), pp. 370-391.

Wu, D., Yang, Z. and Liang, L. (2006) Using DEA-neural network approach to evaluate branch efficiency of a large Canadian bank, Expert Systems with Applications, 31, pp. 108-115.

數據

Figure 1. A typical planning matrix (Tan and Shen, 2000)
Table 1 Results of factor analysis Extracted dimension Statement Factor loading Eigenv-alue Variance explained (%) Cronbach’s 18 Have custome r s ’ bes t i nt er e s t a t he a r t 0.709
Table 2 Results of Experimental BPNN models
Table 3 The conventional planning matrix Customer Attribute RawImportance CurrentPerformance forownproduct CurrentPerformance forCompetitor’s Target ImprovementRatio FinalImportance FinalRelative Importance(%)
+2

參考文獻

相關文件

利用 determinant 我 們可以判斷一個 square matrix 是否為 invertible, 也可幫助我們找到一個 invertible matrix 的 inverse, 甚至將聯立方成組的解寫下.

*More able students and those who have interest may further study the development of popular culture (pop music, cartoons, movies, television, etc.) in post-war Hong Kong to

Then, we tested the influence of θ for the rate of convergence of Algorithm 4.1, by using this algorithm with α = 15 and four different θ to solve a test ex- ample generated as

Numerical results are reported for some convex second-order cone programs (SOCPs) by solving the unconstrained minimization reformulation of the KKT optimality conditions,

Radiographs of Total Hip Replacements 廖振焜 林大弘 吳長晉 戴 瀚成 傅楸善 楊榮森 侯勝茂 2005 骨科醫學會 聯合學術研討會. • Automatic Digital PE

implications for further research.. Characteristics of package tours in Europe. Tourists and aboriginal people. The performance-importance response function: Observations

This two-phase decision-making model includes two major concepts: (1) Analyzing the customer’ s perception of quality practices using the IPGA model, and identifying the service

Subsequently, the relationship study about quality management culture, quality consciousness, service behavior and two type performances (subjective performance and relative