中華大學博士論文

(1)

中華大學博士論文

題目：二維品質模式分類方法與評價應用之探討

系所別：科技管理研究所學號姓名：D09303013 鄧肖琳指導教授：李友錚博士

中華民國九十七年十二月

(2)

謝辭

不免驚嘆時光的飛逝，流竄在身邊的數據、文獻與計畫從未告知我此一階段已近尾聲！人生角色的多重與期望，使我博班生涯有慌亂、有備忘、有躊躇、

也有牽絆。但對選擇踏上教學研究路途的我，這僅僅是起步與開始。

本論文得以順利完成，首要感謝指導教授李博士友錚，及口試委員趙博士志揚、許博士良僑、梁博士綺華與黃博士廷合指導與教誨。恩師李博士友錚，

與我亦師亦友，修業以來承蒙費心指正，精進豐富我在學術研究上的能量。對恩師照顧提攜之感念，實難諸於筆墨，點滴心頭。

也要感謝我親愛的丈夫 Bryan，僅有你能包容我的壞脾氣與忙碌，一路安定我的心性、溫暖關照我的身心。而最是親愛的寶貝女兒孙辰，妳是我力量與喜樂的來源，因為有妳抹去我的汗水淚珠，一切我都願意。

受業期間所有師長之教誨與授業恩情，永存我心；父母親與公婆家人給予的支持與協助，使我無後顧之憂；文嘉、思涵給我如家人般的情誼與幫助，時刻縈繞不曾斷卻。當然，一路以來要稱謝的對象真真太多無法一一列載，怎一個謝字可道盡。謹以此論文獻予所有關心與協助我的師長、長官與親朋，特誌表謝忱與感念！

鄧肖琳謹誌於中華大學科技管理所博士班

中華民國九十七年十二月

(3)

Abstract

Quality attribute classification by applying Kano’s model is determined by evaluation table of author, however, a few scholars has pointed out weakness of proceeding the evaluation table and propose new methods to revise. Most of these methods are measured from the perspective of operational validity which does not sufficiently reflect the non-linear relationship between quality attribute and customer satisfaction. Therefore, one-dimensional quality may be under estimated while attractive quality, must-be quality, reverse quality and indifferent quality may be over estimated. This paper conducts multivariate analysis to propose new method for two attribute classification in Kano’s model. Firstly, expertise meeting and discriminant analysis will build up four quality attributes discriminant function and then to obtain discriminant score of each quality factor in each group. The classification will base upon the score sequence for discrimination. Within the test in case study, it will compare the classification result of this method and difference of traditional method.

As a consequence, the result indicates that the new method is practical and has its validity. Secondly, piecewise regression analysis will be set up to meet the non-linear conceptual work of quality performance and satisfaction in Kano’s model. The result of test shows that the combination of statistic and graph technique will replace original Kano’s evaluation table. Consequently, combination of Tan and Shen (2000) and Matzler and Hinterhuber (1998) will illustrate the non-linear relationship in each quality attribute performance and satisfaction which propose new evaluations’

calculation conform from Kano’s model, to modify the scale value on options to 5-points or 7-points Likert scale and duly illustrate it. As a consequence, the case study will be undertaken to compare the result diversity of new method and traditional method.

Keywords: Kano’s model, Piecewise regression, Discriminant analysis, Likert scale

(4)

摘要

傳統上運用 Kano 二維品質模式進行品質屬性分類是依作者提出之評估表的建議來做決策，但已有許多學者提出以此評估表進行之一些缺點，並提出一些新的方法加以改善。但這些方法多仍是從操作的有效性之角度來衡量，而並未充份呈現品質屬性與顧客滿意之非線性關係，恐可推論其有一維品質被低估，而魅力品質、必須品質、反向品質與無差異品質則高估之可能。本研究以多變量分析提出兩個 Kano 模式屬性分類之新方法，首先以專家會議與區別分析建立四品質屬性區別模式，再求取各品質要素在各群體之區別分數，以區別分數高低來做為分類之依據。經實例之驗證，比較此方所得到之分類結果與傳統方法之差異性，發現此方確實具可行性與有效性。其次，提出以分段迴歸分析建立來納入更貼近原始 Kano 模式對各品質屬性在品質績效與滿意度之非線性概念。經實證驗證，此結合統計與繪圖技巧的方式可取代原 Kano 的評估表。最

後，為說明各品質屬性績效與滿意度之不同非線性關聯，結合 Tan and Shen’s

(2000)與 Matzler and Hinterhuber’s (1998)之概念，提出新的評估應用，本研究將基於二維品質模式將李克特 5 與 7 尺度量表之測量尺度予以修正。最後並以實證比較新方法與傳統處理所得結果之差異。

關鍵字：二維品質模式、分段迴歸分析、區別分析、李克特量表

(5)

List of tables

Table 1 The attribute classification of quality factor ... 9

Table 2 Schvaneveldt, et al.’s 5×5 evaluation table ... 10

Table 3 Matzler and Hinterhuber’s 5×5 evaluation table ... 11

Table 4 Gerrard and Cunningham’s service quality research ... 23

Table 5 Quality factor of banking service ... 24

Table 6 The representative factor of banking services ... 24

Table 7 Proportional analysis table for population statistic ... 25

Table 8 The survey result of the evaluation factors ... 26

Table 9 Discriminating classification matrix result ... 29

Table 10 The discriminating score result of each factor ... 29

Table 11 The judgment result ... 30

Table 12 Performance of each quality factor ... 39

Table 13 Output of piecewise regression parameter ... 41

Table 14 Result of Kano attribution classification ... 42

Table 15 Modified five-points likert scale list ... 48

Table 16 The results of the categorization of quality elements ... 52

Table 17 Results of the patient satisfaction (CS) ... 54

Table 18 Results of the patient satisfaction based on likert scale (CS’) ... 55

Table 19 Modified seven-points likert scale list ... 56

(7)

List of figures

Figure 1 Kano’s two dimensional model ... 4

Figure 2 The distribution chart of Kano’s four quality attributes ... 20

Figure 3 The distribution chart of ideal state ... 21

Figure 4 Two dimensional distribution chart of the mean of quality factor ... 31

Figure 5 The distribution chart of the mean of quality factor ... 32

Figure 6 Output of piecewise regression of each quality factor... 41

Figure 7 The processes of survey ... 50

(8)

CHARPTER 1 Introduction

Section 1 Research Background

In past studies, the higher customer satisfaction which has been perceived by the greater desired quality attributes fulfillment measured in one-dimensional terms.

However, sometimes it is unnecessary to satisfy customer in higher level fulfillment in some specific quality attributes that fulfils individual customer expectations to a great extent (Matzler & Hinterhuber, 1998). Several studies have attempted to figure out how specific attributes of a product or service actually relate to customer satisfaction or dissatisfaction implying the linkage of the physical and psychological aspects of quality, where the physical aspect is concerned with the physical state or extent of the specific attributes, and the psychological aspect is related to the customer’s subjective response in terms of personal satisfaction (Schvaneveldt, Enkawa, & Miyakawa, 1991). As early as 1984, Kano, et al. proposed the theories of

“Motivator-hygiene nature of quality” and “two-dimensional quality” of products which quoted from the “Motivator-hygiene theory” of Herzberg, Mausner, and Snyderman (1959) (Kano, et al., 1984). In a survey of the demand from consumer on TV and decorative purpose table lamp as performed by Kano, Seraku, Takahashi, and Tsuji (1984), it was found that the consciousness of the user on the quality is not one dimensional but two dimensional; therefore, if consumer’s demand is mistakenly estimated, a product that can really satisfy the consumer is going to very difficult to be designed.

He demonstrated that traditional one-dimensional surveys often give partial and sometimes misleading understandings of subjective quality. Some scholars believe that the model can explore customers’ stated needs and unstated desires; on the other hand, it can develop a competitive strategy for products and services development as well (Lee & Newcomb, 1997; Matzler & Hinterhuber, 1998). After Kano’s model is proposed, related application and research are published in many journals one after another. However, the user also has some questions on that model; first, that model is only a conceptual structure, in the judgment of quality attribute, the currently existed evaluation tables are all the subjective experiences or suggestions from the experts or the scholars; until now, there are no evaluation table or evaluation method that is objective and has its theoretical basis; next, although there are some researches

(9)

published on the relationship between each quality factor under different quality attribute and customer’s satisfaction, yet these research results are hard to integrate with other quality topics(eg SERVQUAL, SERVPERF or IPA, etc).

Theoretically, Kano’s model can supplement the insufficient consideration in the traditional quality attribute. In the past, the concepts of tradition on quality attribute all belong to one-dimensional and there is no dispute for a long time, hence, we can say that one-dimensional is the majority among quality attribute. In the mean time, many scholars pointed out that attractive quality will evolve into one-dimensional because of innovation. Therefore, we can deduce that in empirical study, the reasonable result should be the number of factor of one-dimensional attribute will be greater than the number of attractive quality attribute. However, according to the currently existed literature result, it can be found that most of the quality attributes as judged by Kano’s 5×5 evaluation table are not so, this further proves the under-estimation of one-dimensional. Similarly, Kano’s 5×5 evaluation table will over-estimate attractive quality, must-be quality, reverse quality and indifferent quality.

Section 2 Research Limitations

Classifying customer requirements into Kano’s dimensions will focus the providers’ efforts where their customer will notice their effect the most. It would be helpful on which priority service element the latter studies might focus. Those new approach enriched Kano categorization information, for example, there can be better targeting of resources with a better prioritization plan for improving service attribute performance to, first and foremost, the attractive attributes. Several limitations of the foregoing essay are to be noted.

1. Compared with well-developed measurement tools such as Patient Judgment of Hospital Quality (PJHQ), which has a total of 106 items (Rubin, Ware, Nelson, &

Meterko, 1990). This essay has considered creditability and validity in conducting service item in case study. However, the number of items and options may be arbitrary and too limited as well as mostly base on single industrial case in local area. Further research should examine similar research objectives across different nations and industry.

2. The prior possibility of attribute was set the same in discriminant analysis. This is due to the four representing factors in four attribute in retrieved questionnaire has

(10)

equal relationship in plus that 100% of respondents in the assumed factor agree the factor has the same attribute. Therefore, the further study is recommended to set prior possibility as different. One method is not to assume 100% respondents have consensus on the factor. Another method is to assume that the number of representing factors in each attribute has difference.

3. There are two perspectives in determining attribute. One is to focus on independent variable which is the preference of pair-wise questionnaire. Another one refers to dependent variable which is 4 quality attributes. The former one is a continuous data and the most common method is to use average in statistic analysis. The later one is a categorical data and the most useful method is to use mode in statistic analysis. This paper utilizes the perspective of former which is to put average preference in pair-wise questionnaire into segmentation function to obtain the result of one attribute classification. The data will be the principle when determining classification. As to the later perspective, the method of Kano’s two dimensional model is to conduct the questionnaire result of each respondent as one attribute classification. To add up each output, attribute determination will belong to the most common attribute. The further study is recommended to carry on both method of using average into function or calculating each data into function and then add up. The comparison will show whether or not there is difference of classification and their effects.

4. Most independent variable must be independent when using multivariate. Quality characteristics are interrelated, the traditional hypothesis that the quality characteristics are independence will led the wrong policies.

(11)

CHARPTER 2 Literature Review

Section 1 Kano’s Model Basis

Kano thinks that quality factors can be divided into attractive quality element, one-dimensional quality element, must-be quality element, indifferent quality element and reverse quality element; these five quality elements have different performances in the two dimensional space that depends on whether the quality element is sufficient or whether customer is satisfied, it is thus generally called Kano’s two dimensional Model as shown in Figure 1.

Figure 1Kano’s two dimensional model

Note. From “Attractive quality and must-be quality,” by N. Kano, N. Seraku, F.

Takahashi, & S. Tsuji (1984), The Journal of the Japanese Society for Quality Control, 14(2), p.39-48.

Customers’ expectations toward product and service attributes can be grouped into five categories:

1. Attractive attribute as the attributes beyond customer’s explicit expression and expectation which generates surprise and delight to customers. Customers would not be aware of attractive attributes when they are exposed to them; they are not able to articulate them when asked. To create high additional value, product attractive attributes influence customers’ preferences strongly. For new customers, the attractive quality attributes can reinforce competitiveness firmly to draw their attentions (Rao, Ragu-Nathan, & Solis, 1997).

2. One-dimensional attribute raises satisfaction while offered, but lead to Satisfaction

Attractive

One-dimensional

Product Dysfunctional Indifferent Product Fully Functional

Reverse Must-be

Dissatisfaction

(12)

dissatisfaction while less present. The high performance of hybrid One-dimensional attributes leads to satisfaction and vice versa. Customers explicitly demand these quality attributes and compare them with the offerings of competitors. Customers clearly state One-dimensional attributes and specify their level of requirements. Firms must also do their best on the one-dimensional attributes – which are typically articulated by customers as being a functionality they would desire.

3. Must-be attribute means that the property or function of the product or service certainly has otherwise customers cannot accept the product or service. It refers to those whose presence is a must, and their absence will cause dissatisfaction. It is minimum requirement that cause dissatisfaction if not fulfilled, but do not lead to customer satisfaction if fulfilled or exceeded. As must-be attributes are entirely expected and considered as prerequisite, customers do not articulate them when asked about their need.

4. Indifferent attribute refers to those whose presence or absence will not cause satisfaction or dissatisfaction. This attribute will not result in satisfaction or not, whether it is sufficient or not.

5. Reverse attribute refers to those whose presence will cause dissatisfaction whereas their absence will cause satisfaction. Non-satisfaction comes when it is insufficient and on the contrary satisfaction comes when it is sufficient.

Further, it may not be considered in the process of decision-making and analysis for the reverse attribute is easily filtered out. Kano put forward the respective positions of these quality attributes to present the meanings of these quality attributes as illustrated by Kano’s model diagram (Figure 1) (Kano, 2003). As Figure1 shows, the extent to which a quality element is provided is indicated on the X-axis. The more the arrow moves towards the right, the greater the extent to which the quality element is provided, while the more the arrow moves towards the left, the less the left, the less the extent to which the quality element is provided. The customer satisfaction is indicated on the Y-axis. The higher the arrow, the higher the customer dissatisfaction, while the lower the arrow, the higher the customer dissatisfaction.

Based on these axes, Kano’s model divides product or service features into five distinct categories, each of which affects customers in a different way.

Joiner (1996) pointed out that Kano’s model could help manager to realize that

(13)

“no complaint” meant “satisfaction,” is not a correct concept and could help managers to set the priority order of work. It also presents that satisfaction and dissatisfaction are two independent concepts in the mind of the consumer and should be treated separately in quality improvement. Matzler and Hinterhuber (1998) believed that Kano’s model may help the managers to understand market needs, and can easily define the attribute of quality elements influencing the customer satisfaction. The advantages of classifying product attributes are the following (Matzler, Hinterhuber, Bailom, & Sauerwein, 1996; Matzler & Hinterhuber, 1998):

1. Product attributes are better understood: it can identify those which have the greatest influence on the consumer’s satisfaction.

2. Discovering and fulfilling attractive attributes creates a wide range of possibilities for differentiation. A product which merely satisfies the must-be and one-dimensional attributes is perceived as only average and therefore interchangeable.

3. Priorities for product development. According to the Kano’s model, a product induces various distinct types of satisfaction or dissatisfaction depending on whether certain consumer needs are completely fulfilled, only partially met. It is, for example, useless to invest in improving must-be attributes which are already at a satisfactory level. It is better to improve one-dimensional or attractive attributes as they have a greater influence on the consumer’s level of satisfaction. Thus, Kano’s model is used to establish the importance of individual product features for customer satisfaction. Even if each customer has his/her own specific needs and Kano’s model applies to each individual, Griffin and Hauser (1993), underline that we should consider consumer heterogeneity when describing the importance of product attributes on acceptance. In practice, it is more relevant to interpret results concerning product attribute influences by cluster of consumers with the same patterns of preference. Generally, must-be and one-dimensional attributes can potentially generate dissatisfaction and should first be brought under control when developing a product. Once such sources of potential dissatisfaction have been eliminated, attention can be focused on optimizing one-dimensional and attractive attributes to potentially generate greater satisfaction. Matzler and Hinterhuber (1998) also translate Kano’s model into strategic implications for new product development. They hierarchize and prioritize customer needs and establish

(14)

process-oriented product development activities to reach the “ideal” product.

Over the past two decades this theory has gained increasing exposure and acceptance, and it has been applied in product development (Fuller & Matzler, 2007) strategic thinking, business planning, and service management (Watson, 2003).

Miyakawa and Wong (1989) studied Kano’s model in manufactured goods.

Schvaneveldt, et al. (1991) explored the applicability of Kano’s model to four mass-market services – retail banking, cleaning services, family restaurants, and supermarkets. Matzler and Hinterhuber (1998) demonstrated the applicability of Kano’s model, in combination with quality function deployment, using a case study from the ski industry. Sa Moura and Saraiva (2001) used Kano’s analysis to develop an ideal kindergarten. Within service management, the theory of attractive quality has been applied extensively. Mostly it has been presented as a methodology that can be used in service development; its relationships to, and how it can be combined with, other service development methodologies such as QFD and SERVQUAL have also been investigated. Some applications within the service management field are supermarkets (Ting & Chen, 2002), web pages (Tan, Xie, & Shen, 1999), health care services (Jané & Domínguez, 2003; Lee, 2007), bank services (Bhattacharyya &

Rahman, 2004) and e-services (Fundin & Nilsson, 2003; Lee & Li, 2006), teaching quality (Chien, 2007), Centers for Independent Living (Stone, Bauer, Montgomery &

Usiak, 2007). In this essay, the operation process of Kano’s model is viewed as two parts, the first part is to distinguish the element of quality attribute, and the second part is to calculate the contribution from quality attribute performance on customer’s satisfaction.

Section 2 Evaluation of Quality Elements

1 Conventional Way of Categories

The method adopted in the first part was first proposed by Kano, et al. (1984), which uses positive and negative survey questionnaire of five points measurement table; after the interviewer under answers the pair wise questionnaire, the mode of the positive and negative survey questionnaire of each quality attribute is taken, then the 5×5 evaluation table as shown in Table 1 is referred to so as to decide each the category of the quality attribute. M means must-be quality attribute, O means one-dimensional attribute, A means attractive attribute, and I means indifferent

(15)

attribute. This model has turned out to occupy an effective and a specific strategic position in understanding the quality characteristics of products and services.

According to this model, pair-questionnaires are designed in positive (sufficient) and negative (insufficient) directions of “sufficiency”, and the quality characteristics represented in different questions, i.e. attributes, are measured according to Kano’s evaluation table (Table 1).

Generally it requires consumers to categorize the functional (i.e., presence of or high level – e.g., “How would you feel if the bank staff own willingness to help customers?”) and dysfunctional (i.e., absence of or low level – e.g., “How would you feel if the bank staff without willingness to help?”) condition of each quality attribute in terms of the following responses: (1) I like it that way, (2) It must-be that way, (3) I am neutral, (4) I can live with it that way, and (5) I dislike it. By combining the answers to both functional and dysfunctional questions, each response may be represented by one cell in the matrix. At the same time, each cell in the matrix can be categorized as an attractive, must-be, one-dimensional, indifferent, or reverse quality attribute. Suppose the interviewee thought Service A must be included (Choice 2) and disliked no Service A (Choice 5). In a manner similar to critical incident technique, after the transformation in Table 1, we know that Service A belongs to the category “must-be”. The procedure repeats for all the responses. Consequently, there will be a frequency count score for each quality attribute. The final classification of the product/service attribute is based upon which of the five quality categories has the highest frequency count score. 5×5 evaluation table is the tool used by Kano’s two dimensional model in judging quality attribute; in recent years, the 5×5 evaluation table of Berger et al. (1993) is the one that is most frequently used by the scholar (Kuo, 2004; Matzler & Hinterhuber, 1998; Matzler, Hinterhuber, Bailom, &

Sauerwein, 1996; Shahin, 2004), hence, in this essay, in the subsequent discussion of 5×5evaluation table, the 5×5 evaluation table of Berger et al. (1993) will be used as the main target of discussion; all the problems that could possibly occur when this evaluation table is used will be analyzed.

(16)

Table 1

The attribute classification of quality factor Customer

Requirements

Dysfunctional

Like Must-be Neutral Live with Dislike

Functional

Like Q A A A O

Must-be R I I I M

Neutral R I I I M

Live with R I I I M

Dislike R R R R Q

Note. From “Kano’s Methods for Understanding Customer-defined Quality,” by C.

Berger, et al. (1993), Center for Quality Management Journal, 2(4), p.3-36.

2 Others Methods For Categories

When 5×5 evaluation table is used to do quality attribute judgment, one-dimensional attribute judgment will be generated only when the answerer selects

“like” as the feeling when that evaluation factor exists and selects “dislike” as the feeling when that evaluation factor does not exist; except this, other choices will be seen as the judgment of other attribute. It was frequently pointed out in the literature of general market research that the answerer has the trend to select the medium selection item, that is, not to select the selection item in the head and in the end but select as usual the selection item that is between the two semantics; therefore, in the final statistical result, the selection item of extreme semantics is under-estimated and the medium selection item is over-explained (Coulthard ,2004). It can thus be seen that the one-dimensional attribute that is a combination of extreme semantics in the positive and negative survey questionnaire is more likely to be mistakenly judged as other attribute because it can not reach the status of obvious majority during the judgment of the attribute due to the habit of filling such survey questionnaire by the answerer. Meanwhile, there are 25 combinations in the association of five points scale from both positive and negative survey questionnaire; among them, one is set up as one-dimensional, three of them are set up as attractive quality, three of them are set as must-be quality, nine of them are set up as indifferent quality, seven of them are set up as reverse quality and two invalid combinations; since all the above setups are the subjective recognitions from the experts and the scholars or are deduced from the definition of the attributes, hence, there is no sufficient and objective evidence to prove that the division of these 25 combinations is correct and relevant, therefore, when different customer is evaluating different product or service,

(17)

the appropriateness of 5×5 evaluation table is questionable. To sum up from the above discussion, it can be seen that if the attribute judgment is decided by the 25 combinations of positive and negative survey questionnaire, it would be too subjective; meanwhile, the psychological factor of the answerer will also affect the statistical result of attribute judgment, that is, the statistical majority is actually not the real majority; therefore, all these will cause the under-estimation of one-dimensional attribute and the over-estimation of three quality attributes such as attractive, must-be and indifferent quality attribute; therefore, how to avoid the occurrence of such problems has become the focus of this research method.

Kano’s two dimensional model is a conceptual work which must base on evaluation table to proceed from quality attribute classification. Since Kano, et al.

(1984) proposed the two dimensional model; it has been widely accepted and applied.

However, there were still some arguments about Kano’s model. Quite a few scholars have been discussed about Kano’s evaluation table. (Berger, et al., 1993; Matzler &

Hinterhuber, 1998; Tan & Shen, 2000; Tontini, 2000) Furthermore, they amend evaluation table (Jané & Domínguez, 2003; Tontini, 2003). For example, Schvaneveldt, et al. (1991) revised some wordings, see correction evaluation in Table 2, in which E stands for Else. Matzler and Hinterhuber (1998) also revised some wordings, see correction evaluation in Table 3.

Table 2

Schvaneveldt, et al.’s 5×5 evaluation table

Customer’s evaluation response

When quality attribute is absent Would like It’s to be

expected No reaction It can’t be helped

Wouldn’t like

When quality attribute is

present

Would like E A A A O

It’s to be

expected E I I I M

No reaction E I I I M

It can’t be

helped E I I I M

Wouldn’t

like R E E E E

Note. From “Consumer Evaluation Perspective of Service Quality: Evaluation Factors and Two-Way Model of Quality,” by S. J. Schvaneveldt, T. Enkawa, & M.

Miyakawa (1991), Total Quality Management, 2(2), p.149-161.

(18)

Table 3

Matzler and Hinterhuber’s 5×5 evaluation table

Product Requirements

Dysfunctional form of the question I like it that

way

It must be that way

I am neutral

I can live with it that

way

I dislike it that way

Functional form of the question

I like it that

way Q A A A O

It must be

that way R I I I M

I am neutral R I I I M

I can live with it that

way

R I I I M

I dislike it

that way R R R R Q

Note. From “How to Make Product Development Projects More Successful by Integrating Kano’s Model of Customer Satisfaction into Quality Function Deployment,” by K. Matzler & H. H. Hinterhuber (1998), Technovation, 18(1), p.25-38.

In addition, Tontini (2000) increased the choices from five to seven, increased very attractive and very must-be, and renamed category of attribute as very satisfactory, satisfactory, must-be, no influence, acceptable, dissatisfied and very dissatisfied. Jané and Domínguez (2003) pointed out that if the answers on the questionnaire being cut down to three, i.e. attractive, no influence and not attractive, it makes no influence to the identification of quality attribute, thus 3×3 evaluation table may be used to simplify the operation. Since those methods are of great differences from Kano’s model, the research does not make further investigation on it. In the past, there are many scholars using different judgment methods: Ting and Chen (2002) and Matzler, Fuchs, and Schubert (2004) had adopted regression analysis method; they used the positive and negative performance of quality attribute as the independent variable and entire satisfaction as the dependent variable;

meanwhile, after the acquisition of the regression coefficients in the positive and negative sides of quality factor, the attribute of the quality factor is judged from the plus or minus sign of that coefficient. Jané and Domínguez (2003) had pointed out that if the answering items of positive and negative survey questionnaire are reduced to three, that is, “like”, “no feeling” and “dislike”, there is no effect on the judgment of quality attribute; therefore, it is suggested to use 3×3 evaluation table to simplify

(19)

the operation. Tontini (2003), from a point of view of survey questionnaire design, had thought that 7 points measurement table is more suitable for answering purpose, hence, 7×7 evaluation table is proposed to replace 5×5 evaluation table. In addition to correction and addition in the answering items, two attributes of very attractive quality and very must-be quality are also added in the classification result. Kaneko (2005) had re-explained the text description by Kano on must-be and attractive quality attribute in statistical language, then a method that associates statistical with drawing technique is proposed based on this, a satisfaction correlation chart of a comparison between satisfied and dissatisfied groups is used to replace the original evaluation table of Kano and to perform attribute classification. When there is a direct relationship between satisfied group and the satisfaction and there is no relationship between dissatisfied group and satisfaction, it is judged as attractive factor; when there is direct relationship between dissatisfied group and satisfaction and there is no relationship between satisfied group and satisfaction, it is judged as must-be factor; when there is direct relationship between both satisfied group and dissatisfied group and with satisfaction, it is judged as one-dimensional factor; when there is no correlation between satisfied group and dissatisfied group with the satisfaction, it is judged as non-correlated factor.

In the above mentioned research, if Kano’s 5×5 evaluation table is used for quality attribute classification, then only when the answered mode of positive and negative survey questionnaire is “like, dislike”, quality factor will be judged as one-dimensional, the other 24 combinations will all be judged as not one-dimensional; meanwhile, generally speaking, when the survey questionnaire answerer is filling the survey questionnaire, he/she tends not to select extreme answer, hence, if Kano’s 5×5 evaluation table is used to judge quality attribute, it will cause obvious under-estimation on one-dimensional case. The 3×3 evaluation table of Jané and Domínguez (2003) and the 7×7 evaluation table of Tontini (2000; 2003) are similar to Kano’s 5×5 evaluation table. However, Kaneko’s method (2005) can only distinguish must-be and attractive quality attribute but can not distinguish one-dimensional, reverse and indifferent quality attribute; therefore when quality attribute is judged, it could easily judge one-dimensional and indifferent quality as must-be or attractive quality.

Traditional Kano’s model may easily underestimate the one-dimensional

(20)

category of quality attribute for its confusable wording and unsymmetrical judgment standard, however, the three category of quality attributes, attractive, must-be and indifference, may be overestimated, especially the indifference attribute according to 5×5 evaluation table. Thus this essay proposed new identification methods for quality attribute to improve the accuracy of determination. These methods were not only easy to use, but also effective through difference comparison with traditional method by real cases.

Section 3 To Evaluate Customer Satisfaction Basing on Kano’s model

1 The Relationship of Attribute-level Performance and Customer Satisfaction

The traditional view of the relationship between attributes performance and customer satisfaction is symmetric and linear. The overall satisfaction will increase when quality attribute performance is improving underlying statistically calculated by a variety of covariance menthols such as correlation, regression, or structural equation models (Hamson, 1992). Nevertheless, a few scholar claim that different quality attribute performance has different relationship with customer satisfaction.

(Anderson & Mittal, 2000; Brandt, 1988; Jané & Domínguez, 2003; Johnston, 1995;

Kano, et al., 1984; Matzler, et al., 1996).Based on the Kano’s model, it illustrates the relationship between customer satisfaction and quality performance (customer perception); moreover, each category respectively affects customer satisfaction in a different way. Not only all customer requirements are not equal, but also the relationship between customer dissatisfaction and low (or high) performance of service/product (not fulfilling customer needs) would be both linear and non-linear.

That means different quality categories have different impact on satisfaction when they are delivered or not. The shapes are not the same; switch could be concavo-convex function and straight line, according to different Kano categories.

Study the must-be curve to understand this requirement situation: a little or a lot of a given feature leaves the customer unmoved; however, any lack of that particular feature quickly dissatisfies the customer. As the attractive category, changes in the positive-performance domain have a greaten impact on satisfaction than changes in the negative-performance domain. Nonlinearity appears additional one-unit increase

(21)

in an input has a smaller impact than the preceding one-unit increase. Modeling an attribute that has an asymmetric and nonlinear relationship with customer satisfaction as symmetric and linear symmetrically misestimates the impact of that attribute on customer satisfaction. The non-linear nature of the relationship among different categories attributes and overall satisfaction make identify such attributes difficult with standard linear modeling techniques. Also several scholars (e.g. Anderson &

Sullivan, 1993; Oliver, Rust, & Varki, 1997) and practitioners (e.g. Coyne, 1989) have agreed the nonlinear relationship of customer satisfaction and quality performance. Anderson and Mittal (2000) make an extensive literature review exploring the connection of individual attribute performance, customer satisfaction, customer loyalty and companies’ profit. Ting and Chen (2002) use regression analysis to show the non-linear relationship of the performance of different attributes in supermarkets and customer satisfaction. Nilson-Witell and Fundin (2005) study how the attributes of an e-service have different classification in the Kano’s model depending on how often customers have contact with them, as well as their comfort and tendency in using technology. Oliver (1993) shows that attribute-level satisfaction and dissatisfaction significantly affect overall satisfaction with a service (undergraduate course offering). Mittal, Ross, and Baldasare (1998) provide empirical evidence, from medical care and the automotive industry, that overall customer satisfaction is affected asymmetrically by attribute-level performance.

Matzler, et al. (2004) further pointed out that the relationship is not only nonlinear but also asymmetric.

2 Method to Calculate The Contribution about Attribute-level Performance

When the attribute confirmation and classification of quality factor is completed, the second part evaluation of the contribution from quality factor or quality attribute on satisfaction can then be calculated. It was first proposed by Matzler and Hinterhuber (1998) who suggested to use to indexes for the evaluation, that is, Satisfaction Increment Index (SII) and Dissatisfaction Decrement Index (DDI), wherein,

I M O A

O SII A



  , 



 







 

 A O M I

M

DDI O . Later on,

Shahin (2004) and Tan and Shen (2000) had used the definitions from Kano, et al.

(1984) on five quality attributes to deduce a formula between customer satisfaction

(22)

and quality performance, s = c×p^k, wherein k is different depending on the differences in quality attribute; then based on this, the contribution from quality factor on satisfaction is obtained. Erto and Vanacore (2002) had used quality experts to investigate the satisfaction of each quality factor on site, and then probability theory is used to estimate the contribution from each quality attribute on customer’s satisfaction.

For the method of Shahin (2004) and Tan and Shen (2000), since the c value in the suggested formula is unknown, hence, in the definition of quality as expectation divided by perception, c value will disappear, it is thus more suitable to be used in the analysis process of FMEA or QFD, if it is applied in other field, especially the analysis of service quality, it is mostly assumed that service quality is the gap between perception and expectation, in this case, since c in the formula can not be eliminated, it is thus less suitable. In the calculated contribution of each quality attribute on customer’s satisfaction according to the method of Erto and Vanacore (2002), since in the process of accumulating the contribution from quality factor into contribution from quality attribute by that method, different accumulation method is adopted for different quality attribute, hence, its value is only suitable for a comparison among different organizations under the same quality attribute;

therefore, if certain organization is to perform a comparison of the contributions of five quality attributes on customer’s satisfaction, it is then very unsuitable.

In advance, quality attribution is determined by independent variable which is positive and negative performance of quality factor, in addition by dependant variable which is overall satisfaction. (Matzler, et al.,2004; Ting & Chen, 2002) etc.

The above research is to draw or explain the variation of the relational functions.

Therefore, it is simplified as linear relation to undertake regression analysis.

However, the explanatory power of the non-linear regression models is not greater than that of the linear regression models (Mittal, et al., 1998; Ting & Chen, 2002).

Furthermore, Kano, et al.’s (1984) functional and dysfunctional technique represents an ordinal scale (at best) transformed into a nominal scale (one-dimensional, attractive, etc. classification) only after dropping factors that do not cleanly fit into the available categories. Therefore, the data collected through these two techniques are not suitable for model testing analysis. So, what is needed is a truly interval-scaled index, which may be used in more powerful statistical analyses.

(23)

Therefore, Kano’s model cannot be used to perform quantitative analysis and does not take into consideration competitors’ performance. It seems complex to evaluate the interactions of the service providing processes and the results of the customer satisfaction except the customer perception.

In order to solve the defects of conventional method as mentioned above, this essay has adopted the following research method:

1. Piecewise regression

In order to take customer’s satisfaction as a continuous function and to follow the two dimensional thinking of Kano’s model, we use the piecewise regression to reflect the link in the attributed-level performance and overall satisfaction for its asymmetric and nonlinear nature. And we use the change of slopes by piecewise regression configuration to define the quality category. This study thinks that this method can re-explain in statistical language the text description on must-be and attractive quality in Kano’s two dimensional model; meanwhile, a method that combines statistical and drawing technigue is used to replace the original Kano’s evaluation table. In order to prove the feasibility of this method, this research is going to use banking service to perform empirical study explanation.

2. Discriminant analysis

The discriminant analysis of multivariate analysis is the most frequently used prediction tool by the socialist in performing fixed attribute classification. To quality attribute judgment, the answer for positive and negative survey questionnaire can be viewed as independent variable and customer’s satisfaction can be seen as dependent variable; meanwhile, five quality attributes can correspond to five discriminant functions. If quality attribute can be used to classify the most non-disputable quality factor to set up discriminant function, then the classification of other quality factor can be reached through the acquisition of discriminant score by the substitution of the result of positive and negative survey questionnaire into the discriminant function; furthermore, this study uses banking service to perform empirical study explanation.

3. Evaluation application

To reach this purpose of representing the nonlinear relationship between the quality performance and patient satisfaction, the most contribution of this study is, to let evaluations’ calculation conform from Kano’s model, to modify the scale

(24)

value on options to 5-points or 7-points Likert scale and duly illustrate it by conducting Matzler’s definition and referring to Tan and Shen’s (2000) concept.

(25)

CHARPTER 3 Discriminant Analysis

Section 1 Background and Research Purpose

Since Kano’s two dimensional model is only a conceptual structure, in judging quality attribute, the currently existed evaluation tables are all the subjective experiences or suggestions from the experts or scholars, until now, there are no objective evaluation table or evaluation method with theoretical basis developed. For a long time, 5×5 evaluation table is the tool used by Kano’s two dimensional model in judging quality attribute, but the judgment process might be too objective, plus some hidden defects inside the table, hence, it could easily cause mis-judgment among different quality attributes; meanwhile, the psychological factor of the answerer will affect the statistical result of attribute judgment, that is, statistical majority is not the real majority; therefore, these will all cause the under-estimation of one-dimensional quality attribute and the over-estimation of three quality attributes such as attractive, must-be and indifferent quality attributes; therefore, how to avoid the occurrence of these problems has become the focus of this study method.

In this study, other classification methods are tried to be used for attribute judgment, that is, the original 5×5 evaluation table will be abandoned, and the method of multivariate analysis will be used to replace it.

Discriminant analysis is a method for dividing groups, first, in 1938, it was proposed by Fisher; its basic principle is to use certain characteristics in prediction variables to divide the research target into more than two groups, its goal is to find out linear combination of prediction variable and to set up a set of discriminant model so that this linear discriminant model will have the best performance to discriminate groups. That is, to find out the prediction variable combination that can perform optimal classification on the research target (Cooper & Emory, 1995).

Basically, the linear combination of prediction variables X1, X2, …, Xk is used as the basis of the division of observation values; this linear combination is the so-called Discriminant Function. When the classification problem of more than two groups is dealt with, if the prediction variable of each group is of multi-element normal distribution, then the discriminant function of the square of distance from the sample to be classified to the center of the group can be deduced; the smaller the distance, the higher the possibility that it belongs to this group; if the discriminant function

(26)

value between the sample to be classified and certain group is minimal, then the sample to be classified is classified into that group. Through discriminant principle or model, the independent individual in the mixed samples can be clearly divided according to the group it belongs to. However, if we take a look from the point of view of sample’s characteristic, it is the same as using different characteristic to describe the observation value and combine these characteristics to form a discriminating principle or model so that after each sample is judged by this model, it can be very accurately attributed to different group respectively. Quality attribute judgment based on Kano’s two dimensional model also belongs to clustering concept;

hence, if we use discriminant analysis to handle, the process would be able to be conducted in more objective way and the doubt about traditional 5×5 evaluation table during the judgment process can then be avoided.

Section 2 Research Methodology

It was pointed out in a study by Fornell (1996) that the good thing about using 10 points scale is to let the answerer have complete discriminant space and to avoid skewed distribution on the acquired data and make it approach normal distribution better; in addition, Coulthard (2004) thought that in addition to the original evaluation scale, if there is one more “unknown” option that fits better the situation of the person who does not have the knowledge or experience for answering the question, it could prevent this person from answering the question arbitrarily in the evaluation scale, which might affect the real value of the data. Therefore, in the survey questionnaire design of this study, 10 points scale measurement table plus one

“unknown” selection item will be adopted. Only the positive attribute investigation of quality attribute is meaningful; for the product or service provided by the modern enterprise, due to the trend of reinforcing quality, it is less possible for the existence of reverse quality; therefore, this study is the same as Berger, et al. (1993), reverse quality attribute is excluded from the design of this new judgment method. Berger, et al. (1993) treated the selection items of negative questionnaire as X-axis and the five selection items of positive survey questionnaire as Y-axis, all kinds of attribute judgment is then labeled in the form of distribution chart; after neglecting the factor of reverse quality, the inspection scope can be focused to the most likely happened region of the general questionnaire survey result, as in Figure 2. It can be seen from the figure that for the four quality attributes, due to the mutual combination of the

(27)

evaluation scale of the positive and negative survey questionnaire, they are located respectively at four corners.

Figure 2 The distribution chart of Kano’s four quality attributes

Note. From “Attractive quality and must-be quality,” by N. Kano, N. Seraku, F.

Takahashi, & S. Tsuji (1984), The Journal of the Japanese Society for Quality Control, 14(2), p.39-48.

If the forward and negative questionnaire survey result of this study imitates the method used in the distribution chart of Berger, et al. (1993), that is, the selection item of the negative survey questionnaire is treated as X-axis, the selection item of forward survey questionnaire is treated as Y-axis, similarly, only the more likely happened area is used as the inspection scope, then a distribution chart is as shown in Figure 3. Under ideal state, if the attribute performance of attractive, one-dimensional, must-be and indifferent quality is very clear, that would mean that the distribution scope won’t be too large, then the most likely distributed location of average values of four attributes after survey questionnaire evaluation should be similar to Figure 2, that is, located at the areas in the four corners as shown in the figure. It can be found from the figure that the areas can be roughly divided into four groups according to the distance of the relative distance; moreover, since

(28)

classification method of distance concept is adopted by discriminant analysis, hence, the use of discriminant analysis to perform explanation and prediction on all kinds of attributes has become the most easily understood method in real application;

however, under non-ideal state, the distribution location of each attribute might not be so clear, hence, when this method is used, the representative factor of attribute performance needs to be assured first so as to enhance the prediction capability.

Figure 3 The distribution chart of ideal state

In this study, the positive and negative survey questionnaire result is viewed as independent variable and classification attribute is viewed as dependent variable to perform discriminant analysis; moreover, canonical discrimination analysis is going to be used to test the discriminating power of independent variable and discriminating function; after the discriminating power is assured to be significant, to construct discriminating probability function. Four quality attributes can be performed with discriminant analysis through representative quality factor so as to set up discriminant function. In order to let discriminant function can judge four

(29)

different quality attributes effectively, this study is going to adopt the expert meeting that is frequently used in qualitative research to ensure the representative factor of each attribute classification of each research topic, then the quality attribute judgment of each quality factor of the discriminant function and research topic is performed.

The steps are described as in the followings:

Step 1: Hold expert meeting to make sure at least one item of the representative factors of the four quality attributes of the research topic to be used as the source of independent variable of the discriminant function set up in the subsequent step.

Step 2: Design Kano’s positive and negative survey questionnaire by targeting at the research topic, which includes four representative factors.

Step 3: Perform Kano’s positive and negative questionnaire survey.

Step 4: Sum up the average values of the level of favor of the factors in Kano’s positive and negative survey questionnaire of the target under investigation.

Step 5: Treat the answered result of the positive survey questionnaire of the representative factor of each quality attribute as independent variable X₁ and the answered result of the negative survey questionnaire as independent variable X₂ to set up discriminant function.

Step 6: Substitute the average value of the level of favor of each factor of Kano’s positive and negative survey questionnaire into discriminant function to get the occurrence probability of each factor at different attribute.

Step 7: Compare the four attribute probabilities of each factor and take the attribute with the maximum occurrence probability as the basis for judging the classification so as to perform quality attribute classification.

Section 3 Case Study

1 Source of Subjects and Questionnaire Design

In this study, discriminant analysis is used as the judgment basis to perform Kano’s two dimensional quality attribute classification for bank service quality so as to ensure the improvement direction of service quality. MBA student in certain school is used the research target to perform Kano’s two dimensional quality positive and negative questionnaire survey; the results are going to be used as related

(30)

suggestions provided regarding bank service quality. In addition, in this study, in order to make sure the representative factors of the four quality attributes of the banking service, an expert meeting is thus formed; the members of the meeting include three consumer, three enterprise’s, two scholars and experts in the field of banking service and one bank supervisor. This study mainly refers to service quality research content (Table 4) for Singapore’s bank as performed by Gerrard and Cunningham (2000); the content is submitted for discussion in the expert’s meeting, then by considering the national trend, culture and status, the content is integrated or deleted by the experts attending the meeting; finally, 16 small items of service quality elements are obtained as in Table 5. Among them, the representative factors of the four quality attributes are proposed (as in Table 6).

Table 4

Gerrard and Cunningham’s service quality research Quality factors Personalized service

Convenience of the service time Customer’s confidence

Feeling of security Self-confidence Process design Politeness

Fast and accurate service The interest rate provided

Financial stability of the bank itself Personalized service

Appearance of the architecture Internal layout of the bank Etiquette of the employee

Level of professionalism of the employee Level of the customer

Note. From “Gazetted hotels in Singapore: A banking study,” by P. Gerrard & J. B.

Cunningham (2000), International Journal of Bank Marketing, 18(3), p.135-147.

(31)

Table 5

Quality factor of banking service

Quality factor Personalized service

Convenience of service time

The service can be provided within time limit Commitment to the fulfillment of the contract Tell the service content in honest way

Appropriateness

Service willingness of the personnel In-time handling of the complaint Politeness

The feeling that customer is highly valued.

Customer’s confidence Feeling of security Professional knowledge Modern equipment and system Etiquette of the personnel Generalized banking services Table 6

The representative factor of banking services

Quality attribute Representative factor

Attractive quality The feeling that customer feels he/she is valued.

One-dimensional quality Appropriateness

Must-be quality Professional knowledge Indifferent quality Convenience of service time

The survey questionnaire can be divided into two main parts, the first part is the positive and negative survey questionnaire, the second part is personal data, which includes, for example, gender, age, marital status, level of education, monthly income. The first part is the banking service quality factors (A total of 16 items) according to a conclusion from expert’s meeting; it can be, at the same time, divided into two supply statuses for evaluation, one is the positive survey questionnaire when the supply is sufficient, another is the negative survey questionnaire when the supply is insufficient; 10 points scale is used for the evaluation, the score is give 1~10 from

“very dislike” to “very like”. A total of 200 paper survey questionnaires are issued, the effective sample is 135 with effective reply rate of 67.5%.

The sample structure of this performance investigation is based on five items of the population statistics, namely, age, sex, education level, vocation, income, etc.; the

(32)

frequency distribution summed up using SPSS statistical package software is made into a table as shown in Table 7. In the age aspect of this study, age in the range 26~35 and 36~45 years old stays as the majority, which occupies a percentage of 20% and 16.38% of the total sample respectively; however, there are very few people with age below 18 years old. Male and female occupies respectively 52.59% and 47.41% of the total sample, it can be seen that there is not too much difference in the gender aspect for the people to be tested. In the aspect of educational level attribute, the majority is from college and university which occupies 57.78% of the total sample, there are very few sample in elementary school and other diploma, that is, they are all lower than 1.5% of the total sample. In the vocation aspect, manufacturing industry occupies a percentage of 14.82% of the total sample; social and personal service industry, dining and trading industry occupies respectively 12.60%, 11.12% and 10.37%. In the income attribute aspect, the majority is in the range of 40,000~59,999 dollars, which occupies about 35.56% of the total number;

however, there are very few below 19,999 dollars and above 100,000 dollars, which are all below 10.5%.

Table 7

Proportional analysis table for population statistic

Attribute Category Head count Percentage

Age

Below 18 years old 2 1.49

18~ 25 years old 31 22.97

26 ~ 35 years old 27 20.00

36 ~ 45 years old 22 16.38

46 ~ 55 years old 18 13.34

Above 56 years old 35 25.93

Sex Male 71 52.59

Female 64 47.41

Level of education

Elementary school 1 0.74

Junior high school 18 13.34

Senior high (Vocational) school 21 15.56

College and university 78 57.78

Graduate school (including) and above 15 11.12

Others 2 1.49

Income

Below 19,999 14 10.37

20,000~39,999 dollars 26 19.26

40,000~59,999 dollars 48 35.56

60,000~79,999 dollars 35 25.93

80,000~99,999 dollars 22 16.30

(33)

Table 7 (Continue)

Attribute Category Head count Percentage

Income Above 100,000 dollars 12 8.89

Vocation

Agriculture, forestry, fishing and pasturage

industry 4 2.97

Military, government and education

industry 7 5.19

Manufacturing industry 20 14.82

Water, electricity and fuel gas industry 11 8.15 Civil engineering and construction industry 7 5.19 Social and personal service industry 17 12.60 Transport, logistic and warehousing

industry 10 7.41

Banking and insurance industry 8 5.93

Trading industry 14 10.37

Dining and tourism industry 15 11.12

Cultural industry 4 2.97

Housekeeping 10 7.41

No vocation 6 4.45

Others 2 1.49

2 Result and Discussion

When SPSS 13.0 is used to perform the positive and negative survey questionnaire data processing of quality factor of banking service, we can obtain the Cronbach’s α value of positive survey questionnaire to be 0.8163, and the value of negative survey questionnaire to be 0.9102, which shows that this measurement table has the consistency that can be trusted. Moreover, since the content of the survey questionnaire mainly refers to the service quality research content as performed by Gerrard and Cunningham (2000) on Singapore’s bank, and expert meeting is adopted for the addition and deletion of the content, it thus has content validity. The statistical quantity of each factor of the positive and negative survey questionnaire can be seen from Table 8. It can be seen from the table that the representative factor of each attribute as confirmed by expert’s meeting in this study has distribution status approaches the location of ideal status, that is, small standard deviation and mean values very close to extreme values.

Table 8

(34)

The survey result of the evaluation factors

Quality factor Positive

mean S.D. Negative

mean S.D.

Personalized service 8.5 1.17 2.71 0.96

Convenience of service time 6.29 0.85 5.28 0.72

Service can be supplied in time 8.13 1.04 1.55 1.55 Commitment for the fulfillment of the

contract 7.92 0.94 1.88 1.65

Telling of the service content in detail 6.55 1.57 3.15 1.18

Appropriateness 8.92 0.88 2.69 0.72

Service willing of personnel 6.21 0.84 3.27 1.34

In-time complaint handling 8.57 0.88 5.92 0.94

Politeness 6.62 0.87 2.77 1.18

Feeling that customer feels he/she is

highly valued 8.35 0.96 6.75 0.73

Customer’s confidence 7.42 1.42 1.11 1.30

Feeling of security 8.32 1.06 1.23 1.22

Professional knowledge 5.94 0.92 2.35 0.88

Modern equipment and system 6.58 0.95 3.82 0.91

Etiquette of personnel 7.8 0.86 3.66 0.89

Generalized banking service 6.40 0.96 3.36 1.03

In this study, the answered value of positive survey questionnaire is viewed as X1 and the answered value of negative survey questionnaire is viewed as X2, and the 135 respective data of representative factor of each quality attribute are set up as four groups, and the items of these four groups represent respectively four quality attributes of attractive, one-dimensional, must-be and indifferent; after a calculation performed by the program, we can obtain two canonical discriminant function, as in equation 1 and equation 2.

Y₁ = 0.437 – 0.528X₁ + 1.069X₂ (1) Y2 = – 8.843 + 0.984X1 + 0.497X2 (2)

In the test of discriminant function, the Wilks’ Λ value of equation 1 is 0.126, significance P = 0.000; in equation 2, Wilks’ Λ value is 0.407 and significance P = 0.000, which all mean that the discriminating power of these two discriminant functions has statistical significance; in addition, in equation 1, the eigenvalue is 2.238, in equation 2, the eigenvalue is 1.457; when equation 1 is used, the variance that can be explained is 60.6%, if equation 2 is added, it can then reach 100%

中 華 大 學 博 士 論 文