AKADEMIJA IV. F*.......a j—i ^HHIT Herbert Kotzab izredni profesor Oddelek za menedžment Copenhagen Business School Danska Thomas Reutterer izredni profesor Oddelek za trgovino in marketing Wirtschaftsuniversität na Dunaju Avstrija The design of logistics systems by logistics practitioners - Optimal results by the use of the conjoint-analysis? Izvleček Oblikovanje logističnih sistemov v logistični praksi -Optimalni rezultati z uporabo conjoint analize Študija s pomočjo referenčnih sodb 39 logistični menedžerjev proučuje način, po katerem naj bi logistični menedžerji oblikovali idealno nabavno verigo. Literatura, ki pokriva SCM (Supply Chain Management) oz. upravljanje nabavnih verig (in še posebej ECR -Efficient Consumer Response), priporoča oblikovanje nabavnih verig bolj ali manj neodvisno od položaja posameznih členov verige, iz rezultatov naše »conjoint« analize pa izhajajo štirje idealni tipi za postavitev nabavnih verig. Abstract Using preference statements of 39 logistics managers, our study investigates the way, how logistics managers would design an 'ideal-type' supply chain. Although SCM-related (and especially ECR-related) literature suggests to design supply chains more or less independently from the position of respective supply chain members, we have derived four 'ideal-type' setups for supply chains based on the results of a conjoint analysis. 1. INTRODUCTION Since the introduction of Efficient Consumer Response (ECR) in Europe in 1994, general thoughts on (re)designing a supply chain become more and more 'popular' amongst logisticians. Thereby, a number of different ECR-proposals assume a similar understanding of the different members of the supply chain on the way their supply chain is working1. In our study we investigated the validity of this assumption. Based on an experimental design, we asked logistics managers of manufacturing, retailing/wholesaling companies and managers of third-party-providers to rate various versions of supply chain setups with respect to the question "How would you design a logistics system?" based on four 1 Efficient Consumer Response Europe, "CEO Overview — Efficient Consumer Response", 1997. MM AKADEMIJA Figure 1: The general Supply Chain Management Model as suggested by Cooper, Lambert and Pagh Information flow \ \ \ \ Tier 2 Tier 1 Purchasing Materials Production Phisical Marketing & Customer Consumer Supplier Supplier Management Distribution Sales cn CD CO CO CD CJ> O CO CO CD ' CO =3 QQ CI ' ca C-D Productflow I I I I I I Customer Relationship Management I I I I I I Customer Service Management I I I I I I Demand Management I I I I I I Order Fulfillment Manufacturing Flow Management I I I I I Procurement Product Development and Commercialization I I I I I I I I Return Channel J_I_I_I_I_I_I_L Supply Chain Management Components r Planning and Contrnl * Product structure * Work structure ' Managementmethods * Organization structure * Culture and attitude " Risk and reward structure * Information flow facility (IT-structure) " Power and leadership * Product flow facility structure sttuctute predetermined parameters. This approach promised to determine so-called 'ideal-type' logistics systems established on the impressions of the interviewed managers. Our study shows that there are significant differences in the way a supply chain would be redesigned. The differences identified are not due to the position within the supply chain, but more to the size of the supply chain member (measured in sales and number of employees). 2. CONCEPTUAL CONTEXT Supply Chain and Logistics system Handfield and Nichols define a supply chain as "all activities associated with the flow and transformation of goods from the raw material stage (extraction), through to the end user, as well as the associated information flows2". This definition seems to be very closed connected with the -within the German literature - widely used term of a logistics system3. 2 Handfield, R. B„ E. L. Nichols, Introduction to Supply Chain Management, Upper Saddle River, NJ, Prentice Hall, 1999, p. 2 3 e.g. Pfohl, H. C„ Logistiksysteme, Berlin, Heidelberg et al., Springer, 1995. The difference between a logistics system and a supply chain is according to Bowersox and Closs twofold4: a) a supply chain is an extended logistics system, whereby the extension refers to the inter-organizational integration b) the trigger of a supply chain is the end user (pull instead of push orientation). Managing the Supply Chain/Logistics System The art of managing a supply chain (= Supply Chain Management or SCM see e.g. Houlihan5) is seen as the key area in increasing the overall performance of the business. Within the literature, one can find several attempts of defining the phenomena SCM. The definitions reach from »the integration of business processes from end-user through original suppliers that provides products, services, and 4 Bowersox, H. J., D. J. Closs (1996), Logistical Management: The Integrated Supply Chain, New York, McGraw-Hill, p. 101. 5 Houlihan, John, "International Supply Chain Management", International Journal of Physciai Distribution and Logistics Management, 1987, 51-66. Figure 2: US ECR Model1 A Single ECR Grocery Supply Chain Without Buffers C Demand Flow -X- \ Supplier f --- Distributor / Retail Consumer Warehouse Warehouse / Store ^ Household \ \ > Product Flow information that add value for customers6« by The Global Supply Chain Forum to "interfirm linkages designed to attain joint cost savings, product enhancements, and competitive services" by Cavinato7. In our paper, we follow the definition of Cavinato and understand SCM as a special form of strategic partnership between members of a supply chain with positive effects on the overall performance of the logistics system. Regarding to the various processes to be used for SCM we follow the SCM-model of Cooper, Lambert and Pagh, who have presented a very broad SCM-model (presented in Figure 1). SCM aims to reduce the cycle time in the channel, to reduce total channel inventory, to avoid the duplication of costs and to increase the overall customer service9. This goal can be achieved by integrating and coordinating different flows of merchandise and related information, which consist between all members in a supply chain (starting with tier suppliers and ending with the final customer). Thus, SCM is replacing the 'old' logistical paradigm, where advancements in logistics service are accompanied by increased costs. The reason for turning the traditional principal upside down lies in the use of modern information technology by realizing SCM and thus realizing a just-in-time orientation within the supply chain10. The presented empirical SCM-examples up to now, seem to be very promising for logisticians to redesign their traditional logistics system. Within the grocery business the concept of 6 The Global Supply Chain Forum, quoted by D, M. Lambert, J. R. Stock, and L. M. Eilram, Fundamentals of Logistics Management, Boston et al„ Irwin, 1998, p. 504. 7 Cavinato, J. L, "identifying interfirm total cost advantages for supply chain competitiveness", International Journal of Purchasing and Materials Management, 1991,10-15. 8 Cooper, M„ D. Lambert, J, Pagh, "Supply Chain Management: More than a new name for logistics", International Journal of Logistics management, 1997, pp. 1-14. 9 Lalonde, B. J., "Distributing inventory. More speed, less cost". Chain Store Age Executive, 1994, 18 MH - 20 MH; K. O'Laughlin, W. Copacino, "Logistics Strategy": J. Robeson, W. Copacino (ed.). The Logistics Handbook, Toronto et. al., The Free Press, 1994, 57-75. 10 Zentes J., "Effizienzsteigerungspotentiale kooperativer Logistikketten in der Konsumgüterwirtschaft": Isermann, Heinz (ed.), Logistik. Beschaffung, Produktion, Distribution, Moderne Industrie, Landsberg/Lech, 1994, p. 351. SCM has been (partially) realized by using the concept of Efficient Consumer Response (ECR)11. ECR - definition, model and goal ECR is defined as a logistical partnership between retailers and manufacturers in the grocery business with the goal to increase the performance in this business12. Comparing the very first ECR-model introduced by the FMI and developed by Kurt Salmon Associates (Figure 2) with the presented SCM-model (Figure 1) we can discover some similarities. ECR also tries to integrate different members of a channel, from the vendor to the final customer. The goal of ECR is to minimize costs and increase value for the final customer14. Starting with the introduction of an ECR-model for the US-American grocery industry16, the ECR-idea has been presented for different European markets, e.g. in Austria. The ECR-ideas and realizations are carried out by nationally organized ECR-movements, consisting of companies representing all member of the supply chain (in Austria: manufacturers, retailers/wholesalers and third-party-providers16. Problem definition - The design of a supply chain SCM and ECR are discussing and questioning the way distribution channels are yet organized and propose the way the various supply chains should be organized in the future17. " Von Tucher F., H. Wiezorek "Efficient Consumer Response": Klaus, Peter, Krieger, Winfried (ed.), Gabler Lexikon Logistik. Management iogistischer Netzwerke, Wiesbaden, Gabler, 1998, pp. 93-99. 12 Sherman R„ "ECR Vision to Reality, Creating Innovative Strategies to Astonish Customers", Annual Conference Proceedings Council of Logistics Management, 1994 13 Same reference as 12, p. 12. 14 Kurt Salmon Associates, "Effici ent Consumer Response: Enhancing Consumer Value in the Grocery Industry", Washington 1993. 15 Same reference as 1. 16 Franzmair P. (1999), "Efficient Consumer Response", Presentation to students at the Fachhochschule Ufr Marketing and Sales, February 26, 1999. 17 Stern L„ A. El-Ansary and A. T. Coughian, Marketing Channels, 5th Edition,Upper Saddle River, NJ, Prentice Hall, 1996: M. Cooper, D. Lambert and J. Pagh, "Supply Chain Management: More than a new name for logistics", International Journal of Logistics Management, 1997, 1—14. iVi ll AKADEMIJA Especially in the ECR-reiated literature, the redesign-process is concentrating on four key questions (Table 1), proposing one 'perfect' supply chain, accepted by all members of this channel. Table 1: Key questions for designing the 'ideal' ECR-supply chain Key question..,._.... relates to...._ (1) Who should be responsible ... the problem area of make or buy of logistical for logistical activities? activities and the concentration to the core _businesses of the respective channel member. (2) How should the concept .... the problem area of centralization resp. of consolidation be implemented? decentralization of logistical activities and is closely related to the logistics organization of _the respective channel member._ 13) How should be vertically integrated? .... the problem area of consideration resp. non- consideration of the activities of all members within a channel in organizing the logistics _activities of the respective channel member. (4) How should the ECR-performance ... the problem area of defining the key numbers be measured? tD measure the logistical performance, esp. how to evaluate the special input-output-relations within e channel. Who should be responsible for logistical activities? The focus of this key question is whether the individual supply chain member should make or buy supply chain activities within the supply chain. While the traditional logistics literature is concentrating this questions to the areas of inventory carrying and transportation18, a number of authors are expanding this question to all activities19. A example for expanding the Make-or-buy-question to all supply chain activities is given by the Supply Chain Operations Reference Model (SCOR) of the Supply Chain Council20. Within this framework, SCOR examines if the activities performed in the chain should be made, delivered or sourced by the individual supply chain-member. How should the concept of consolidation be implemented? This strategic question aims on the way the supply chain activities are organized, especially if these activities should be consolidated (= centralized) at one point of the supply chain or broken up (= decentralized) between the members of the channel. There is some criticism in decentralizing supply chain activities. Bowersox and Closs are recognizing a causal relation between the degree of centralization and the companies profit21. But they suggest the concept of consolidation more for transportation capabilities within a 18 e.g. Schulte C„ Logistik. Wege zur Optimierung des Material— und des Informationsflusses. München, Vahlen, 1995; D. M. Lambert, J. R. Stock, Strategie Logistics Management. 3rd edition. Homewood, IL./Boston, MA, Irwin, 1993. 19 Same references as 6. 20 http://www.supply-chain.org/html/scor_overview.cfm, 1999—03—23, 01.13 p.m. 21 Same reference as 3 and 16. supply chain. For Pfohl there is also a relationship between the size of the company and the centralization of supply chain activities22. But there is a lack on clear recommendations whether to centralize or to decentralize. It seems that the concept of centralization is very dependent on the individual situation of the supply chain member. One example for consolidation (= centralization) as a successful SCM-strategy is given by the Consolidation Work Group of the ECR Best Practices Operating Committee. The committee admits that the realization of JIT-principles within the ECR-environment consequently leads to smaller order quantities attended by an increasing of the number of deliveries within the supply chain. The only way to solve this 'order puzzle' in an economic way for all parties participating is seen in the "opportunity for industry implementation of Consolidation"23. How should be vertically integrated? This key area is focussing on the integration aspect of SCM and ECR. For Bowersox and Closs, the system integration is the key for reengineering a supply chain. Persson strengthens this suggestion by demanding both, internal and external integration as well. The power of coordination and integration is also recognized by Cooper and Ellram. The authors have thereby made a comparison between SCM and traditional logistics approaches. SCM is thereby emphasizing on a supply chain-wide integration of all activities of the supply chain members24. Morehouse and Bowersox call this strategy the breaking down of functional and enterprise silos towards a supply chain orientation26. According to Persson the success of SCM is dependent on the way integration is realized26. A promising example for the way integration could work can be seen in the many ECR-proposals by the various ECR-movements. The savings potentials within the different grocery industries (USA, Europe, Austria) lies between US-$ 100 million (in Austria) to US-$ 30 billion in the US. The key to achieve this potential lies in the industry-wide use of standards and processes to avoid duplication and triplication of workload27. 22 Pfohl H.C., "Logistik als Überlebenshilfe in den achtziger Jahren", Zeitschrift für Betriebswirtschaft, 1983, 719-34. 23 Consolidation Work Group ECR Best Practices Operating Committee and CSC Consulting, "Consolidation, Strategies to Maximize Efficiency and Minimize Costs", 1996, p. v. 24 Cooper M.C., L. M. Ellram, "Characteristics of Supply Chain Management and the Implications for Purchasing and Logistics strategy", International Journal of Logistics Management, 1993, 13-24. 25 Morehouse J. E., D. J. Bowersox, Supply Chain Management. Logistics for the Future, FMI, Washington, DC, 1995. 26 Persson G„ "Logistics Process Redesign: Some useful insights", International Journal of Logistics Management, 1995, 13—26. 27 Same references as 1, 12 and 14. How should the performance be measured? This question is concentrating on the question how to evaluate the success of SCM or ECR. The way, how the supply chain performance is going to be measured is, according to Hewitt, a combination between efficiency and effectiveness evaluation28. Because of the special goal of SCM/ECR to increase service while decreasing costs, it seems that these two metrics are defined to be the cornerstones of the evaluation whether SCM fails or succeeds29. Handfield and Nichols also point out that the final outcome is of the greatest importance from the measurement perspective30. From the perspective of the analyzed literature, the SCM/ECR supply chain design problem can be reduced to the following (Table 2). Table 2: Design components for designing a supply chain Design component Strategic {1) Make or buy decision Making Dr buying of supply chain activities (2) Degree of Centralization Centralizing or decentralizing supply chain activities (3) Degree of vertical integration Fully integrating or not integrating supply chain activities (4) Performance measurement Measuring of costs or services or both Although the relevant literature seems to give logical answers to the design of a supply chain, we have not identified any empirical results referring to preferences and/or judgements of logisticians in deciding how to evaluate different alternatives of supply chain design. 3. THE METHODOLOGICAL CONCEPT OF CONJOINT ANALYSIS APPLIED FOR PREFERENCE ANALYSIS OF SUPPLY CHAIN DESIGN COMPONENTS In order to study managers' perceptions towards the relative importance devoted to the SCM/ECR components outlined in Table 2, we reference on a conjoint measurement framework. In the field of marketing research, conjoint analysis and related techniques of experimental choice analysis represent widely used methodologies for measuring and analyzing consumer preferences. Excellent reviews of the numerous technical improvements in this approach to preference measurement during the last three decades are provided by contributions of Green and Srinivasan or Carroll and Green31. In addition, a paper of Wittink and Cattin or more recently one of Wittink, 28 Hewitt F., "Supply Chain Redesign", International Journal of Logistics Management, 1994, 1—9; or C. Caplice and Y. Sheffi, "A Review and Evaluation of Logistics Metrics", International Journal of Logistics Management, 1994, 11-28. 29 Same reference as 20. 30 Same reference as 2. 31 Green P. E., V. Srinivasan, "Conjoint Analysis in Consumer Research: Issues and Outlook", Journal of Consumer Research. 1978, 103-123; P. E. Green, V. Srinivasan, "Conjoint Analysis in Marketing: New Developments with Implications for Research and Practice", Journal of Marketing. 1990, 3—19; J. D. Carroll, P. E. Green, "Psychometric Methods in Marketing Research", Journal of Marketing Research. 1995, 385-391. Vriens, and Burenne document the widespread diffusion of conjoint analysis in marketing practice32. According to the authors more than 300 conjoint studies are conducted in the U.S. and Europe per annum. The majority of these studies focus on new product evaluation, competitive and/or product positioning analysis as well as market segmentation. However, applications to the business-to-business field (as intended here) are very rare. In contrast to multi-attribute models frequently employed for measurement of product images, conjoint measurement represents a decompositional technique for deriving part worth estimates associated with selected aspects or attributes of a choice alternative on the basis of overall preference statements of respondents. Consequently, the task of conjoint analysis is to 'decompose' the holistic information about respondents' reactions (e.g., statements or choices) to a set of stimuli into the relative importance of each level of each factor (or attributes) according to a pre-specified utility model. For subsequent analysis these part worth estimates can serve as a basis for predicting the choice probabilities of various combinations of attribute levels. Figure 3 provides a brief outline of the steps involved in a conjoint study. First, the analyst is required to specify a set of (salient) attributes of the stimuli under study as well as the number of specific level values for each of the attributes (step 1). A possible combination of such attribute levels is frequently referred to as a 'profile'33. Hence, according to the attribute-level-combinations depicted in the example of figure 1 a complete factorial design would comprise 3x4x2x3 = 72 different profiles. To obtain individual-level part worth estimates, respondents are required to evaluate these profiles. For this purpose conventional data gathering procedures utilize well-known techniques such as (metric) rating techniques, ordinal ranking of profiles, pair comparisons or choice of the most preferred profile ('choice-based' conjoint analysis) out of the corresponding set of stimulus profiles34. However, especially if there is a large number of attributes and levels, even for the most involved respondent the task is often characterized as excessively demanding, time consuming, boring, and frustrating35. Therefore, it is usually advisable to use some 'suitably' reduced sub-sample of the complete or full set of all possible profiles for the conjoint experiment. First of all, such a reduction of stimulus space can be simply achieved by a limitation of the number of attributes and/or levels. The reduction can be done by giving the primary focus on a few 32 Wittink D„ P. Cattin, "Commercial Use of Conjoint Analysis: An Update", Journal of Marketing, 1989, 91-96; D. Wittink, M. Vriens, and W. Burhenne, "Commercial Use of Conjoint in Europe: Results and Critical Reflections", International Journal of Research in Marketing, 1994, 41—52, 33 Green P. E„ "On the design of choice experiments involving multifactor alternatives .Journal of Consumer Research, 1974, 61—68. 34 Same reference as 29. 36 Malhotra N. K, "An Approach to the Measurement of Consumer Preferences Using Limited Information", Journal of Marketing Research, 1986,33-49. Figure 3: The Conceptual Framework of Conjoint Analysis (2) (3) (4) \ Level X X X =3 X X X X t=J • »buy«(Z) (2j Degree of Centralization • »centralized« (1) • «decentralized« |2| (3) Degree of vertical integration • »fully integrated« (1) • iinnn-integrated« (2| (4) Performance measuring • »cost-oriented« (11 • »cost-and-service-eriented« (2) • »service-oriented« (3| Given the attributes and levels as depicted in Table 3, a complete or 'full' factorial design would require the logisticians to discriminate between 24 different profiles. Of course, this represents a heavy burden on respondents' willingness and capability to join in the evaluation task. In order to discharge respondents with this respect, we further reduced the conjoint design via construction of an orthogonal main effects plan for the attribute-level combinations from Table 3 at the cost of neglecting interaction effects between attributes. However, this results in the 'fringe benefit' of a considerable reduction of the original full factorial design to a handsome set 35 e.g Steenkamp J. B., D. R. Wittink, "The Metric Quality of Full-Profile Judgements and the Number of Attribute Levels Effect in Conjoint Analysis", International Journal of Research in Marketing, 1994, 275—86. of stimuli profiles. Assuming further an additive-compensatory utility model without interactions between attributes37, we arrive at the following basic utility function for profile evaluation: Vj = ßo + ß,Xrj + ß2X2j + ßaXaj + ß&tw + + er <1 > where represents the evaluation of profile j = \,K,m observed from a specific person i = 1 ,K,n, ß0 is a constant term and ß.,X,ß5 the model parameters indicating the effect of attribute level variation on the profile evaluation to be estimated: The indicator variables represent a set of dummy variables reflecting the effect-coding for the attribute-level combination of profile j and finally e. is an exogenous stochastic nuisance term. Notice, that the quadratic functional form for the attribute "performance measuring" (x4) as chosen in (1) is indicative for a concave part worth function with peak part worth values for the level 'cost-and-service-orientation'. That is what we expect for this attribute, if we are willing to assume rational choice behavior of respondents38 (remember furthermore that each categorical variable with k categories can be re-coded into a set of k-\ dummy variables: hence, the effect-coding x4 of the "performance measuring" attribute requires two digits). Since we observe m profile evaluations for each of the n respondents input data arrive in an elongated mxn two-way matrix or a so-called 'stacked data' format and the dummy-regression-type model formulation described by equation (1) can be more compactly rewritten as: y = Xb + e, (2) 37 cf. e.g. Shocker A. D., V. Srinivasan, "Multiattribute Approach for Product Concept Evaluation and Generation. A Critical Review", Journal of Marketing Research, 1979, 159—180: J. J. Louviere, "Conjoint Analysis", R. P. Bagozzi (ed.), Advanced Methods of Marketing Research, Cambridge 1994. 38 Cf. same reference as 29. where y is the column vector containing the profile evaluation values of respondents, the matrix X resembles the (binary) indicator variables as row vectors of the predictors and b the column vector of the parameters to be estimated; e again gathers the stochastic disturbances. The remaining task for the conjoint analyst is to fit equation (2) with the observed profile evaluations collected in the empirical study. 4. EMPIRICAL STUDY AND PARAMETER ESTIMATION For our research we have developed a two-paged questionnaire consisting of eleven ECR-Supply-Chain-design related questions and nine questions to the person interviewed. The research was conducted during the 13th Annual Conference of the Austrian Council of Logistics Management. Table 4 gives an overview to the methodological design of the research. Table 4: Methodology of the empirical research Data collection and survey design • personal interview with standardized questionnaire (closed and open questions) Population • attendees of the 13th Annual Meeting of the Austrian Council of Logistics Management (BVL Osterreich) - 136 persons duration of the data collection • 13. And 14.6.1997 number of collected questionnaires • 41 number of analyzed questionnaires • 39 With this approach chosen, the study concentrated on the 'real' logistical decision maker. The total sample of 39 data sets represents manufacturers (8), retailer/wholesaler (12) as well as third-party-providers (19). Using this sample-mix, we have covered all possible members of a supply chain. Regarding to internal and external validity of the empirical results, the data show all of the bias of the Austrian-CLM-membership base. That is, the findings can be easily transferred to the members of the Austrian CLM. The businesses of the logistics managers interviewed represent a total sales volume of approx. 1.2 billion US-D and employ more than 80,000 people all over Austria. These companies can be regarded as leading companies within Austria's economy. The eight profiles (i.e., different combinations of attribute levels) resulting from the main effects plan plus two additional 'holdout' profiles were presented to the logistics managers for evaluation. The interviewees were asked to rank them in order of preference for adoption with respect to their individual needs (1 = most preferred, and 10 = least preferred). Since the 'holdout' profiles serve for validation purposes they were excluded from the following parameter estimation. Thus, a remaining total of 39 x 8 = 312 profile rankings can be used for estimating the part worth values of the above utility model. Although the metric assumptions, which are underlying standard OLS-estimation procedures are likely to be violated by the use of ordinal ranking data (which, of course, is also of concern for categorical or pseudo-metric rating data with response-style bias), the results of a comparative study by Jain, Malhotra and Mahajan39 support evidence that OLS-estimates may provide remarkably robust part worth values with respect to their predictive power. Furthermore, the nearly perfect reproduction of the original rankings for the holdout profiles (as reported below in the tables of results) based on estimated part worth values justify the usage of OLS-fitting of the data in the present study. 5. RESULTS In this section, we first present aggregate-level results of the conjoint analysis (derived via OLS-estimation of the model parameters) by using the total sample of respondents. Subsequently, in order to account for the heterogeneity of respondents, we present and discuss the estimates resulting from two different approaches of a 'pooled' or segment-specific analysis, which in the market segmentation literature are usually referred to as a-priori or a-posteriori segmentation schemes, respectively40. The first 'pooled' analysis arises from an a-priori grouping of respondents according to the respective position they occupy in the supply chain. Finally, we investigate the results derived from 'pooled' analyses based on an a-posteriori or cluster-based segmentation of the sample of logistics managers. Aggregated results from the supply chain management decision-makers perspective Using the total sample of respondents, aggregate part worth estimates and relative importance value results for the supply-chain design components under study are as given in Table 5. Table 5: Relative importance of the examined supply chain design components Supply Chain design parameters part worth Relative Range standard estimates Importance deviation (n—39} (per cent) 1. Make-or-buy-decision 29.93 0-67 .2122 a) make -.3812 b) buy .3812 2. Degree of Centralization 13.75 0-57 .1404 a) centralize .2750 b) decentralize ,2750 3. Degree of vertical integration 24.73 0-80 .2083 a) integrated .8500 b) not integrated -.8500 4. Performance measuring 31.59 0-91 .1956 a) cost orientation 4.0500 b) cost and service 5.6625 c) service orientation 4.8375 5. Constant value -.1500 Kendall's t (Profile) .982 Kendall's t (Holdout) 1.000 In contrast to the absolute value of the part worth ug contribution associated with a certain attribute or supply-chain 39 Jain A. K., N. K. Malhotra, V. Mahajan, "A Comparison of Internal Validity of Alternative Parameter Estimation Methods in Decompositional Multiattribute Preference Models", Journal of Marketing Research, 1979, 313-322. 40 For an up—to—date literature review cf„ e.g., T. Reutterer, "Competitive Market Structure and Segmentation Analysis with Self-Organizing Feature Maps", P. Andersson (ed.), Proceedings of the 27th EMAC Conference, Track 5: Marketing Research, Stockholm 1998. design element denoted a, the corresponding relative importance wa of the same attribute measures the impact of the feature on the total utility of a specific supply-chain setup in relative terms. Hence, the w values makes the relative a impact or importance of the supply-chain design components comparable to each other. For each attribute considered in our study the associated importance value is computed as follows: max{t/a} - min[t/a) W" Y. max{t/J - min{t/J ^ J=m* , a , a a 4 4 Notice, that with the only exception of attribute xA ("performance measuring") the part worth values ug are equivalent to the ßa-coefficients for respective attribute levels /. Due to it's concave nature, the partial utility contribution of the "performance measuring" attribute is given as ua = ß4%4(i)j + ß5X4(2)j2