.fl/{ru’ ..I I. .7 .. 1 ..c.L>-.. 93. u 4.1.1.9. AIPIJ) o4... 4;). . irrit. . v 3 {.2 (Ir. .. ‘ on.) .Do‘ kit 1. I ‘ a ‘ r...” ... .1 m3, . h 5.2.. uMFA-tshwwmu. ”a. 3.3mm Mata" MANIA” o ... 9t! 4. .. tn. INA» I; . ,x. .\ Hampt’ ’ 59.. 8:12 LIBRARY Michigan Siam University This is to certify that the thesis entitled Business Logistics Service As A Determinant in the Industrial Buying Decision presented by Richard E. Mathisen has been accepted towards fulfillment of the requirements for ___P_h_._D_._degree in New Cyndi!) & (i HALL] Major professor Date 11,933+)? 0-7539 HI- .- O ‘\ .fflbswnsair .' ‘_ ‘ a t; x tr»: 1 ~. is: , . . . . l .. l ' l . ‘1 ‘ ‘ I u ‘i a . . 'L-. | t" ‘ K r' I I w is: - iv” ‘ "‘- A“ . ‘, ‘ .F ’ ' ’u‘r‘; ‘9" w p ~ I ’ .. K l - A .' ).“'l',{ 1|” “| 1 - x1 ~4 \. " ‘1 - u ‘ v Link 1.4 ’ ‘ “-136. L~ \ - 2 g seleczad was; m fcigi 33,3 .£.:.,:‘._.“; ~51" “31.1'afi‘i'ousl ~. 2g; “was #1:; fgfigh m .V_“;’\V .\\ ‘.".‘q7 _Iy .- ’ J ABSTRACT BUSINESS LOGISTICS SERVICE AS A DETERMINANT IN THE INDUSTRIAL PURCHASE PROCESS by Richard E. Mathisen This research investigates the relevance of business logistics service variables in the industrial purchase proc- ess. It is an attempt to provide greater understanding of the role of the logistics component in the marketing mix of a firm of manufacturing industrial installations. The problem addressed by the research was the identi- fication of market segments where the determinance of busi- ness logistics service in the purchase decision making proc- ess varied. This entailed the identification of market seg- ments which were significantly different in terms of the importance profiles across several buying criteria or fac- tors, the ability of customers to perceive differences in the performance of suppliers in the market with respect to the buying criteria, and finally the overall preferences of cus- tomers relative to their attitudes towards individual sup- pliers on all criteria. A two-phase approach was used to investigate the problem. In the first phase, exploratory data was collected to determine the market segmentation dimensions and the cri- teria used by purchase to select suppliers of the product. The product selected was commercial and industrial air con- ditioning installations, a part of the industrial plant and ¢“) . Richard E. Mathisen equipment category. The first phase identified a series in the purchase process leading to the ultimate selection of a supplier. The dimensions chosen for segmenting the market were the buying influence centers consisting of contractors and engineers, the job type, the application of equipment, and the product type. All dimensions were investigated in the second phase. A list of nineteen purchase decision mak- ing criteria was developed for the second phase. The validation phase tested several hypotheses about the level of determinance and importance of the supplier se- lection across market segments delineated along the dimen- sions identified in the first phase. A random sample of nearly five hundred contractors and engineers was selected from fifteen major metropolitan areas. The data was collected.with a mail questionnaire directed to principal operating officers in contracting and engineering firms. A followup mailing to non-respondents was completed within four weeks of the initial mailing to all sample elements. Using this procedure a response rate of 35 percent was obtained. A five point importance scale was employed to meas- ure importance of the buying criteria. Attitudes towards suppliers were also measured with a five point scale from S - excellent to l - poor. Factor analysis was used initially in the validation phase to reduce the nineteen buying criteria to a more man- ageable set of variables. The importance ratings on the wru- Fl . Richard E. Mathisen nineteen criteria were factored across the entire sample. The factor solution was rotated using the varimax method and a combination of variables with significant factor loadings was accomplished using zero-one weighting. Four factors were de- rived from the analysis and named using the composition vari- ables. These were 1) operating lifetime of equipment, 2) sales service, 3) distribution service, and 4) miscellaneous. Discriminant analysis was utilized to test the sig- nificance of the factor importance profiles across segments. Those dimensions found to be significantly different were 1) contractors versus engineers, 2) job type (contractors only), and 3) product type application (contractors only). Within the engineers group no further breakdOWn along dimen- sional lines was significant. The logistics or distribution service factor was found to be most important in differenti- ating on the contractor-engineer dimension, second most impor- tant for job type, and third for the product application group. Distribution service received a high importance rat- ing from the contractor, traditional plan and spec job, and chiller product. Once the significantly different segments had been identified, they were used as a framework to study the de- terminance of the four factors listed above in the prefer- ences of customers toward a group of selected suppliers. The four suppliers used in the analysis were those receiving the highest number of overall preference votes. The inde- pendent variables were the relative ratings of the individual Richard E. Mathisen supplier versus the average rating on each criteria. The dependent variable was the preference versus toward the sup- plier on an overall basis. In sixteen of the possible twenty combinations of segments and suppliers, a significant dif- ference was found between preferences and non-preferences based on supplier ratings. Distribution service varied in determinance from first to fourth in relationship to the two other factors. The conclusion was draWn that the determi- nance of the logistics factor in supplier preference varied by market segment and supplier. The ability of suppliers to perform on the various criteria was therefore significantly different enough as perceived by the respondents to make the factor determinant in some segments and not in others. The results indicate that the ability of suppliers to perform varies by segment and likewise the importance of the purchase decision making factors. The marketer of industrial equipment must therefore investigate his competitive position with respect to logistics service across market segments. Where all suppliers are perceived as competitive but the fac- tor is important, Opportunities for increased performance in that segment exist for the supplier who is able to improve his logistical performance. In segments where his perform- ance is not perceived as competitive, he must upgrade to at least the level of the other suppliers if he desires to neu- tralize their advantage on this factor. If he is unable to do this he may opt to devote his efforts to those segments where he can remain competitive. SUSINESS LOGISTICS SERVICE AS A DETERMINANT IN THE INDUSTRIAL PURCHASE PROCESS BY {\2 09: Richard E. Mathisen A DISSERTATION Submitted to Michigan State University ‘ in partial fulfillment of the requirements q for the degree of ~ DOCTOR or PHILOSOPHY . - q -. '- Department of Marketing and Transportation Administration 1977 "1'1 ._ ms,- *im‘rz'e grni and; for ._.i 3 n._.~"‘-'. {4}“ “ice? Miiih ' * '~ Ll fr iJth*‘t!inu'-.1:- ("" :r‘ 7 . ( 1 ~$.“fielpTC>p ' i L g ‘ Ill . J" r til-.0303 ) 5 : 1- , (>00pyright by RICHARD ERLING MATHISBN 1977 3.1cannor 30 nice n! {PCCfbf;:; .2 ovidcx: mace 1.918 masonry) H». ' ., -.~' . .- enlig'nterfing ACKNOWLEDGEMENTS Many people are deserving of my sincere gratitude for helping make this dissertation a reality. The piece which you hold is not the product of the author alone. Unfortunate- ly, to express thanks to everyone would require yet another chapter. A few people stand out as the brightest influences, however. First and foremost, my sincere thanks and apprecia- tion must go to Hedy, my wife. Without her steadfastness, tolerance, and encouragement a fraction of the work would not have been possible. The members of my dissertation committee deserve a great thank you. Dr. Donald S. Henley, Chairman, not only provided support and guidance throughout the effort, but was responsible for helping secure the aid of the sponsor. Dr. Gilbert D. Harrell provided me with both technical and con- ceptual guidance which allowed a straight course to constant- ly be followed. Finally, Dr. Richard J. Lewis, Dean of the Graduate School of Business Administration, has not only been a valuable member of my committee but a guiding light through my entire graduate career at Michigan State University. The sponsor of the research desired to remain anony- mous, but cannot go without recognition. The financial sup- port he provided made the research possible. It is my hope that the study was as enlightening to the sponsor as it was to the researcher. II TABLE OF CONTENTS List of Tables List of Figures Chapter I - Research Overview Introduction Problem Background Problem Statement Research Hypothesis Overview of Methodology Research Limitations Contributions and Potential Research Extensions Organization Overview Chapter II - Research Background Introduction Industrial Purchase Behavior Models Attitude Measurement and Brand Preference Market Segmentation Physical Distribution and Marketing Chapter III - Research Methodology Introduction Phase I - An Exploratory Study iii Page vi ix \IMH 10 13 14 15 17 17 25 35 41 52 S4 Research Hypotheses 68 Sample Design 71 Alternative Sampling Methodologies 72 Sample Size, Selection and Response 78 Data Collection Instrument 84 Data Analysis Techniques 90 Computer Data Analysis Tools 112 Summary of Research Design 113 Chapter IV - Analysis of Market Segments Introduction 115 Step One - Factor Analysis 115 Analysis of Segment Differences Based on Factor Importance 125 Discriminant Analysis of Factor Importance Scores 128 Second Level Analysis of Importance Ratings 134 Summary of Segments Based on Factor Importance 157 Chapter V - Ratings of Suppliers within Market Segments Introduction 160 Supplier Ratings within Segments 160 Summary of Ratings and Factor Determinance 173 Chapter VI - Research Conclusions, Limitations and Recommendations for Future Research Introduction 179 General Conclusions 179 iv I 1" ' 1 Supplie: Sci: .. law-Pi 9... 3 Snap: I" v ‘ 3118.1“:th .-. Fun... 'l ,i . S Summr“ -. x‘ I \ .t ‘1 Chapter {V .11“ . . ,s 4 Eigenvalue; m u, ..' R t.‘ .ftel .Al 1‘. .. 5'7} , ‘ -14. . I Cannons: ‘13 _ FQCTFV' “J I' Vern-an RPJfK if? I‘M Inpomm I ,r. v ”ii-‘1. _‘ i PIC‘OI Lama’- I. .ms I "- here: Inparu Gt doles we in ‘ ' "are (m- Contractor Walked and Um; mndariized (incrimaedfl ' Q -. ' gufiieients for Cont.-34cii£x"¢‘¥““' ‘ ‘3‘ ~ I! of Correctlfi‘m _. 333m h‘flmu w I“. 188 _Il 1: ' 'L’; . " an? - 9-3 . '- ( 3'6 7. . .‘7 ‘r. .' g; 1. ' . 'f,"o 3f; .8 6:! Versus “she'd. ‘ g LIST OF TABLES Chapter III 1 Buying Process Stages and Activities 61 2 Supplier Selection Factors 69 3 Sampling Regions and Corresponding Cities 76 4 Breakdown of P0pulation and Sample Size by Region 82 5 Summary of Level of Measurement 94 Chapter IV 1 Eigenvalues and Percent of Variation Explained 116 after Varimax Rotation of Importance Ratings 2 Communalities of Variables in Varimax Rotated 117 Factor Analysis 3 Factor Loading Matrix for Importance Rating after 119 Varimax Rotation 4 Mean Importance Rating for Nineteen Buying Criteria 121 5 Correlation Coefficients for Criteria Importance 122,123 Ratings 6 Variables Loading Higher than .60 on Factors 125 7 Mean Factor Importance Scores and Tests of 129 Difference for Contractors Versus Engineers 8 Standardized and Unstandardized Discriminant 130 Function Coefficients for Contractors Versus Engineers 9 Confusion Matrix of Correctly Classified Respond- 131 ents on Factor Importance Ratings Mechanical Contractor Versus Engineers 10 11 12 13 14 15 16 17 18 19 20 21 22 Mean~Factor Importance Scores and Tests of 135 Difference for Plan and Spec Versus Design Build-Team Managed Jobs (Contractors) Standardized and Unstandardized Discriminant 136 Function Coefficients for Plan and Spec Versus Design Build-Team Managed Jobs (Contractors) Confusion Matrix of Correctly Classified 137 Respondents on Factor Importance Ratings for Plan and Spec Versus Design Build- Team Managed Jobs (Contractors) Mean Factor Importance Scores and Tests of 139 Difference for Plan and Spec Versus Design- Build-Team Managed Jobs (Engineers) Standardized and Unstandardized Discriminant 140 Function Coefficients for Plan and Spec Versus Design Build-Team Managed (Engineers) Confusion Matrix of Correctly Classified Respond- 141 ents on Factor Importance Ratings for Plan and Spec Versus Design Build—Team Managed (Engineers) Mean Factor Importance Scores and Tests of Dif— 142 ference for Institutional Versus Commercial Job Applications (Contractors) Standardized and Unstandardized Discriminant 143 Function Coefficients for Institutional Versus Commercial Job Applications (Contractors) Confusion Matrix of Correctly Classified Respond- 144 ents on Factor Importance Ratings for Institu- tional Versus Commercial Job Applications (Contractors) Mean Factor Importance Scores and Tests for Dif— 146 ference for Institutional Versus Commercial Job Applications Standardized and Unstandardized Discriminant 147 Function Coefficients for Institutional Versus Commercial Job Applications Confusion Matrix of Correctly Classified Respond- 148 ents on Factor Importance Ratings for Institu- tional Versus Commercial Job Applications Mean Factor Importance Scores and Tests of Dif— 150 ference for Chiller Versus Rooftop Applications (Contractors) 23 Standardized and Unstandardized Discriminant 151 Function Coefficients for Chiller Versus Rooftop Applications (Contractors) 24 Confusion Matrix of Correctly Classified Respond- 152 ents on Factor Importance Ratings for Chiller Versus Rooftop Applications (Contractors) 25 Mean Factor Importance Scores and Tests of Dif- 154 ference for Chiller Versus Rooftop Application (Engineers) 26 Standardized and Unstandardized Discriminant 155 Function Coefficients for Chiller Versus Rooftop Applications (Engineers) 27 Confusion Matrix of Correctly Classified Respond- 156 ents on Factor Importance Ratings for Chiller Versus Rooftop Applications (Engineers) Chapter V 1 Supplier Mentions by Segment 164 2 Discriminant Analyses of Supplier Mentions Versus 166 Non-Mentions Contractor - Plan and Spec Jobs 3 Discriminant Analyses of Supplier Mentions Versus 168 Non-Mentions Contractor - Design Build-Team Managed Jobs 4 Discriminant Analyses of Supplier Mentions Versus 170 Non-Mentions Contractor - Chiller Applications 5 Discriminant Analyses of Supplier Mentions Versus 171 Non-Mentions Contractor - Rooftop Applications 6 Discriminant Analyses of Supplier Mentions Versus 172 Non-Mentions Engineers 7 Frequency of Importance Ranks in Discriminant 175 Functions for Purchase Factors Based on Supplier Ratings Chapter VI 1 Level of Purchase Decision Making Control on 183 Seven Stages by Job Type - Contractors 2 Level of Purchase Decision Making Control on 184 Seven Stages by Job Type - Engineers viii LIST OF FIGURES Chapter II 1 Summary of Selected Market Segmentation Dimensions Chapter III 1 Buying Center Roles in A-C Purchase Process 2 Comparison of Study Process and Webster and Wind Model 3 Relationships of Buying Groups for Commercial and Industrial Air-Conditioning Equipment 4 Sampling Methodologies 5 Classification of Multivariate Methods 6 Confusion Matrix for Two-Group Discriminant Analysis Chapter IV 1 Analysis Sequence for Factor Importance Rating 2 Summary of Significant and Non-Significant Dif- ferences for Groups Based on Factor Importance 3 Summary of Relative Importance of Factors in Discriminating Between Groups Chapter V 1 Summary of Significant and Non-Significant Dif- ferences for Groups Based on Factor Importance Scores 2 Summary of Significant and Non-Significant Func- tions and Relative Determinance of Factors ix 40 57 63 67 73 92 100 127 157 159 161 174 CHAPTER I RESEARCH OVERVIEW Introduction This research deals with the relevance of business logistics variables in the industrial purchase process. It is an attempt to contribute a greater understanding of the role of distribution components in the marketing mix. As such, the subject matter cuts across both the marketing and business logistics management functional areas. 0f particu- lar interest is the inclusion of physical distribution serv- ice variables in the formulation of marketing strategy for industrial goods. Strategic market planning consists of "the definition of market targets (or segments) and the composition of a mar- keting mix."1 In formulating his market strategy the marketer attempts to realize his firm's overall and market objectives through this two stage process. Stage one consists of iden- tifying as many alternative customer profiles as exist in the market place. These profiles or market segments define the universe of all customers into sub-universes or groups hav— ing similar characteristics. The various customer character- istics become dimensions along which these sub-groups may be located for classification. The marketer focuses his effort on one or more of the segments. Stage two consists of vary- ing the elements of the marketing mix over which he has 1Alfred R. Oxenfeldt, "The Formulation of a Market Strategy", Managerial Marketing: Perspectives and Viewpoints, ed. by Eugene J. Kelley and William Lazer, (Homewood, 111.: Richard D. Irwin, 1967), pp. 98-108. 2 control to win the market loyalty of customers in the select- ed target markets or segments. The marketing manager decides how to set levels of the marketing mix elements to most efficiently achieve the firm's market objectives in each market segment. The strategy of segmentation in the market place "represents a rational and more precise adjustment of product and marketing effort to consumer or user requirements. Successful application of the strat- egy of market segmentation tends to produce depth of market position in the segments that are effec- tively defined and penetrated." In pursuing a positive market segmentation strategy the mar- keting manager identifies all of the relevant segments, chooses those he feels may be cultivated within the scope of the available resources, and tailors his marketing effort to the segments selected. By adjusting his marketing mix to each segment, the manager more effectively concentrates to the needs of the market place. When the marketing manager adjusts the marketing mix to the segments selected, he specifies the level of the sev- eral variables over which he has control. Many lists of these variables have been proposed in the marketing litera- ture. One of the most complete is Borden's from his original 3 conception of the marketing mix. The physical handling 2Wendell R. Smith, "Product Differentiation and Mar- ket Segmentation as Alternative Marketing Strategies", Jour- nal of Marketing, Vol. XXI, No. 1, (June, 1956), pp. 3-8. 3Neil H. Borden, "The Concept of the Marketing Mix", Journal of Advertising Research, Vol. IV, No. 2, (June, 1964), pp. 2-7. 3 element includes those activities which are included in the area of study called physical distribution or business 10- gistics (these terms will be considered synonymous). Some authors add the channels of distribution element to this list while others include order handling and processing in defining the set of logistics or distribution activities.4 These activities create temporal and spacial utility in goods whether they be destined for consumer or industrial market segments. It is the integration of these activities into the marketing mix, which this research addresses. Problem Background The physical distribution or business logistics ac- tivities are called demand servicing activities by Lewis and Erickson. They comprise one of the two functions of mar- keting "to obtain demand and service demand".5 The demand creating activities are those such as product planning, pricing, personal selling, advertising, etc. In some firms the marketing manager may not have complete control over the PD activities, if he has any at 4For more specific descriptions and definitions to physical distribution refer to: Ronald H. Ballou, Business Lo istics Mana'ement, (Englewood Cliffs, N.J.: Prentice-Hall, 1973); Donald 5. Bowersox, Edward W. Smykay, and Bernard J. LaLonde, Physical Distribution Management, (London: The Macmillian co., 1968); James L. Heskett, Robert M. Ivie, and Nicholas A. Glaskowsky, Business Logistics, (New York: The Ronald Press, 1964); and James F. Magee, Physical Distribution Systems, (New York: McGraw—Hill, 1967). 5Richard J. Lewis and Leo. G. Erickson,”Marketing Functions and Marketing Systems: A Synthesis", Journal of Marketing, Vol. XXXIII, No. 3, (July, 1969), p. . 4 all. However, as Lewis and Erickson also propose, the demand obtaining and demand servicing activities are interrelated in the sense that "the firm's ability to service demand can be used as a demand obtaining force.” The degree to which this relationship holds is dependent upon the level of impor- tance that the purchaser places on having a product at the time and place where he wants it and how well the competitors in the market place can meet his needs in this area. Ballou refers to the end result of logistics activi- ties as customer service. "Customer service is a complex collection of demand related factors under the control of the firm, but whose importance in determining supplier patronage is ultimately evaluated by the customer receiving the service." He further comments that "there is a general lack of research in this area, but researchers have attempted to define the factors that make up customer service."7 Customer service may be measured in several different dimensions. Sometimes the availability of inventoried items or in-stock level is measured. Other measurements include time between order placement and receipt of merchandise, out of stock frequency and levels, and percent of customers receiving complete orders in a given time frame -— for exam- ple, 48 hours. Although these and many others may be used as measures of customer service it is the manifestation of 6Ballou, p. 96. 7Ibid. 5 performing the PD activities as viewed by the buyer which is the only relevant measure of customer service. The customer is not interested in the percent of all orders filled within twenty-four hours when his is not one of them. The frequency with which his personal orders are filled within this time is more relevant to him and more useful in his evaluation of alternate sources of supply. It is these customer services, resulting from the performance of physical distribution ac- tivities, that are relevant to the marketing manager. Ideally, the marketing manager who has control over the specification of the PD activity output should know the responsiveness of customers to varying levels of service. If he could devise a customer service function "that ex- presses revenue as a function of the customer service fac- tors...”8 he could readily set the levels of these activities which result in order cycle time, order cycle variability, frequency of back orders, etc. Ballou cites a hypothetical customer service function as: Sales Due Average Order Order Cycle 9 to Cycle Time (A) Variability (A) Customer = + Service Competitor Average Competitor Order (Firm A) Order Cycle Time Cycle Variability This model assumes that the customer faced with a supplier decision which includes logistics services compares each supplier's order cycle time and variability with those of 81bid., p. 102. 91bid., p. 103. 6 that supplier's competition. Sales are partially determined by the ability of each supplier to provide a level of serv- ice relative to the market place. This function directly relates the demand servicing activities to the demand ob- taining activities. Realistically, a function such as the above might be different for every customer and product. While tailoring the level of service to each customer would not be profitable to the firm, tailoring the level to specified market segments might be worthwhile. The quantitative measurements necessary to establish a function such as the one illustrated above are at best very difficult to gather. This information is not readily available in even the most data abundant information system of a large firm. A behavioral rather than a purely mathematical ap— proach to the revenue function is more straight forward and more readily attainable given most firms' data bases and collection capabilities. By determining how the buyer or purchaser perceives the customer service (of PD activities) in his purchase process, a function relating sales or market presence to the various marketing activities can be con- structed. More specifically, a purchase decision for an industrial product might be based upon several sets of fac- tors, one of which is the set of supplier and/or product attributes used to evaluate alternative sources of supply. A discussion of this purchase decision making model will appear in Chapter II. 7 The behavioral approach to a sales function would include information on, 1) the importance of logistics serv- ice versus other supplier rating criteria and 2) a measure of how well each supplier meets the various criteria. With this information plus the buyer's expression of overall pref- erence for a supplier, the marketing manager may then select those criteria and services upon which he will put most em- phasis in his strategy planning. The most determinant cri- teria in the purchase decision would be those where a high level of importance is attached and where a significant dif- ference in competitors' performance existed. When coupled with purchase preference information a predictive model of supplier loyalty versus market performance should result. The theoretical basis for this model will be presented in Chapter II. Problem Statement The principal problem addressed in this research may be stated as follows: How determinant is business logistics service in the process of selecting a supplier of industrial goods? Of primary interest is whether or not this service is a de- terminant factor in the selection process. An extension of the problem is to determine the degree to which logistics service is or is not a determinant for each of several market segments. If so, a strategy emphasizing better logistics service in some market segments results. For industrial products, the solution of the problem entails several subproblem solutions. 8 These subproblems are as follows: 1) Z) 3) 4) 5) 6) 7) What are the relevant dimensions by which the market may be segmented? Who is (are) the key purchase decision maker(s) for the product in each market segment? What criteria for supplier selection are used by the purchase decision makers? What is the relative importance of logistics service criteria when considered with other marketing and/or product criteria? How do the decision makers rate the performance of competitors? Does the importance rating and competitor rat- ing on the selection criteria vary by market segment? Is the choice of supplier related to the determinant criteria especially the logistics services for each market segment? The answers to these subproblems should provide in- sight to the principal problem. A framework that serves as a guide to specifying logistics service levels for market segments will emerge. The factors upon which to concentrate marketing effort for each market segment are brought into focus. Research Hypothesis The guiding hypothesis of this research that results from the problem statement is: 9 market segments may be identified for which business logistics services are of varying degrees of deter- minance in the industrial purchasing decision. This general or overall hypothesis was tested in relationship to other selection criteria. Thus a relative indication of determinance was obtained. Two sub-hypotheses are critical to this overall hy- pothesis. These are: 1) the relative importance attached to supplier selection criteria varies by market segment, and 2) the performance of competitors in the market relative to the selection criteria varies by both market segment and individual firm. Furthermore, if a particular selection criteria is highly determinant in the supplier selection process, that criteria is very important to the buyer and a high variation in per- formance of the competitors is perceived by the buyer. Thus the purchaser's choice preference among suppliers should be related to both the importance of the criteria and the per- formance of supplier relative to these criteria. The specific market segmentation dimensions along which the determinance of logistics versus other product se- lection criteria are expected to vary are discussed in detail in the methodology chapter. These segments are the product application area and the nature of the purchase decision making center, each of which will be expanded further. 10 Overview of Methodology Prior to beginning an empirical investigation to verify the research hypothesis, a real market place was se- lected from which to collect information. Because the pri- mary objectives were to expand understanding of the indus- trial marketing area, the choice was limited to a product in the raw materials and components sector, the supplies and services sector, or the plant and equipment sector. The general criteria for selecting the product were that a mul- tiple decision making influence structure was present, there were various applications for the product, and information for the study was readily available. The final selection of a product was in the plant and equipment area. Specifically the marketing process for industrial and commercial air-conditioning equipment was investigated. This product area falls into the mechanical equipment area for commercial and industrial buildings such as offices, stores, hotels, restaurants, warehouses, manu- facturing facilities, and similar buildings. There is a dearth of material in the literature upon marketing in the construction area. This research will provide some under- standing of the relationships and procedures in this impor- tant industry. Several factors made the market for mechanical equip- ment (air-conditioning components) attractive for this re- search. First, a multiplicity of purchasing influences is prevalent in the market. A preliminary investigation 11 indicated that no less than six individuals potentially in- fluence the purchase decision for the selected product. These are the building owner and/or occupant, the architect, the consulting engineer, the general contractor, the mechan— ical contractor, and the sheet metal contractor. Each of these has a varying influence upon the purchase decision depending upon several factors. Secondly, a single piece of equipment has many po- tential applications across various types of installations. This feature yielded some attractive dimensions for seg- menting the market as well as a varying set of conditions under which the selected product was purchased. A limited amount of information was available on the purchase patterns in secondary sources. However trade asso- ciation contacts were eager to provide background information and contacts with some individuals. A collection of primary survey information was dictated by the relative unavailabil- ity of previously published research in this specific area. The research was conducted in two phases. Phase one had as its objective the collection of information describing the purchase process for the selected products. Its major thrust was to familiarize the researcher with the market and the generation of testable hypotheses. The second phase was designed to verify the hypotheses generated in phase one. Statistical data was collected and analyzed in this phase to test the research hypotheses, and lead to conclusive state- ments about the market place. 12 In phase one, a series of personal interviews was conducted with individuals associated with the purchase of industrial and commercial air-conditioning equipment. These interviews were conducted with building owners, architects, engineers and contractors. The interviews yielded a descrip- tion of the purchase process from the perspective of each interviewee, an enumeration of the criteria used in the sup- plier selection decision, and classificatory data to group the respondents. The information was collected by utilizing an open-ended question format and general discussion with the interviewee. From the first phase information a struc- tured format for gathering statistical data was constructed. In phase two a questionnaire was mailed to principals in engineering and contracting firms. The selection of these individuals resulted from the interviews in phase one; the other influence centers were not found to be active in the purchase process. A statistical universe of 800 was enumer- ated in fifteen metropolitan areas, and a sample was selected from this universe. The primary analysis tool for determi- ning market segmentation relevance was discriminant analysis. The market was first segmented based on importance ratings for the buying criteria selected in phase one. These ratings were factored to reduce the data set to more man- ageable proportions. Next, discriminant analysis was uti- lized to test for differences among segments. After segmen- tation based on importance ratings, the attitudes of the market place in relation to the ability of selected suppliers 13 to perform was studied. This second stage measured the determinance of the factors in the segments. Research Limitations Each piece of research using a specific situation to test general hypotheses carries with it some restrictions and limitations. This research is no exception. To apply the results of this research effort without at minimum con- sidering its limitations could yield erroneous conclusions. The limitations of this research result from three sources. First, the circumstances surrounding the specific product market in which the study was conducted. Second, the nature of a statistical or probability study such as this; and finally the restrictions of budget and time which constrict the full range of investigation and scope of any study. Specifically the limitations of this study are--- Only one product area was studied. This resulted primarily from time and budget constraints. The inferences drawn from studying a single part of the entire plant and equipment sector of the industrial goods market will be generalizations. Thus state- ments about industrial goods marketing are made at the second level of generalization. A limited geographical coverage was used in the re- search. While only a small number of metropolitan areas were selected, these include some of the larg- est markets in the United States. Inference about the total U. S. market for the study product are based upon this group of markets. The major purchase decision making influences were considered in the study. While these often serve as proxies for other influences in reality, extraneous influences may persist in specific cases. The minor purchase influences are a subject for extension stu- dies. 14 A precise quantitative relationship between sales and service levels was not obtained. Once the most determinant criteria are found for each market seg- ment further study would be required to estimate demand elasticities. This was beyond the scope of this research. Finally, the information obtained in the study was from contact with only part of the total universe. As in all statistical studies, certain cases can be found which are exceptions to generalizations made herein. The validity of the results is tempered with the accuracy of the statistical tools available. Contributions and Potential Research Extensions As was stated earlier one objective of this research is a contribution to understanding the relevance of physical distribution services in the industrial purchase process. The results should indicate first whether or not these serv- ices represent determinant criteria in selecting a supplier of a specific product and second whether or not the level of determinance varies by market segment. Although the precise level of service does not result from this type of investigation for any of the criteria, the results serve as a guide to more detailed study of the indi- vidual criteria. Where large differences exist between per- formance levels on a particular criterion, the "best" level in the market place might be used as a target for future marketing effort. Where some criteria are highly determinant and concentration on them is within the scope of the firm's resources, an extension study measuring the demand elastic- ities and/or trade-offs between criteria would be in order. The segments of the market where this extension research should be conducted are determined. 15 The question of whether or not the level of physical distribution service should be varied by segment in the plant and equipment sector is reviewed in this research. The results would indicate whether or not a market segmenta- tion strategy is applicable to certain industrial markets. As extensions of this research beyond measuring the demand elasticities and criteria trade-offs, markets for other types of equipment should be investigated. The format of the investigation used here could be applied in other markets. A more general understanding of industrial market segmentation would result. An extension such as this would provide knowledge of how the determinance of supplier selec— tion criteria might vary across many product marketing situ- ations. Organization Overview The remaining chapters will be a detailed discussion of this research. They will deal with the theoretical back- ground and conduct of the investigation. As background material, Chapter 11 covers the liter- ature available in both the marketing and business logistics areas of study, which is applicable to the research. This chapter will discuss the behavioral models which form the basis of the investigation, market segmentation, the indus- trial purchase process (including determinants of supplier selection), and the relationship of physical distribution service to marketing. Chapter III presents the research methodology in 16 detail. The results of the first phase of the study will be outlined as well as their implication in designing the sec- ond phase. Statements of detailed hypotheses, a data col- lection method and the statistical analysis of phase two data will also be outlined. Additional background literature which relates to the selected methodology will be brought to bear on this study. Chapter IV reviews the findings dealing with segmen- tation of the market place for the selected product. The implications for strategic planning resulting from the veri- fication of research hypotheses relating to segmentation of the market place will be discussed in this chapter. Chapter V summarized the findings related to the attitudes toward suppliers of the study product and their ability to perform with regard to the choice criteria devel- oped in Chapter IV. Chapter VI reviews the results of the research in light of the existing research. The general conclusions and recommendations for future research are included in this summary chapter. CHAPTER II RESEARCH BACKGROUND Introduction This chapter presents the theoretical base upon which the research was founded. It reviews the literature pertinent to the accomplishment of the research objective and problems in the previous chapter. The first section deals with models of industrial purchase behavior in the marketing area. The second section discusses the behavioral material which led to the supplier selection measurement methodology utilized. Next a discussion of market segmenta- tion covers perspectives which were used in addressing the research problem. Finally a section which presents thoughts on the interface between marketing and physical distribution services closes the chapter. Industrial Purchase Behavior Models Several models of the industrial buying process have been proposed in the marketing literature. While specific applications of the various models have a wide range, some are peculiarly suited for the present research. A great deal of the recent work in this area has been done by Wind, 1,2 Faris, Robinson, and Webster. The model developed by these researchers predominately serves as a guide to the 1Patrick J. Robinson and Charles Faris, Industrial Buying and Creative Marketing, (Boston: Allyn and Bacon, nc.’ I 2Frederick E. Webster, Jr. and Yoram Wind, Or aniza- tional Bu ing Behavior, (Englewood Cliffs: Prentice-Hall, Inc., 197772 18 factors affecting an ”organizational" buying decision. Sheth3 has posited another somewhat analogous model as a result of his work. The prominent features of both models lend themselves well to this research effort; and these will be integrated with the product selected for investigation. The adequacy of the models for describing the realities of the industrial purchase process is beyond the scope of the research. They will, however, provide a set of inputs which serve as guidelines for segmenting markets and identifying differences across the segments. In proposing their model for simulating the indus- trial buying process, Wind and Robinson suggest that market- ing strategies directed at the buyer "require knowledge of the buyer's behavior ... his decision processes, buying motives, and the relevant forces which affect his behavior."4 This model breaks the buying process down into a series of phases and outlines a framework for evaluating each phase in various classes of buying problem situations. This framework suggests that a meaningful difference in buying behavior might be found as the decision maker operated at each 3Jagdish N. Sheth, "A Model of Industrial Buyer Behavior", Journal of Marketing, Vol. XXXVII, No. 1, October, 1973), p. 52. 4Yoram Wind and Patrick J. Robinson, "Simulation of the Industrial Buying Process", Marketing and the New Science of Planning, ed. by Robert L. King, (Chicago: American Mar- keting Association, 1968), p. 444. 19 sequential phase. The phases presented are: 1) Anticipation or recognition of a problem (need) and a general solution. 2) Determination of characteristics and quantity of needed item. 3) Description of characteristics and quantity of needed item. 4) Search for and qualification of potential sources. 5) Evaluation of prOposals and selection of suppliers. 6) Selection of an order routine. 7) Performance feedback and evaluation.5 Although these phases may not occur in distinctly identifiable steps, they serve as a useful guide in ordering the myriad of decisions made by the industrial buyer. As will be seen in more detail in the report of phase one re- search results (see Chapter III) the purchase process for the selected product closely parallels this series of buying phases. Principal purchase influencers recognize the process as being multistaged and exhibit varying behavioral charac- teristics depending upon the phase where they are most in- fluential in decision making. A more "integrated" version of this model is proposed by Webster and Wind6 when they identify the classes of fac- tors which influence the purchase decision. This model of 51bid. 6Webster and Wind, p. 27. 20 organizational behavior is: B = (I, G, O, E) Where the symbols are B - Buying Behavior - Individual Characteristics - Group Factors - Organizational Factors - Environmental Factors [TIOOH The significance is that multiple sets of factors may influ- ence the purchase decision and can account for observed dif- ferences in behavior. These factors potentially determine what criteria might be used in choosing among alternatives at various stages in the purchase process. Once more, phase one results indicated that the buyer of mechanical equipment is influenced in his behavior by many factors which conven- iently fall into the above groups. Examples of such factors include the type of building into which equipment is placed, the building owner type, and the contract situation by which the project is Operated. The Wind and Webster organizational model also sug- gests that meaningful variations in behavior might be ac- counted for by the three dimensional factors; these being the buying center identity, the nature of the decision pro- cess and the buying situation.7 Organization of an investi- gation for market segmentation strategy along these dimen- sions could prove to be fruitful. The buying center is the group of "members of the organization who interact during 71bid., p. 110 21 the buying decision process."8 The buying center roles are users, influencers, buyers, deciders, and gatekeepers. These roles may be performed by one or more than one individual. It is important that the marketer identify the relationships between the buying center roles and particular individuals within the buying organization. In addition, the individuals and their buying center roles which are most influential in the supplier choice decision must be determined. Finally, the stage in the purchase decision making process where the final decision is "actually" made should be identified. The decision making process integrates the decision making roles and the type of information needed at each stage. In some cases the actual alternatives which are considered in the latter stages of the process are defined at earlier stages, therefore constraining the range at these latter stages. Finally, the nature of the buying situation determines the amount of information needed by decision makers and to some extent the probability of a new supplier being considered. Webster and Wind characterize the buying situation by l) the newness of the problem, 2) the amount and type of information required, and 3) the number of new alternatives considered.9 Each individual decision maker may approach the problem dif- ferently with respect to the buying situation. As will be illustrated in Chapter III, the key purchasers in the buying 81bid., p. 77. 91bid., p. 115. 22 process selected for study do approach purchase decisions differently from the standpoint of the buying situation. In his model of industrial buyer behavior, Sheth also emphasizes various phases of the industrial purchasing pro- cess as well as the multiplicity of individual influences on the final purchase decision. He prOposes that purchase de- cision making behavior consists of "the psychological world of the individuals ..., the conditions which precipitate joint decisions among these individuals ..., and the inevi- table conflict among the decision makers and its resolu- tion ...."10 Sheth's model lends some understanding to the interactions among the various role players and individuals as they move through the phases of decision making. Of major significance is the recognition of how conflict resolution leads to a final purchase decision. According to the Sheth model of industrial buyer behavior, three types of individuals are "continuously in- 11 These volved in different phases of the buying process." are engineers, purchasing agents, and users. The primary factor accounting for variations in behavioral patterns among these individuals is their expectations. The sources of these variations are the model variables of 1) the back- ground of the individuals, 2) information sources, 3) active search, 4) perceptual distortion, and 5) satisfaction with 1OSheth, p. 27. 11Ibid., p. 52. 23 past purchases. These factors, which influence expectations of how well a supplier can perform relative to various cri- teria, provide a framework for predicting the determinant supplier selection criteria for a product as the influence level of the decision maker changes in the buying phase. If engineers tend to be highly influential in a particular buy- ing segment, their expectations are relevant for designing market strategies. The engineer's expectations as to which manufacturer's offerings can meet his criteria can constrain the final choice by limiting the number of suppliers who might be specified in his design of a mechanical system or whose product will meet the specifications. The second factor in Sheth's model, that of joint versus autonomous decision making, is determined by six fac- tors . 12 These are grouped as product specific factors and company Specific factors. Product specific factors relate to those connected to putting the product into use. The per- ceived risk, type of purchase and time pressure determine whether the decision making is predominately performed by a single individual or a group. In the purchase process for mechanical equipment, the risk perceived by contractors and engineers may vary depending upon their relationship with the building owner and the building use. While the product tends to be a one time purchase for some owners, it may be routine for the key decision makers, thus varying the type of 121bid., p. 54. 24 purchase. Time pressure may or may not be a factor to these decision makers. The company specific factors of orientation, size and degree of centralization also are relevant to the problem. Finally, the process of joint decision making through the resolution of conflicts between individuals is the key concept of the Sheth model. Conflicts between the decision makers result from differential expectations. "What matters most from the organization's vieWpoint is how the conflict is resolved."13 For purposes of this research, the relevant concern is that conflict resolution plays a varying part in the decision making process depending upon the number of in- dividuals influencing the purchase decision and their degree of influence. This concept lends understanding to the role behavior encountered in the various phases of the process. An engineer who specifies certain product configurations has particular expectations relative to a group of suppliers' quality, performance, etc.. On the other hand a contractor more interested in price, delivery, etc. may have a set of divergent expectations. These create a conflict which must be resolved in the final phases of the process or a reitera- tion might have to take place. The combination of these models of industrial pur- chasing behavior (Webster -- Wind and Sheth) provides a framework from which to investigate some dimensions for 13Ibid., p. 55. 25 segmentation of an industrial product market, specifically mechanical air—conditioning equipment. The roles, degrees of involvement in purchase decision making, and situations in which the individuals are found to varying degrees in the market place suggest dimensions for investigation. Varying marketing strategies aimed at segments delineated along these dimensions would be designed on the basis of the purchase decision making criteria which are most determinant in each segment. In each market segment a measure of which criteria are most relevant or determinant in choosing among the SUp- pliers to that segment is needed. This will lead to a frame- work by which marketing strategies aimed at each segment are designed. As the amount of influence in the purchase deci- sion making process, their expectations resulting from their roles as decision makers, and the purchase situation is varied; a difference in the pattern of determinant factors should be revealed. Attitude and Measurement and Brand Preference It is generally accepted by marketing researchers that attitudes toward various market offerings are closely associated, although in varying degree, with purchase be- havior. Rosenberg defined an attitude as "relatively stable affective response to an object."14 This psychological definition of attitude has been translated into marketing 14Milton J. Rosenberg, ”Cognitive Structure and Atti- tudinal Affect", Journal of Abnormal and Social Psychology, Vol. VIII, (November, 1956) p. 367. 26 as a positive or negative disposition toward a particular product or brand of product. This section will briefly re- view the basic attitude model pr0posed by Rosenberg, Fishbein and Sheth as well as some research efforts attempting to apply these models to marketing situations. In his model of attitude structure, Rosenberg theo- rized that an attitude was made up on two components related to personal values. He tested the association of these com- ponents summed over all relevant values with an overall feeling toward an object. The components of Rosenberg's attitude model were: 1) Value importance (VI) - the relative importance of the stated value in relation to other salient values; and 2) Perceived instrumentality (PI) - the extent to which the stated value ygs "attained or blocked" by the attitude object. Thus Rosenberg's model stated that the overall attitude to- ward an object was a sum of all value importances times the perceived instrumentality or A =i‘:a1[cv11) (min where A - attitude towards the object n - number of values VIi - value importance of ith value h value Pli - perceived instrumentality of it Rosenberg's research indicated that the sum products of value importance and perceived instrumentality were "significantly 151bid., p. 368. 27 related to attitude position."16 Thus, a case was made for a two component model of attitude structure toward an object. The Rosenberg model was tested in a marketing situa- tion by Sheth and Talarzyk.17 Respondents were asked to rate VI and PI as well as rank preferences for brands based on several attributes. The findings of this research indicated that perceived instrumentality was a "better" surrogate for attitude towards a brand than either value importance alone or perceived instrumentality times value importance. These authors note however that the results may have been confound- ed by several factors. Particularly, the implicit inclusion either consciously or unconsciously of the importance of an attribute in rating the perceived instrumentality of a spe- cific attribute.18 Also, the similarity of value importances across products in a class, and the aggregation of purchasers in one group. This research suggests that perhaps segmenting the study group by such things as product usage patterns, demog- raphics, and purchaser types might explain the low associa- tive properties of the two component model of Rosenberg. 16Ibid., p. 369. 17Jagdish N. Sheth and W. Wayne, Talarzyk, "Per- ceived Instrumentality and Value Importance as Determinants of Attitudes", Journal of Marketing Research, (February, 1972), pp. 6-9. 18 Ibid. 28 19 took these factors into consideration when Bennett and Scott they tested the Rosenberg model in another setting. The mar- ket was segmented into two groups who were hypothesized to have differing value importance profiles for similar product attributes. This research held that value importance when used as a weighting factor in fact contributed little to the association of attitude with perceived instrumentality. However when the market was divided by importance profiles, the predictive ability of PI was significantly enhanced. The results of the Bennett and Scott study tend to support a meaningful market segmentation base utilizing the value im- portance profiles for individual customer types. Relating these results to the models of industrial purchase behavior suggests that market segments delineated by influence center and the degree of control over purchase decision making by these centers in specific purchasing sit- uations would provide revealing results. While the differ- entiating variable between segment would be the value impor- tance profiles by product and/or supplier attributes, the divisions could be made on demographic dimensions and char- acteristics of purchase situations. Brand preference within each segment could be measured in association with perceived instrumentality ratings as Bennett and Scott suggest. 19Peter J. Bennett and Jerome E. Scott, "Cognitive Models of Attitude Structure: 'Value Importance' is Impor- tant", in Combined ProCeedings: 1971 Spring_C0nferenCes, ed. by Fred C. Allvine, (ChicagozAmerican MarketingAssociation, 1972)., pp. 346-350. 29 Bass and Wilkie20 observed that value importance con- tributes to predictability for brand preference attitude models. Using the identical data bank as in the Sheth and Talarzyk study, this effort held that both the beliefs and importance scores should be normalized across brands for each consumer.21 In doing so, Bass and Wilkie found a greatly en- hanced predictability in the two component model and addi- tionally in the "beliefs only" model as well. The controver- sies surrounding the one versus two component model are thoroughly reviewed by these researchers, who conclude that the analytical techniques utilized can have a significant influence on the study results, while a single data set is investigated. Although the controversy has not been decided, a new perspective, that of choice in analytical tools, has been interjected. While not providing a direct resolution to the con- troversy, the propositions of Myers and Alpert in their article on determinant buying attitudes perhaps reveals some of the cause for mathematical differences in testing the two model types, (Single and two component). The majority of studies dealing with prediction of brand preference have measured the associatiOn of preference ratings with impor- tance scores and instrumentality ratings. As a result one 20Frank M. Bass and William L. Wilkie, "A Comparative Analysis of Attitudinal Predictions of Brand Preference", Journal of Marketing Research, Vol. X, No.3, (August, 1973), pp. 262-269. ZlIbid. 30 might expect high correlations between brand preference and attribute importance scores. However as Myers and Alpert observe the attributes of the product and/or supplier that are most determinant, are not necessarily those which are most important. They refer to attitudes toward each feature of the product or brand which "are most closely related to "22 as determinant. preference or to actual purchase decisions The most relevant contribution to understanding the relation- ships of multi-attribute models lies in the observation that although a particular feature may be rated highly on the importance scale, the degree to which all competing brands provide that feature may be perceived as equivalent by the purchaser. If this situation holds for a feature, that fea- ture cannot conceptually be considered as a determinant in the purchase selection or in expressed preference. This concept also provides some insight into the en- hanced predictability of both the one and two component models when the importance and instrumentality scores are normalized before being correlated with purchase preference. Since the measures of importance and instrumentality, in the above studies were summed across attributes or features prior to correlation with brand preference, the determinance vari- ation among the attributes was lost. This suggests either a stepwise or at least multiple variate analysis rather than 22James H. Myers and Mark I. Alpert, "Determinant Buying Attitudes: Meaning and Measurement", Journal of Mar- keting, Vol. XXXII, No. 4, (October, 1968), pp. 13-29. 31 an aggregated measure of overall attitude. Another conceptual matter which needs attention is the relationship between attitude behavior and behavioral intent. While the Rosenberg model deals primarily with the structure of attitudes, two other models namely the Fishbein and Sheth models treat attitudes as part of the model for predicting purchase preferences and/or behaviors. Research studies have been conducted on all three models to test their effectiveness in predicting purchase behavior. Since the expressed behavior is of prime interest to marketers, rather than simply attitudes alone, some discussion must be directed toward how attitudes, whether investigated via a one or two component model, are related to purchase behavior. In their investigation of the relationships between attitudes, belief, behavior intentions and behavior, Harrell and Bennett23 found that behavior as measured by a panel diary form of the dependent variable was not highly corre- lated with preference data. While only a .4 coefficient of correlation resulted, there may have been methodological circumstances which precluded higher values. This research also tested the effectiveness of including normative beliefs 23Gilbert D. Harrell and Peter D. Bennett, "An Evalu- ation of the Expectancy Value Model of Attitude Measurement for Physician Prescribing Behavior",'Journal of Marketing Research, Vol. XI, No. 3, (August, 1974), pp. 269-278. 32 in the extended version of the Fishbein model. The model tested was 3231 =5£§1CBiaiWfl + [(NBHMcHWz where B - BehaV1or BI - Behavioral Intent B. - Belief About ith Outcome a. - Evaluative Aspect of ith Outcome NB - Normative Belief Mc - Motivation to Comply with Norms Wj - Beta Weights n - Number of Relevant Outcomes Although more complex than the basic attitude model of Rosenberg, this extended version of the Fishbein model ”includes a measure of social consequences.”24 of a particu- lar behavior. While the study found that including the normative factors in the model did not enhance the predic- tability of brand preference, the authors express caution about excluding these factors from other marketing studies. The results of the Harrell and Bennett investigation suggest that further research is necessary to determine how closely behavioral intent and actual behavior are in fact related. The authors suggest that situation specific factors might well be the major intervening force between these vari- ables. This would lead to a research design which offers various situations more closely defined by familiar purchase 24Ibid. 33 circumstances among which the respondent would be allowed to choose. This research recommends the use of behavioral in- tent as a dependent variable in attitude studies, until more is known about the variables which intervene between behavi- oral intent and actual behavior. Furthermore a disaggregated form of the model which weights all product and/or supplier attributes individually is recommended for use when studying marketing situations. A detailed comparison of the Rosenberg and Fishbein models as well as a model proposed by Sheth was conducted by Raju, Bhagat, and Sheth to measure the relative predicta- bility of all three models. This work provides a sound re- view of the models and accurately outlines a testing criteria scheme. The Sheth model as presented in this research in- cludes four dimensions of behavioral intent toward a partic- ular brand. Behavioral intention is "a function of 1) evaluative beliefs about the object's potential to satisfy needs, wants, and desires, 2) perceived social stereotype of the object, 3) predisposition resulting from past satisfaction, and 4) situational influences that :he person anticipgges will be effective at the 1me of behaV1or. In contrasting the three models with regard to pre- dictive validation, cross-validation, and validity generali- zation, complete discussion of these criteria may be found 25p.s. Raju, Rabi s. Bhagat, and Jagdish N. Sheth, "Predictive Validation and Cross Validation of the Fishbein, Rosenberg, and Sheth Models of Attitudes", in Advances in Consumer Research, ed. by Mary Jane Schlinger, (Chicago: ASsociation for Consumer Research, 1975), p. 405-425. 34 in the authors' paper. Generally these forms of validity test the predictability of each Specific model across various samples from one or more populations. The conclusions of this research support the inclusion of situation specific components in predicting behavioral intentions. In addition it appears that a disaggregated format in the model again tends to enhance predictability. While this study utilized variates derived from a factor analysis of product related items, the use of a disaggregated form of an attitude model is again indicated. This research also concluded that in all three models, "attitudes were effective predictors of behav- ioral intentions."26 The cited research in general concludes that while not all models perform equally in a specific situation, atti- tudes toward a purchase object or act may be used as pre- dictors of behavioral intentions. Whether Specifically in- cluded in a mathematical sense or not, situation specific factors are generally agreed to influence both the purchase intention and purchase behavior. These conclusions strongly suggest designing attitude research which rec0gnize varying situational factors in the purchase decision. These factors may be in the form of applications for the product, circum- stances surrounding the purchase act, random unexpected events ‘which change the purchase environment or others relevant to the specific research problem. The evidence of 26Ibid., p. 422. 35 previous research strongly upholds the inclusion of situa- tion Specific variables in a model of behavioral intention. Market Segmentation Smith's pioneer work with market segmentation was primarily oriented toward consumer products. As a marketing strategy, segmentation has received wide attention in the marketing literature on consumer goods. On the other hand only a small group of researchers have devoted their efforts toward segmentation of industrial markets. The reasons for this diversity is not documented; however, one vieWpoint might render the question moot. Frank, Massy, and Wind maintain that -- "the choice of segmentation as a marketing strategy for industrial goods and services in a domestic market and both consumer and industrial goods in international markets is predicated on the same assumptions and criteria as segmentagion for con- sumer goods in the domestic market." The implication is that the principles of designing different strategies tailored to individual market segments, might be equally applicable in both consumer and industrial goods mar- kets. The problem that emerges is therefore not one of find- ing new ways to plan strategy, but rather new methods of segmenting various types of industrial markets. These au- thors suggest that segmentation be carried out in a two step process. The first step involves segmenting the organiza- tions (buyers) in the market place according to whether or 27Ronald E. Frank, William F. Massy, and Yoram Wind, Market Segmentation, (Englewood Cliffs: Prentice-Hall, Inc., 1972), p. 91. 36 not they may or may not use the firm's product. After this ”initial screening" a more detailed analysis of the purchase characteristics of potential customers is performed in seg- menting the market. Cardozo in his survey of the literature on industrial market segmentation found "only six sources which carried the concept of segmentation beyond end use and geography."28 The bases which he found were: 1) The type of buying situation. 2) The phase of the decision process. 3) The primary role of the purchaser and his commitment to it. 4) The purchasing strategies employed by different buyers. 5) The interest of, or problems faced by different industrial buyers. and 6) The self confidence of particular buyers.29 In a later piece of research, Cardozo and Cagley demonstrated that "industrial purchasers held clear preferences for types of bidders and bids, responded to the amount and type of risk in the purchase situations, and exhibited identifiable behavior patterns which cougg form the bases for segmenting industrial markets." 28Richard N. Cardozo, "Segmenting the Industrial Mar- ket", in King, p. 433. 291bid., pp. 433-434. 30Richard N. Cardozo and James W. Cagley, "Experi- mental Study of Industrial Buyer Behavior", Journal of Mar- keting Research, Vol. VIII, No. 3, (August, 1971), p.432. 37 What this research suggested was an approach to delineating market segments that did not utilize merely demographic in- formation on buyers such as age, income, education, etc. A set of situation specific characteristics is rec- ommended by Frank, Massy, and Wind as one choice of segmenta- tion bases for organizational buyers. Borrowing from the Webster and Wind model of organization buyer behavior; these authors suggest such bases as the composition of the buying center; the buying situation; the attitudes, perceptions and preferences of the buyer towards alternate sources of supply; and the determinants of the buying decision.31 The attitude and decision determinant profiles for market segments delin- eated by the prior dimensions would serve as a direct link between the target segments and design of market strategies directed to them. It has been noted above that Bennett and Scott found significantly different value importance profiles for mili- tary and industrial users of a product. This would lead to a research design for segmentation which accounted for dif- ferences in product use by buyers as suggested by Frank, Massy and Wind. The former study investigated: "Whether the structure of the relationship between brand appeal and instrumentality remains the same across segments where there are significant dif- ferences in perceiveg importance of the attributes among the segments." 2 31Frank, Massy, and Wind, pp. 98-101. 32Bennett and Scott, p. 347. 38 The results indicate that a difference between segments did in fact exist. The observed difference lends SUpport to the proposition that brand preferences are more predicatable by perceived instrumentality scores after segmentation. This would indicate that profiles of preferences, attitudes, and perceptions are relevant bases for segmenting industrial markets. Furthermore within a type of purchaser delineated by demographic variable, several importance and attitude profiles may even exist. By first segmenting on importance or attitude profiles, a strategy might be more accurately designed for the target segments. Another study of market segmentation, although not in the industrial goods area, also demonstrated the viability of using attitude profiles as segmentation bases. Cunningham and Crissy in their investigation of market segments for foreign and American compact automobiles, found that demOgra- phic and socioeconomic variables could effectively be aug- mented by attitudinal and motivational variables.33 Lehmann and O'Shaughnessy were also able to Show that different types of buyers attach varying degrees of impor- tance to product and supplier attributes in the act of se- lecting a supplier of industrial products. While their re- search was not directly focused on market segmentation, the evidence points toward attribute importance profiles as a 33William H. Cunningham and William J. E. Crissy, "Market Segmentation by Motivation and Attitude", Journal of ' Marketin 'Research, Vol. IX, No. 1, (February, 1972), pp. 100- 02. “c—I1 I-.. Q. -'—fl‘_-n , :v‘oag..- w 39 segmentation base. This study concentrates primarily on the purchaser's perception of a product utilization situation. In defining the four product types Studied the authors em- ployed a classification scheme which entailed the buyers perception of "problems likely to be encounted if the pro- duct is purchased."34 This study showed both that the im- portance of several attributes was Significantly different across product types and that the two groups of purchasers varied Significantly in the amount of importance attached to several attributes by product type. Existing research, in particular that cited above, indicates that non-demographic bases are viable for segment- ing markets for industrial goods. Extensions of this re- search Should attempt to expand the number of types of pro- ducts, buying influences, and purchase situations. The present research will investigate potential non-demographic segmentation dimensions for a single product type. A set of buying influences and purchase decision making stages will be incorporated with application situations for the selected product, maintaining an equivalent functional use. Attitude and preference profiles for the various market segments after being identified will be compared across segments. 34Donald R. Lehmann and John O'Shaughnessy, "Dif- ference in Attribute Importance for Different Industrial Products", Journal of Marketigg, Vol. XXXVIII, No. 2, (April, 1974), pp. 36-42. F" __ 40 Figure 1 summarizes both demographic and non-demo- graphic dimensions for segmenting the industrial market place. When utilizing any of the dimensions an analysis of buying criteria importance profiles should be employed to establish the significance of any single dimension or combi- nation of dimensions for a single market place. A combina- tion of both demographic and non-demographic dimensions might exist for any product market situation. Any dimensions which are therefore relevant based in a difference in importance profiles should be used. In addition combining non-demo- graphic classes with the traditional demographics may lead the researcher to explanations of differences across the latter dimensions. As such, each study of the industrial goods sector must include segmentation dimensions in both cases. Figure 1 SUMMARY OF SELECTED MARKET SEGMENTATION DIMENSIONS Demographic Non-Demographic Geography Buying Situation End Product Use Product Application SIC Category Stage in Purchase Decision Process Sales Volume Buying Center Roles Employment Risk Level in Purchase Product Classes Position or Title of Buyer 41 Physical Distributipn and Marketing Whether termed physical distribution, business 10- gistics, distribution, or materials management, the demand servicing activities referred to by Lewis and Erickson en- compass an area of business management which create time and place utility for the products of the firm. Primarily the objective of physical distribution (hereafter referred to as PD) activities is to have the right product at the right place at the right time in the right condition. While ap- pearing as a straight forward charge, the accomplishment of this objective conceivably involves the management of a myr- iad of activity centers. The National Council of Physical Distribution Man- agement outlines these activities as "...freight transportation, warehousing, material handling, protective packaging, inventory control, plant and warehouse Site selection (and site logg- tion), order processing, and customer service.” From a marketing perspective it is the cause and result of the importance of the PD activity center which is relevant. As Ballou observes: "The activities which are referred to as logistics activities are a consequence of the distance and time gaps between production's location and the point of consumption and the inability or the economic un- desirability of having production output respond 36 instantaneously to the needs of the market place." 35Defined by the National Council of Physical Dis- tribution Management as cited in Donald J. Bowersox, Edward W. Smykay, and Bernard J. LaLonde, Physical Distribution Management, (New York: The Macmillian Company, 1968), p. 4. 36Ronald H. Ballou, Business Lo istics Management, (Englewood Cliffs: Prentice-Hall, Inc., 973), p. 8. 42 When viewed in terms of a national sales grid, the spatial and temporal relationships between buyers and the seller be- come increasingly complex. Thus the orientation of manage- ment towards the PD activities can be an important considera- tion. Because of the unique nature of PD as an interfact r-t activity between production and marketing, the control place- ment of the demand servicing functional area can be a per- plexing decision for top management. As Schiff states: "The logic would suggest that this independent function because of the site of expenditures in- volved, the uniqueness of its functions, and its significance as a kind of balancing mechanism between manufacturing and marketing particularly as it relates to inventory management, would place it in the organization structure on a level equal to Manufacturing, Marketing, Engineering, and Finance.” According to LaLonde, the perspective of top management will determine where the control of the PD function is placed in the organization. He observes that "in a company where tOp management is primarily financial, the distribution function is often viewed as a means of cost reduction. If marketing or sales predominate, the emphasis is frequently on service capability and demand responsiveness." 8 The proper organizational posture for PD would appear to be governed by l) the proportion of total cost accounted for by 37Michael Schiff, Accountin ‘and Control in Physical Distribution Mana ement, (ChiEago: e National Council of Physical Distribution Management, Inc., 1972), p. 6. 38Bernard J. LaLonde, "Strategies for Organizing Physical Distribution", Transportation and Distribution Management, (January, February, 1974), pp. 21-22. 43 the performance Of PD activities and 2) the relative respon- siveness of demand facing the firm to variations in the level of service provided by these activities. The present re- search is directed towards formulation of suitable groundwork for this latter consideration. If PD services are signifi- cant determinants of demand response through brand preference r‘ and purchase behavior, the level of service should at minimum be specified by marketing and included in their strategic planning. Wherever the control of the PD activities is placed in the organization, the knowledge of demand responsiveness of their output is desirable. There are two dimensions measuring the output of these activities which appear to be relevant. As Smykay, Bowersox and LaLonde observe: "A firm's physical distribution capability is meas- ured in terms of speed and consistency. ...a fast delivery cycle is of little value to customers un- less it is Sonsistently met from one order to the next ...."3 One measure of how well PD activities are Operating is the time it takes to place a customer's order into his physical possession. This is referred to as order cycle time, and is composed of all the portions occupied in communicating, proc- essing, and transporting the Order. Although monitoring the levels Of each of those is important from the firm's view- point, the customer sees only one result -- the total time from placement to receipt of his order. He is oriented 39Bowersox, Smykay and LaLonde, p. 14. 44 ummdsthis total time in his evaluation of his supplier. Another aspect of customer service resulting from PD :3 Hm bvel of stock availability. If an ordered item is notawnlable from inventory or if production capacity is chdkaum to another order, the company may be said to be out- Failure to provide availability might result in While of—stock. eifiwrziback order, a lost sale, or a lost customer. the affect on demand is merely temporary in the first in- stanan a lost sale and a lost customer have lasting effects on flwedemand structure. Both lost revenue and Opportunity costcufall marketing activities directed at that customer result. Some firms establish service standards based on both service time and stock availability. An example of such standards is: ”1) the system will be designed to provide 95% in- ventory availability for category A products, 92% inventory for category B products, and 87% for category C products; 2) desired delivery of all customer orders will be within 48 hours48f order placement for 98% of all orders .... Standards such as these are only a result of information pro- vixieci b)’ customer contact in the form of marketing research. Tn estimating service standards, measures that are meaningful o the customer must be utilized. Servicing 98% of all cus- omers within a selected time frame is not relevant to the 4°Ibid., p. 41. “mammmt¢m v .ul-m I I .F. l I 4S shmle customer. He only sees the service with which his or- demsare delivered. As a result he measures the service in tame of the number of orders that he personally receives wiflfin.a given time frame. If he is in the two percent not senmd.98% of the time, it is probable that he may shift his supplier loyalty. Several alternative lists of physical distribution services are available in the literature. One which is com- plete and appears to be customer oriented is proposed by Willett and Stephenson:41 1) Order cycles length: The time elapsed between placement of an order and receipt of goods, de- fined in terms of specific customer's expectations based on a history of orders from a supplier and/or a supplier's guarantee. 2) Consistency of order cycle length: The degree of variation in the lengths of a history of order cycles from a specific supplier to a Single cus- tomer. This variable is measured in terms of the absolute deviation from the mean of a history of order cycle lengths and/or a supplier's guarantee. 3) Order preparation: The way in which orders are formulated and the medium by which orders are transmitted from customer to supplier. 4) Order accuracy: The degree to which items re- ceived conform to the specification of the order. 5) Order condition: The physical condition in which the goods are received. 6) Order size: A service restraint consisting of the minimum size of an acceptable order. 41p. Ronald Stephenson and Ronald P. Willett,"Sel- ling with Physical Distribution Service", Business Horizons, (December, 1968), p. 78. m... . - an! I :F‘-‘ 9-D _-hi . Y 46 7) Order frequency: A service restraint consisting of the maximum frequency with which orders can be placed in a given period of time. 8) Billing accuracy: The degree to which billing is accurate with regard to actual order. 9) Billing efficiency: The degree to which the bil- ling procedure facilitates the customer's handling of accounts payable. 10) Back order: The quality of the supplier's proce- dures for handling back orders. 11) Claims: The quality of the supplier's procedures for handling buyer's claims. The above authors examined the first three factors, claiming that they are the "most potent in terms of their in- fluence on demand."42 A study was conducted that measured 1) reorder cycle service times received by retailers on comparable orders 2) retailers ratings of service times, and 3) conditions under which orders were placed. The results indicated that ”ratings of satisfaction with service received were a linear function of service time."43 1\ measure of customer response to the time dimension of PD :service was established in this research. Ballou and DeHayes found that consistency of service :is inore important to customers as a differentiating factor tflran.is pure Speed. This study indicated that customers are 421bid. 43Stephenson and Willett, "Determinants of Buyer Ileaslaonse to Physical Distribution Service", J0urnal of Mar- keting Research, Vol. V, No. 3, (August, 1969), p. 279. IT. 47 inclined only to alter their order size in response to changes in average delivery time.44 Both of the cited studies indi- cate that a change in the level of PD service can influence the demand level for a particular product. They do not how- ever Study the responsiveness in demand when PD service is varied in relation to other product and/or supplier attrib- utes. Various other factors which made up part of the pur- chaser's buying criteria are such things as price, quality of products, reputation Of the selling firm, service on repair or adjustment for faulty products, etc.. These factors should be included in a study aimed at designing market segmentation strategies. Several other research efforts have attempted to Show the relative positions of a general list of product or supplier attributes in terms of importance and determinance. These studies lead to assumptions or hypotheses about the universe of criteria or factors which should be included in the set of variables for an industrial product market. In his study of the factors which industrial buyers considered most important Klass listed the following: 1) maintaining quality consistent with specifications; 2) on time delivery performance; 3) honest and sincere attitude on the part of the salesman; 4) price; 44Ronald H. Ballou and Daniel W. DeHays, Jr., "Trans- port Selection by Interfirm Analysis", Transpgrtation and Distribution Management, (June, 1967), pp. 33-40} 48 5) keeping buyers informed of new product and product development; and 6) effective handling of requests for samples and in- formation.45 'ste factors were considered most important by a general cupss section of buyers. The presentation of this research dhinot however specify the importance ratings by buyer type (e.g. - purchasing agents, managers, engineers, etc.). What is relevant from the research are the relative positions of ”WA. A. ———-—.——1’ I product quality, delivery, price, and sales related factors. It may be concluded from this work that marketing factors as well as product quality rank among the most important for industrial purchases. Dickson conducted a study of relative importance of 23 product and supplier related factors rated by purchasing personnel. He concluded that: "the ability of potential vendors to meet quality standards and delivery schedules, stand out as the two most Egitical factors in the vendor selection I! process. 1\ Similar study by Wind, Green and Robinson found the quality- pxrice ratio and delivery reliability to rank substantially 45Bertrand Klass, "What Factors Affect Industrial Buy- :irig’iDecisions", Industrial Marketing, (May, 1961), p. 34. 46Gary W. Dickson, "An Analysis of Vendor Selection Systems and Decisions", Journal of P'ur‘chas‘i'n , (February, 1966),p.9. 49 above all other factors in the vendor selection process/17 Although this research studied only the importance ratings of ten product and supplier attributes, some important conclu- sions from a methodological perspective are relevant. The respondents, in addition to being presented with a list of all attributes to rate singly, were also presented with. groups of attributes (three (3) at a time). The conclusions from com- paring the results were that with a large number of attri- butes a linear model of the ratings without interaction terms was a good measure of overall performance. When a small num- ber of attributes was presented, the interaction terms were important.48 When Bennett and Scott examined the importance of supplier attributes across market segments, they found the ordering to be significantly different between segments. Their research concluded that an analysis "conducted across total markets where there may be inter-segment differences in a ttribute importance”9 might seriously affect results. Four- fifths of the most important attributes changed relative positions across the two segments studied. These attributes 47Yoram Wind, Paul E. Green, and Patrick J. Robinson, The Determinants of Vendor Selection: The Evaluation Func- Lon Approach", Journal of Pugrchasipg, (August, 1968), ). 29-41. 48Ibid. 49Bennett and Scott, pp. 346-350. 4 "‘fii‘m '” a- 1 Huh‘ .. 'Td‘ 1F’._4¢ -u 50 were reliability, non-flammability, quality, and load life.50 The rating by the total market and the segments of delivery lead time was not specifically reported. In their study of the importance of product and sup- plier attributes by product type, Lehmann and O'Shaughnessy also found a variation in the average attribute importance across segments. For two of the four product purchase types, reliability of delivery was ranked most important and fourth most important by the remainder.51 Price was second, eighth, eighth, and first across all four types. The evidence sug- gests once more that importance profiles change across seg- ments and further that a PD related factor (delivery relia- bility) was not only rated highly but changed in rating across the defined product types. In summary, these studies demonstrated two important points. First, PD services rank highly as important supplier selection criteria in industrial markets; and additionally, that some purchasers consider consistency of delivery more important than delivery time. Secondly, the relative impor- tance may change across segments of the market. The present research attempts to extend the findings of the existing re- search, in the following ways: 1) Consider more than one element of PD service in relation to other product and supplier attributes; and solbid. 51Lehmann and O'Shaughnessy, pp. 36-42. 51 2) Measure the determinance of the various supplier selection criteria, especially those related to PD, across relevant market segments. The remaining chapters will present a research meth- odology and the results of an empirical study aimed at ex- tending the existing research in the industrial market seg- mentation literature. The design will incorporate portions of the industrial behavior models discussed in the beginning of this chapter. The results of a preliminary investigation of a Specific product market will be presented from the per- spective of these models. Further the research design will utilize the contributions of the research cited above in measuring attitude profiles of market segments in the selected product market. CHAPTER III RESEARCH METHODOLOGY Introduction Chapter Three presents a research design intended to guide the collection and analysis of empirical information relevant to the role of physical distribution service fac- tors in the industrial purchase decision. A specific indus- trial market was selected for study which was closely aligned with the characteristics desired for the investigation. These characteristics, as alluded to in the previous chapter, were: 1) A multiplicity of purchase decision making influ- ences; 2) A multi-stage purchase process; 3) A set of purchase decision criteria by which sup- pliers are evaluated; and 4) A set of various purchase Situations. The market place selected for study was the commer- cial and industrial air-conditioning industry. As will be discussed below, this industry exhibited the characteristics desired for the study. The research was Sponsored by a mem- ber of that industry whose identity will not be disclosed. As a result this research was designed to help meet the sponsor's needs as well as the academic interests of the re- searcher. The research design consisted of a two phased ap- proach to the problem. To reiterate, the central problem of the research was to identify relevant market segments for an 53 industrial product, and further measure the determinance of physical distribution service factors in the purchase deci- sion for that product. Phase I of the research was an ex- ploratory effort designed to aquaint the researcher with the industry under study, and certify the existence of the de- sired characteristics (as listed above). The primary focus Of Phase I was determination of the relevant buying influ- encers, a set of purchase situations, a universe of decision making criteria utilized by purchase decision makers, and finally the sequence of events in the purchase process. At the end of Phase I, the findings were compared to the theo- retical concepts detailed in Chapter II. The detailed plan of Phase I will be presented below in conjunction with the results of that exploratory effort. Phase II of the research was an empirical study aimed at verifying the relationships which were hypothesized to exist in the market place. A set of research hypotheses was tested based on Phase II data. These hypotheses, generated as a result of the exploratory findings in the first phase were tested at the end of the second phase. The analysis plan, explained in a later section of this chapter, was bas- ically designed to accomplish two tasks. First, the statis- tical difference in buying segments was evaluated on the basis of the importance of the selected purchase criteria. The primary tools employed in this analysis were factor anal- ysis. Second, an analysis of the predictability of supplier preferences based on attitudes towards suppliers on the part S4 of purchasers was conducted. The latter analysis also utiliz- ed the discriminant analysis technique. In the sections which follow, the research design and corresponding techniques are presented in detail. The first section presents the design and results of the Phase I ex- ploratory investigation. In the succeeding sections the de- sign Of Phase II is outlined. These sections include the sampling design, the construction and administration Of the data collection instrument, and an explanation of the data analysis techniques. The results and findings of the Phase II validation study are presented in Chapters IV and V. Phase I — An Exploratory Study The first phase of the research was designed to help familiarize the researcher with the market chosen for empir- ical study. In the process of familiarization, the desired research characteristics were also examined with respect to their existence in the study market. Thus the exploratory phase was to have as its output: 1) the determination of whether or not a multiplic- ity of purchase influences existed; 2) the identification of the stages which existed in the purchase process; 3) the generation of a universe of supplier selection criteria; and 4) the identification of "typical" purchase situa- tions which might affect supplier choice. SS Exploratory Design Phase I data was collected through a judgemental sam- ple of air-conditioning equipment purchasers in four mid- western cities. These cities were Lansing, Michigan; Detroit, Michigan; Chicago, Illinois; and Dallas, Texas. A selection of respondents was made in each city based on the size of the organization and the perceived familiarity of the re- spondent organization with the purchase process involving the study market. Personal interviews were conducted with 60 buying organizations in the four cities mentioned above. The organizations interviewed in this phase were: 1) Architectural firms; 2) Consulting Engineering firms; 3) Mechanical Contractors; 4) General Contractors; 5) Sheet Metal Contractors; and 6) Building Owner/Occupants. These organizations appeared to contain the full range of buying roles as perceived by Webster and Wind in their model of organizational buying behavior. The proposed roles were users, influencers, buyers, deciders, and gate- 1 Users are the individuals who come in physical keepers. contact with the product either through handling in the pro- duction process (component parts or raw materials) or through the consumption of the product or its service in the process of their work activity. Influencers are those who may or may not come in direct contact with the product or its service, 1Frederick E. Webster and Yoram Wind, Or anizational Bu in Behavior, (Englewood Cliffs; Prentice HaII, Inc., 19725, p. 77. 56 but provide input into the purchase decision making process. Buyers are those who are responsible for making the actual purchase and have the formal authority to represent the buy- ing organization as the purchaser. Deciders make the buying choices in relation to the selection of suppliers and/or al- ternative sources. Finally, gatekeepers are the controllers of information inflow and outflow between the buying organi- zation and suppliers, both potential and actual. The above models propose that these roles may exist Simultaneously within an individual's preoccupation in the buying organiza- tion. Thus, a single individual, or organization for that matter, may act in more than one role. For example, the building owner who has A-C equipment installed may be a user, influencer, and decider simultaneously, while he may not per- form the actual buying role. An effort was made in the exploratory phase to iden- tify which roles were predominately performed in each type of organization listed above. Figure 1 presents the buying center role relationships that were believed to exist in the purchase process. As is illustrated, all of the organization types play gatekeeper and influencer roles. The problem in the exploratory phase was to determine which organization had the greatest impact on the final purchase decision. It was also hypothesized that one organization could reflect the buying roles of another through normal interaction in the decision making process. Thus, one organization would tend to represent another in the actual decision. While the 57 Figure l BUYING CENTER ROLES IN A-C PURCHASE PROCESS Architectural Firms Engineering Firms Gatekeeper Gatekeeper Influencer Influencer Decider Decider Buyer (sometimes) Buying MechaniCal Contractor Decision General Contractor Gatekeeper’/////’/7 a\\Gatekeeper Influencer Influencer Buyer User Building Owner Sheet Metal Contractor Gatekeeper Gatekeeper Influencer Influencer User User Buyer (sometimes) IF-J ....t-‘ .k‘...’ .' ‘9‘..- m 58 purpose of the research was not to determine the interrela- tionships in the purchase process, some understanding of these was deemed necessary to understand the overall problem. The specific findings with respect to this will be presented in the next section in conjunction with a more detailed explan- ation of the nature of the interrelationships. Within each respondent organization, contact was made with a principal operating officer. This individual ranged from the president or owner of the organization to a general manager. It was felt that an individual of this type could accurately represent the organization's role in buying, if not directly reflect his own involvement in the purchase process. In many cases, particularly in smaller organiza- tions, the chief Operating officer is also the primary pur- chasing Officer. Each interviewee was approached in the eXploratory phase with a series of Open-ended questions about the Overall purchase process for the study product group. He was ini- tially informed that the researcher was attempting to under- stand how purchases of A-C equipment were made. At the beginning of the interview, the individual was asked to describe how purchases were made and how he perceived his role in the overall process. Once he had identified a point where a decision relating to equipment suppliers had to be made, he was asked about the variables which were used to make the decision. The events which followed his decision were then traced. Next, the interviewee was asked to place ._ A A... A-“-fl.¢~—_m 59 an importance rating on the selection variables. Finally he was asked to indicate whether or not the importance attached to each variable could vary depending upon the specific job type or application. At this juncture, he was asked to ex- plain the various applications Of a single equipment model and how his decision making might vary. Analysis and Summary of Phase I Data All interviews conducted in the first phase were con- tent analyzed to deve10p information for the second or vali- dation phase of the research. Several information items were needed from the phase one content analysis. These were: 1) A description of the purchase process 2) A universe list of the variables or criteria used to make buying decisions 3) A list of equipment applications and/or job types 4) A description of the types of buying organizations involved in the purchase process, and 5) A description of the roles of each buying organi- zation in the process. Purchase Process and Job Types The purchase process for the study product was read- ily perceived to be multistaged. The stages in the process were perceived as separately identifiable by the respondents. While the names of the stages varied somewhat the activities or functions occurring at each stage were nearly identical in all cases. Webster and Wind in their model of organizational buying behavior suggest five stages in the purchase decision n‘nre'l. ~cv-fi .. 1‘.- 60 making process. To reiterate these stages are l) identification of need 2) establishing objectives and specifications 3) identifying buying alternatives 4) evaluating alternative buying actions, and 5) selecting the supplier.2 From the exploratory interviews seven distinct stages were identified. These stages and the activities taking place at each one are detailed in Table 1. Figure 2 illus- trates the similarity between the stages identified in the research and the Webster and Wind model. By comparing the two descriptions it is evident that the market selected for study coincides with the theoretical behavior model from the literature. The seven stage process was utilized in phase two in determining the level of decision making involvement and control by buying group at each stage. In what might be classified as a "traditional" buying process, the seven stages occur in chronological order with the various buying groups or organizations becoming involved to different degrees in each stage. During this phase most respondents indicated that new job types are evolving in which the patterns of decision making involvement are chang- ing. The primary job type, which most industry respondents term "traditional", is the plan and spec category. This job type follows the seven step process sequentially, with 2Ibid., p. 31. 61 Table l BUYING PROCESS STAGES AND ACTIVITIES Stage Activity 1) Building Conception At this stage the idea for con- struction is generated con- sisting of the potential use for the building, the type of occu- pants, who is to own and finance the building, its approximate Size, method of construction, I etc. 22) Preliminary At this stage, the overall de- Investigation Sign constraints are usually established. Budget, time 1 horizon for construction, pro- L. posed occupancy date, feasibil- ity of construction, preliminary design plans, environmental needs, etc. 15:) Design of A-C At this stage the engineering System Needs parameters which will later determine the Size and type of A-C equipment are established. The environmental constraints (temperatgre and humidity) are converted to system specifica- tions in conjunction with pro- posed uses of the building. 4*) Specification of At this stage the exact speci— A-C Equipment fications for A-C equipment are established. These may include ductwork, piping, and air reg- ister measurements; pump and fan capacities and ratings; power specifications; and weight and size. 3Gilbert A. Churchill, Jr., Marketing_Research: Methodological Foundations, (Hinsdale: The Dryden Press, 1976), p. 263. Stage .5) Bid Proposal Solicitation 6) Bid Proposal Evaluation 77:) Contract Award 62 Table 1 (continued) Activity At this stage the specifications from the preceeding stage are distributed to manufacturers and installers of equipment for bids giving the cost and ability of the supplier and/or installer to accomplish the physical task of placing the system and making it Operational. At this stage the bids received above are reviewed pending se- lection of a supplier. Ideally the bids are rated in relation to the specifications and needs established in the early stages of the process. At this stage the final selec- tion of the supplier is made and an agreement to purchase signed or a contract let to the spe- cific supplier whose bid was acceptable. ..a‘4 '_ .P—i .q \ 1) 2) 3) 4) 5) 6) 7) 63 Figure 2 COMPARISON OF STUDY PROCESS AND WEBSTER AND WIND MODEL Study Building Conception Preliminary Investigation Design of A-C System Needs Specification of A-C Equip- ment Bid PrOposal Solicitation Bid Proposal Evaluation Contract Award 1) 2) 3) 4) 5) Webster and Wind Identification of need Establishing objec- tives and Specifica- tions Identifying buying al- ternatives Evaluating alternative buying actions Selecting the supplier . '—.v_-n 'umfi l lr‘m-Mli' 64 engineers, architects, and building owner/occupants being in- volved primarily in stages one through four. In stage five through seven the contractor group typically enters the proc- ess, with the engineering group again influencing decisions in stage Six. The contractor typically will actually make the equipment purchase and arrange for delivery to the con- struction site. Several mentions were made Of three "new" job types during the Phase I investigation. These were 1) Negotiated-Team Managed - a job type where the owner, architect, engineer, and contractor form a team and work together through all Stages Of the purchase process. Irvin—Lana...— ”New i“);- u-‘“ 2) Design Build - where either the engineer becomes the primary contractor for the job and subcon- tracts the equipment installation to a specialist contractor or the contractor does his own design work (specification, etc.) and hires an engineer- ing firm to accomplish this task. 3) Owner Prepurchase - where the building owner/occu- pant Specifies (sometimes with the help of an en- gineer) and purchases equipment and the contractor only provides installation. Most interviewees expressed the fact that their in- I>1Its to the final purchase decision were not the same for all Li<>t> types. This became the basis for a test of differences 3111 'both decision making involvement and importance rating of 't}1€> buying criteria during the validation phase of the re- search. Mng Groups Several types of organizations were interviewed to determine which ones were most relevant to the decision mak- JLT‘E! jprocess. As stated above these organizations were: 65 1) Architectural firms; 2) Consulting Engineering firms; 3) Mechanical Contractors; 4) General Contractors; 5) Sheet Metal Contractors; and 6) Building Owner/Occupants. In questioning all organizational representatives about their involvement in the purchase process for commer- cial and industrial air-conditioning equipment, the conclu- sion was reached that mechanical contractors and consulting mechanical engineers (a subset of GrOUp 2) were the most P—m.r ...-fl - -q . . ‘l 1 relevant buying groups. The orientation of both architects 21nd building owner/occupants tends to be represented by the terigineer group. In the typical working relationship, it is 'tlie engineer's function to design, specify and approve all Ineechanical equipment, including air-conditioning. The ar- czllitectural firm either employs a full time engineering staff (311‘ retains an outside engineering firm to perform the design, ssIDecification, and approval functions. Similarly, the mechanical contractor represents the C>lfientation of both the general contractor and the sheet metal contractor (if one is involved). He is responsible for ‘tlle: installation of any mechanical systems utilized in the t>11:i.21ding. As such, he may subcontract or work in parallel vVith related trade contractors such as sheet metal and elec- til‘itcal. The air-conditioning contractor is a Specialized 3EC>IWn of mechanical contractor who is involved with air-con- dj~tlioning equipment installation only. The two primary buying groups which were therefore l(lentified are mechanical contractors and mechanical 66 engineers. Figure 3 illustrates the relationships among all buying groups involved in purchasing. Nearly all interview- ees concurred that these two buying groups were the most significant influences in the purchase decision. Several architects, building owner/occupant, and general contractors explicitly stated that an involvement in mechanical systems was relegated to those having practical expertise in that area; specifically these were mechanical engineers and me- chanical contractors. FF-“ mt. ' --.-‘._n- In", .Purchase Decision Criteria The criteria used by purchase decision makers to ervaluate suppliers, were also identified in the preliminary Ioliase of the research. Each individual was asked to list ‘tlle criteria that he used to evaluate suppliers and equip- :nleent. From the analysis of the personal interviews a uni- xreerse list of criteria was generated. Each respondent was asked to rank or rate each cri- tzearia in terms of its importance to him in making the final Cieicision. Secondly he was also asked if he could rate vari- ous suppliers on each of the criteria and if these ratings wOuld be different for each supplier. At this stage nearly all respondents felt confident that they could rate suppliers 3iif' asked to do so. Few respondents felt that all suppliers :r;11:e equally on the majority of the criteria. The objective ‘DIE' this series of questions to the respondent was the deter- mination of whether or not a multi-attribute model could be appliedto the supplier selection process for the study 67 Figure 3 RELATIONSHIPS OF BUYING GROUPS FOR COMMERCIAL AND INDUSTRIAL AIR-CONDITIONING EQUIPMENT Building Owner/ Architect occupant \ / Mechanical Engineer A-C Buying Decision Mechanical Contractor I\ (Seneral Contractor Sheet Metal Contractor Electrical Contractor 68 product. It was concluded that the multi-attribute approach was logical for this form of purchase process. The criteria listed in Table 2 are those which were mentioned by one or more respondents in the exploratory phase. In Phase II, as is detailed below, a total of 19 cri- teria were used. These criteria were the result of combining two or more of the original criteria, and eliminating others. The final selection was made by the research sponsor and re- searcher agreeing on variables of common interest. All of ‘the variables may be grouped into areas. These areas relate t<> product, sales service, distribution, and cost (first and <31)erating). In the second phase the criteria were grouped ss1:atistically through the use of factor analysis. Iteesearch Hypotheses From the exploratory phase of the research, a set of Iljrjpotheses were developed for testing in the second or vali- dation stage. Three research hypotheses were to be tested 1111 this stage. These were: 1) Physical distribution service characteristics are important criteria utilized by purchasers to evaluate suppliers; 2) Suppliers are rated differently in their ability to perform by purchasers and therefore these characteristics are determinants of supplier selection; and 3) The level of determinance varies depending upon the specific purchase Situation and buying influ- ence center. The research was designed to validate these hypoth- eSes with empirical information, gathered from the market for (:CDHRDJercial and industrial air-conditioning equipment. WNOU'I-thH 69 Table 2 SUPPLIER SELECTION FACTORS Price-first cost . Operating cost . Maintenance cost Installation cost . Ease of installation . Equipment reliability Equipment construction . Equipment size . Equipment weight . Ease of maintenance . Life expectancy of equipment Space required for maintenance Noise and vibration level . Availability of pre-wired control panels . Regular contact by salesman Established relationship with salesman . Assistance in design Assistance in startup . Assistance in writing specifications . Availability of salesman with hours to help with problems Regular catalog updates by personal call by sales- man Factory service in first year of operation . Availability of parts within 36 to 48 hours Catalog descriptions of equipment specifications Catalog descriptions of installation specifica- tions . Delivery time (average) Consistency of delivery on past jobs . Delivery expediting capability of manufacturer Back order response of manufacturer . Availability of parts and service on a nationwide basis . Availability of parts and service in the locality of the project (30 miles) . Manufacturer's guarantee or warranty 70 A series of subhypotheses were also needed for the research. These hypotheses referred to the dimensions along which the determinance of the set of physical distribution service factors was expected to vary. Specifically, the set of subhypotheses was: the determinance of physical distribution factors is different for - 1) mechanical contractors versus engineers (buying influence center 2) plan and spec versus design build - team managed jobs (job application) 3) commercial versus institutional jobs (job appli- cation) 4) rooftop versus chiller (product type). The problem became one of evaluating whether or not a difference existed along any of the above dimensions. If a difference was found (i.e. the level of determinance in each segment varied) the conclusion would be that varying emphasis should be placed upon the logistics service factor depending upon the target market segment. That is to say the strategic plan Should recognize differences in market seg- ments and adjust those variables on which differences are significant to each segment. The research methodology that follows details the study design that was formulated to test the research hypoth- eses stated above. 71 Sample Design The sampling process which provided statistical in- formation for the research consisted of three stages. First, the research population was defined and enumerated. This stage involved a description Of the pOpulation which led to a determination of who was and who was not a member of the statistical universe. Next, a sampling method was selected. This required the selection of a means to choose a sample from the above universe which was both descriptive of the population and manageable from a size point of view. The determination of sample size was also part of this Step. Finally, the population was enumerated and the actual sample was selected. This step also included the collection of per- sonal contacts within sampled buying organizations and cur- rent addresses of the organizations. Each step in the sam- pling process will be discussed below. Sample Frame The definition of the sample frame outlines the boundaries of the research population about which inferences may be directly made from the sample information gathered. For this research the population was defined to in- clude only those individuals who influenced the purchase of a product in this category. The exploratory phase helped define the research population. AS described in the previous section, this phase indicated the market for commercial and industrial air-conditioning equipment that fell within the guidelines used to select an industrial product market. That 72 phase also revealed two primary purchase influence centers in the buying process, mechanical contractors and mechanical engineers. In the mechanical contractors group are all firms performing work on any system within a building related to water distribution either for heating or cooling purposes, sanitary systems, or environmental control systems. Mechan- ical engineers are firms who are primarily involved in the design and Specification of the above systems. Within each of these organizations several individuals were identified who influenced the purchase decision. For purposes of the research, an individual within each organization who repre- sented the overall purchase behavior was further identified. The exploratory research revealed that the principal officer of the organization adequately reflected that firm's purchase decision making behavior and process. Only one official in each organization received a data request. The definition Of the sample frame was: The principal executive officers of all mechan- ical contracting and all consulting engineering firms in the United States. This universe was not totally enumerated prior to the selec- tion of the sample. The justification for this choice will be explained below in the section on sample selection. Alternative Sampling Methodologies Many sampling methods are available for use in a re- search product such as this one. The broad classification of sampling technologies presently used is nonprobability versus probability sampling. Within both of these categories 73 several subclasses of sampling techniques exist. Figure 4 shows the alternative sampling methodologies.3 Figure 4 SAMPLING METHODOLOGIES Sample Designs l 1 Nonprobability Samples Probability Samples Convenience Simple Random Judgement Stratified Quota PrOportionate Disproportionate Cluster Systematic Area Source: Gilbert A. Churchill, Jr., Marketing Research: Methodological Foundations. Of the two broad alternatives, a probability sampling methodology is the most valuable to the researcher because he can attach an estimate to the sample element that it will be a part of a given population. This allows the researcher to apply statistical methods in estimating and testing hypoth- eses concerning population variables or parameters. Two potential sources of error may be present in any sampling plan. These are systematic error and experimental error.4. Systematic error is the difference between the true pOpulation parameter or property (in the case of non-para- 3Gilbert A. Churchill, Jr., Marketing Research: Methodological Foundations, (Hinsdale: The Dryden Press, 1976), p. 263. 4Paul E. Green and Donald S. Tull, Research for Mar- ketin Decisions, (Englewood Cliffs: Prentice-H511, Inc., 1975), p. 213. 74 metric statistics) which results from the process by which the data pertaining to the population is collected; it is reduced by careful preparation of the research objectives, hypotheses, sampling designs and procedures, analysis and inferences from the analysis. The second error type, experi- mental or sampling error, results from the selection of a sample which does not accurately reflect the true population parameter. Several methods are available to reduce sampling error. One means is an increase in sample size. This in- cludes more members of the total population and therefore increases the probability that the sample estimate and the parameter will be equal. Another method which aids in re- ducing experimental error or at least in understanding it, is the use of probability sampling. As shown in Figure 4, the types of probability sampling designs are 1) Simple Random 2) Stratified 3) Clustered Churchill5 explains these sampling designs quite ex- tensively. In addition the determination of sample size and methods of simple estimation are also presented. A fourth design exists called multistaged sampling.6 This design uses a random selection process at more than one level or stage and may include two or more of the above designs. For SChurchill, pp. 268-297. 6Green and Tull, p. 227. 7S example, in a study gathering data on market segments based on income from the entire United States, several cities might be randomly selected as in a clustered sampling technique, respondents might be stratified by income level, and the sam- ple selected randomly from each income strata. The sampling design used for this research combined all three of the basic sampling techniques in a multistage plan. Because the research pOpulation for the validation phase was defined as all mechanical contracting and con- sulting engineering firm representatives in the United States, a wide geographic sample had to be selected. Therefore, the sample was drawn from the entire population using a multi- stage sampling design. Two considerations influenced the design of the sampling plan; the economics of sampling the entire country and the potential variation in attitudes based upon geographical location and organization type (contractor- engineer). It was impossible to identify a particular con- tractor or engineer with a job application or type prior to contact with the respondent. As a result, job type and application was not included in the sampling design. To reduce the potential systematic error in the sampling design, the entire pOpulation was stratified on the first level by geographic area. The geographic regions se- lected were 1) the northeast 2) the southeast 3) the midwest 4) the far west. The justification for geographically stratifying the sample 76 lay in the existence of varying needs and uses of the study product by region. Within each region a cluster sample of cities was selected. This selection was made on the bases of judgement on the part of the researcher and the sponsor. The cities selected were considered typical markets from two points of view. First, the usage of air-conditioning equipment was considered extensive and the number of users or buyers was large. A total of fifteen cities was selected for the sam- ple. The list of regions and corresponding cities from which the sample was drawn is presented in Table 3. Table 3 SAMPLING REGIONS AND CORRESPONDING CITIES Region City Northeast New York, New York Boston, Massachusetts Washington, D. C. Southeast Atlanta, Georgia Miami, Florida Fort Lauderdale, Florida Midwest Chicago, Illinois Milwaukee, Wisconsin St. Louis, Missouri Dallas, Texas Fort Worth, Texas Houston, Texas Far West Phoenix, Arizona Los Angeles, California San Francisco, California The subpopulation of all mechanical contractors and engineers in each of the cities listed in Table 3 was enumer- ated next. Finally a random selection of both mechanical contractors and mechanical engineers was made in each city 77 from the subpopulation listing. Each contractor and each en- gineer in all four regions was assigned a number. A total sample was selected at random from each region independently. Thus eight independent samples were selected across all four regions and the contractor-engineer groups. The population enumeration in each city was supple- mented with listings from two trade journals which survey member activity on an annual basis. The two publications are Engineering News Record and DE Journal. These supplied ad- ditional firm names that did not appear in the telephone directory. Engineering News Record annually enumerates the levels of sales activity (billings) of the top 500 design- engineering firms in the United States. This listing served as a cross check of the coverage of the telephone directory listings. DE Journal lists the top 200 mechanical con- tracting firms in the United States on the basis of billings. For both groups (contractors and engineers) any firms listed in either publication which were located in or near the four- teen cities and not included in the directory listings were included in the population. Very few cases were found where this condition existed, however. The cross check did provide a verification of the adequacy of the telephone directory method of population enumeration for this type of study. An alternate method commonly utilized in developing population enumerations is trade associations such as local mechanical contractors' associations and professional engineering societies. One 78 drawback exists in using these sources. Not all firms belong to these groups in all cities. Most mechanical contractors' organizations exist for the primary purpose of negotiating in trade union matters for example. Thus non-trade union shops do not typically become members. In some cities, par- ticularly those in Texas and the south, many non-union shops exist. Therefore those sources were ruled out for use in developing population lists. Sample Size, Selection, and Response Methods of Sample Size Determination AS with sampling design, several methods are avail- able for determining sample size. The considerations in sample size computation are twofold: l) the statiscal accuracy of the sample should be optimized by balancing the cost of sampling against the cost of poor information 2) the economics of sampling must fall within the budgetary constraints of the research. A balance between the above considerations is necessary in any practical marketing research application. As the need for statistical accuracy becomes greater, the cost of infor- mation rises while the cost of wrong decision falls. Con- versely, the value of the information gained through in- creasing accuracy may be readily outstripped by the cost of providing that accuracy. Tull and Hawkins7 suggest three specifications which 7Donald S. Tull and Del I Hawkins, Marketing Re- search, (New York: Macmillian Publishing Company, 1976), 186. p. 79 must be made prior to determining sample size. These are: l) the allowable error 2) the level of confidence and 3) a measure or estimate of the standard deviation of population. Once these specifications are made, the sample size required to either estimate parameters or test hypotheses may be de- termined via a mathematical formula. The appropriate choice is determined by the importance of the statistic being esti- mated in relationship to the success or failure of the re- search. Since most research involves several variables and statistics, the one which either is most critical or that appears most often might be chosen. A third method of sample size determination exists, which does not directly involve the use of probability or estimates of pOpulation parameters. When cross-tabulation of data is a part of the data analysis, the number of cate- gories and level of cross tabular analysis may easily require a sample size larger than that needed with the probability techniques referred to above. Several "rules of thumb" exist for determining the appropriate sample size for cross-tabular analysis. The lower limit seems to be five elements per cell 8 in at least eighty percent of the cells. This is suggested as the lower limit when using a chi-square test on nominally 8Sidney Siegal, Nonparametric StatiStics: For the Behavioral Seiences, (New York: TheTMcGraw-HilIiBookTCompany, Inc., 1956), p. 178. 80 scaled data. On the other hand, if parametric statistics are to be compiled means or percentages for example, the lower limit of the cell sizes may be as large as 50 elements. The larger sample Size per cell is suggested when utilizing the normal approximation to estimate probabilities. Considering these limits, a minimum cell size of thirty elements would serve as a guide to determining sample size for analyses involving both parametric and non-para- metric statistics. Several authors suggest a sample size of 30 as the threshold for moving from the sampling distribution of the non-parametric statistic to the normal probability 9 The cell Size re- distribution for statistical analysis. ferred to here is the expected cell frequency for cross tabu- lation. Sample Size Determination The sample size for this research was determined by the latter approach. The cross-tabular sample size estima- tion method was used. One objective of the analysis was to compare responses across several segments in the market. Specifically, the responses Of contractors versus engineers for plan and spec versus design build team managed, commer- cial versus institutional, and chiller versus rooftop appli- cations were desired. If all dimensions were compared simul- taneously, a total of 960 (2 X 2 X 2 X 2 X 2 X 30) elements 9See Sidney Siegel, Non-Parametric Statistics; William Mendenhall, Introdpction to PrOBability and Statis- tics, and Dick A. Leabo, Basic Statistics. 81 would have been necessary in the sample. Since the total pOpulation of interest in the sample frame was less than the required size for a five level comparison, the decision was made to reduce the number Of Simultaneous comparisons. The final sample Size for the study was set at a minimum of 120, 30 per cell in four cells, a two by two classification com- parison. Given equal prior probabilities of group membership, this minimum fell within the constraints of the methodology discussed above. Sample Selection The actual selection of the sample was accomplished through a step-wise process. After the minimum sample size was determined, an enumeration of the population was done to determine the proportion of the sample that would come from each geographic and organizational strata. Since it was recognized that the response rate for mail interviews usually runs between 20 and 40 percent,10 more than 120 questionnaries had to be placed in the field. A conservative estimate of the expected response rate was set at 25 percent. With this figure, at least 480 questionnaires had to be placed into the field. Table 4 shows the total mailing in each region by mechanical contractor and engineer and the response rate of usable returns. The slight difference in percentage of mailings over total pOpulation versus the target percentage (480/882) 10Green and Tull, p. 152. 82 resulted from the inadequacy of some addresses for popula- tion elements and total failure to reach the remaining twelve elements. Only 468 mail interviews were attempted. AS shown in Table 4, the response rate was significantly higher than expected, yielding a total usable sample size which was larger than the minimum established. Table 4 BREAKDOWN OF POPULATION AND SAMPLE SIZE BY REGION (BY CONTRACTOR AND ENGINEER) . Population Usable Region 'Size Mailing . Responses MC E MC E MC E Northeast 111 189 62 68 20 27 Southeast 78 102 34 46 l4 13 Midwest 148 106 88 62 26 18 Pacific 80 68 52 56 29 lg Total 417 465 236 232 89 77 (37.7%)(33.2%) To select and contact the potential respondents, the population was enumerated city by city in each region. A number was assigned to each organization. The organizations were then chosen randomly until the needed size was selected in each region for both groupings (contractors and engineers). Once the organizations were chosen, a personal contact was established in each one. Two methods were utilized for this information. The field sales force of the sponsor provided the bulk of the personal contacts and the remainder were made through trade publications or local association rosters where available. The names of the chief executive officers of all 83 sample organizations were successfully obtained by these methods. Three mailings were made to generate the total re- sponse. An initial mailing was made to all selected recip- ients. After a two week period, a second mailing was made to those organizations selected but not responding to the initial mailing. A coding system was used to determine the organizations which had not responded. A third and final mailing was made three weeks after the second mailing, only to those organizations which had persisted in not responding. After the final mailing, many non-respondents were contacted by telephone concerning the interview. The purpose of this follow-up was to determine either the potential re- spondent's intent or reason for non-response. Many ques- tionnaires which were returned either totally or partially unanswered provided a means of determining the reasons for non-response. The primary reason appeared to be the time and detail required to complete the questionnaire. Several un- answered, but returned, questionnaires carried comments about the length of the instrument. In balance many usable responses also carried comments on the length and detail. These returns were roughly equally divided in terms of posi- tive and negative positions. Several respondents commented (a few by personal letter) on the completeness and depth of the research and offered further assistance if needed. The conclusion after reviewing all comments was that no unusual bias resulting from the length or subject matter effected the 84 response rate. By most standards, the response rate would be considered "typical" for a mail interview format. Data Collection Instrument Two areas had to be considered in designing the data collection instrument for the study. First, the media for collecting the data had to be chosen and second, the general question type was selected. Each of these areas will be treated in turn. A survey format was initially chosen for data col- lection as Opposed to either Observation or a true experi- mental design. Since the attitudes and preferences of "professional" purchasers were being sought, the survey meth- od was deemed adequate. The preliminary investigation in- dicated that officials of purchasing organizations were able to readily express both the buying process and purchase cri- teria. The problems associated with survey bias such as inability and unwillingness of the respondent to express his attitudes appeared to be absent in Phase I. The next choice regarding the instrument concerned the media. Each of the three basic media, personal, tele- phone and mail interviews were considered. While each type has its own advantages and disadvantages, the research ob- jectives and data content needs were the determining factor in the choice process. Prior to the final media decision a subjective test of each was conducted by the researcher. The primary data collection effort was centered around the rating of both criteria importances and evaluation of 85 suppliers. AS stated in a previous section an extensive list of supplier selection criteria emerged from the Phase I investi- gation (See Table 2). Since the set of criteria was the focal point of the study, it was necessary to design the data collection technique around this section. Several personal and telephone interviews were conducted as a test to deter- mine the feasibility of collecting importance ratings of the buying criteria and attitudes of how well nine competitors in the market place rated in relationship to the same cri- teria. In both personal and telephone interviews presenting the list of criteria to the respondent was found to be ex- tremely difficult if not impossible in some cases. After obtaining his impressions on the first several criteria and suppliers, he often would need the entire list or a portion of the list repeated. This process was very time consuming and the respondent rapidly lost interest. During a personal interview, the respondent could be handed a list of both cri- teria and manufacturers, however the data recording process became extremely mechanical and could have been accomplished as well without personal contact. The economics of personal interviews were also a drawback. On the positive side, both personal and telephone interviews potentially made the data collection effort more personal for the respondent. The personalization of the interview format helped motivate some individuals to respond. However this advantage did not outweigh the disadvantages 86 presented above. Another potential advantage of both tele- phone and personal interviews is flexibility both in question wording and ordering. With the amount and type of data that was needed for the research, flexibility did not prove to be a desirable characteristic. The control lost over ordered responses to the questions was a critical factor from the perspective of statistical accuracy in the succeeding analy- Sis. Mail was selected as the data collection media for the research. The choice was influenced primarily by the considerations discussed above. Due to the large volume of information needed, the hard copy provided in mail and per- sonalization through a personally signed and addressed cover letter was selected. Both the cover letter and questionnaire are reproduced in Appendix I. No incentive, monetary or otherwise, was used with the questionnaires. Although some studies11 Show increased response rates with monetary incentives, it was concluded that an incentive should not be used in this data collection effort. Since the information was collected under the cover of Michigan State University, an incentive was deemed in- appropriate. While the effect of using a non-incentive pro- gram could not be measured, the response rate was not atyp- ical for a mail questionnaire. 11See Paul L. Erdos, Professional Mail Surveys, (New York: The McGraw-Hill Book Company, 1970), pp. 94-100. 87 Another concern in designing the data collection in- strument was the type of questions to utilize. The impor- tance rating of each buying criterion and the attitude of the respondent with respect to how well various suppliers were rated on that criteria was needed. Several methodolo- gies are available to measure this information, including ordered ranking of criteria and suppliers, semantic dif- ferential and attitude scales, and sorting techniques. Of the three types, attitude and ranking scales were chosen as the most viable for the study. The choice was determined by two major variables, the total time and space required for the respondent to completely provide the data and the rela- tive ease with which the respondent could understand the questions and relate his attitudes. Ranking methods provide measures of how various ob- jects are related to each other on an ordered scale. Some problems are inherent in rank ordering methods which pre- clude the use of some statistical analyses as well as the ability of the respondent to supply accurate information. First, respondents cannot indicate the degree of difference between Objects being compared. Only the relative position is indicated. If several are perceived as equal, the rank- ing method is also ineffective. Secondly, respondents can- not handle a large number of objects with ranking methods. Methods of sorting and/or paired comparisons are another means of studying buying criteria and brand prefer- ence. In a sorting technique such as the Q-Sort, the re- 88 spondent creates a scale value by placing statements in piles, along an interval scale. Another sorting method re- quires the respondent to create pairwise comparisons of brands which are most similar in one or more characteristics to those which are least Similar. Both methods are extremely difficult to execute in a mail interview format. They also are extremely time consuming for the respondent and because they are not often utilized in research, many if not most respondents require extensive instructions on how the proc- ess operates. The alternative chosen for this research was the equal appearing interval scale. The specific questions uti- lized are illustrated on the questionnaire reproduced in Appendix 1. Five point interval scales were used both to measure importance ratings of buying criteria and attitudes toward suppliers. Because the second section of the ques- tionnaire was the most critical portion the question types used were most important. The remaining section contained questions similar in format to those in section two to main- tain continuity for the respondent. Debate continues over the appropriate number of scale points to be used in marketing research. Jacoby and Matell maintain that reliability and validity of attitude scales are not substantially affected by the number of points or cate- gories on a scale. They infer that dichotomous or trichoto- mous scales may be used in scoring or recording attitudinal data after the respondent has expressed his feelings on "an 89 instrument that provides for the measurement of direction "12 Lehman and Hulbert13 and several degrees of intensity. contend that increasing the number of scale points reduces rounding error when measuring attitudes and individual as Opposed to group behavior. They suggest that a minimum of five points should be used. The controversy if it does in fact exist seems to lie in the trade off between the ability of the respondent to express his true feeling along a contin- uum and his ability to truly differentiate between points on a scale. Hulbert14 in his review of several marketing re- search efforts found a mean of Six to ten points being used. However, he also cites situations where a respondent was unable to comprehend a scale with more than ten points. The range would therefore be from two to ten scale points. Looking at the problem from the perspective of the respondent and his problem of expressing his true feeling the minimum number of points necessary might be determined. He may have an extreme feeling in either a positive or negative direction, this locates the end points of the scale. If he is indifferent and perceives the scale as a continuum from 12Jacob Jacoby and Michael S. Mattell, "Three Point Scales are Good Enough", JoUrnal of Marketing Research, Vol. VIII, No. 4, (November, 1971), pp. 4951500. 13Donald R. Lehmann and James Hulbert, "Are Three Point Scales Always Good Enough", Journal of Marketing Re- search, Vol. IX, No. 4, (November, 1972), pp. 444-446. 14James Hulbert, "Information Processing Capacity and Attitude Measurement", Journalyof”Marketing‘ReSearch, Vol. XII, No. 1, (February, 1975), pp. lO4-I06. 90 positive to negative, he requires an intermediate point placed near the midpoint. So far three scale points are necessitated. If he is not indifferent, but he also does not have an extremely positive or negative feeling, he needs two more intermediate points. One point is halfWay between indifferent and positive and the other halfway between in- different and negative. Thus it would appear that most re- spondents Should be provided with a minimum of five points to express their true feelings. For this research, a five point scale format was selected for use in measuring both attitudes and importance ratings. Data Analysis Techniques The primary focus of the research was the analysis of group differences of buying criteria importance and de- terminance. As stated in the section dealing with research hypotheses, the analysis of differences was to be conducted along several dimensions. To review, these dimensions were: 1) mechanical contractors versus engineers (buying influence center) 2) plan and spec versus design build team managed job types (job application) 3) commercial versus institutional jobs (job application) 4) rooftop versus chiller (product group). The analysis plan was designed to identify group differences along these dimensions. Each dimension was analyzed inde- pendently in the following sequence. First, a comparison 91 between mechanical contractors and engineers was made. If the difference in buying criteria importance was found to be significant the entire sample group was to be split on this dimension. Next each subgroup was analyzed along the job application dimension and the product group dimensions. The objective of this part of the plan was to identify the varia- bles or criteria upon which the groups differed to the great- est extent with respect to importance ratings. In addition, whether or not the between group differences were statisti- cally significant also had to be known. If there was a Significant difference on importance ratings between two or more subgroups, the next step was the identification of the variables which determined supplier choice. Since the data collection instrument asked respond- ents to indicate the supplier (or suppliers) from which they would purchase the product, the dependent variable for eval- uating determinance was the mention of a supplier name. Several statistical techniques are available to ana- lyze group differences in marketing research. The selection of the prOper technique is dependent upon the characteristics of the variables under study. Figure 5 is an example of one decision tree which might be used to select the statistical methodology. Because the relationships of all selected cri- teria were to be studied Simultaneously a multivariate sta- tistical methodology was chosen as opposed to a univariate or bivariate technique. The level of measurement attained in the data and the Specification of the independent and 92 Figure 5 Dependent on Others ? CLASSIFICATION OF MULTIVARIATE METHODS Methods - Some of the Dependence Methods How Many Variables are Dependent ? One Several ls Are it They Metric .7 Metric ? Yes NO Yes No I_-]_—j Multivariate . . _ Metric Multiple AnalySIS Factor Cluster . . . Regression of AnaiYSIS Analysis ‘ Mumd'msns'oml . lung Variance .M”."'P'e Canonical Nonmetric Discriminant Anal sis Scalin Analysis y g Latent Structure Analysis 93 dependent variables for analysis were the two controlling factors in selecting Statistical technique. Table 5 summarizes the characteristics of the various levels of measurement for statistical analysis. The proper or desired level of measurement is determined by the data collection method and instrument. Whether the nominal, or- dinal, or interval scaling level is reached is determined by the design of the questions and the underlying assumptions of the researcher with respect to interpretation of the scales by the respondent. In the study, an equal appearing interval scale was used to collect both importance ratings and attitudes towards suppliers. Both the importance ratings and attitudes were considered the independent or predictor variables. These variables were considered to be intervally scaled. The dependent variable in the first part of the analysis was the group to which each respondent belonged (contractor-engineer, etc.). The second part of the analysis used supplier mentions versus nonmentions as the dependent or criterion variable. Both dependent variables were therefore assumed to be nominally scaled. With this information and the decision framework illustrated in Figure 5, the primary statistical analysis technique was selected. Following the steps through the framework, the decision was made to utilize discriminant analysis for evaluating the group differences based upon both importance ratings and attitudes. The next section will explain this technique and explain its applica- tion in this research. Level of Measurement Nominal Ordinal Interval Ratio 94 Table 5 Operations Allowed Equivalence Equivalence Ordered Rela- tionships (rankings) Greater than/ Less than Equivalence Ordered Relat. Arithmetic Oper. addition subtraction multiplication division Equivalence Ordered Relat. Arithmetic Ratio of Two Scale Values Absolute Zero Point SUMMARY OF LEVEL OF MEASUREMENT Statistics Mode Frequency Dist. Chi-square Median Percentile Rank Correlation Mean Variance (std. deviation) Product Moment Correlation Geometric Mean Coefficient of Variation 95 Discriminant Analysis Applications Discriminant analysis is a technique used to evaluate classifications of observations into groups. As applied to marketing research, the purpose of discriminant analysis (DA) is to study group differences based on the observation of several variables. The use of DA is two fold. First it may be used to predict group membership based on a set of obser- vations on known group membership. In this process, a set of observations are taken on individuals whose membership is known, and subsequently individuals whose group membership is unknown are classified based on the same set of observations. A second use of DA is the study of group differences based on a set of observations. This use evaluates the importance of various observations in distinguishing between the groups. As such it is predictive in the sense that it may be used to classify unknown individuals, but greater emphasis is placed upon whether or not the observations can discriminate among groups and which ones are most effective. The objective of the latter approach is to identify the variables which are the discriminating variables. The use of DA for marketing research has been rather limited. However, several studies have used the technique to study users versus non-users of a brand, good versus bad credit risks, adopters versus non-adopters, and readers ver- sus non-readers. For the most part these studies used demo- graphic characteristics to predict group membership. 96 15 used socioeconomic variables to predict FM station Massy selection. He found that the DA approach could be used to identify dissimilarities in station audiences. Sweeny and 16 used a combination of store attribute vari- Reizenstein ables, customer shopping variables, and customer demographic variables to predict preferences for store types. They suc- cessfully discriminated groups with roughly one-half of the original twenty variables. Lehman and O'Shaughnessy17 de- termined the attributes which were most important in dif- ferentiating between industrial buyers in the United States 18 clustered and Great Britain. Finally, Scott and Bennett buyer types with DA, on the basis of product attribute im- portance ratings. They suggest that buyers should be clus- tered on the bases of attribute importance prior to evaluat- ing attitudes toward suppliers. 15William F. Massy, "Discriminant Analysis of Audi- ence Characteristics", Journal of Advertising Research, Vol. V, (March, 1965), pp. 39-48. 16Daniel J. Sweeny and Richard C. Reizenstein, "De- veloping Retail Market Segmentation Strategy for a Women's Specialty Store Using Multiple Discriminant Analysis", in Proceedings, Fall Conference, American Marketing Association, I971, pp. 466-472. 17Donald R. Lehmann and John O'Shaughnessy, "Dif- ference in Attribute Importance for Different Industrial Pro- ducts", Journal of Marketing, Vol. XXXVIII, No. 2, (April, 1974), pp. 36T42. 18Jerome E. Scott and Peter D. Bennett, "Cognitive Models of Attitude Structure: 'Value Importance is Impor- tant'", in Proceedings, Fall Conference, American Marketing Association, 1971, pp. 346-350. 97 DA seems to be a viable methodology to use in evalu- ating the importance of individual product and/or supplier attributes in determining the results of the purchase deci- sion. In the next section, the foundations of discriminant analysis as a statistical technique will be discussed. Discriminant Analysis - Statistical Foundations Assuming that systematic differences exist among groups, discriminant analysis develops a decision rule, based on observations of several variables, to classify individuals into one of two or more mutually exclusive groups. If a single variable could be used to predict group membership there would be no need for a technique like discriminant analysis. However, one variable is usually not adequate for prediction. If no variables are available to predict member- ship, a simple probability or chance model might be used to predict membership. The objective of discriminant analysis as a formal technique is to improve the level of predicta- bility over the chance model. In order to predict group membership, a Single vari- able is needed to Simplify the decision process. Given a value of the discriminator or predictor variables, the re- sulting value of the criterion (dependent) variable is used to classify an individual. To create the criterior varia- ble, a linear combination of the predictor variables is specified. This combination takes the form 98 or Y = ab + aixi + azxz + a3x3 + ... anxn Y - criterion variable ab - constant ai - predictor variable weights >< I predictor variables The technique attempts to find the linear combination (places values on ai) such that the equation is better than the chance model for predicting group membership. TO accomplish an efficient prediction the weight are found such that the ratio of between group variation - Y within group variation - Y is maximized. The mathematical sc0pe of this manipulation will not be presented here. Several sources 19 give detailed explanations of the process. The mathematics involve manip- ulation of the cross product matrices of the independent 120 variables. Green and Tul present a straightforward approach which uses the cross product matrix to generate a series of simultaneous equations whose unknowns are the 19See; Cooley and Lohnes, Multivariate Data Analysis; Maurice M. Tatsuoka, Multivariate Analysis; and Ben W. Bolch and Cliff J. Huang, MuItivafiate StatiStical Methods for Business and Economics; fOr mathematical explanations of the discriminant analysis technique. 20Green and Tull, p. 450. 99 discriminant coefficients. Once the values of the discriminant coefficients have been determined, two analyses follow. One evaluates the ability of the function truly to differentiate the groups based on the prediction variables. The other studies the relative importance of each prediction variable in discrimi- nating. These analyses will be dealt with separately. To evaluate the ability of the function to discrimi- nate group membership, a confusion matrix is constructed. Figure 6 shows the format of a matrix for a two group analy- sis. In the evaluation, a value for each member of the sam- ple based on the values of the prediction variables for that individual. The values are then used to place the individual into a predicted group. Some methods calculate the proba- bility membership in each group and assign the individual to the group with the highest probability. Other methods find the mean value of the criterion variable (Y) for each group and establish a midpoint (in the case of a two-group analy- sis) between the values referred to as group centroids. Assignment is based on the position of an individual's dis- criminant score (value of function) relative to the midpoint. With either method the number of correctly classified indi- viduals becomes the statistic for evaluating the worth of the discriminant function. Since both the known and predicted group memberships are readily found, the percent of individ- uals correctly classified may be computed. This percentage is then evaluated through the formula 100 t = Pcc -.Pec \‘ Pechc n where t - Student's-t statistic Pcc - Percent correctly classified Pec - Percent expected by chance (equal to l/number of groups assuming equal probabilities of membership) Pec - l - Pec n - Total sample size (Groups 1 and 2) The critical value of t is found in a Student's-t distribu- tion with a degrees of freedom at the selected alpha level (the probability of judging the null hypothesis false when in fact it is true). Figure 6 CONFUSION MATRIX FOR TWO-GROUP DISCRIMINANT ANALYSIS Predicted Membership Actual Group 1 Group 2 Membership Group 1 I111 n12 N1 Group 2 n21 n22 N2 n11 + n22 % cOrrectl classified = y N1 + N2 101 Morrison21 poses two criteria for testing the propor- tion of individuals correctly classified. He suggests using either the proportional chance criterion or the maximum chance criterion. The hypothesis is tested by setting up a critical value based on either of these criterion and check- ing the proportion correctly classified against it. If the test prOportion is larger than the criterion, the function is said to discriminate between the groups. The only deci- sion involved is the selection of either the proportional or maximum criterion, and this depends Upon whether the Objec- tive is to study both groups or maximize the prOportion cor- rectly classified. The second type of evaluation used with discriminant analysis is determining the predictor variables that serve as the "best" discriminators. The objective is to decide which variables are most important in distinguishing between one group versus the other (or all others in the case of n-group analysis). AS in multiple regression the variable with the highest function coefficient is judged most important in discriminating. Before this judgement is made however, the coefficients must be standardized to account for both meas- urement units and variation. The correction is accomplished by adjusting the coefficients through multiplication of each one by the sample standard deviation (across all groups) for 21Donald G. Morrison, "On the Interpretation of dis- criminant Analysis", JOurnal of Marketing Research, Vol. VI, No. 2, (May, 1969), pp. 156-I63} 102 each independent variable.22 Only after the adjustment is made can the relative importance of the variables in the func- tion be judged. Both positive and negative signs normally appear in the function, whether the raw or standardized co- efficients are used. These Signs merely indicate the direc- tional nature of the coefficient and are dropped when eval- uating the relative importance of the variables. Several assumptions are associated with the use of discriminant analysis. These are l) the predictor variables are intervally scaled 2) the subgroups in the population (subpopulations) are mutually exclusive and exhaustive 3) the variables are normally distributed 4) the subpopulations (subgroups) from which the samples are drawn are multi-variate normal with common (identical) covariance matrices which are unknown 5) the costs of misclassification are equal for all subgroups 6) for most analyses, the probability of group mem- bership is equal across all groups. The assumptions are similar to those of other multivariate parametric analyses. While it is possible to statistically prove whether or not the assumptions are all met, that proof might require an exhaustive analysis. What is most important is the realization that the assumptions are present and a logical explanation of whether or not they may be reasonably considered to exist in the data used for the analysis. 22Ibid., p. 159. 103 An upward bias problem may be present in discriminant analysis when testing the Significance of the function in dif- ferentiating between groups. AS Frank, Massy, and Morrison23 point out, the ability of the function to discriminate may be misinterpreted when it is evaluated with the same data used to generate the function. This problem is also inherent in multiple or even Simple regression. The above authors sug- gest two methods of measuring the bias. One entails parti- tioning the sample into two halves, the first to generate the function and the second to evaluate how well the function discriminates. A second method would use the entire data set to generate the function which would then be tested with a randomly generated set of values the result of which would help estimate the level of bias which existed. The latter alternative is recommended when the sample size is too small to be halved. Both techniques minimize the bias by elimi- nating the condition where identical data sets are used to generate and evaluate the discriminant function. The purpose of this study includes both the determination of which vari- ables are important in differentiating among group and how Significantly the groups differ. It is recognized that an upward bias problem exists in evaluating the worth of the function in discriminating with the identical data used to 23Ronald E. Frank, William F. Massy, and Donald E. Morrison, "Bias in Multiple Discriminant Analysis", Journal of Marketin Research, Vol. II, No. 3, (August, 1965 , pp. 250-258. 104 generate it. The problems inherent in over biasing are out- weighed by the fact that if only half of the sample is used to generate the function a reduction in the accuracy of the discriminant coefficients in demonstrating the relative im- portance of the predictor variables results. Since the dis- criminant technique has no real means of estimating the co- efficients in the pOpulation from the sample data, the only alternative seems to be reducing the sampling error by in- cluding the maximum number of sample elements possible in the analysis. Therefore, the problem of upward bias will be treated as a limitation in the results rather than being ana- lyzed and nullifying part of the data set. Another problem also exists in discriminant analysis when a large set of predictor variables is included in the analysis. If the correlations between several of the vari- ables are high, the predictive power of the function as well as the evaluation of variable importance is hampered.24 One method of relieving this problem is by factor scores for individuals to predict group membership. Not only is this an attractive alternative from the above point of view, but it makes the interpretation of group differences easier due to the lesser number of resulting predictor variables. A brief explanation of factor analysis and the particular technique selected for this research follows. 24Churchill, p. 531. 105 Factor Analysis The purpose of factor analysis is the reduction of a large number of observations to a smaller and more manageable set. A family of factor analytic techniques exist which all Operate on the assumption the set of underlying factors for which several observations might be surrogates, exists in the process being studied. The Objective of all factor analytic techniques is the generation of linear combinations of vari- ables which reduce the number of total dimensions of the variable set (independent variables) while preserving the original relationships in the data. The methods of factor analysis are the ways in which the coefficients of each lin- ear combination are determined.25 For each factor a set of weights is determined for the equation Fj = ai’js + 32,sz + a3,jx3 + ... + an’jxn where Fj - jth factor ai j - Factor weight for ith variable and jth factor 9 x- - ith variable in original data set. 1 One mathematical procedure in factor analysis uses the correlation matrix between variables to derive principal factors.26 This method actually factors the correlation ma- trix rather than the original data set itself. Two methods 25Green and Tull, p. 529. 26Ibid., pp. 535-541. 106 may be used.27 First, principal factors using communalities (the proportion of variance accounted for by common factors) in the diagonal of the correlation matrix; and second prin- cipal components setting the diagonal values to unity. The first method was chosen for this research. The principal factors are extracted from the cor- relation matrix such that the first factor accounts for the greatest amount of variance in the original variable set. The second and successive factors are selected such that they are uncorrelated with the previous factor and account for the next largest proportion of variance. Although a discussion of how the number of factors is determined is beyond the scope of this discussion, the procedure performs matrix ma- nipulations on the correlation matrix until the last compo- nent or factor accounts for at least the total variance of a single variable.28 Once the principal factors have been found an inter- mediate step is performed which computes the correlation between the principal factors and the individual variables. These correlations are the factor loadings.29 The factor loadings tell the researcher which variables are highly correlated with which factors. This aids in interpreting and 27Richard L. Gorsuch, Factor Analysis, (Philadelphia; W. B. Saunders Company, 1974), pp. 85-92. 28Norman H. Nie, et a1, Statistical Package for the Social Sciences, (New York: McGraw-Hill Book COmpany, 1975), p. 479. 29Green and Tull, p. 531. 107 naming the factors. The problem with the principal factor loading as they result from the above procedure is that one variable may load highly, be correlated with more than one factor. The set of methods for overcoming this interpreta- tive problem is referred to as principal factor rotation. The objective of this process is to Simplify the interpreta- tion by rotating the axes describing the loading of the var- iables on the factors such that each variable loads highly on some factors and has near zero or zero loading (correla- tion) on others. Several rotational schemes are available for per- forming this Simplification step. The selection is based upon the research objectives. A broad classification of rotations is Specified by orthogonal or Oblique rotations. To simplify the explanation the axes are Simply graphical scales illustrating the correlation of factors and variables. Orthogonal rotations maintain uncorrelated factors while oblique rotations allow some correlation between factors to result. Graphically on orthogonal rotation maintains a perpendicular relationship between factors. Oblique rota- tions result in factors being graphically represented in relationships which are at angles other than perpendicular. The former class, orthogonal rotations, is most often used due to simple interpretation. Within each class several rotational methods are present. In orthogonal rotation quartimax, varimax, and 108 equimax rotations are commonly used.30 Quartimax rotation attempts to Simplify the factor structure to the point where a variable loads highly on one factor, on near zero on others. Varimax, on the other hand, attempts to rotate the factor such that several variables load highly and others load near zero on each factor. The third, equimax, attempts to achieve the best of both situations. In cases where R factoring is used (factoring observations) and the descrip- tion of underlying factors is the objective, the varimax criteria is the most popular of the three. That criteria was used in this research. Once the rotated factor loading matrix is determined, the factors can be named and utilized in further analysis. The naming of factors is an arbitrary process based on the researchers judgement. Often the name is a combination or summary of the variables which load highly on the factor. The naming of the factors is not as important how- ever as their use in further analysis. Since the purpose of factor analysis is to simplify the original data set without substantial information loss, the knowledge gained from the technique used most be put to use. Usually factor scores are develOped for each individual which are combinations of the original variables. These combinations are then used for an analysis of individual, group, or other differences, or to predict some phenonmenon. A decision rule must be estab- lished concerning how the original data is to be combined to 30Nie, et al, pp. 484-485. 109 produce factor scores. Actually factor scores are estimated since the factors themselves rarely account for the total variance in the original variables. Gorsuch31 presents sev- eral methods for estimating factor scores. One method often used is a zero-one weighting system. All variables are given a weight of zero or one depending upon their loading on the factor. A commonly used critical loading is .60. All var- iables loading greater than or equal to .60 are included and all other excluded. The factor score on each factor for each individual is simply the sum of the raw variable values for those variables with high loadings. This method makes the computation Simple and assumes that all variables included in a factor are equally weighted. All variables which do not have high loadings on any factor may either be dropped from further analysis or lumped into a miscellaneous factor. If the Objective is to use a set of uncorrelated variables in further analysis the latter alternative is chosen. This was the case in this research. Factor Analysis of Importance Scores The first analysis step applied a factor analysis to the importance scores for the set of buying criteria used in purchase decision making. (See Section II of the question- naire - Appendix I). The importance ratings on the nineteen criteria were factor analyzed for the entire sample. No separate groupings were used in this step. 31Gorsuch, pp. 231-239. 110 The purpose of this step was twofold. First, the re- duction of the nineteen buying criteria to a set of buying factors was sought in using the technique. Second, if a set of factors was found which retained the original information in the data, while explaining or accounting for a large pro- portion of the total variance, the problem of covariance among independent variables used in the subsequent discrimi- nant analysis would be relieved. The factor analytic tech- nique was used in a data reduction rather than in a predic- tive mode in the research analysis plan. As discussed aboVe, several techniques in factor analysis may be performed. Principal components analysis was chosen for this analysis to provide a set of factors de- rived from the correlation matrix of the buying criteria. This correlation matrix is shown in Chapter Four - Table 5. The correlation coefficients between the independent varia- bles indicate which sets of variables potentially may become factors. The final combination and number of factors is de- termined by the prOportion of variance explained by the fac- tors and the eigenvalue. The eigenvalue is the amount of variance explained by the factor and in the SPSS factor t.32 The variance analysis package is set to 1.0 as defaul accounted for by each factor is found by summing the squared loadings of all variables on the factor. To reiterate, the factor loading is the correlation between the variable and 32Nie,“et‘al, p. 479. 111 the factor. Factors accounting for a variance of less than 1 are not included in the analysis. Stated another way if a factor does not account for a prOportion of variance equal to or greater than critical eigenvalue = l number of variables it is not included. This criteria may be altered if neces- sary, however. Another method of determining the number of factors is through the use of the percentage of total variance ac- counted for by the factors.33 This may be determined by the researcher and is directly related to the eigenvalue by sum- ming the prOportion of total variance represented by the actual eigenvalue divided by the number of variables. Com- paring these methods when twenty original variables are in- cluded in the analysis and the critical eigenvalue of 1.0 is involved, approximately 95 percent of the variance will be explained by the factors which remain. Another set of summary statistics of importance in the factor analysis sequence are the communalities of the variables. These indicate the percent of total variance in each variable that is explained by the common factors. These sequentially replace the values in the diagonal of the cor- relation matrix during the analysis. The final communality estimates are then evaluated to measure the variance 33Ibid. 112 explained by the factors. The communalities are calculated by summing the squared loading for all factors across each variable. In the SPSS routine, the estimate of the com- munalities, eigenvalues, and factor loadings are presented both before and after factor rotation. The post rotation statistics are the more important for an analytical summary. Computer Data Analysis Tools The bulk of the statistical analysis was accomplished with the use of the Statistical Package for the Social Sci- ences (SPSS). This package allows the researcher to perform many types of parametric statistical analyses on a common data set. The mathematical procedures used in the SPSS pack- age are adapted from popular references in the social science literature and have been utilized in several fields for anal- ysis. The package allows for several Options in handling mis- sing data in a survey, as well as a wide range of statistics in most routines. It was assumed that SPSS was free Of gross methodological errors. The data from the research questionnaire was coded and recorded in computerized form for use with SPSS. Two verification steps were used in the coding process. To eliminate excess error data was recorded onto card images directly from the questionnaire. After coding a random sam- ple of questionnaires was selected for verification. This was in addition to a recheck of each case (questionnaire) while coding. The card image coding was then permanently recorded on computer cards. A verification (mechanical) 113 step was also performed in the keypunching process. This process was designed to reduce or eliminate any mechanical errors in the analysis resulting from transfer of data from the form where the respondent recorded it to the actual anal- ysis algorithm in the computerized SPSS package. Each analy- sis was cross checked with Similar analyses to certify that the correct number of respondents were being included as group members as the successive runs through the package were completed. Summagy of Research Desigp A step by step process was used to analyze the re- search data. First, the importance scores for all buying criteria were factored across all respondents. This step served to reduce the covariance problem encountered in using a large variable set for discriminant analysis. Factor scores for each respondent were developed as linear combina- tions of the importance ratings. A zero-one weighting pro- cedure was used to develop the factor scores. Once factor scores for all respondents were develop- ed from the importance ratings, discriminant analysis was employed to evaluate differences between groups on the buying criteria importance ratings. The analysis looked at the con- tractor-engineer dimension first for a significant differ- ence. Next the plan and Spec versus design build-team man- aged job types were evaluated in both the contractor and engineer groups. The job type and product type dimensions were also evaluated. 114 Once dimensions were identified where significant differences existed, a second level discriminant analysis was conducted for buyers versus non-buyers of three selected suppliers. These second level analyses were only conducted in groupings where significant differences based on impor- tance ratings were found in the first level analysis. Chapter IV presents the results of the factor analy- sis of importance scores and the discriminant analysis of segments based on the factor importance scores. The signifi- cance of the differences in segments based on factor impor- tance is discussed in that chapter. Chapter V examines the significance of the factors as determinants in the buying process. The overall preference of respondents is studied in relation to the ratings of sup- pliers on the buying factors. CHAPTER IV ANALYSIS OF MARKET SEGMENTS Introduction Chapter IV presents, in detail, the research findings related to the analysis of differences among segments of the market for the study product. The importance scores for the buying criteria are used to develop a profile for the vari- ous segments which are significantly different from each other. Chapter V uses the market segments derived in the present chapter and analyzes the determinant criteria for supplier choice. To reiterate the methodology set forth in the pre- vious chapter, the steps in the total analysis process are: 1) Factor analyze the entire data sample based on importance ratings for the buying criteria 2) Evaluate the statistical significance of group differences using discriminant analysis based on factor scores developed in Step 1 3) For all groups found to contain significant dif- ferences in Step 2, use discriminant analysis to determine the determinant factors in purchase preferences. The first two steps are covered in Chapter IV and the third in Chapter V. The managerial implications and conclusions will be presented with the findings. Step One - Factor Analysig Table 1 presents the eigenvalues and percent of ex- plained variation following the varimax rotation chosen for the present research. To summarize the varimax rotation, the factors are rotated orthogonally to a solution where some variables load highly on each factor and others load 116 highly on each factor and others load near zero. By exami- ning the information presented in Table 1, it is noted that 92.3 percent of the total variation in the original data is accounted for in factors one, two, and three. Factor four may be considered a miscellaneous factor explaining the re- maining variance in the data. The composition of the fac-. tors (variables which are highly correlated with the factors) is explained below under the factor loading discussion. Of importance at this point is the fact that three underlying factors account for over ninety percent of the total vari- ance in the data. Table l EIGENVALUES AND PERCENT OF VARIATION EXPLAINED AFTER VARIMAX ROTATION OF IMPORTANCE RATINGS Percent of Cumulative Factor Eigenvalue Variation Percent Variation 1 4.627 47.4% 47.4% 2 2.731 28.0 75.4 3 1.652 16.9 92.3 4 .747 7.7 100.0 Table 2 presents the final estimated communalities for the nineteen importance variables. As discussed above, the communality is the variance in each variable accounted for by the common factors, one through four. By examining the communalities the potential variables are determined for which the minimum amount of information is lost in reducing the number of predictors from nineteen to four. Several of Variable 1 2 10 11 12 13 14 15 16 17 18 19 117 Table 2 IN FACTOR ANALYSIS ngg Price-First Cost Operating and Maintenance Cost Reliability-Life Expectancy Ease of Installation Ease of Maintenance Size, Weight, and Construction of Equipment Noise and Vibration Levels Delivery Time Consistency of Delivery Delivery Expediting Capability Local Availability Of Parts and Service Regular Contact by Salesman Long Term Established Contact Salesman's Assistance Regular Catalog Updates Availability of Salesman Availability at Full Line Factory Service Support Catalog Descriptions COMMUNALITIES OF ORIGINAL VARIABLES Communality 0. O 0. 16973 .54875 71678 .87990 .65978 .27249 .34017 .57638 .82368 .81932 .53836 .61385 .62948 .22924 .44443 .42848 .26157 .47539 .32903 118 the original variables have a high percentage of variation explained. The factor loading matrix Shown in Table 3 is used to determine which variables are combined to estimate the factor scores from the original data. As stated in Chapter III (Re- search Methodology) a factor score is a linear combination of the original variables of the form '11 l i - wlxl + WZXZ + W3X3 .... + ann where F- - ith factor score for sample element wl to n - weight Of jth variable x1 to n - raw score on jth variable. A factor score is calculated for each member of the sample with the four new variables. The weights used in computing factor scores are zero or one depending upon the factor loading of the variable. The weights are determined by a simple decision rule: 1 if factor loading EE:.6O weight weight 0 if factor loading¢::L.60 The factors are named by studying the variables which load highly on each factor or those which have weights of one. By examining Table 3, the four factors might be named operating and/or equipment lifetime, sales service distri- bution service, and miscellaneous, respectively. Thirteen of the nineteen original variables are included in the fac- tors. The remaining six variables have variances which are not accounted for by the factors to be included in the 119 Nomno.o omauo.o mmnov.o Neoflm.o mmuao.o nenmo.o wmeN.o NNHmo.o ommoa.o eooco.o mooov.o omoo~.o vmmmH.o NommH.o vm~oo.o Hewoa.o mmmvo.o vwmvo.o- oaooo.o veonN.o eawmo.o Hmeo~.o muonm.o mvmma.o Nmoeo.o nmwmo.o ewamn.o oooma.o- waowo.o wow~o.o ovnnn.o mummo.o- mo~wo.o emmow.o oomem.o eammo.o mweoa.o wmmom.o omva.o mnnmo.o mmnvo.o Nonmw.o nmmoa.o maooo.o mwOOH.o omomn.o ooomo.o BHNNH.o- vomfiv.o mMOHo.o- mmmma.o emaom.o vaNe.o moama.o mmeam.o mewH.o Hanav.o omaoo.o omwom.o QHNoo.o whoaw.o moamv.o emomo.o omvno.o- womOH.o nwomo.o- meomo.o- mammm.o ommoa.o mwNNH.o- mnmoo.o ommam.o mmnno.o mmHeN.o emavo.o- omHNm.o- e hepumm m pOpumm N houumm H hOpumm whouumm mcoflumanomom moaeumu unommsm oufl>nom AHOpomm dean Haze no sunfiapeana>< and: mdamm mo abnaapafiae>< moaned: mofimpmu neHSwom ooemumfimm< moamm pomucou moamm Show weoq ouhom moamm an pumpnou headmom oufi>pom can mused Suwafinmaflm>< kuog zuflaflnmmmu mcflpwpomxm ano>fiaon Seo>fiaom mo xocopmflmeou mane Ano>fiaoa schemanfi> can omfioz cofiausspmcou wee .ugmfioz .onwm ouemeoenfiez mo ommm :ofiumfiampmeH mo ommm xpfiafinmflaom ueoemflncm amou oonm:0pn«ez mzflumpomo pmou umnflm-oufipm osmz oflnmfiam> ZOHH mmhm< UZHH 120 computation of factor score estimates. These variables are: l) Price-first cost 2) Size, weight and construction 3) Noise and vibration 4) Sales assistance 5) Availability of full line 6) Catalog descriptions. All six of the variables listed above are confounded with all four factors. Of particular relevance is the fact that the price- first cost variable did not load highly on any one factor. An examination of the mean importance ratings and the cor- relation matrix, and the communalities is necessary to help explain this phenomenon as well as that observed with the remaining five variables. Table 4 presents the aggregate importance ratings for the entire sample. Price-first cost appears to be an important criteria in the choice process, thus the confounding of the variable is not a result of unimportance and therefore a random rela- tionship to other variables on the part of the respondents' ratings. In fact price-first cost is rated as second in importance by the entire sample. On the other hand, the variation in this criteria rating is not well explained by the factors. The communality of price-first cost is only .16973 (See Table 2). The correlation matrix indicates that this criteria is not highly correlated with any other cri- teria (Table 5). The highest coefficient is -.31, the other 121 Table 4 MEAN IMPORTANCE RATING FOR NINETEEN BUYING CRITERIA Mean Importance Criteria Rating Price-First Cost 3.96 Operating Maintenance Cost 3.95 Equipment Reliability 4.21 Ease of Installation 3.71 Ease of Maintenance 3.67 Size, Weight, and Construction 2.89 Noise and Vibration 3.93 Delivery Time 3.64 Consistency of Delivery 3.73 Delivery Expediting Capability 3.46 Local Availability of Parts and Service 3.94 Regular Contact by Sales Force 3.20 Long Term Sales Contact 3.10 Sales Assistance 3.23 Regular Catalog Updates 3.26 Availability of Sales Help 3.93 Availability of Full Line 3.05 Factory Service Support 3.78 Catalog Descriptions 3.61 122 mH.0 50.0 5H.0 0N.0 NH.0 0N.0 0H.0 0H.0 0N.0 00.H Nw.0 00.0 0H.0 0N.0 0H.0 Hm.0 «0.0- 00.0- HN.0 0H NH.0 0H.0 0H.0 HN.0 No.0 NN.0 HH.0 mH.0 v~.0 Nw.0 00.H 50.0 00.0 5H.0 0H.0 Nv.0 No.0- 00.0- va.0 00.0 H0.0- no.0 0H.0 00.0- HH.0 0H.0 00.0 00.0 00.0 50.0 00.H 50.0 00.0 H0.0 0v.0 NH.0- mH.0- 0N.0 m mozHH 00.0 Hm.0 Hm.0 Nm.0 0N.0 Hm.0 00.0 0H.0 00.H 0N.0 0N.0 00.0 Nm.0 mN.0 00.0 NH.0 Hm.0 Nv.0 mH.0- HH ”opoz 0H 0H 5H 0H 0H 0H mH NH HH WON P‘NMQ’ 124 criterion variable being equipment reliability. The second highest correlation coefficient is .26 that being with de- livery time. While price-first cost is on one hand an important criteria, but is related to several other variables both directly and inversely, the results indicate that price is considered in conjunction with several other variables. There does not appear to be any clearly singular relationship with any other criteria. As a result it may be concluded that price is a mitigating factor in the choice process. As a result the price variable is not highly correlated with a single factor. Recalling that the purpose of factor analysis is data reduction with minimal information loss, the inclusion of only those variables which are highly correlated with the underlying factors needs some explanation. Principal compo- nents analysis yields a new set of variables for which the most variation exists from one sample member to another. In creating linear combinations, it is those variables con- taining the largest amount of variance explained by the fac- tors, which are of interest to a researcher. The value of factor analysis to this research was the reduction in the total number of independent variables used to predict group membership in segments of the market. Table 6 presents a summary of the four factors derived in step one of the research analysis and the variable names in- cluded in each factor with weights of one. 125 Table 6 VARIABLES LOADING HIGHER THAN .60 ON FACTORS Factor 1 Factor 2 Factor 3 Factor 4 Operating and Regular Contact Delivery Time Ease of Maintenance Instal- Cost lation Equipment Long Term Consistency of Reliability Contact Delivery Ease of Main- Regular Catalog Delivery Expedi- tenance Updates ting Capability Local Avail- Availability ability Parts Sales Help and Service Factory Service Support Once the composition of the factors was determined, factor scores were used as inputs for the discriminant analy- sis phase of the research. The factor scores and factor composition was utilized in two areas. First, segments of the sample were delineated using the factor importance rat- ings. Second, the supplier ratings on the factors were used to predict purchase preferences within the identified seg- ments. The following section discusses the results, the first of these analysis steps. Analysis of Segment Differences Based on Factor Importance In this analysis step, the discriminant analysis technique was used to evaluate the differences between seg- ments. The factor scores for all group members on each fac- tor were used as independent or predictor variables in the analysis and group membership was the dependent variable. As stated in the Research Method010gy chapter, groups were 126 analyzed two at a time according to the tree diagram shown in Figure 1. At the first level mechanical contractors ver- sus engineers were studied. In the second level analysis job-type, equipment application, and products were evaluated within each of the above groups. This sequence resulted from the hypothesis that the greatest difference in impor- tance ratings was expected in the contractor-engineer break- down. Group Assignment A method for assigning the sample elements to the groups outlined above was designed into the data collection instrument. Each questionnaire was coded prior to mailing to indicate whether the respondent was a mechanical contrac- tor or engineer. This was necessary since identical ques- tionnaires were mailed to both groups. A process of self-placement was used to group the respondents into job type and job-application categories. As may be noted on the questionnaire (see Appendix I) each respondent was asked to choose a combination of job type and application. Further, each was instructed to answer the sec- tion on importance and supplier ratings in terms of his se- lection. The reason for using this methodology hinged on the lack of information as to what job types and applications were predominately accomplished within any one respondent's organization. Thus each respondent could not be asked about a specific type and application due to the fact that no prior knowledge of his familiarity with that combination could be 127 moumoom / moumoom H Lofififieu \ umflfifleu \ commemz Enos oomw HmHopoEEou HmcoHHSQHumcH 0HHsm :mHmom 0cm :mHm / \ / \ wmmmcmz Emoe oomm HmHupoEEou HmcoHusqumcH 0HHsm :mHmom 02m :mHm / \ / \ r mozHHHHom mmHmm OOH>pom :OHuanhumHQ oEHumqu mcHumgoao msoocmHHoumHz moumoom .m> HOHHHAU Houumpucou mnoocmHHoumHz oEHuoqu mchmaomo OOH>hom coHpanhumHQ OOH>Hom mOHmm wommcmz EmOH-0HHsm :meon .m> 00mm 0cm cmHm HOpumhucou maeomo zmmzemm queHahom mOHmm msoocmHHoomHz oEHpomHA wcHumhomo OOH>HOm aofiaanflaamfln hOOcHwnm .m> nonumhpcou CHAPTER V RATINGS OF SUPPLIERS WITHIN MARKET SEGMENTS Introduction Chapter V presents the analysis of the attitudes to- wards selected suppliers Of the study product. The determi- nance Of the four supplier selection factors is examined with respect to loyalty towards each supplier. The four factors derived in Chapter Four are utilized in the analysis. The respondents in the study were asked to rate each of nine selected suppliers on each of the nineteen purchase decision criteria. Linear combinations Of the rela- tive rating of each supplier were used as the predictor or independent variables for discriminant analysis. The analysis was conducted within the segments de- veloped in Chapter IV. The entire sample was split into these sub-groups. Supplier Ratings within Segments Once the entire sample was split into groups based on the importance rating Of the buying criteria, an analysis of supplier ratings was conducted in each group. The groups used were those which were found to have significantly dif- ferent importance profiles in the first step of the analysis. Figure 1 shows the groups where differences were found to be significant. The second step of the analysis used the discriminant analysis technique. The independent variables were the rat- ings for selected suppliers on each buying factor, while the dependent variable was the mention versus non-mention Of the 161 supplier, in question one, part three of the questionnaire (see Appendix I). The steps in the discriminant analysis used in this chapter are similar to those in the previous chapter. The Objective of this analysis step was the identi- fication of the factors which are determinant in the actual purchase decision. Figure 1 SUMMARY OF SIGNIFICANT AND NON- SIGNIFICANT DIFFERENCES FOR GROUPS BASED ON FACTOR IMPORTANCE SCORES Contractor (Significant)-———-——-Engineer I //////////Plan and Spec Design Build- Team Managed / Significant Non-Significant \ II ////////,Institutional Non-Significant\\\\\\\\\\ Commercial / Non-Significant \ III / Chiller RooftOp Significant Non-Significant \/ In computing the values of the independent variables, a conversion was made from the raw supplier ratings to a relative rating. Each respondent was asked to rate all 162 suppliers on all criteria. For each respondent, an average supplier rating was computed across all rated suppliers for each criterion variable. The relative rating was found by subtracting the average rating from the actual rating. This procedure was used to adjust for a reSpondent who rated all suppliers identically and thus indicated no difference among them on a given variable. If this were the case, he would see no difference among them and thus the variable would not be determinant in his decision. If the reSpondent rated all suppliers differently, some high and others below average on one factor, that factor could be considered to be determinant in the purchase decision. The factors which are most impor- tant in the discriminant function are then most determinant in the purchase decision. The order of determinance was therefore developed from the discriminant function. Recall from the discussion in Chapter III the standardized discrimi- nant function coefficients are used to evaluate the impor- tance of the independent variables in differentiating between groups. The higher the value (absolute) of the coefficient relative to the others the more important was the variable in differentiating. This implies that the variable accounts for the greatest variation between the groups. If the var- iable is highly important in differentiating between pref- erences and non-preferences toward a supplier, then that variable may be assumed to be determinant in the preference function. Since the independent variables in each analysis of supplier preference versus non-preference were the 163 relative ratings on only the supplier in question, this tech- nique was valuable in indicating the individual determinance of each variable to each respondent. Factor ratings were the actual values used as inde- pendent variables in this analysis step. AS such, they were computed in a manner similar to that used for factor impor- tance ratings. After the individual ratings of all suppliers by each respondent were converted to relative ratings, the factor ratings were computed by summing the relative ratings Of the variables which loaded highly On each factor (see Chapter III). The weights used in this summation were zero and one as in the factor importance analysis. If a respond- ent rated a given supplier below average and several others above, the score on that variable would then be negative for that supplier. Whether the composite factor rating was pos- itive or negative depended upon the relative ratings of the supplier on each variable loading on the factor. Finally, some respondents did not rate all suppliers. Any supplier who was not rated on a given variable was treat- ed as a missing case and assigned a value equal to the mean of all other variables. If all ratings are missing for a given respondent on a supplier, that case is dropped from the analysis, but included in the classification phase by substituting mean values for each variable. The four suppliers selected for analysis were those which received the greatest number of reSpondent ratings. Table 1 presents the frequency of Supplier mentions by 164 segment. Table l FREQUENCY OF SUPPLIER MENTIONS BY SEGMENT Segment Plan and Supplier Spec DBTM Chiller Rooftop A 28 29 42 15 56 B 30 26 40 16 53 C 1 3 4 0 12 D 11 11 17 5 27 E 3 4 0 7 7 F 6 3 4 5 4 G 4 4 0 5 5 H 2 l 2 1 2 I 1 0 1 1 2 J 0 0 0 0 1 Two hypotheses were tested in the analysis of sup- plier preference within each group. First, that the discrim- inant functions significantly differentiate between mentions and non-mentions of a supplier. Second, that the business logistics factor varied in importance in the function de- pending upon the segment. The discussion which follows presents the results and implications of the analyses across 165 all significant segments. Some analyses were not conducted due to the small sample size in either the mentions or non- mention category. These are indicated in each analysis as "no reSponse". The analyses in each segment follow. In the contractor-plan and spec segment, only one supplier mention - non-mention analysis was significant. Table 2 presents the summary information for this segment. As may be noted for Suppliers A and B the discriminant func- tion did not significantly differentiate between groups. For Supplier C the sample size was too small for analysis. Finally, the discriminant function was significant in dif- ferentiating mentions versus non-mentions. The order Of im- portance of the factors was: 1) Distribution Service 2) Miscellaneous 3) Sales Service 4) Operating Lifetime Contractors evaluating this supplier indicate that the dis- tribution service factor is most important in selecting the supplier. By examining both the directional nature of the coefficients and the centroids, indicates that above average ratings on the distribution service factor tend to be signif- icant in determining the choice of Supplier D. As such the conclusion is drawn that logistics service is indeed a de- terminant factor for this supplier by contractors in plan and spec jobs. 166 eeeefimflewfim-eez - mz DEEUALAEMAm - m Hmv Hw.mu n aho.mh Hmzvmma.fiu e RNH.me Vflmzvmwm.fiu a He amfi.me 05Hm>-p onmHmmmHD prOophou uaoopmm msoocmHHmOmHz OOH>Hmm :OHuanhumHm OOH>uom mOHmm oEHpomHH mchmnomo H0 chow maH0m0h0 m3oHHom mm mH mucOHonmooo mo nacho Hm ”Opoz MN.0- 0N.0 00m.0- Nm 50N.0 00.0- 00.0 m50.0- HH H55.0- H5.H- 00.0 NHm.0 00.H- NH.H H¢0.0 0 Omcommmm Oz - HOHHmmsm mng how mHmAHmc< Oz 0 00.0- mm.0 0mm.0- mH Nm0.0 0m.0 No.0 m00.0 0m NOH.0- 00.H mN.e 0mv.0- Hm.H 00.N HOH.0- m H0.0- Hv.0 05m.0- eH 000.0 HH.H- 5m.0 HH0.0- 0N H5m.0- 5m.0 N¢.N wmv.0 m0.0- mm.N Hm0.H- < a meOHuCOU mcoHucmz-:oz mcoHpcoz Hmvmu:OHunwoou HOHHmmdm macho coHuOcsm mwcHumm ucmcHEHhOmHo pouumm wO memo: woNHwhwwamum mmOH ummm QZ< z monHzmz mmHHmmDm mo mmm>H-u a .mefiwmmmefiu NHHOOHSOU accused 00.0- 0Hv.0- 0N.0- 500.H mN.N- 00.0 H0.0 N5N.0 Hm.H- N0m.H- 00.N- N0.N- 00.0- 500.0- 00.0 050.0 mm.H NH.0 NN.0- 0N0.H 5m.0- vmm.0- em.0- 0H.0- 00.0 H0.0 N0.N N0.N MH.0 mH.0 05.H 0N.0- NH.0 00.0 m0.m N5.N 0N.0 00.H mm.m Nv.m meOHucmu mcoHucoz-coz mcoHucoz Hmv anchu mwcmpmm pOpumm mo memo: mmDh Qmu monHsz mmHHmmDm mo mmm>H-H C meOHHHHOU mGOHuGQZ-QOZ mcowpaoz macho Wmm-mwmmmb mmcmamg xwuomsuou paouhom monumm HO memo: mone mon92m2 mmHHmmDm mo mmm>4-u : enHmmmmeHu pruophou ucmuyom mwcHumm .HOHumnH WC mfimmz mmOHmoom - moeu monHzmz mmHHmmDm mo mmm>H-H G wwwohudmu mfiOHuGOZ-floz mfiowucmz my mucmHUmeoou Ho-mHanmHm machw H cOHpO:Sm onMHmmmHD mwfiwwmfl pawfiHEHHUmHQ mpuohhou unmoumm Hopumm m0 mcmoz VQNMUHmmeHm mammzHuzm monezmz-zoz .m> monezmz mmHH00pm 00 0H000Hom mmHmm .msoocmHHmumHz .ouH>Hom :OHuanHumHm mH oocmuHomEH mo H0000 on“ n HmHHmmsm .0000 0:0 :mHm-HOpomHuaou - x6 mzoocmHHoomHz - 0 .OOH>How GOHuanHpmHm - m OOH>hmm monm - N .OEHpmmHH 0aHumHomo - H - mowoo :oHHOGSM ucmaHEHHOme :H maouumm mo Howho enmeHmHemHm-eoz - mz eeaeHHHemHm - m 0.H.m.N m.H.0.N 0.m.H.N H.N.0.m m m m m omcommom oz 0.H.N.m H.m.0.N omcommom Oz 0 m 0.0.m.H 0.0.H.N 0.0.m.H m.H.N.0 m m m 02 0.0.N.H 0.H.m.N 0.0.N.H H00m.0.N.H m m 0 00002 Hm .HOHmu 2000 .wM mHOuumHucou HszOmm mmoeuHHm<2230 N ohsmHm Hm ”mopoz < HOHHmmsm 175 the basis Of the respondent's relative comparison of sup- pliers On each variable, the importance of the factor in the discriminant function is a direct indication of the determ- inance of the factor in the purchase decision. Table 7 summarizes the importance ranks of the fac- tor in the discriminant function across all suppliers and all segments. As may be seen by inspection Of this matrix, the level of determinance of the various factors varies by mar- ket segment. While the factor of interest in this research is most frequently second or third, there were two segments where the factor was first and one fourth. The results in- dicate that the level Of determinance of the distribution or logistics service factor does vary by market segment. Table 7 FREQUENCY OF IMPORTANCE RANKS IN DISCRIMINANT FUNCTIONS FOR PURCHASE FACTORS BASED ON SUPPLIER RATINGS (EXCLUDING NON-SIGNIFICANT AND EMPTY CELLS) Rank in Function lst 2nd 3rd 4th 1. Operating 6 2 5 3 Lifetime 2. Sales 7 S 2 2 Service 3. Distribution 2 6 7 1 Service 4. Miscellaneous 1 3 2 10 176 The implications of these results lead to a design of a marketing strategy which might emphasize the ability of the supplier to provide acceptable service, as well as a plan to actually provide a specified service level. Buyers do perceive a difference across suppliers in their ability to provide an above average level of service. This percep- tion then determines in combination with the other decision factors, the buyers overall preference for a given supplier. From a marketing strategy perspective, the supplier who wants to influence purchase preference should stress the distribution factor. In some segments where little perceived difference exists in the ability of suppliers to provide ade- quate service, Opportunities may exist for one supplier to gain a competitive edge Over the others by providing a higher level of service. This strategy would be particularly bene- ficial in those segments where the importance placed on the factor is also great. In other segments where the supplier has below average service he Should bring his service level up to at least average or above in order to remain competi- tive. How far above average he should go again depends upon the importance level attached to the factor in the segment. Considering the engineer group of purchase decision influencers, the level of importance attached to the dis- tribution service factor is only moderate. This leads to the conclusion that to raise the perceived level of service above average might not prove beneficial. However in the contractor plan and spec and chiller segments, where the 177 level of importance of the factor is high and the determi— nance of the factor is also significant every effort should be made to both design a high level of service into the sys- tem and to convince purchasers that the ability of the sup- plier to provide this service is above average. This supplier must identify the segments in which his performance is below average and develop a strategy to im- prove in those segments. As may be noted in Figure 2, the determinance Of the distribution service factor varies not only by segment but by manufacturer. This indicates that each supplier may have to determine his own relational model of the determinance of the various factors. His inability to provide competitive service levels may be overcome by better performance than competing suppliers on other factors. As a result, his rating on only one factor cannot be inde- pendently considered. The level of determinance Of one fac- tor is related to all other factors and as such the pur- chaser will trade Off performance in one area against another. This may in part explain why the importance of the distribu- tion service factor varies by supplier within a given segment. In planning his marketing effort a supplier may then effect purchase preference in two ways. First he may adjust the levels of the determinant factor such that he is competi- tively superior to other suppliers. This must be done par- ticularly with factors rated as highly important by pur- chasers. Second, if the competitive level of the factor is already high, the supplier must then raise his service level 178 to equal the competitors and thereby neutralize the determi- nant effect. Once neutralized, the supplier then works from a position of strength in the other factors. The primary implication for an individual supplier is therefore that the market must be segmented to identify sig- nificantly different importance profiles. Once the various factor importance profiles have been determined, the per- formance of the supplier in relationship to competition on all buying factors should next be determined within each segment. It is the level of importance in conjunction with the relative rating that determines the final purchase pref- erence . CHAPTER VI RESEARCH CONCLUSIONS, LIMITATIONS AND , . D N" _. "“E E R Introduction Chapter VI discusses the conclusions Of this research in relation to existing research as they relate to industrial marketing. The limitations of the research are presented followed by some suggested paths for future investigations of the relevance of logistics system service in the industrial buying decision. General Conclusions As stated in Chapter 11 (Research Background) several studies have identified logistics or distribution service as an important variable in the industrial buying decision. Klassl; Dicksonz; 3 4 Wind, Green, and Robinson ; and Lehmann and O'Shaughnessy found that some form of delivery and/or dis- tribution service was rated as very important in a list of industrial buying decision making criteria. All of the cited studies investigated the importance of the factor to a single lBertrand Klass, "What Factors Affect Industrial Buy- ing Decisions", Industrial Marketipg, (May, 1961), pp. 33-40. 2Gary W. Dickson, "An Analysis of Vendor Selection Systems and Decision", Journal of Purchasing, (February, 1966), p. 9. 3Yoram Wind, Paul E. Green, and Patrick J. Robinson, ”The Determinants of Vendor Selection: The Evaluation Func- tion Approach". Journal of Purchasing, (August, 1968), pp. 29-41. 4Donald R. Lehmann and John O'Shaughnessy, Difference in Attribute Importance for Different Industrial Products", Journal Of Marketing, Vol. XXVIII, NO. 2, (April, 1974), pp. 36-42. 180 group of industrial buying personnel. These individuals were primarily formal purchasing officers. These studies are re- viewed in Chapter II (Research Methodology). 5 As Bennett and Scott suggest, the entire market must first be segmented on the basis of attribute importance. Once this step is accomplished and the importance profiles isolated, the evaluation of suppliers may take place. But added to this, the actual evaluation of suppliers' ability to provide acceptable levels of performance must also be anal- yzed. It is not the absolute level of performance but the relative level to other suppliers that causes a purchase cri- teria or attribute to be determinant as Myers and Alpert6 suggest. TO reiterate the hypotheses developed in Chapter III, the research was designed to test whether or not: 1) business logistics service is an important cri- teria used by industrial purchasers to evaluate suppliers; 2) suppliers are rated differently in their ability to provide some level of service or performance of the criteria, therefore making the criteria determinant; and 3) the level of determinance varies by market seg- ment. 5Peter J. Bennett and Jerome E. Scott, "Cognitive Models of Attitude Structure: 'Value Importance' lg Impor- tant”, in Combined Proceedin s: 197l§pring_§onferences, edited by Fred C. Allvine, ( Icago: AmerIcan Marketing Association, 1972), pp. 346-350. 6James H. Myers and Mark I. Alpert, ”Determinant Buy- ing Attitudes: Meaning and Measurement", Journal of Marketipg, Vol. XXXII, NO. 4, (October, 1968), pp. 13-20. 181 The results indicate that not only is the business logistics service factor important but the importance level varies by purchase influence center and market segment. The level of importance was different within the purchase influencer di- mension as delineated by the contractor-engineer breakdown. The engineer tends to be more concerned with long run product performance criteria while the contractor is more concerned with installation of the equipment. The importance does not appear to change across pur- chase situations and product applications when presented to engineers. Their role as a representative of product user interests appear to be unchanged from the Operating lifetime characteristics including Operating and maintenance cost, service and parts availability, and service support. Con- tractors on the other hand seem to change their importance ratings depending upon equipment application and their in- volvement in the job type. Where the job characteristics in- volve potentially costly scheduling error possibilities their rating of the importance of logistics service is higher than in jobs where these problems are not as important. Perreault and Russ7 recognized that the general pur- chasing environment had an effect on the relative importance of physical distribution service in the industrial purchasing 7William D. Perreault, Jr. and Fredrick A. Ross, Physical Distribution Service in Industrial Purchase Deci- sions", Journal of MaIketing, Vol. XXXX, No. 2, (April, 1976), pp._3410. 182 decision. They grouped respondents on the basis of importance ratings of eight buying criteria including distribution serv- ice. While several patterns were observed which were signif- icantly different, the position of the physical distribution service variable remained constant. It is difficult to ascer- tain how the purchase situations were delineated, but the en- tire sample was Of purchasing agents for various products. One conclusion drawn from this study was that many purchasing agents are insensitive to poor service and a minor number are sensitive to poor service. The present research has helped identify which segments of the industrial installation mar- ket for air conditioning equipment are sensitive to service. The identification of two purchase influence centers has also added to this area. Those influencers whose roles are dif- ferent in terms of the application of the product and their representation in the buying center have different sensitivi- ties to the level of service as evidenced by the combination of importance ratings and individual supplier ratings on the various buying criteria. The evidence points to the fact that the purchase situation is just as important as the prod- uct in delineating market segments where logistics service must be adjusted to a competitive level. The interrelationships among the members of the buy- ing center also determine which factors must receive atten- tion. Tables 1 and 2 present the analysis of the various levels of decision making control by both contractors and engineers within the job types Studied in the research. The 183 000.00 5H5.00 000.00 HN0.05 000.05 NOH.50 000.0N .Hehgeou oz - H 500.H 0N0.H 000.N 000.N 00H.N 0H0.H 000.H mmmAOHsmohm gonzo 500.0 500.0 000.0 00N.0 00N.0 000.0 000.N 0HH50. -a0Hmom mmOBum mmumm zo Homhzou DZHM<2 onmHUmQ mmmH H OHan 500.0 000.0 005.0 005.H 005.H 000.H 550.H oomw 0cm :mHm .Hoppcou HmHOH - 0 ” 0cH000 acoEmHsvm mo mcHhowho ommcuhsm paoemHzcm How me0 mo coHumsHm>m unoEmHscm 000 me0 mo :oHumpHOHHom pdoEmHscm mo :OHumuHmHuomm mwmoz Eoumxm O\< we :mHmmn GOHumepmo>cH thcHEHHonm manHHam mo :oHpmoocou mmoOOH0 :H uwmum .H 184 02 02 02 02 02 mz\m 0N5.0 000.H N00.0 000.0 0N0.0 HNO.N 5N0.H Houucou oz - NOH.N 00N.N NOH.0 055.N 050.N 000.N 0N0.0 H00.0 000.0 005.0 50N.0 00H.0 000.0 000.N ommnupsmohm WMMNMI Hoazo -:0Hmon 000.N 000.0 000.N 005.0 000.0 000.0 000.0 .HOHucou oEom - 0 mammzHuzm Hm mmomm zo Homhzou UZHKmH N mHan .Hoeeeou H0060 - m 000.H 500.0 055.N 500.0 00N.0 055.0 00N.0 wommcmz-Emoe \eeumHuomez 0:000 oomm 0cm :me ”wcHwou pcosmHsvm mo 0:Hhovao ommcuhsm pcoEmHSUm How me0 mo coHpmsHm>m psoEmHsvm How me0 mo GOHumpHOHHom unoEmHscm mo :oHumoHMHommm mwooz Empmxm u\< we emHmea :OHummemo>cH HumaHEHHohm meHeHHam we coHumooaou mmoOOHm cH owmum .H 185 data for these tables was derived in section one of the data collection instrument. The level of decision making control shared by engineers and contractors at various stages of the purchase process changes in composition with job type. In the plan and spec job category (traditional method) the en- gineer has primary control over the preliminary phases where needs are determined and alternatives specified. The con- tractor has more control in the latter stages where the al- ternatives are narrowed down to the final selection. Mov- ing to the next two job types (Design Build and Team Managed) contractors and engineers trade some decision making control in earlier stages Of the process. Contractors become more involved in the purchase decision process earlier. It is at these stages where alternative suppliers are specified. The main implication of these findings is that in the job types where contractors become involved in selection Of alternatives at early stages, the determinance and importance of the logistics service factor becomes relevant not only for specifying the final supplier but also for determining the feasible alternatives. Thus the higher level of importance placed on the delivery factor, combined with the experience of the contractor with various suppliers may either exclude or include a Specific supplier from consideration as alter- native. Once the alternatives have been determined the final choice is constrained. As a result, the evaluation of the determinance Of the logistics service factor as well as all other factors is not adequate when conducted with only a 186 single purchase decision making influencer. When the deci- sion makers who consider logistics service a determinant get involved in the earlier stages of the decision making process the attitudes of engineers may be affected. As a result the provision of adequate or competitive service may have to be stressed to both groups in these latter segments. Thus the marketer of industrial equipment must identify several vari- ables in designing his strategy. These are: 1) The key purchase decision making influencers in the buying process; 2) The buying factors which are important in the process; 3) The segments of the market which have signifi- cantly different buying factor importance pro- files; and 4) The determinance Of the buying factors for various market segments. He must then work from a position Of strength where his capa- bility to provide competitive levels Of logistics service in the market segments where this factor is determinant. Research Limitations The conclusions drawn in this research have some limitations. These are in the areas of product specificity, selection of the research population and sample frame, and finally the scOpe of the study in relation to the full range of purchasing influencers and product applications. A single product category was the subject of the in- vestigation in this research. As such, generalizing the results to other product types might be approached with some 187 caution. The nature of the relationships and identification Of segmentation dimensions may vary from product to product. A product in the plant and equipment category, Specified in the design stages of a building and installed by a Specialty contractor such as that studied here would fit in the frame- work readily. The roles of the engineer and contractor are not necessarily Similar for component parts, raw materials, supplies, and accessories as subgroups of the industrial product markets. As such Similar studies need to be conduct- ed in these areas. The research population included a nationwide cross section of contractors and engineers. The sample for the research was selected from fifteen major U. S. cities. Those metrOpolitan areas bypassed in the sample may have charac- teristics which are different than the research population. The conclusions should not technically be extended beyond the fifteen city area. However, these cities are the largest concentrations of population and industrial activity. They are furthermore representative of the geographic areas where they are located. NO substantial error is therefore expected in making inferences to the entire U. 8. population of con- tractors and engineers. Finally, the investigation was limited to the two major influence centers - contractors and engineers. Sev- eral other potential influences in the purchase decision were assumed to be represented by these groups. It should not be concluded that these groups are the sole influences. Some 188 information flows from other groups directly related to these. A third highly important group whose input was not solicited was the owners and occupants grOUp. This group consumes the service of the product, but has little direct influence in the purchase decision process. These individuals need to be investigated further for their potential influences and cor- responding results in terms of marketing strategy and suc- cess. Recommendations for Future Research Although this research has hopefully made some con- tribution to how the business logistics service factor fits into an industrial purchase decision, many areas exist for further research. Initially the area Of actual service measurement across market segments and suppliers needs to be developed further. The statistical relationship between real service levels and overall performance of suppliers in terms of sales, market share, or profits is yet to be evaluated fully. The parameters of the actual service time and variation, fre- quency of failure, and related service components need to be evaluated within market segments in relation to the perceived service level as well. Both the level of tolerance and the conditions sur- rounding the reaction to failures in the system to provide consistent service or deliver complete orders, remain open for investigation. The frequency with which the promised delivery schedule is not met and the change of preference 189 from one supplier to another is another relationship which needs investigation. How many times can the system fail to provide acceptable performance before a customer switches to another supplier permanently. Finally, a methodology for estimating the current level of service which is acceptable needs development. This research has identified the fact that service performance is not only perceived differently in various segments, but has also established it as a deter- minant purchase criteria which should be studied further with respect to its quantitative nature. The trade-Offs between the level of logistics service and the composition of the service mix must also be studied. How much additional cost will be borne by the customer to affect faster delivery time and less variation in delivery remains unstudied. These are separate issues which need fur- ther investigation. Evidence exists that the absolute time is not as important as the variability (see Chapter II). The customer may simply readjust his schedule to reflect the ab- solute time. However, variability is another matter. Whether or not he views the reduction of variability as a responsibility of the supplier in his normal performance or as something which he is willing to substantially reduce by paying for better service is still the question at hand. The second trade off question involves product qual- ity and delivery performance. Is the quality reputation of a supplier sufficient enough to overcome poor delivery serv- ice. The research indicates that quality is determinant in 190 preferences toward some suppliers and not others within the same importance profile, thus other factors where little dif- ference exists between suppliers are non-determinant. Con- versely, how far above the average level Of logistics serv- ice must a supplier perform in order to overcome a low quality image. Finally, the interaction between members of the buying center needs further investigation. The information trans- ferred from one member to another concerning the service lev- e1 may be investigated more fully. To what degree does the experience of one member of the buying center with a given supplier effect the preference of other buying center mem- bers. In addition other influencers such as product users, owners, architects, and other contractors should be studied with respect to their inputs to the buying center. A more clear-cut distinction is needed for the members of the buy- ing center and their roles in various purchase situations. APPENDIX SECTION II 3:13 SECTION DEALS WITH YOUR FEELINGS REGARDING FACTORS THAT MAY BE INVOLVED WHEN YOU SPECIFY PURCHASE INDUSTRIAL OR COMMERCIAL AIR CONDITIONING EQUIPMENT. NOTE: PLEASE ANSWER THE QUESTIONS IN THIS SECTION IN TERMS OF THE JOB YOU SELECTED ON THE LAST PAGE. IIPORTANCE RATIIG EXTREMELY IMPORTANT PLEASE INDICATE THE RELATIVE IMPORTANCE OF EACH OF THE FACTORS LISTED BELOW IN YOUR SPECIFYING 0R PURCHASE SOMEWHAT DECISION MAKING. THE IMPORTANCE RATING IS TO BE BASED IMPORTANT ON THE SCALE TO THE LEFT. SEVERAL FACTORS MAY SE OF APPROXIMATELY EQUAL IMPORTANCE, THEREFORE THE SAME RATING No NUMBER MAY BE USED FOR MORE THAN ONE FACTOR. T IWORTANT EXCELLENT PLEASE RATE EACH OF THE FOLLOWING COMPANIES ON THE FACTORS LISTED BELOW ACCORDING TO THE SCALE ON THE RIGHT. You MAY USE AVERAGE THE SAME RATING FOR MORE THAN ONE COMPANY, IF YOU FEEL THAT THIS IS APPLICABLE. NOTE: RATE ALL COMPANIES UNLESS YOU HAVE ABSOLUTELY NO INFORMATTOR'ON A PARTICULAR ONE. POOR a} Q‘ ‘8. ‘1 x <\ (329% 0‘7 «F 0°? 09* 0“ 03> 0° (9“ 0 FACTOR $§Q