MARK)?! SEGMENTATiON BY CONSUMER PERCEPTION: A CASE STUDY Dissertation for the Degree of Ph. D. MICE-{EGAN STATE UNWERSHY JORGE F. GONZALEZ - ARCE 1973 LIBRARY M id! 155311 3 :3. 9.7-6 {' .n‘s‘CY-‘iiy I This is to certify that the thesis entitled MARKET SEGMENTATION BY CONSUMER PERCEPTION: A CASE STUDY presented by Jorge F. Gonzalez-Arce has been accepted towards fulfillment of the requirements for Ph . D degree in Marketing \ Major professor Date 74 A 3’ // { 0-7639 leate x» . m r. . 1 L a "V fi~ d p 2.. ”Re the: - '2‘“.- “'“uu Ac o- ”I ;.. o 2. QCOROT . II «J w . _ . ~ _ .1 .mu 8 a .n.. a» w t «C «a Q» '1‘ .o « Wu T,‘ «.3 tn .. v..." YL Do C ".1 MI; C l VHIN Cw n O S t ~44 a: .1 Q» Q» A» Nu «Nu A: Q» A: nJ #L ‘IL A u r; L... nu. «.3 x v. -\ \ v o .s .. Q. .1 ‘4. :u A v u . 2» A: .: V. o a» 2h . y or“ ~= I ~ I 7 r; .V. fix. v.v« ‘1‘ u h Cc 1.. H . he CL an 2m .1‘ re at no ¢L c. ‘15 “fit “it." A» hvv .Vl« a; .7. h; L, .. A“ L‘. a e A» r .. ~\~ a; I s A . 3 o 1. .. .. N u n ‘ oA ‘ urn ABSTRACT MARKET SEGMENTATION BY CONSUMER PERCEPTION: A CASE STUDY BY Jorge F. Gonzalez-Arce The main objective of this research is to examine and validate the application of newer marketing research techniques in a developing country. The intent is to reinforce the hypothesis that technology does not recog- nize economic or national borders as long as it is correctly adapted to local environments. 7 The methodology which was selected for testing, although widely discussed in academic circles, has not been described step by step. Nor has it been used, at least in published materials, in large and probabilistic samples. It is thus our aim to present a practical business case in which a multicity (30 areas) survey was conducted by personally interviewing a large and probabi— listic sample of respondents (14,309). A detailed description will be presented of each of the methods which were used and of the validating comparisons which wt": made in ' o- - *1 lEJ :9?.e~.ts ”"1‘Vant bY'aV‘I V‘.‘bo$ ‘.\ I I I ‘ a" 'h‘fin ‘ .::-n imam. . Jorge F. Gonzalez-Arce were made in order to generate information about market segments integrated through consumer perception. It is our hypothesis that other segmentation bases, although useful for sOme marketing purposes, are more descriptive and less problem oriented. Through the use of this, or similar methodology, for market segmen- tation, management could find out not only how the market is structured, but also how dissatisfied it is with current brands within a product line. Management also may learn which product attributes effectively can position a specific brand closer to a meaningful market cluster, an idea which seems much in consonance with the modern mar- keting concept. After an exploratory survey was conducted to test the attributes to be scaled and to establish the questioning procedure, the sampling method, and the programs best adapted to our computer facilities, the field work was car- ried out. The market of a consumer product was segmented in every city according to consumers' perceptions of an ideal brand, and a geometrical space was structured through the consumers' perceptions of current brands in the market. Several analytical techniques were combined and adapted in this research: analysis of variance, factor analysis, multiple discriminant analysis, and cluster analysis. Additional samples of 960 retailers were obtained from five cities, and an ex post facto survey was conducted fit In C' L ,- a t'wYO—S- Val .ing t "LN.“ rcutznes a L ouvu“ ‘V *0“ 3 fun: o.‘ N' FAR ‘ ~." 1 I .L.’ 'H ‘1‘“ *e_‘v.‘ gar Jorge F. Gonzalez-Arce in a city where the marketing mix of a brand had been modified. Likert type scales were integrated into the questionnaire, and respondents were selected statistically by a two-stage area sampling method. Validity tests were attempted not only by con— trasting the various individual city reports, but also by modifying the parsimony levels in the factor analytical routines and by introducing a third dimension for ideal brand locations. These later were to be clustered on the reduced geometrical space. The conclusion of this research is that, at least for this particular product and in Mexico, the method employed seems superior to other currently used bases of market segmentation. Furthermore, although more research is necessary to complement the information thus obtained, management now possesses a very useful tool for determining marketing strategy, and a tool which can be tested par- tially in different areas since a separate estimate was obtained for each area. In our opinion it also is valid to conclude that these techniques usefully can be adapted to other products and to other developing areas. Only when management is pro- vided with this type of information will it begin to believe in marketing research and begin fully implementing the marketing concept. MARKET SEGMENTATION BY CONSUMER PERCEPTION: A CASE STUDY BY Jorge F. Gonzalez-Arce A DISSERTATION Submitted to Michigan State University in partial fulfillments of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Marketing and Transportation Administration 1973 v.‘ \1; «Q '1 :5 . c: it 7‘ ‘I 3‘ ":1 {h . 6“ COPYRIGHT BY Jorge F. Gonzalez-Arce 1973 To Yolanda To Yolanda ... and our parents ... and for Jorge Alejandro Roberto Yolanda Adrian Alberto ii The be the most It is very c3 incere grat ticns, who I". tire and pat the to carry '0 . A,V'V “‘vL us ’4' (1" E1 ACKNOWLEDGEMENTS The very last stage of this dissertation seems to be the most difficult to write, even in my own language. It is very difficult to find words to express my most sincere gratitude to all the people, and their organiza- tions, who have believed in the author and have had the time and patience to sponsor, teach, assist, and encourage me to carry through. Repayment hardly can be achieved with words; it must be accomplished by deeds. It is my hope that throughout my life I will be able to do for other people what has been done for me. My.special gratitude goes to Dr. Leo G. Erickson, who never relinquished the idea that this dissertation would be finished. His encouragement and guidance made it possible. He and Professors Richard Lewis and Richard F. Gonaalez had the patience to listen to several of these ideas and provided me with the insight and theoreti- cal framework with which to construct from them. And Yolanda, GRACIAS. iii ACKNOWLEDGEMENTS . . . . . . . . LIST OF LIST OF Chapter I. II. III. TABLE OF CONTENTS TABLES . . . . . . . . . . . FIGURES O O D O O O O O O O C THE APPLICATION OF QUANTITATIVE TECHNIQUES TO MEASURE CONSUMER PERCEPTION AND BEHAVIOR O O O O O I O O O O 0 Introduction . . . . . . . . . General Considerations . . . . . . The Problem . . . . . . . . . . Research Objectives . . . . . . . Methodology . . . . . . . . . . Outline of the Dissertation. . . . . MARKETING PHILOSOPHY AND MARKETING RESEARCH IN DEVELOPING COUNTRIES . . . . . . Introduction. . . . . . . . . The Role of Marketing Research in the Implementation of a Marketing Philosophy . . . . . . . . The Role of Marketing Research in Less Developed Countries . . . . . . Introduction . . . . . . . Stages in the Development of Marketing Research . . . . . . Limitations to Marketing Research in Developing Countries ' . . . . . THE THEORY OF MARKET SEGMENTATION . . . Market Segmentation and Marketing Planning .. . . . . . . . Evolution to a Market- Oriented Business Firm . . . The Product Differentiation Approach . iv Page iii vii H deU‘H—‘H 16 18 27 27 28 33 42 42 43 44 er f‘kgrb t-Mz’ " v.‘ v. t e i . V \s . cu. ARV A”. in“ D“ G U... 1.. \g .... \ .. S . :1: I. Chapter Page The Market Segmentation Approach . . . 45 Involving Market Segmentation in the Marketing Planning Function . . 46 Methodology for Market Segmentation . . 49 Alternative Bases for Market Segmentation. 51 Segmentation by Market Characteristics . 52 Geographical Segmentation. . . . . 53 Demographic Clustering. . . .- . . 55 Socio- economic Characteristics . . . 59 Personal Characteristics . . . . 61 Segmenting the Market for a Product Class . . . . ,. . . . . . . 62 Institutional Segmentation . . . . 63 Segmenting the Intermediaries . . . 64 Consumer Behavior and the Product . . 65 The Brand within the Industry . . . 67 IV. AN ALTERNATIVE APPROACH: SEGMENTING THE MARKET BY CONSUMER PERCEPTION . . General Objectives . . . . . . . . 69 Methodology . . . . . 74 Advantages of This Methodology over Other Market Segmentation Approaches . 97 V. MARKET SEGMENTATION BY CONSUMER PERCEPTION: A CASE STUDY IN MEXICO . . . . . . . 100 Content and Method . . . . . . . 100 Description of the Case Study . . . . 100 Development of Research Objectives . . . 103 Exploratory Methodological Survey . . . 106 Questionnaire Design . . . . . . 108 Preliminary Research on Attribues to Be Measured by Consumer Per- ception on Selected and "Ideal" Brands . . . . . . . . . . 109 Design . . . . . . . . . . 110 Analysis . . . . . . . 111 Structuring the Question for Attribute Measurement . . . . . . . . . 111 Types of Scales . . . . . . . 112 SelectiOn of the Scaling Technique. . 113 Selection of Scaling Technique . . . 114 Selecting the Number of Intervals to Use in the Likert Type Scale . . . 116 V Chapter ' Page Final Draft of Questionnaire . . . . 118 Final Questionnaire Form . . . . 126 Sampling Procedures and Field Work . . . 126 Sampling Method . . . . . . . . 126 Sample Size . . . . . . . . . 129 Integrating a National Sample . . . . 130 Field Work Procedures . . . . . . . 133 VI. MARKET SEGMENTATION BY CONSUMER PERCEPTION: DATA ANALYSIS . . . . . . . . . 136 Analysis of Variance . . . . . . . 140 Factor Analysis . . . . . . . . . 148 Multiple Discriminant Analysis . . . . 159 Cluster Analysis . . . . . . . . . 174 Classifying Market Segments . . . . . 179 VII. MARKET SEGMENTATION BY CONSUMER PERCEPTION: NATIONAL SAMPLE, VALIDATION OF RESULTS . . 182 Structuring a National Sample (Weighted Averages Technique) . . . . 182 Structuring a National Sample (Stratified Random Sample) . . . . . 194 Regional Validation . . . . . . . 203 Alternative Analytical Steps Used for Validating Results . . . . . . . 211 Parsimony to Seven Factors . . . . . 213 Parsimony to Five Factors . . . . . 217 Analyzing the Ten Original Product and Brand Attributes . . . . . . . 221 Comparative Results from a Three- Dimensional Analysis . . . . . 225 Ex Post Facto TeSt in City X . . . . 227 Retailers' Perception. . . . . . . 230 Contrasting Segmentation Approaches . . 231 VIII. EVALUATIONS AND CONCLUSIONS . . . . . . 235 General Conclusions ' . . . . . . 236 Market Segmentation by Consumer Perception . . . . . . . Potential Uses of These Research Findings. 245 Additional Research to be Conducted in the Future . . . . . . . . . 248 BIBLIOGRAPHY . . . . . . . . . . . . . 252 vi LIST OF TABLES TABLE PAGE 5-1 Alternative Scaling Techniques . . . . . 114 5-2 Likert Scales Used in the Mexico City Study, 1971 . . . . . . . . . . 117 5-3 Fictitious Name Assigned to Perceived‘ Attributes . . . . . . . . . . . 123 6—1 Brand Preference Share and Sample Size in City X o o o o o o o o o o o 141 6-2 Variance Analysis of the Attribute Higher-Priced . . . . . . . . . . 143 6-3 Relative Brand Retail Prices . . . . . 144 6-4 Means and Standard Deviations of Rated Brand Attributes in City X . . . . . 150 6-5 Correlation Coefficients of Rated Brands in City X 0 O O O O O - O O O O O 152 6-6 Eigenvalues from Factor Matrix (City X) . . 153 6-7 Factor Matrix (Principal Component) for City X C O O O O C O O O O O 155 6&8 Selected Eigenvectors in City X . . . . 157 6—9 Rotated (Varimax) Factors for City X . . . 158 6-10 Percentage of Variation Extracted by Each Root in City X through Multiple Discriminant Analysis . . . . . . . 161 6-11 Factor Correlations for the Three Main Axes of the Perceptual Space, City X . . 162 6-12 Brand Location in Perceptual Map for City X 0 O O O O O O O O O O O 165 6-13 F—Ratios and Percentage of Discriminating Power of Each Factor in City X . . . . 169 vii TABLE 6-14 6-15 6-16 6-17 7-1 7-4 7-5 7-6 7-7' 7-8 7-9 7-10 7-11 7-12 7-13 PAGE Factor Discriminant Weights in City X . . 171 Brand Average Rating on Each Factor in City x O O O O O O O O O O O O 173 Selected Consumer Ideal Points in City X. . 176 Market Segments in City X . . . . . . 177 Attribute Weights for Each Factor in ' National Sample . . . . . . . . . 195 Factor Correlations for the Two Main Axes of the Perceptual Space, National Sample . 196 Brand Location in Perceptual Map, National Sample . . . . . . . . . . . . 196 Factor Discriminant Weights in National sample O O O O O O O O O O O O 197 F-Ratios and Percentage of Discriminating Power of Each Factor in National Sample . 197 Market Segments in National Sample. . . . 200 Percentages of Factors Discriminating Power for Cities X, Y, W and Z and National Sample . . . . . . . . . . . . 203 Test of Goodness of Fit for Factor Discriminating Power in Cities X, Y, W and Z . . . . . . . . . . . 204 Ideal Brand Perception in Cities X, Y, W and Z c g o o . o o o ' o o o o o 206 Brand Perception in Cities X, Y, W and Z for the Attribute Sweetness . . . . . 210 Brand Perception in Cities X, Y, W, and Z for the Attribute Density . . . . . 210 Comparison of Cluster Integration by Using Parsimony . . . . . . . . . 216 Rotated (Varimax) Factors for City X. (Parsimony to Five Factors) . . . . . 218 viii TABLE 7-14 7-15 7-16 PAGE FéRatios and Percentage of Discriminating . . Power of Each Factor in City x, Ten Attributes . . . . . . . . . . . 222 Ideal Brand Perception in City X. A Comparison of Two Different Survey Dates O O O O O O O O O O O O 228 Percentage of Discriminating Power of Each Factor in the Ex'Post Facto Test in: City X O O O O O O O O O O O O 230 ix LIST OF FIGURES FIGURE PAGE 3-1 Brand and Market Matrix . . . . . . . . 46 3-2 Want and Product Matrix . . . . . . . . 47 3-3 Want and Segment Matrix . . . . . . . . 47 4-1 Methodology for Data Analysis on Market Segmentation by Consumer Perception . . . . 79 4-2 Consumer Brand Perception Along Specific Attributes . . . . . . . . . . . . 80 4-3 Factor Correlations in the Perceptual Map . . 86 4-4 Brand Positioning in the Perceptual Map . . 87 4-5 Brand Positions and Factor Correlations in the Perceptual Map. . . . . . . . . 88 4-6 Cluster and Brand Positioning in the Perceptual Map . . . . . . . . . . . 93 6-1 Methodology for Data Analysis on Market Segmentation by Consumer Perception . . . . 137 6-2 Brand Ratings According to the Attribute "Higher-Priced." . . . . . . . . . . 144 6-3 Brand Means as Perceived by Consumers Of City X O O O O O O O O O O O O 145 6-4 Factor Correlations in City X . . . . . . 164 6-5 Brand Position and Factor Correlation in City X o o o 0' o o o o o o o o o 166 6-6 Discriminant Weights in City X . . . . . 170 6-7 Market Segments in City X . . . . . . . 178 'x FIGURE .- PAGE ‘6-8 Market Segments, Brand Locations and Discriminant Weights in Reduced Space. . . 180 7-1 Brand and Cluster Locations in City Y . . 185 7-2 Factor Correlations in City Y . . . . . 186 7-3 Brand and Cluster Locations in City W . . 187 7-4 Factor Correlations in City W . . . . . 188 7—5 Brand and Cluster Locations in City Z . . 189 7-6 Factor Correlations in City Z ‘. . . . . 190 7-7 Brand and Cluster Locations in National Sample. (Weighted Averages Procedure) 191 7-8 Factor Correlations in National Sample. (Weighted Averages Procedure) . . . . . 192 7-9 Factor Correlations in National Sample . . 198 7-10 Discriminant Weights in National Sample . . 199 7-11 Brand and Cluster Locations in National Sample . . . . . . . . . . . . . 202 7-12 Selected Brand Means for the Attribute Density. . . . . . . . . . . . . 207 7-13 Selected Brand Means for the Attribute Healthfulness. . . . . . . . . . . 208 7-14 Selected Brand Means for the Attribute Mildness . . . . . . . . . . . . 209 7-15 Brand Position and Factor Correlations in City X. (Parsimony to Seven Factors) . . . 214 7-16 Brand Position and Factor Correlations in City X. (Parsimony to Seven Factors) (Rotated Dimensions) . . . . . . . . 215 7-17 Brand Positions and Factor Correlations in City X. (Parsimony to Five Factors) . . . 219 xi FIGURE , ' PAGE 7—18 Perceptual Map in City X by the Use of the Ten Original Attributes. (Rotated Dimensions) . . . . . . . . . . . 224‘ 7-19 Comparison between Cluster Integration by Use of a Two- and Three-Dimension Procedure . . . . . . . . . . . 226 7-20 Brand Position and Factor Correlation in City X. (Ex Post Facto Sample.). . . . . 229 7-21 Perception of Attribute M Along the Different Market Segments in City X . . . 232 7-22 Perception of Attribute M2 Along the Different Market Segments in City X . . . 233 xii CHAPTER I THE APPLICATION OF QUANTITATIVE TECHNIQUES TO MEASURE CONSUMER PERCEPTION AND BEHAVIOR Introduction This dissertation attempswto examine and validate newer research techniques for market segmentation in developing countries. It is our firm belief that techno— logy has no economic or national borders and that it is possible to implement different and better methodologies if the analyst adapts them to the local socio-economic environment. General Considerations Marketing research, both as an academic discipline and as a pragmatic field has evolved very rapidly during the last ten years. It has borrowed heavily from the com- puter, quantitative and behavioral fields. Marketing research in the less developed countries has lagged behind that in more developed nations for several reasons. The most important one is the lack of experimentation with newer concepts of, and methods for, data analysis. The modern marketing philoSOphy theoretically has been adopted 1 by the largest business firms in these areas, but its implementation in real business situations has been hin- dered by the lack of sound market data that measure the impact of marketing decisions. Researchers in developing countries, instead of providing general descriptive market data, should provide management with more problem—oriented information which requires better and more sophisticated analytical tools. It is our belief that once a decision maker under- stands and tries to implement the "marketing concept," he immediately discovers that his market is heterogeneous, and that his previous "shotgun" approaches to obtain larger sales volumes need to be transformed into specific actions directed to targeted market segments. His marketing plan comes to be the integration or the sum of the specific marketing mixes planned to cater to selected and profitable market segments. An important problem for a company's management becomes that of knowing which groups of consumers are more apt to purchase its brand. To learn this, management requests market research with such a purpose. Studies of this type are called market segmentation surveys. Cross classifying consumers into geographic, demo- graphic, socio-economic, or psychographic groups has been attempted before. However, it has been our experience that these bases for segmentation, although useful for some marketing purposes, show only partial pictures of a given situation; through this type of analysis management can find out who their customers are and an explanation for their divergent purchase behavior. A decision maker, for example, can learn‘through such an analysis that, in a given area, a specific age group.from a certain socio- economic stratum prefers brand A x percent of the time. However, he must know what specific action he is to take and its probable impact on each group in order to improve sales and profits with current or new brands. An alternative approach to market segmentation is that of structuring clusters through specific con- sumer behavior, or consumer perception. If it is pos- sible for a decision maker to learn which sectors of a market are dissatisfied with his current market offerings within a product line, and, specifically, what it is that consumers expect a brand to have in terms of features in order to be better satisfied, then that information can be used to integrate market segments as a function of common— ality of consumer pattern recognition, or in other terms, as a function of consumer wants; This procedure for market segmentation can offer a marketer several distinct advantages: (1) Clusters are integrated by consumers who have homogeneously perceived wants from a product, perceptions which might not be related to market classifications. By this knowledge a decision maker is able to prepare specific marketing mixes for each of those groups. (2) It can also be learned which product features or attributes are significant for each market segment, permitting ad hoc communica- tions campaigns for each of them in terms of what they expect a brand to have. (3) It is also possible to become selective, when choosing among alternative market clusters or niches, to cater to them with current, modi- fied, or even new brands. In addition to these advantages, the measuring procedure can also be very useful in (1) learning the impact of a change in the marketing mix of a brand, either the company's or its competitor's, as the perception of current brands that has been obtained in time ti can be comparable to results at time ti (2) In the same way, +1' it is possible to compare perception among different areas since the same procedure was used to measure the same items in several places. (3) In contrast to its fore— runner, motivation research, this approach is quantita- tively measured, and the final results are less dependent on the researcher's subjectivity, and (4) it requires simpler field work. The task of measuring perception and assigning weights to variables ot attributes is removed from interviewers and respondents and is passed on to the computers. The Problem The present researcher contacted a manufacturer of censumer products in Mexico who was seeking a national, city by city, study of consumer preferences. The firm also sought descriptive analysis of the market for its several brands within a product line. Having observed that,in the current marketing literature, heavy emphasis was being placed on market segmentation through the use of consumer perception, this researcher asked the company if he could experiment with such methodology in order to examine its validity in the country. The company, one of the leaders in its field, agreed to sponsor the experiment. When the final results were analyzed, the firm altered its plans for their previous research project and decided upon one conducted along these lines. Research Objectives The examining and validating of the newer techniques for market segmentation by consumer perception in a prac- tical case in a developing country is to be considered the main objective of this dissertation. However, there are several more research objectives in this project: The methodology which was selected for testing, although widely discussed in academic circles, has not been described step by step, nor has it been used, at least in published materials, in large and probabilistic samples. Therefore, it is our aim to (1) present a real life business case in which a multicity (30 urban areas) survey was con- ducted by personally interviewing a large and statis- tically selected sample of consumers (14,309); (2) to present a detailed description of each of the techniques which were combined and adapted in order to conduct the survey; and (3) to describe the validity of comparisons which were attempted to test the goodness of the metho- dology. This seems necessary in order to convince management and fellow scholars that this method for market segmentation, at least for this particular pro— duct and in this country, seems preferable to alternative ones which had been attempted before. It is our belief that once research along these lines has been initiated in developing countries, and once it has been proven that computers, as small as they are in such countries, can be used for marketing research, practitioners will start contributing with additional research by testing alternative approaches and surveying different products in these areas. It is only through pro- blem oriented surveys, more than through descriptive market analysis, that marketing research can contribute to the full implementation of the marketing concept. Methodology In this research of market segmentation by consumer perception several different techniques were used for data gathering and analysis. The purpose of each one can be briefly and nonmathematically explained in the following paragraphs. First, in order to find out which attributes con- sumers recognized or perceived in the particular product under study, an exploratory survey was conducted that involved several stages: (1) Revision of previous des- criptive and motivation research questionnaires and reports; (2) meetings with production control and marketing executives of the firm; (3) search in published materials; and (4) exploratory blindfold taste tests among a sample of consumers to learn about relevant product features and characteristics. Second, in order to formulate the final question— naire a decision was made to use Likert type scales. A phrase was formed with each of the selected attributes and the respondent was asked to rate on a four point scale each statement about a selected, but known by the respondent, brand within a product line. In each one of the inter- views, ten different attributes or features were asked about three locally marketed brands, and the same attri— butes were used for an "ideal" brand, one if in existence would fully satisfy each consumer perceived want. During the experimental test period, conducted by interviewing 648 consumers, it became clear that respondents were able to configurate an ideal brand and that the type of scale being used was adequate because it was differentiating among brands. Third, in order to select respondents in each one of the 30 urban areas, a two stage area random sampling method was used. This procedure first involved the unrestricted sampling of city blocks from the city map, and in each of them, a systematic selection of respondents. Fourth, three different methods were combined for data analysis: factor analysis, multiple discriminant analysis, and cluster analysis. Each of these is described below. Factor analysis is a procedure for pattern recog— nition and it was intended to reduce redundancy in the original information from respondents. If two or more of the attributes which were rated for all brands seemed to be measuring the same quality, they could be grouped together into a factor. For example, let attribute one be speed, attribute two be velocity, and attribute three be slow. If we are measuring attributes for automobiles, the three seem to be redundant (although with a different mathemati- cal sign). Through factor analysis, all of them would be grouped together in only one factor. This method can be used for cases when the analyst is undecided about which of the n attributes or features he should include in the questioning process, and he is able either to eliminate redundant ones, or to question about all of them and the program would group them into a single factor. The factoring routine begins by establishing the similarity or correlation among the original rated attri— butes for all brands. It sorts all the measurements into a number of groups or "factors" based on the extent to which they are measuring common ground. This is accom- plished by examining each individual's pattern of answers for all attributes and identifying those which show com- monality. In this way it is possible to reduce the original dimensionality of m attributes to a k dimensional— ity, where k is equal to or less than m, but where the original redundancy has been eliminated. A set of factor loadings is generated and it indicates the extent to which an attribute is associated to a given factor; those close to +1 and -l are heavy components of it. The principal component analysis, which is the method used in our factor analytical procedure, integrates those factors in such manner that they are mutually uncorrelated and each accounts maximally for a decreasing portion of the total variation among the origi— nal data. In this way, the factor chosen to be number one is the one that accounts for most of the original variation. 10 Factor two takes the larger proportion of the remaining variance, and so forth until the last factor has been dealt:vfithd It thus is possible for the analyst to elimi- nate: those factors which account for the least variation, and Ilast ones to be extracted, without a serious loss of information. This action is called parsimony. However, as some of the factors show high loadings for sneveral of the original attributes, the analyst can have 61 difficult tast in properly identifying them. To provirie a meaning for all of the factors, a varimax routine was pmerformed. It rotates, without any loss of information nor curthogonality, the previously extracted factors into a new nuitrix of rotated factors in which a given attribute has alxhgh loading (near one), and the rest of them have lower ones (near zero). This permits the identification of each factor through naming it similarly to the original attri- butes with which it is correlated. Most of the available computer programs for factor analysis would permit the researcher to perform these pro- cedures. He must decide the type of rotation to be attempted as well as the extent of parsimony which is deSired. In this manner, the analyst obtains a set of uncorrelated factors in which redundancy has been elimina— ted. The similitude of all the ratings provided by each individual respondent for every brand have been considered in thee structuring of the factors. 11 Multiple discriminant analysis was the second technique used in our procedure. Its objective is to find out which factors account for most of the difference among brands. Furthermore, once a perceptual map has been developed by using factor correlations, this method locates each brand in either two— or three-dimensional reduced geometrical space. This technique uses the ratings for each 2 brand provided by each respondent along the ten original attri- butes. First, attributes are transformed into factors by using the loadings obtained in the previous program. The discriminating power of each factor then is determined and brands are segregated according to those factor weights. A factor with the strongest weight is the one which dif- ferentiates most among existing brands. For example, if each brand of automobiles is perceived similarly along an availability scale, but differently on a quality of service scale, the availability attribute would not differentiate or discriminate among brands, but the latter attribute would. For advertising purposes, for example, the use of a theme related to availability would not produce as good results as using a theme of quality of service, as this last one is accounting for more of a differentiation among existing brands. The matrix of factor correlations thus produced enables the analyst to locate each one of the dimensions 12 on a perceptual map which is to be drawn. The factor correlations represent the names that would be assigned to each of the axes on this reduced geometrical space. For presentation purposes the analyst might choose either a two- or a three-dimensional space, but the com- puter works with the entire k dimensionality produced in the factor analytical procedures. An output from this program is the location for each a brand rated by res- pondents. A point along each one of the k dimensions is produced, and the research may position each brand in the perceptual map by using those points in the first two or three axes or dimensions. Discriminant analysis thus finds a set of formulas. If these are applied to the original ratings for each brand, they will assist the analyst in learning the weight of each attribute in brand differentiation. They also will enable him to construct a perceptual map; through the correlation of each attribute to a given axis, brands can be positioned according to the relative brand means of those attributes. If an analysis of variance is performed, those attributes whose means differ the most will be those with a larger discriminating power; those in which the means are clustered together will not have a heavy dis- criminanting weight among brands. The use of multiple discriminant analysis allows the researcher to observe 13 simultaneously the performance of all of the selected attributes in producing brand differentiation as per- ceived by the consumer. Cluster analysis was the third procedure used in this study. In contrast to the methods explained above, this technique uses the ratings for the ideal brand, not the ratings for current brands. Its objective is to posi- tion each of the respondents on the perceptual map. Once this has been accomplished, respondents are grouped according to their proximity in that reduced geometrical space. Ideal brand ratings are assumed to be what con- sumers desire from a particular product. When the regions of "desired characteristics" are established, the analyst can determine to what degree each of the current brands differs from each one of the ideal regions, or market segments. The objective of this clustering technique is to integrate market segments as a function of consumers' desires or wants from a given product. These clusters may or may not accord with geographic, demographic, socio- economic, or psychographic patterns of consumer grouping. However, such clustering would indicate which brand attri- bute should be emphasized in the campaign to reduce the distance between a specific brand and a target cluster. The ratings of the attributes of the ideal brand given by each respondent are transformed into factors 14 by using the previously mentioned loadings. Through the use of discriminant weights, each ideal point then is positioned on the perceptual map drawn with the aid of the multiple discriminant analysis program. A clustering routine next is used to place con— sumers in market clusters according to their proximity in that space. An alternative clustering technique was tested. Once each respondent had been placed on the per- ceptual map, the analyst would integrate the groups visually according to density. This later procedure, although allowing the analyst to stay close to his data, was not reliable; different eyes would produce differences in cluster integration. 1 Cross tabulations later were used to learn more about each one of the clusters and about the descriptive data surveyed. Reports were prepared for each individual city. In order to provide better basis for management's action, a national sample was obtained thrcugh a strati- fied sample, weighted by city consumption, of all reSpondents. Interurban and interregional comparisons then were possible. Outline of the Dissertation The objective of chapter 2 is to present a picture of marketing research in developing countries. First, an attempt is made to relate the role of marketing research 15 in the implementation of the marketing concept, and later, once that a description has been made of the several stages in the development of marketing research, some of the most important limitations that a researcher faces when conducting surveys in these areas are analyzed. In chapter 3 an introduction to the theory of mar- ket segmentation is presented. Also discussed are a research methodology and the alternative bases that can be used to achieve this purpose. Chapter 4 introduces the rationale for market segmentation by consumer perception as well as the tech- niques which were used in this study. These techniques are more fully presented in chapter 5, where data gathering procedures also are described. Chapter 6 outlines the data analysis, step by step, for a given city. In chapter 7, a summary of the validity tests which were conducted to verify the reliability of the technique is presented. Chapter 8 is devoted to an evaluation of, and conclusions to be drawn from, this study. CHAPTER II MARKETING PHILOSOPHY AND MARKETING RESEARCH IN DEVELOPING COUNTRIES Introduction Marketing as a business philosophy has found its basic foundation in the more economically advanced econo- mies, where private enterprises struggle to gain and main- tain consumer acceptance as their basis for continuous growth. The most advanced enterprises in these countries have not stayed within the borders of their respective nations, but have become international in their approach. When Jean Jacques Servan-Schreiber, in The American Challenge,l describes multinational business firms, he is implicitly contrasting a marketing philosophy to a production-oriented point of view. Imbedded in this marketing orientation is the emergence of a marketing—oriented information system; consumer wants and desires must be detected in order to best serve them. During the laSt decade, marketing research has been directed toward extracting useful mar- keting information for decision making from market data. Jean Jacques Servan—Schreiber, The American Challenge, trans. by Ronald Steel (New York: Atheneum House, Inc., 1968). l6 17 The "new" marketing concept would be only a good idea if management could not be provided with an intimate knowledge of the market and its dynamic change. Although marketing research, as an academic sub- ject, evolved very rapidly to cope with these new concepts, it was not until the late sixties that it began to analyze data systematically for better decision-making purposes. However, we shall observe that whereas in developed coun- tries, where larger business firms began to demand more SOphisticated data handling to obtain better information, in developing countries many limitations stood in the way of market researchers. Their ability was used less in data analysis than in descriptive research and the struggle with management to sell this information. It is our thesis, and our firm belief, that as long as researchers in developing countries remain descriptive and narrative Hreating broad and undefined problem areas), most business concerns will not have the basic marketing information, as a product of data analysis, necessary to really and fully adopt and adapt a marketing philosophy. Their view will remain product oriented, thus widening the gap between the home industries and the multinational giants, and between the less developed and the economically advanced countries. It is also our firm belief that technologies do not have national borders, and we will test and validate 18 marketing techniques in a less developed.ecomony. Obviously, when a method is adapted from industry to indus- try, there will be changes in its usage when crossing borders. Our corollary is that the basic marketing con- cept also will need adaption, but not in its basic philo- sophy, as much as in its relative impact within business structures. The newer marketing research, the more quanti- tatively oriented methodology which we will be validating, will accelerate the development of the marketing concept. In contrast to the more traditional research, oriented toward exploratory description, the newer methodology is directed toward obtaining conclusive evidence as to product and brand position in a market; it not only des- cribes it, but also prescribes specific changes necessary to the achievement of market goals. The Role of Marketing Research in the Implementation of a Marketing Philosophy A marketing business philosophy implies four main elements.2 First, the consumer is placed at the beginning and at the end of the enterprise cycle. Activity begins at a firm when it discovers an unsatisfied consumer, and it ends when the consumer is finally satisfied with the pro- duct or service acquired from the firm. The business firm 2Jorge F. Gonzalez-Arce, "Concepto Operacional de la Empresa," Administracion, ITESM, XIV, No. 79 (February, 1968) 1-3. 19 'will survive and grow as long as it continuously repeats 'this cycle. Its operational goal is to serve a market, not -tO produce a given item. Second, a market is not a geographical or demo- gyraphical set of people; it is an aggregation of consumers with different wants and purchasing power abilities whose warrts can be satisfied by patronizing different suppliers. The; business firm is only one of those suppliers. Even in the; rare case that the firm enjoys a monopoly, the con- snnner might not, in the long run, continue to demand its gocxis if he is not satisfied with the firm's performance. Third, a marketing orientation encompasses the phiIlosophy of the entire business organization, not simply the: activities of the sales, advertising,and research deguartments. The market (sum of individual consumer deci- sicni makers) is, in the end, the sole judge of every decxision made within the firm, as long as a change is pro- ducxed, objective or subjective, in the final output of the company. Fourth, a consumer orientation, although morally and_ socially accepted, should not be equated with profit reduction in the long run. Profits are the measure of : effective and efficient marketing. Having described the marketing orientation, our nex1; task is to discuss its basic input, marketing 20 information, and to investigate the role that marketing research should play in providing that input in a manner useful for better marketing decisions. Martin Bell has presented an expliCatiVe model: R = f (Er 0! R) (2-1) Results (R) in a marketing-oriented firm are a function of E, the kind and degree of effort made by the firm, that is, how well decision makers play the control- lable cards in the marketing game. Results also are a function of 0, opportunities not only observed and measured by the firm, but also taken advantage of at the proper time. Such opportunities include innovation, efficiency, obtain- ing and conserving a market niche, and obtaining a differential advantage. Finally, results are a function of R, resistances found and overcome in the marketplace, among these technological, economic, and human factors. Although from a descriptive point of View Bell's could be a useful model, a researcher must be more specific. A useful research model might be the following:4 R = f(ca,cb,...,cx,na,nb,...nx) (2‘2) 3 Martin L. Bell, "Marketing: Concepts and Strategy,‘ The Functional Concept of Marketing, Chapter 2. (Boston: Houghton Mifflin Company, 1966) pp. 26—54. Gonzales-Arce, op.cit., p.8. 21 In this model it is stated that the results (R) of a marketing decision system are a function of two types of variables: controllable and noncontrollable. The first specific task of marketing research thus becomes to obtain information regarding which variables affect results, that is determining which of them are at least partly controllable. . Concerning this first task of marketing research, there usually are two extreme points of view. The traditional manager usually is convinced that there are only a few variables which might affect results. He generally believes that sales, at most, will be a func- tion of price, credit or promotion policies, competitive action, and governmental expenditures (GNP in the most developed countries). But as the only variables which can be affected are the first two, and since pricing frequently is not easy to change, he decides only on the credit and/or the promotion solution. In contrast, the sophisticated scholar overempha- sizes methodology and usually thinks of constructing a "complete model" which will consider all of the variables that might contribute to results. It is unfortunate that many of these scholars are not more realistic, which would help close the gap between their approach and the needs of decision makers. Very sophisticated and ex post facto 22 normative models never will accomplish the task of turning a traditional decision maker into a more scientific manager. The role of the marketing researcher thus becomes one of being the intermediary between managers and scholars. He must understand not only the problem on hand and the timing of the decision, but also that better methods, if properly adapted, will produce a better knowledge of the market and better information for decision making. The second task of marketing research is to dis- cover the contribution of each variable to the final result. Expressed another way, what is the elasticity of each controllable variable? This task involves the introduction of measurement methods. Many of these are borrowed from other sciences, and it is necessary to adapt them if con- sumer response is to be known and predicted. Several theories of market response have been presented in the past,5 but a final measure of each con- trollable variable must be objectively or subjectively determined if executive action is to occur. To alter the level of any marketing variable, the decision maker must know, either from basic information produced by researchers, 5See James F. Engel, Henry F. Fiorillo and Murray A. Cayley, Market Segmentation: Concepts and Applications (New York: Holt, Rinehart and Winston, Inc., 1972). See also Robert J. Holloway, Robert A. Mittelstaedt and M. Venkatesan, Consumer Behavior: Contemporary Research in Action (Boston: Houghton Mifflin Co., 1971) for summary readings on consumer behavior and measurement. 23 or from his own experience, what the expected return of his decision might be. Common sense indicates that resources first should be allocated to that factor that contributes the most toward increasing profits, either under ceteris paribus conditions or under the hypothesis of no inter- factor correlation. From basic economic theory decision makers have learned that, when facing diminishing marginal returns, they should increase inputs until the marginal costs equal their marginal returns. In the case of a multivariable effect on profits, money should be allocated until MP1 = MP2 = ...MPX = o (2-3) The equation means that, in order to maximize profits, resources should be assigned to marketing variables such that the marginal profit obtained by the movement of any variable should be equal to themselves and to zero. For the marketing researcher, the foregoing means that his task is that of providing management with infor— mation regarding the expected returns of any outlay in any of the variables. In other words, his task is that of knowing the elasticity of each variable that may affect final results. If the two tasks already assigned to marketing research do not seem easy, the matter is further compli- cated by the fact that changes in any one variable will 24 cause an effect not only on results but also on the weight of other variables. Stated in other terms, in a real life marketing situation, the ceteris paribus hypothesis never holds true; the market variables are autocorrelated, or, in economic language, they have cross elasticities. The existence of cross correlations necessitates adoption of multivariate analysis in marketing research since unidimensional descriptions of markets are only partial views of the problem. The traditional technique of cross tabulation, although an advanced view, presents only roughly approximated schemes of the market reality. The multivariate revolution in marketing research seems 6 here to stay. "One is pushed to a conclusion that unless a marketing problem is treated as a multivariate problem, it is treated superficially."7 The marketing research problem seems to be even more complicated when managers observe differences in consumer response in different groups. A decision maker realizes he is not facing one market, but a multitude of 6Jagdish N. Sheth, "The Multivariate Revolution in Marketing Research," Journal of Marketing, XXXV (January, 1971) 14.; John A. Howard and Jagdish N. Sheth, The Theory of Buyer Behavior. (New York: John Wiley & Sons, 1969). 7Sheth, op.cit., p. 17, quoting from Ronald Gatty, "Multivariate AnalySis for Marketing Research: An Evaluation," Applied Statistics, XV (November, 1966) 158. 25 submarkets (segments). These not only are of a different size and importance for his firms, but also tend to behave differently as a result of a change in a marketing variable. His market is not a unit, but is the sum of several dis- tinct units. Market segments exist when, for decision- making purposes, the entire market does not show constant elasticity for the most important marketing variables. A segmented market multiplies the tasks for market researchers. First, they must determine which are the meaningful segments; second, they must gather information (similar to that previously described) for each segment. A firm facing a segmented market, which is almost universally the case, needs information about each segment if it wants to develop a sound marketing strategy. That strategy is composed of the aggregate of decisions made concerning each segment. Market segmentation thus is not the division of a market to facilitate attack; rather, it is the integration of a strategy based on decisions made for individual segments of the market.8 Market seg— mentation, as theory and as applied research, will be the theme of the next chapter. The last, but not the least important,task for marketing research as explained in our model of a business firm, is that of being attentive to change. Economic 8Ronald B. Frank and William F. Massy, "Marketing Segmentation and the Effectiveness of a Brand's Price and Dealing Policy," Journal of Business, XXXVIII (April, 1965). 26 development occurs within a dynamic market which seems to be changing faster all the time. Alvin Toffler, among others, has discussed how slowly human behavior changed in the past as compared to how rapidly it is changing in today‘s world.9 If marketing researchers are to provide informa- tion for decision makers on a continuing basis, their tasks acquire a different perspective. They should not concentrate merely on present structures but should pre- sent relevant information regarding actual market dynamics and expected future trends. When market response elas- ticity results, it must be tested to verify its consistency through time. Human wants, beliefs, and behavior, or the entire socio-psycho-economic structure is not, by any means, constant for a long period of time. More rapid changes are to be expected in the future; if a business firm is to sur- vive and to grow, it first must comprehend that dynamic quickly and must adapt to it. In summary, if marketing research is to become decision oriented rather than simply describe situations, it must do several things. It must provide meaningful information regarding which variables have an effect on results and must measure their relative importance for a giVen problem. Measurement should be made through the use 9Alvin Toffler, Future Shock (London: The Bodley Head, Ltd., 1970)- 27 of multivariate analysis, applied not to the whole market, but to the most relevant segments. Finally, researchers should bear in mind that the dynamics of the market can alter significantly the information obtained about the previous items. The Role of Marketing Research in Less Developed Countries Introduction Marketing research as a source of information for decision making purposes seems to be positively correlated to economic development and to the marketing orientation of businessmen. Business firms in less developed countries are smaller than those in developed countries, and many of them, even those dealing with consumer products, tend to be production oriented. In these circumstances marketing researchers hardly can expect a growing demand for their Services. I In developing nations market research tends to be descriptive. In an effort to cover a large field, broad problem areas simultaneously are surveyed within one pro- ject, or "cooperative" studies are conducted in order to reduce Costs. Unfortunately, these attempts have a greater probability of failure since the information thus produced only incidentally coincides with the needs of a decision maker. 28 In other instances market researchers are called upon only when the problem approaches utter failure, and too much is asked of them. The decision maker does not want information for the solution to his problem, but wants the entire problem of the firm to be solved by the researcher. Growing government intervention, increased com- petition from international companies, and the emergence of businessschools and management courses have had a posi— tive effect on business firms. They have been forced to re-evaluate their role in society. The larger companies, managed by the new generation, have begun to look at more advanced managerial methods, and it is not unusual to find many of these firms using a technology as advanced as that of their counterparts in the developed countries. A demo- cratic, or people-oriented, organization has been one end result of this new management philosophy, and it has been within these organizations that a marketing orientation has emerged and a decision-oriented brand of marketing research has found its application. Stages in the Development of Marketing Research When a decision maker must select among alterna- tives, he must have a way of acquiring marketing information. This function of a business firm always has existed, but it is only in the last fifteen years that we have witnessed 29 the emergence of marketing researchers. Only recently have business firms begun to feel the need for a more systematic, objective, and scientific method for obtaining better information in order to reduce uncertainty in their decisions. The development of this information seeking acti- vity has followed several stages. The steps have paralleled the growth of the business firms and their increased marketing orientation, both in developed and in developing countries. Stage one of marketing research is conducted by a man with line or field experience who is capable of summar- izing both his knowledge and current reports from the salesmen in the field. He acts as a secretary and adviser to the sales manager. Occasionally the firm will buy market information from outside specialized sources when they press hard enough to sell their services. In stage two, sales analysis begins to be con— ducted by using the existing accounting data within the organization. The introduction of mechanical accounting processing equipment produces, as a by-product, data which, if properly classified, serves as a basis for comparing results obtained by different salesmen or about clients, products, or regions. Trends begin to be studied and used for market projections or quota assignments. 30 During stage three, a country continues to develop, both government agencies and private organizations begin to publish a larger amount of economic data. This the company begins to consider by relating it to its per- formance. Stage four finds the company facing the "marketing shock." It discovers: that marketing and innovation are the key business functions;10 that there is a marketing concept and that the center of the whole industrial struc- ture is not the product, but the consumer;11 and that it is highly possible that it is experiencing a marketing 12 myopia. The business firm multiplies its efforts to "look outside, and its philosophy changes from a basic product to a basic marketing orientation. The formerly uncoordi- nated activities of sales, advertising, public relations, and distribution are gathered under a department of divi- sion head, and a marketing research organization enters the field, either organized within the firm or contracted through outside consultants. 0Peter F. Drucker, The Practice of Management (New York: Harper and Row, 1954) pp. 37-41. 11 J. B. McKitterick, What Is the Marketing Manage- ment Concept? 'Proc. Americatharketing'Assobiation’f (Chicago: The Association, 1957) pp. 71—82. 12Theodore Levitt, Innovation in Marketing: New Perspectives for Profit and Growth (New York: McGraw Hill Book Co., 1962). 31 In this stage, the market researcher faces a huge descriptive task. His function becomes that of sys- tematically beginning to construct a data bank of basic marketing information. His emphasis is on secondary sources of published data and on constructing an organization to systematically gather data from primary sources. Question- naire construction, sampling procedures, and efficient field work are the basic concerns, as are middlemen, advertising content, the media, and the ultimate consumer. Needless to say, most of this information is not actually used by management, since it tends to be descriptive and not directed to problem solving. The researcher will have to ensure that correct data are provided if comparisons are to be made and if decision-oriented information is to result later. In stage five the marketing research organization, after establishing effective and efficient data gathering procedures, now is in a potition to integrate a marketing- oriented information system with the aid of the basic decision makers in the firm. The firm's research organiza- tion is in the position of feeding it more precise decision-oriented information. .A group of models have been constructed which present not only a picture of the market, but also of its change through time and of its implicit behavior. Measured historical trends and basic correla- tion of the different market variables, in addition to 32 specific research, are used to predict, within an expected deviation, future market behavior as a response to changes in marketing inputs. Attempts to arrive at this stage without first at least going through stage four, or introduction of stage five to a product-oriented firm will produce frustra- tion both for the researcher and for the decision maker. External consultants or internal researchers, when over- selling their services by promising too much from their projects (as did motivation researchers in the past), have been responsible for the "overcautious" attitudes of many decision makers regarding research, researchers, and their methodologies. Finally, stage six is reached. Larger and faster electronic data processing equipment and basic hardware, including simpler computer languages and more advanced programs, enable market researchers to perform more sophis- ticated data analysis and thereby produce better informa- tion for decision makers. Techniques imported from the behavioral sciences and the statistics fields rapidly have been adopted and adapted to analyze quantitative and qualitative market data. The researcher thus can perform multivariate analysis instead of using the former unidimensional approach. This information explosion, which has affected almost all large-scale organizations in developed 33 countries, soon will spread to their counterparts in the developing nations, thus changing the role of marketing researchers there. The increasing complexity of the home markets and the need to compete with manufactured (instead of raw materials) goods in international markets, will force organizations in developing countries to understand and validate this new information technology rapidly so as to be able to remain and grow in these markets. Market research has played a minor role in devel- oping countries, due to reasons which will be detailed later. It is our belief that at present it is at stage four, as most of the published material seems to be at the descriptive (exploratory or theoretical normative) level. If the gap between advanced and developing countries is not to be widened in the future, we believe it is necessary not only to introduce and implement a marketing-oriented philosophy in these areas, but also to change the role of marketing research so as to produce decision—oriented information. Limitations to Marketing Research in Developing Countries In an attempt to integrate the previous points 'with the present study's market research experience in a - especially when in need of an interdisciplinary approach, one is forced to resort to the current literature to seek new ideas and to validate older ones. Previously conducted studies, mainly those classified as motivation research, seem very useful in the formulation of the first list of attributes, along which perception later will be measured. A decision must be made as to which type of measurement scale will be used, as well as to the specific computer programs for data analysis. An additional set of problems arises if research is to be conducted in a develo— ;ping country. These include the absence of literature, ;poorly conducted previous motivation research, the limited availability and capacity of existing computers, and the Iieed to adapt the techniques to be utilized to the idio— syncracies of both management and consumer respondents. Stage three is questionnaire construction. The (questionnaire must be designed so as to provide at least 76 interval scaled perception ratings of respondents regarding two items: the "ideal brand," that which, if it existed, would satisfy consumer wants on each of the rated attributes, and at least one brand currently on the market and known by the respondent. Additional information about preferences, both present and past, about current objective buying lbehavior--what, when, where, how much, and soforth, about ! demographic and socio—economic characteristics, and about L personality traits also could be included. Such items t would provide descriptors of the clusters formed by con- é sumer perception. The quantity of information to be gathered from respondents is limited by the type of field work conducted and by the type of respondents from whom the information is collected. The quality of responses, rather than their relative quantity, should guide the structuring of the questionnaire. - The fourth stage is sampling. Three basic problems arise regarding this stage of the methodology: determin— ation of the unit to be sampled, the sample size, and selection of respondents. Obviously, if research is to be used for decision-making purposes, a probability sample must be drawn to assure valid generalizations about the universe. Most of the published research that follows a similar methodology has been conducted on an experimental basis, without using probability sampling, and on a very limited number of observations. 77 In order to determine the unit to be sampled, a decision must be made about whether a multistage segmen- tation approach is to be followed.7 Every consumer type can be included in the sample, or a pre-segmentation can be attempted, for example, rural areas, older people, women, and so forth, may be excluded. In a segmentation study sample size must be larger than would be the case in descriptive projects since a sufficiently large sample must be obtained from each mar- ket segment. The sample size for smaller clusters should be at least 30 in order to provide statistical confi- dence. This minimum size will produce an error of 18.2 percent at a 95 percent confidence level which still is very high for comparison purposes. The selection of respondents should be on a random basis since perception is not evenly distributed among demographic groups. The nonresponse and not—at-home rates must be kept under control since they could bias or influence total results from the sample. Stage five is field work. Consistent procedures are to be followed to secure unbiased responses. Field supervision is recommended on a fairly large scale in order 7For a detailed discussion on the multistage approach to market segmentation see Chapter III, "Alterna- tive Bases for Market Segmentation." I." « 4. 1} w-r 78 to validate answers from respondents. Personal inter— viewing, rather than telephone or mail surveys, is recom- mended in order to assure lower nonresponse rates. Interviewers have to be selected and trained accordingly. The sixth stage involves administrative work. Questionnaire coding should be made, taking into considera- tion the computer programs which will be used for data analysis. A separate and distinct code should be assigned for each attribute rating on each one of the different brands, as well as for the "ideal brand.; The use of a computer data validating program is suggested in order to eliminate key punching errors. Stage seven is data analysis. In this stage the main differences arise between a descriptive market survey and a market segmentation study. Various researchers have conducted data analysis by making different assump- tions about their information, thereby using specific analysis techniques. A summary of our methodology is pre- sented on Figure 4-1. The various kinds of data analysis discussed below are analysis of variance, factor analysis, multiple discriminant analysis, cluster analysis, testing for validation, and cross tabulation. Analysis of variance tests respondent ratings on attributes of each brand for statistically significant dif- ferences. Unidimensional analysis is conducted for each (ran-.4:- MILD NOIK ban-u A? has: TL All“ 0 52/4407 79 Kuuwsm .ov VA!\Q0r data analysis; (3) combination of several methods in Orrier to arrive at meaningful market segments; and (4) the \Efilidity of the approach from the decision maker's point Of View. 104 Several explicit potential objectives to be achieved through a market segmentation by consumer per- ception research were established. The first objective was to learn and to quantita- tively measure how the consumer market perceived each one of the brands within the product category. Second, information was sought about which attri- butes consumers were considering when evaluating the available brands. The validity of previous motivation research findings were to be tested and, later, the weight of the attributes as assigned by the consumer market were to be determined. Third, conclusive evidence, from management's point of view, was to be gathered regarding the consumer's ability to differentiate unbranded products in order to learn if brands were objectively discriminated against or if their relative market share was a product of subjectively perceived differences among them. A fourth objective was to test the validity of the "average consumer" assumption as an adequate target market, OI? to learn if the existing and measured demographic and sx>cio-economic segments tended to have homogeneous product arud brand perceptions within each cluster. If the first asssumption was true, mass advertising would be justified; 105 if the market was segmented according to perceptions, a spe- cialized and targeted communications mix would be called for. Fifth, it had to be determined to what extent current brands, both Company X's and its competitors', were satisfying consumer wants. This would help decide if new brand(s) were or were not necessary and, if so, what objective and subjective characteristics new brands should possess. Sixth, in contrast, evidence had to be found regarding proposed changes in image (position) for existing brands in order to make them more competitive in the marketplace. Seventh, the market had to be segmented according to perceived wants and tests made to determine whether this clustering procedure was in accordance with previous seg- mentation bases. If not, a more market-oriented decision tool would be available for marketing planning and strategy formulation. Eighth, evidence had to be gathered regarding andvertising themes as related to relevant product attri- kn1tes. By knowing the weight of each attribute (and their ixitercorrelation) as perceived by each relevant market Sehgment, alternative advertising themes, copy, and media COIild be used. If advisable, such material had to be 106 tested in selected experimental areas where "before" measures had been obtained. The final objective was to discover whether significant differences existed between retailers' and consumers' brand perception and attitudes. Changes might be necessary in retailers' public relations and promo- tional strategies. Similar comparative research was pro- posed, but not carried out, to contrast salesmen's perception and attitudes toward Company X's brand and those of its competitors. Exploratory Methodological Survey An exploratory survey was sponsored in order to test the operational tools which originally were designed to segment the market by consumer perception. Several self-imposed limitations restricted the undertaking. Among these, the one which would become the most important was the fact that no additional field work would be car- ried out, although questionnaires might be changed to .include necessary information without excluding previously Emithorized questions for demographic and socio-economic segmentation. However, there were no limitations on the pro- cxedures to be utilized for data analysis, and there was no tiIne pressure as long as the rest of the information was Pressented according to schedules previously approved. The 107 researcher was able to use field work to be conducted in the main urban concentrations of the country by adding a few relevant questions in the questionnaire which was going to be used. Coincidentally the decision was made at the Instituto Tecnologico de Monterrey to replace its IBM 1620 with a CDC 3300. This provided a larger and faster unit for calculations, but, above all, made more computer time available. Two computer programs for data classification were developed by the Systems Department of the institute, "TAB-ENC" and "TAB-CRU." These reduced the burden of data tabulation to a minimum and opened the door for data analysis. Prior to 1968, hand tabulation, Royal McBee pre- punched cards, and mechanical sorting of keypunched cards were the only methods for tabulating market reSearch data. Although several larger companies had been using electronic computing facilities since the late 19505, the machines' memory capacities were limited, and most of their avail- able time was being used for administrative and accounting purposes. No general market research data tabulating pro- grams, such as the ones developed at ITESM, had been used before, although some specialized data handling routines had been privately utilized by some researchers. The main limitation of specialized programs, as opposed to general 108 use ones, such as that mentioned above, is that for every new type of information desired from the original data, the program must be modified. It thus becomes a nonroutine (test) program for the computer departments, and these programs usually are run only once a day which would limit the time available for data analysis, simulation and experimentation. At the same time, and serving as a basis for this study's approach to market segmentation, there was a growing amount of literature on multivariate analysis for market data appearing in marketing and business journals and even in specialized publications. Systems' salesmen of the competing computer firms also were instrumental in acquiring and implementing special programs for market data analysis.1 Questionnaire Design Jean Morton-Williams has remarked that Every stage of a market research survey is of vital importance if valid conclusions are to be drawn from it. But the design of the ques- tionnaire is certainly one of the most criti- cal phases. If the required information is not covered or if the questions are posed in such a way that they make no sense to the 1For a summary of the methodology for data analysis, see Figure 4—1 in the previous chapter. 109 to the informant, no amount of clever inter- viewing or ingenious analysis can produce useful results. Several steps had to be taken before arriving at the final questionnaire used in the exploratory test con- ducted in Mexico City in May, 1971. That survey was fully analyzed, and by the time the national sample was taken, the questionnaire had suffered more modifications. These will be explained later. Preliminary Research on Attributes to Be Measured by Consumer Perception on Selected and "Ideal"_Brands Noted below are the main areas researched before arriving at the attributes to be included in the question- naire in order to rate consumer perception. First, motivation research reports previously con- ducted in Mexico regarding the product class under investigation were probed for product attributes perceived by consumers. Second, since some of the questions included in previous market descriptive surveys were "open" questions, an extensive reviewing of their respective answers was conducted for similar purposes.' 2Robert M. Worcester, Editor in Chief, 'Consumer Market Research Handbook- (London: McGraw-Hill Book Company, Limited, 1972) p. 69. Chapter 4, "Questionnaire Design" is written by Jean Morton-Williams. 110 Third, some of the published literature contained parts of questionnaires that had been used. Several of the rated attributes found on these were included in the first extensive list out of which the final attributes to be measured finally were produced. Fourth, conferences were held with experts from the Product and Quality Control Departments of Company X in order to learn about the physical attributes of the product and about the differences among brands. They were asked what differences consumers objectively might be expected to perceive, both on branded and on unbranded items. Fifth, an informal opinion survey was conducted among salesmen and marketing executives regarding product and brand attributes, including both their personal per- ception and what they had observed in consumers. Finally, an exploratory consumer survey was considered in order to arrive at the final attributes to include in the questionnaire form. Design: A nonprobability sample of 100 consumers was selected. One by one, and in different consumption situations, they were presented-with four unbranded items (Df the product. They were asked to look at, feel, and ‘taste each one for as long as they chose. They were told 'that questions regarding a new brand, which was included annong the four, would be asked. 111 The four unbranded products actually were only two different, but existing and strongly locally demanded brands produced by companies X and Y. An open question was asked the consumers about which of the four they liked the best and the least. After the selection was completed, open "why" questions were asked, probing for, but not suggesting, differences in the attributes observed. When P ‘ In: more discriminating information was obtained, the (nunsumer was questionned about which he considered the Wbest” brand in the market and which was his "most" pre- , ferred brand. Inquiries were made regarding reasons for his;;preferential choice as well as attributed advantages over: other brands. Finally, each interviewee was asked to snaggest to the producer and marketer of the new brand ways; it could be better adapted to his wants. Analysis: All of the interviews, which had been recomxied by permission of the respondent, were tabulated accortiing to perceived attributes. A final list, which will lye disclosed later, was constructed to be included in the prwaliminary questionnaire. StruCthring the Questions for fiEEEggflgte Measurement Measurement of attributes, real or perceived, imp1i£as the construction of a measurement system. Such a Systfifln has been developed in the behavioral sciences, and 112 it has been widely applied and adapted by marketing 3 Open researchers. This system is the scaling technique. questioning, although very useful in an exploratory search for attributes, will not provide adequate measures. Types of Scales.--Green and Tull have stated that Scales can be classified into the following major categories: (a) nominal, (b) ordinal, (c) inter- val, and (d) ratio. Each scale possesses its own underlying assumptions regarding the correspon- dence of numbers with real world entities and the meaningfulness of performing various mathematical operations on these numbers. It can be mentioned that the measurement of real-world entities can progress from scale to sgale as our knowledge of the phenomena increases. Nominal scales are used for classifying items in diffkerent and mutually exclusive categories. Examples inclnide male-female, age, socio-economic strata, and geographical region groupings. Ordinal scales assign numbers to ordered elements to represent their relative rank. Examples are "the best liked brand" and "the least important attribute." Interval scales, in contrast to the former, involve a consrtant but arbitrary unit of measurement. As no "natu- ral" zero point is involved, and it is arbitrarily set; no ¥I 3See Russell L. Ackoff, Scientific Method (New York: John VViley and Sons, Inc., 1962); Paul E. Green and Frank ‘?° CaITInone, Multidimensional Scaling and Related Techniques $2.5E255etingAnalygis (Boston: Allyn and Bacon, Inc., 1972» and Ekiul E. Green and Donald S. Tull, Research for .MEEEEEging Decisions, Second Edition (Englewood Cliffs: Prentice Hall, Inc., «1970). 4Green and Tull, op. cit., p. 177. 113 value on an interval scale is some multiple of another. However, it is possible to perform linear transformations, as the relative magnitude of the new values are the same as they were originally. This is an important and neces- sary characteristic for our study as it enables the numerical transformations explained in chapter 4 to be performed. Ratio scales "represent the 'elite' of scales, in that all arithmetic operations are permissible. . . .they Imossess a unique zero point. . .and, as the name suggests, equal.ratios among scale values correspond to equal ratios anmnag the entities being measured. . . . We can move from one :scale to another by merely applying an appropriate mulixiplicative constant."5 Types of Scale Techhiques.--Several types of scaling techniques are available to a researcher. Their description and usefulness is found elsewhere, but a list containing the ones most commonly used is included in Table: 5-1. In the next section the reasons for selecting a SPeCLific technique will be presented. 51bid., p. 180. 6Delbert C. Miller, Handbook for Research Design Eflgfiéfigggal Measurement, Second Edition (New York: David PflcKay Co., Inc., 1970). 114 TABLE 5-l.--Alternative Scaling Techniques 1. Thurstone's Law of Comparative Judgement 2. Thurstone Equal-Appearing Interval Scale 3. Paired Comparisons Technique 4. Guttman Scalogram Analysis 5. Ranking Individual Products »_1 6. Ranking Pairs of Similar Items 7. Differentiating the Most Alike and the Most Different Items 8. Descriptive Rating Scales 9. Semantic Differential Scaling 10. Likert Type Scales Selection of the Scaling Technique.--After analy- zing 'the differences, advantages and limitations of the abovsrementioned scaling techniques, the Likert scale method was cfliosen for the measurement of perceived brand attri- butes. The bases upon which this decision was made are Preserrted below. First, the researcher had used most of the methods for scxale measurement except (1) Thurstone's Law of Compainative Judgement, (4) Guttman Scalogram Analysis, (6) Rankirng Pairs of Similar Items, and (7) Differentiating the .Most Zklike and Most Different Items, but a partial appli- cation of this latter method was made in the exploratory Stage of the project. 115 Second, techniques 1, 2, 3, and 4 have been designed as complete packages to measure consumer motiva- tions regarding a given product and not for the measurement of individual perceived attributes. Third, technique 5, Ranking of Individual Products, 'which previously had been used by this researcher, would only produce ordinal scales. Fourth, methods 6 and 7 have been reported in the current literature as those used for products attribute L gnarception, but their inclusion would involve a longer L qnnestioning time with each respondent and thus either higher cxnsts or smaller samples. Let us remind the reader that one: limitation was that only one set of questions could be addexi'without altering the probability of obtaining answers in tflie remaining questions already approved in the survey. Fifth, method 8, Descriptive Rating Scales, is very liseful when the respondent is given the questionnaire forfllaand himself selects an answer for each question. As a Prcuportion of the sample to be selected was assumed to inclruie a sector of the least educated market, and since COmPaJLable data had to be obtained, it was decided not to incltuie two different sets of questionnaires, regardless of time advantages implicit in this scaling technique. Sixth, a final decision had to be made between teckunique 9 and technique 10. Our criteria led us to 116 select the latter since it involves the use of only one descriptive adjective for the attribute that will be rated by the respondent on an "agreement-disagreement" scale. The semantic differential technique involves selecting a set of polar adjectives for each attribute, but it runs the risk of their being misinterpreted by the extreme sectors of the market. In any case, it seemed safer to ask respondents to understand only one, and not two,adjectives regarding each attribute, so the Likert Type Scale was chosen. In a privately conducted survey where bOtJI techniques were applied to the same individuals on reliitively easy to rate items, such as colors, responses shovned no statistically significant results. Selecting the Number of Intervals to Use in the Likeurt Type Scale.--Having decided on the attributes to be measnired and the scaling technique to be used, a decision had txa be made as to how many scales to include, and whetfuer the "undecided" level would be used. Privately conducrted surveys in Mexico through the use of part-time interwriewers, which did not probe for definite answers, had Sruown that a very large proportion of answers would fall inn that interval. In contrast, in the "Investigacion del (kinsumidor Regiomontano" study,7 where Thurstone Equal \ 7Jorge F. Gonzalez-Arce and Guillermo Marcos, gnvfflyti.acion del Consumidor Regiomontano (Monterrey: eSarrollo Industfial y Comercial Mexicano, A.C., 1968). 117 Appearing Interval scales were used, thus forcing the respondent to either agree or disagree with the selected statements, a very small percentage of the population fell in the "undecided" interval. Regarding the number of scales, a decision was made to use six different alternatives in the manner shown in Table 5-2. 'LABLE 5-2.--Likert Scales Used in the Mexico City Study, 1971. Scale Value In Complete Agreement + 3 .Mucri in Agreement + 2 Agree + 1 (No Answer) 0 Disagree - 1 Much in Disagreement - 2 In Complete Disagreement - 3 ‘ If the respondent provided no answer for one or two 0f the; ten attributes, those variables were assigned a value Of (b equivalent to a "no answer," but if he did not res- pond the three or more questions regarding a specific brand, all ITESponses related to a given brand were rejected. 118 After the Mexico City exploratory test was com- pleted the extreme interval scales "in complete agreement" and "in complete disagreement" were deleted from the questionnaire.. It was found that consumers, especially those from the less educated strata, did not understand the meaning between the levels "in complete. . . and 'finuch in. . .," and that consumer data tested by combining those two scales into one through analysis of variance resulted in very insignificant variations which did not :fiistify the risk of receiving "garbage" data from the market. As a result of the exploratory test, field work questioning was reduced to four scales: "SI-SI," "SI," "NO" and.‘"NO-NO" intervals, but including the 0 level for "no answer" in the manner explained. FinaJ.jDraft of Questionnaire After the first exploratory test in Mexico City durixu; May, 1971, which included 648 personal interviews, seversil minor changes were made in the questionnaire forms. All tlirough 1971, and as the national survey continued, eightmore urban areas were researched, and 2982 personal interWriews were conducted. During this period the entire methCKiology, including data analysis techniques, were teStEKi, as well as management's receptiveness to this 119 In January, 1972, a final questionnaire form was drafted for the national survey. Its content next will be described. One part of the questionnaire constituted a pro- duct and brand descriptive survey. 1." "What brands have you purchased during the 1 last two months?" If the respondent had not bought any brand during this period, the interview ended, since he xmould be considered a “non-consumer." The order in which 1 'the brands were mentioned was used as an indicator of brand awarene s s . 2. "What is the one that you purchase most often?" Tune answers to this question were considered indicators of lxrand preference. An alternative, "which is your most prefkarred brand?" might have produced answers such as "any? one," "all of them," "the one that is available," and 'so fcurth. "Which is the one that you purchased last time in tkua market?" would have assumed no brand switching among consuuners. Although we agree on not having a perfect indi- cator, this method of formulating the questions seemed Superinor to alternatives. 3. "What is the one that you used to purchase mOSt c>ften a year ago?" By contrasting answers to this queStxion to those on the last one, an indicator of brand SwitCfliing was obtained. For the particular product 120 studied this behavior is higher among the younger and 11eavy consumers than among the rest of the population. It 51180 was observed that during the 1971 test period, when tzlue price of all brands had been increased by 25 percent, czcansumers classified in the lower socio-economic strata 11£ad been changing from premium brands to lower priced JkDJrands. The opposite behavior was observed by analyzing éizata obtained during 1973, when the general economy of the c:<:>untry was growing, thus validating "elasticity" figures otherwise obtained for thebrands of the company. 4. "How many units of this product do you buy ;E>eer week? (assume a normal week period)" Answers to this ‘QILlefition were taken as indicators (biased) of quantity EDLlrchased. However, the recorded figures were grouped erito three wide categories to represent "light," "moderate,' Eirid."heavy" buyers. Contrasting these results to figures <>k>tained otherwise, the total percentages were not very C1j.fferent. 5. "The product is sold in the market in packages (>13 different sizes. Which is the one that you purchase the Answers to this question were very approxi- While the InC>st often?" Inlétte to current sales figures in every city. gilleestion was used as an indicator of the validity of the Survey's results, it provided an additional basis for CIJLEassifying consumers. Much as expected, lower income 121 :purchasers prefer larger size packages. It is interesting to note that in one of the cities, where this figure was IlOt in accordance with real sales quantities, a cross tabu- Jnation was made by the interviewer, and it was found that 1:]iree of them were biasing the results. A new survey was czconducted in that area which showed "acceptable" results. .ngfter this "test," the remaining cities also were tabu- ZLsated to discover this type of error, not only for this [cgtnestion, but also for the entire questionnaire. 6. The "image question" was: "As you know, there 613:8 several types of brands being sold of this product. ‘VVIjich is in your opinion SEE 'premium' brand? And the "vvorst' of them? And a brand 'just in the middle'?" By vveeighting responses on a 3-1—2 scale an image rating was c>1>tained. Three very interesting results were produced Eiifter cross tabulating by the most preferred brand. First, <3c>nsumers buying a regular priced brand did not consider 5_t: the "best" on the market. Second, image thus measured, Eilthough ranking brands according to price for the €3>ctreme cases, ranked brands according to relative local 11market share. Third, when further classified by clusters, tléihmlation produced figures correlated to brand position (Drl the perceptual map, but if tabulated along traditional <2lusters (demographic and socio-economic) brand preference ifCDI:esch of those groups was explained. Although additional 1:63:3earch is necessary to arrive at "real image" ratings, 122 ‘this questioning technique produces good and acceptable results which can be correlated against advertising and pricing strategies. "Effective" advertising for brands Vdcithin the same price range undoubtedly will produce a 11:igher image rating in the targeted clusters for the given brand. 7. "What brand do you purchase when yours is not available?" A second preference figure was the outcome of this question, and it can be usefully applied in order to measure brand "cannibalizing" in each of the market segments. Perceived attributes of second choice brands should be the most similar to the one most preferred. 8. "In what type of outlet do you purchase the product?" Results should be in accordance with sales ffjngures for each brand.‘ If so, answers usefully can be Ei£>pflied for segmented distribution strategies. A second set of questions concerned the attributes (Di? the "ideal" brand. In order to learn perceived satis- fied and unsatisfied consumer wants the Coombsian approach 11C) ideal brands was taken. Eleven different attributes VVf package, which was included for the ideal brand, was caseleted for the brands currently in the market, since the izjype of package is the specific characteristic of some o f them. . R 'IIZXBLE 5-3.--Fictitious Name Assigned to Perceived : Attributes. Number Name 9 1 Sweetness 2 Density 3 . Higher-priced 4 Availability 5 Aftertaste 6 Mildness 7 Odor, smell 8 Strength 9 Flavor 10 Healthful For obvious reasons, the specific attributes InEiasured have been changed for publication purposes. In (Drkier to identify them, a given and fictitious name is 124 assigned to each one, but the reader should be aware that they would not describe the market and that specific mar- }ceting action developed from the results presented herein might produce undesirable marketing effects. Furthermore, as expected, the mathematical relation for some has been inverted to obscure "real market" pictures. Each one of these attributes was rated on the " +2,-2" scale previously described, and these answers were fed into the cluster program as explained in chapter 4. A third group of queries involved perceived attri- butes for the "most purchased" and two additional brands sold in the local market. The first part of this section contained a phrase in which the interviewer would reinforce the answer obtained for the "most purchased" brand. He would begin by saying: "Now, let's talk about brand x, the one you said you purchased the most often, OK?" If no Correction was made by the respondent, the interview would Continue. If a correction was made, an indication had to be included for supervision purposes. A quota was set according to the local market share Of each brand except in cases where more answers were de sired for a given one. Brand ‘names were randomly written on each questionnaire form before giving them to the interviewers, but before asking for a given brand, the inter- viewer would make sure that the respondent knew about 125 its existence, so that a perception rating could be obtained. Phrases regarding each attribute as described in Table 5-2 were used in talking to the respondent in order to obtain his "agreement" response. In many cases complete interviews were rejectedlif repetitive ratings were found along the scales for a given respondent or when an interviewer's work showed equal responses for a certain phrase. In a particular city, interviewer biases along a given attribute would have been responsible for an incor- rect decision regarding a product's attribute. A fourth set of questions related to classification data from respondents. Several types of classificatory and identification data were obtained regarding observed socio-economic class, age or age group, sex, marital Status, schooling, occupation, name and address. A fifth group of inquiries produced advertising data. In order to have additional, but not conclusive, material for advertising purposes, questions were asked regarding in which media the respondent would like to see advertisements for the product and where he had actually Seen or heard them. An additional question was asked as t0 the place where or occasion when this particular pro- dL’lct would better fit into his way of life. This would Serve .as a guide to advertising media and themes. 126 Additional advertising research can be conducted as these described measures would not provide definitive answers for advertising strategy. Einal Questionnaire Form The final questionnaire form, which for obvious zreaasons is not included, was printed on a single standard. sheet (11" by 8.5") of paper of a different color for each city. As all the interviews were to be personally conducted, no more space was necessary, nor advisable. Previous experience had led to the conclusion that when a potential respondent is asked for 5 or 10 minutes of his time, and he sees many sheets of paper, he might be reluc- tant to believe such a statement and refuses to answer. As a matter of fact, the actual interviewing time is from 15 to 20 minutes. The final form also included coding guides to expedite final key work in the office. Sampling Procedures and Field Work @pling Method Obtaining a probability sample in developing cOuntries is not an easy task. Much of the secondary infor- mation available in more developed countries is not obtainable nor purchaseable in those areas. No census tracts exist, and block or street telephone directories are not published. 127 However, if one is to attempt a market segmen- -tation research for decision-making purposes, one must make asure that the selected sample is reliable and that generali- zzations made from it are unbiased estimates of the total population. The only available probability sampling technique thus becomes area sampling. All that is needed is a fairly cztal sample was divided by five. In previously conducted Ikesearch in the various cities it was found that five was 'tlle average number of interviews obtained from every block, 128 whenever systematic sampling, with an interval of five is thed for the second stage, which was going to be the case in our survey . In order to select households within a given block, ‘tzhe interviewer was directed to go to the northweSt corner ¢:>f the first selected block on his program. He was to IIImflmu as one the first house, and walking counterclock- anise, he was to make the first call on the fifth house. 3E:f successful, he was to continue the same procedure, c:<3nducting an interview at every fifth house until he Ireeached the end of the block. If the last house was aisssigned a number lower than five, he was to start the sseacond block with that number until reaching five again. In 1:11is manner each house in every block had the same proba- lazility of being selected. As no “call-backs" were to be attempted, if there V0615 no response in the fifth house, the interviewer was t2c> select the following house, and he would start his new <2<>unt from it. This substitution was to be allowed for <3111y two houses. If in three houses (the systematically Esealected one and two more alternatives), he had not aChieved an interview, he was to begin counting again. If Earl apartment building was included within the sampled hfllock, the interviewer was to start counting and inter- 'Vtiewing systematically on the first floor, then the second, 311d so forth. 129 This sampling methodology assures an unbiased sample. As the number of interviews per block is a function of the block's size, those with more houses, basically in the lower economic strata, would receive more interviews. If the same number of cases were taken from every block, on the density of its respective block, and those inhabited the probability of a house being sampled would depend l E by higher socio-economic families usually would contain 1 a lower number of houses per block. Another method was attempted for the second t' stage selection. The sampled blocks were censused, and from that list an unrestricted sample was chosen. This method, although theoretically eliminating the systematic sampling bias (each combination of units does not have a probability of being chosen) increases the total cost and time of the survey. Furthermore, if houses, and in our case, product perceptions, are randomly distributed through— out the population and not according to house position within a given city block, we might assume that the syste- matic sampling bias is nil. Sample Size The minimum sample size to be obtained from every city was 300 interviews. This number could guarantee a maximum error of 5.7 percent with a probability of 95 per- cent, assuming, of course, a binomial distribution and unrestricted sampling. However, and in order to reduce 130 management's groundless uncertainty due to small samples, in most of the cities, but especially in the larger ones, the sample size was increased up to 1,000 interviews. The maximum error therefore was reduced to 3 percent, also within the 95 percent probability estimate. As clusters were to be formed from the sample, and since 30 was the minimum accepted for each one, in the cities where the sample size was 300, a market segment had to be worth ten percent of the total sample in order to be included as such. In those cases, clusters of 18 or more people were chosen, but care has to be exercised when referring to them. It is our firm belief that if management is to take action from a market segmentation research then the analyst should be very careful when selecting his sample. A probabilistic method must be used in order to be able to generalize findings about the total universe or population from which the sample was drawn. Although this particular paragraph seems obvious, in most of the published literature in this field, most of the data seem to have been obtained from nonprobabilistic or convenience samples. Integrating a National Sample It is always convenient to integrate total results from partial surveys into a national portrait of the market for a given product, but this task should be attempted 131 only when data from the entire nation is available. In our case, as only the 30 main urban concentrations were included in the sample, it would be fallacious to call the integra- tion of the results a national sample. It is possible that product perception in smaller cities and rural areas might differ from that in the largest ones. Therefore, national ! sample should be understood as the area covered by this study. A total national picture is desirable so that any deviations can be observed among the different regions of the country, but with a different weight given each rated attribute. For example, it is possible that in smaller cities availability may not play an important role, but relative price might. It is also possible that in northern cities a given attribute, such as healthfulness, might play an important role, whereas in costal cities thickness might be more important. For analytical purposes, cities were grouped into nine different geographical regions within the country, and in such way they were further analyzed. The selection of these regions was in accordance with population and consumption parameters discussed herein, and a reference to them will be in accordance to nominal scales. Two methods were used for the integration of the national sample. The first weighted the averages found 132 in every region by its relative ccnsumption of the pro- duct class. Sales were not used because each firm holds a different market share in each region, and reference must be made to industry, not company, if market analysis is to be properly conducted. The second method selected a systematic sample from the computer tape where all the questionnaires had been stored. As the sample for each city and region was not a function of industry sales, the same weighting pro- cedure was utilized. By using the second method it was possible to obtain better results, and even comparable ones, as several samples can be obtained similarly and comparisons can be made. The burden of outside computa- tions is reduced since the computer produces all the results. It is also possible to integrate other types of clusters. For example, the young market can be segmented, or the upper class, or the heavy user cluster, thus allowing further analysis, which can be beneficial for advertising purposes. A given segment can be further targeted through the use of specific media catering to those demographic clusters, which in this way are further segmented by consumer perception, both concerning product and existing brands. 133 Field Work Procedures It is needless to elaborate on the idea that every step in the research procedure is important, but from the standpoint of the analyst, he must be provided with "good" consumer data if useful information is to result. Data gathering thus becomes a crucial point for the researcher, as upon it depends the "goodness" of the entire P project. It is only in this particular stage that he must i work through other people, and he must have a foolproof a method for assuring collection of unbiased data. Most of i the currently available marketing research textbooks pro— vide good discussions of this particular point, and in this section we will only mention some of the highlights of, or differences in, procedures which were included in this study. First, all of the interviews were conducted by part- time and locally hired college students. They were recruited from the fields of business, economics, psychology and sociology wherever available. In the smaller cities, where there are no universities, grammar school teachers, students from normal schools (teachers' colleges), or preparatory school students were hired. Second, a different set of research supervisors was sent to every location. This was done to distribute more widely their personal methods of training and supervising field workers. 134 Third, forty percent of total interviews were sniperviSed in every city by visiting the respondents again and asking him a diversified set of questions from the form Exreviously filled in. When, in the supervisor's judgement, aJI interview was biased, either because of differences in 'ajiswers or because of incorrect house counting, the form 1 vnas rejected. If an interviewer had several rejections, rue was automatically dismissed and all of this work removed from further analysis. Fourth, training of interviewers included sample ;‘ fixeld tests, taped biased questionnaires for their analysis, auui retraining sessions. In most cases several recruited people were not hired, and firing ranged from 5 to 10 percent. Finally, as questionnaires were cross tabulated by interviewer, the entire work of some of them was rejected when systematic biases were found or when his results accorded too closely with his previously measured attitudes or preferences regarding brands. (Each interviewer, before being trained, was interviewed by the supervisor using the same questionnaire form. This was used as the basis for further rejection of the forms which he brought :from.the market if much similarity to his own opinions was encountered . ) 135 It is our firm belief, and it has been our research experience, that attempted savings in this stage of the process will only increase the probability of defective data. If costs are to be reduced, steps should be taken in other stages of the methodology. If good data is to come from the market, interviewers must be the best available, L fully supervised, and, most of all, paid accordingly. We ‘“ have witnessed that in many cases some research agencies I have attempted to increase their margins by reducing field work expenses, either through the hiring of lower paid 1}: interviewers or by low rates of supervisors' callbacks. Defective field work cannot be saved with good questionnaires, sound sampling methods, or sophisticated analysis. A researcher always must remember one of the most important principles of electronic data processing, which undoubtfully can be applied to market research: "Garbage in, garbage out." CHAPTER VI MARKET SEGMENTATION BY CONSUMER PERCEPTION: DATA ANALYSIS Having described in the previous chapter the case (nontent and the strategy of inquiry, it is the objective (If this chapter to conduct a stepwise analysis of the djgfferent techniques which were utilized for constructing true geometrical space in which attributes, brands, and consumer wants were located. Several techniques were used in conjunction for the achievement of this purpose. Given the complex structure of most segmentation analyses, it would seem quite likely that 'multi- method' designs will become quite commonplace as familiarity with modern analytical techniques increases. Figure 6-1, a replica of Figure 4-1, shows a summary of the "multimethod" strategy followed in this study in order to develop a market segmentation by consumer percep- tion research. It should be pointed out that a general methodology LHP'to now has not been developed nor universally accepted lFrank,et al, op. cit., p. 169. 136 137 Kutwsu .OV VAUMK-I. V07. krrtuauua at (new snubs Mu.» we BK bun AT 1"“! TL ”Faun. Nan-v.95 1” la‘flJBUTBs OF $09.28“? BRANDS t ktVOt L OKD‘DQS M 0 map“. stcmumm'r Au sun‘s .w‘_‘. .—a—...a— “Dusclzm‘mm wemuws 101mm Pam To loam: £10501- Ets av W44. AMP 8m 00 loan MD ”7106 Ct «we: 404ml: F0 1 (“Janet ”can. Po w!" ' 62:55 75m»... I! ”an; m 400 - ‘ 72-37sz F" 056/1700 Sat 9 303106.: m lunar/on Figure 6-l.--Methodology for Data Analysis on Market Segmentation by Consumer Perception. 138 by marketing scholars or practitioners. The approach herein described, although taking most of its bases from similar ones described in current marketing literature, is an attempt to adapt some of those methods to a case study in a developing country; many changes had to be imple- mented and assumptions to be made in order to fully use these newer quantitative techniques in a large scale. The following are two examples of those changes. ‘1 A First, questioning techniques had to be simplified, both 112"“. “‘u "u ‘4‘; V. yap-o I in limiting the time involved in the interviewing process and in making questions comprehensible to all respondents. Second, computer programs had to be adapted to the equip- ment available in the country. Adaptations also had to be made to achieve better man-machine communication in relation to the level of analysis required and to enable several assumptions about the parametric structure of the input obtained in the market.2 The methodology described here should not be con- sidered "the" one to be used; rather, it is "a" system which can fully use consumer perception data for the structuring of meaningful market segments for decision-making purposes. It has not been our intention to develop original methodology, and what is used here already has been 2"Metric methods provide good approximations to nonmetric techniques when the nonmetric function can be clearly approximated by a linear one." Ibid., footnote 8, p. 168. 139 3 Our combined in a similar fashion by other researchers. aim is to fully describe the methodology by presenting a sequential analysis of every step. The present survey, in contrast to those available in the current literature, was conducted in a developing country and in several cities using the same questionnaire. It thus validates the method- ology, not only by its application to a developing economy, but also by contrasting the results obtained by its appli- cation on a multicity research study of the same product. The possibilities of its application, as well as the con- sistency of its results among different individual surveys thus is demonstrated. The entire research project included 14,309 per- sonal interviews conducted with individual consumers in 30 of the largest urban concentrations in Mexico. Work began in May, 1971, with the exploratory survey described in chapter 5, and this field work phase was concluded in February, 1973. In addition to this sample, information was obtained from 960 retailers in five of the areas by using the same questionnaire, that is, assuming they were ultimate consumers, in order to test differential product and brand perceptions. Finally, in June, 1973, as one of 3For instance: Richard M. Johnson, "Market Segmen— tation: A Strategic Management Tool," Journal of Marketing Research, VIII (February, 1971) 13—18, and Lester A. Neidell, "THE—USE of Nonmetric Multidimentional Scaling in Marketing Analysis," Journal of Marketing, XXXIII (October, 1969) 37-43. 140 the brand's marketing mix had been changed, an additional survey was conducted in one of the cities. In order to contrast changes in perceived behavior, both toward the product and the modified brand, 658 consumers were interviewed. For the purpose of simplifying description of the multistage methodology, in this chapter the results obtained from one of the surveyed cities will be fully analyzed. In the next chapter, in the structuring of the national sample, results from this city will be contrasted to those in other areas. There were several reasons for selecting City X as the analytical example. First, it contains the largest single sample, 999 interviews. Second, it was here that the additional, the ex post facto consumer survey was taken. Third, it was one of the selected areas for measuring retailers' perception. Finally, as it was the city where the national survey started after the exploratory research was concluded, several alternative techniques were used to validate its results. These will be described in chapter 8. City X was surveyed originally in January, 1972, and the 'exppost facto test was made in June, 1973. Analysis of Variance Nine specific brands (A, B,......I) were selected to be studied in City X along the ten attributes previously described. Table 6-1 presents their sample size and rela— tive market preference. CF‘J" 141 TABLE 6-1.--Brand Preference Share and Sample Size in City X. Brand Name Sample Size Market % Preference A 225 6.5 B 345 7.2 I C 347 10.9 3* D 126 1.8 E 584 39.2 . F 94 2.9 L G 355 18.4 H 213 6.5 I 198 4.1 Totals 2,487 97.5 The reason the sample size does not total the 2,997 pos- sible ratings has three possible explanations. (1) Some respondents were undecided about three or more of the ten attributes of a given brand, thus invalidating the entire response for that brand. (2) Brands from J to Z also received consumer ratings. (3) Some consumers rated less than three brands per questionnaire. Analysis of variance was used to test the brands' means for every attribute. The null hypothesis was that all brands had the same value or mean on each of the 142 measured characteristics. An F-value greater than two would mean that the difference among the means of any two given brands was statistically significant. An n by n matrix was formed for every attribute showing the F—ratios in the interbrand comparison as well as each brand's mean. Table 6-2 shows the F-ratios for the attribute "higher-priced" in City X. Figure 6-2 represents the means of the nine brands included in the analysis according to the same attribute, using data from the last column of Table 6-2. In order to test the validity of this measure, Table 6-3 contains the relative prices of the different brands in City X. By contrasting the value of the means (Figure 6-2 and Table 6-2) with the relative retail prices, it can be observed that this kind of measurement is a reasonably good portrait of the market, at least regarding this particular attribute, which could be contrasted with real market figures. Figure 6-3 contains the means for all the brands along each of the ten attributes being measured in this survey. ‘F.. m>.m o . , H ma.m m.mm o m po.m H.mm ~.o o o mm.~ m.mm o.o ~.o o m mo.m m.Hm «.0 H.o m.o o m m mm.m 0.3 «.2 mi: 0.... 9: o . a HH.~ H.~mm n.ma m.mm m.Hm m.HHH m.oa o o am.m m.om H.H v.0 H.o m.o m.m H.Hn o m nm.a m.m~m a.mm m.moa ~.ms v.HmH m.mm o.m e.aa o a cam: H m w m m a o m a ammum =.pmoflumlnm£mflm= musnflupud may mo mammamcfl cosmeum> I|.mlm mamme 144 OOH I m cm I H OHH I Q OHHIOOH I m oma I U OOH I 0 OOH I m 00H I m omH I < .mmoflum pcmum Hampmm m>HHMHmmII.mlw mqmmam b mummunmumfi I m rumcmuum m mmwcoaflz I m HHmEm .HOUO m mmmcum®3m I H "mmbsnfluuum "mumnz ooooo.H mummo.OI vmmmm.o wmomo.o Hammo.0I mvomo.o mmmmo.o nmmoa.o m mnmmo.ol ooooo.H mevmo.0I mommo.o vmmma.o mmmmo.o mmbao.ol mNOHH.OI h vmmmm.o mevmo.ol ooooo.H whmho.o thmm.ol mmmho.o mnwma.o vmmma.o m vmomo.o mommo.o_ mnmho.o ooooo.H namoo.ol Hmmmm.o hmmoa.o mmmmo.o m Harmo.oI vamH.o awhmm.0I hamoo.oI ooooo.H mmHNo.oI mmnoo.on mmmmm.o| w mvomo.o mmmmo.o mmmno.o Hmmwm.o mmamo.0I ooooo.a mmmma.o NmHmH.o m mmmmo.o mmnao.OI mhvma.o nmmoa.o mmhoo.oI mmmma.o ooooo.a mmmma.o m hmmoa.o mmoaa.ol vmmma.o Nmmmo.o mmmmm.0I mmHmH.o ommma.o ooooo.H a mmusn w h m m w m m a Ifluuufl .x mnflo cfi mwcmum Umumm mo mucmfloflwmmou cowumamquUII.mIo mamae 153 The third output of a factor analytical procedure is that related to eigenvalues and cumulative percentage of eigenvalues. An eigenvalue is the sum of the squared factor loadings, which have been rotated through a principal-component method, that is: E1 =_ z EL11)2 + (L12)2 +....+ (L18)2] (6-1) These two vectors are included in Table 6-6. TABLE 6-6.—-Eigenva1ues frtmI Factor Matrix (City X). Level Eigenvalues Cumulative Percentage of Eigenvalues 1 1.79849 0.22481 2 1.30107 0.38745 3 1.00732 0.51336 4 0.92256 0.62868 5 0.85643 0.73573 6 0.75628 0.83027 7 0.69381 0.91699 8 0.66404 1.00000 For operational purposes, several comments are important regarding eigenvalues. First, their sum is equal to the number of variables included, eight in this 154 particular case. Second, the contribution of each one to total variation is continuously decreasing: the first one accounts for 22.48 percent of the total, the second one for 14.26 percent, and the last one for only 8.3 percent of the total. Third, if parsimony is desired, let us say, to reduce to seven factors, a number between 0.69381 and 0.66404 should be set as a limit for this purpose. In the same way, if five is the total number of factors that are wanted for final analysis, a number between 0.85643 and 0.75628 should be set as a limit in the input of this program, thus retaining 83 percent of the original information. If no information is to be lost, then the analyst would have to use as many factors as original attributes that came into the program, in this case eight. A limit point lower than 0.66404, the eigenvalue of the last factor in this analysis would have to be used. For practical purposes, however, a value of zero will assure that no parsimony will be produced. A factor matrix was produced (see Table 6-7) where every Li' is a factor loading, such that in Equation 6-1 3 the eigenvalue 1 is: (0.63047)2 + (0.44641)2 +. . . .+ (0.43611)2 155 mmmmo.ou smema.ou oommm.on Haemm.on momeH.o mmemw.on mmmoa.ou Hamme.o asexuammm monHH.oI memma.ou emamo.on emoem.o mmmao.o mmmme.on mmeme.o mommm.on uo>mHm meamm.o oammm.o mmemv.o mommH.o mmmma.o smamm.ou monmm.ou ommmm.o numcmnum menma.o mmmme.ou mmfimm.o mmHmH.ou momme.o mamma.o mqnmm.o mmmnm.o uowo memom.o mmmma.o mmmao.ou mmmom.ou mmmmm.ou ommma.on oomam.o mommv.ou mnflmcmo meemm.on mmmam.o oomem.ou mommo.o enema.o mmnmm.o ovqmm.o ommmv.o mummuumuma momem.oI mnmmo.on mavmm.o «HomH.oI mmevn.0I Nmono.0I mmmmm.o vaww.o mmmcpaflz vomov.o mnwma.ou mooae.on Hawam.o ommmm.ou mmnem.o nvmmo.ou nvomm.o mmmcpmmzm m w m m e m m H mmusn Houomm H0pomm Houomm Houomm Houomm Hogomm Houomm uouomm IHHuum Hmaflmfiuo .x muflu How ApcmcomEoo Hmmflocflumv xfluumz HoaommII.hIm mqmde 156 Additionally a factor becomes Fl = (L11) Vli + (L12) V21 +’°'°'+(L18)V8i (6-2) where L1i are the factor loadings for factor 1, and Vji are the original attributes j as assigned by the consumer to brands i. For example, Fl = (0.63047) Vli + (0.44641) V2i +. . . . . .+ (0.43611) v8i That is, factor 1 for each brand will be the product of multiplying the factor 1 load vector by the original attri- bute scaled by the consumer for brand i on attributes 1, 2, . . .., 8. If consumer 432 had rated brand.£>as 5, 1,...,3, those ratings would become the Vji in Equation 6-2. In addition to Table 6-7, an Eigenvector matrix was produced such that if the square root of an eigenvalue is multiplied by a column in this matrix, a column in the fac- tor matrix is produced. (Lij) = (452) wij (6-3) Inhere Wij is a number of row 1, column j in the Eigenvector matrix (0.63047) = (J1.79849) (0.47012) Table 6-711)=(Table 6-611) (Table 6-811) 157 TABLE 6-8.--Selected Eigenvectors in City X. Vector 1 Vector 2 Vector 3 Vector 4 0.47012 -o.05477 0.24642 -o.26264 0.33287 0.24892 -0.07036 -0.77486 0.34263 0.49481 0.22640 0.19546 -0.28324 0.49751 0.16535 0.44565 As it can be observed, and as was explained in chapter 4, although the matrix of factor loadings (Table 6-7) pro- duces orthogonality for the new factors out of the original values, interpretation is very difficult. Factor one, for example, has high loadings in all the attributes, but eSpecially in l, 4, and 6, but factor 8 also has high loadings for the original values 1 and 4. In order to be able to name the new factors, a Varimax routine was per— formed, which produces the last significant output from this program. Results from this rotating program are presented in Table 6-9. The advantage of this technique is that it produces high loadings (near one) for a few attributes (only one if no parsimony is obtained) and very low loadings for the rest of the original attributes, and without any loss of information. It thus permits better interpretation of the factors integrated by this method. 158 00000.0 00000.0 00000.0 0 0000060 000000 00000.0 00000.0 00000.0 0 0>000000 00000.0 00000.0 00000.0 0 03600 00 I 00000.0 00000.0 00000.0 0 00000.0 00000.0 00000.0 0 00000.0 00000.0 00000.0 0 00000.0 00000.0 00000.0 0 0000000 000000 00000.0 00000.0 00000.0 0 m>0uwmom mBOSm uH + wOCmVHOMMHQ Hannah HMGHOHHO OHQMHHM> 0000000202200 20 00000 + + + + I I + + \wflflmcmfl mpmmuhmpmd nuumocmhum 000.00.00.70 mmmGUHHE Hflmflpdmwm HOUO mmmcfimmzym mEmz 302 00000.0I 00000.0 00000.0 00000.0I 00000.0I 00000.0I 00000.0 00000.0 000000000 00000.0 00000.0 00000.0I 00000.0 00000.0 00000.0 00000.0 00000.0I 0o>000 00000.0I 00000.0 00000.0 00000.0I 00000.0I 00000.0I 00000.0 00000.0 00000000 00000.0 00000.0 00000.0 00000.0 00000.0I 00000.0I 00000.0 00000.0 0000 00000.0 00000.0I 00000.0- 00000.0 00000.0I 00000.0 00000.0 00000.0I 0000000 00000.0I 00000.0 00000.0 00000.0I 00000.0I 00000.0I 00000.0 00000.0 0000000000 00000.0 00000.0 00000.0 00000.0I 00000.0I 00000.0I 00000.0 00000.0 00000002 00000.0- 00000.0 00000.0 00000.0I 00000.0I 00000.0I 00000.0 00000.0 000000030 0 0 0 0 0 0 0 0 0000 .HOuhumm .HOfiUmm Houomm Houumm Hopvmm .HOHKVMM .HOubmrm HOHUmm lflhflfld Hmswmwho .x >000 000 muouomm Axmeflum>v pmpmuomll.mlw mqmae 159 Table 6-9 should be interpreted in the following manner. Factor 1 becomes "sweetness," as the loading for this attribute is 0.9795, and loadings are very low for the rest. Factor 2 becomes "odor," as its loading is 0.9877, and in this manner the entire table is inter- preted in the last row by looking at the highest factor loading at the intersection with a given variable. In this example, as no parsimony was desired, there are as many factors as original attributes. In the next chapter, where we will present parsimony for this same City X up toiseven factors and up to five, we will find that a given factor acquired a double or triple meaning as it achieved high loadings for two or three original attri- butes and very low ones for the rest. The results of this rotated matrix thus became factor loadings to be used for the next computer program. By the use of these tech- niques we have transformed the consumers' original current brand perceptions into sets of uncorrelated (orthogonally constructed) factors without losing information. We also have retained a name for each one of them, which will be an important feature in interpreting results from the following procedures. Multiple Discriminant Analysis As was described in chapter 4, multiple discrimi- nant analysis is the appropriate technique for finding 160 several items: the weight of each attribute in brand perception; the construction of a geometrical space through the use of those correlations; brand positioning within the map; and, finally, a basis for locating consumer "ideal". positions within the perceptual map. The first three. objectives are achieved through the direct results of multiple discriminant analysis, whereas the locating of ideal points will be considered later through the employ- ment of another set of computer programs, which make use of the space generated in this analysis. The reader should be reminded that whereas in the factor analytical routines all the original attributes for all brands are taken together in order to transform them from an attribute space to a factor space (where every factor is a weighted combination of original ratings), in multiple discriminant analysis, through taking this factor space, brands are segregated by their differences, and the weight that each factor has in this discrimination is found. A new set of axes (discriminant weights) results as a linear Combination of factors. This can be used in order to simulate desired changes in brand perception and, alternatively, will be used for locating consumer ideal brand ratings in this geometrical space. An explanation will be offered later in the cluster analysis section. Input data for this program are the original cur- rent brand ratings on each attribute and factor loadings 161 and their mathematical sign, which will be used to trans- form the original ratings of every respondent for each brand into factors. . The procedure and output of the multiple discrimi- nant analysis is described below. First, the percentage of variance, which was extracted by each of the roots, describes how much explana- tion is found by using one axis (53.8 percent), and up to eight axes for a full multidimensional expression. Table 6-10 presents these numbers along with their relative chi-squares, degrees of freedom, and probability limits. TABLE 6—10.--Percentage of Variation Extracted by Bach Root in City X through Multiple Discriminant Analysis. Root. % Variance Chi-square D.F. P 1 53.78 271.230 15 0.000 2 33.06 170.244 13 0.000 3 6.93 36.671 11 0.001 4 ‘ 3.34 17.732 9 0.039 5 2.06 10.968 7 0.141 6 0.57 3.062 5 0.693 7 0.25 1.318 3 0.729 8 0.01 0.039 1 0.838 D.F. = 64, and 14259 F-Ratio = 8,112 P 0.000 162 In order to provide simpler explanations for management purposes, a two-axis space was chosen; it explained 86.8 percent of the total variation. A second output of the multiple discriminant analysis is the table of factor correlations. These pro- vide the basis for labeling the axes of the two dimensional space and for indicating the direction of each one of the factors (perceived attributes which were transformed into factors in the previous program). Table 6-11 presents the factor correlations for the first three axes of the dimensional space. TABLE 6-ll.--Factor Correlations for the Three Main Axes of the Perceptual Space, City X. Name Of the Dimension I Dimension II Dimension III Factor Sweetness 0.2820 -0.5783 0.3551 Odor 0.0426 -0.2642 -0.0030 Healthful (*) —0.0592 0.4856 —0.7467 Mildness (*) 0.3790 0.8129 0.1265 Flavor -0.1897 0.4610 0.0834 Strength 0.6472 -0.4681 0.1644 Aftertaste 0.1223 ~0.339l 0.1207 Density ’ -0.6704 0.5750 0.1145 * Shows negative factor loading. 163 These factor correlations were plotted on the per- ceptual map in the following way. First, the mathematical signs of those with a positive factor loading sign were changed so that all factors fall along the same agreement scale. Remember that a one means a "SI-SI" answer, that is "in complete agreement," whereas a five would mean "in complete disagreement." Thus, a lower mean would repre- sent more agreement, and a larger one, more disagreement, with the phrase containing a given attribute. Second, each factor was plotted in the geometrical space formed by dimensions I and II by assigning those numbers from Table 6-11 in Figure 6-4. Factor 1, sweetness, was plotted as -0.28 on the vertical scale and as 0.57 on the horizon- tal scale. The last factor, density, would be 0.67 on axis I and -O.58 on axis II. The next step in the analysis of the multiple discriminant output was that of positioning each rated brand in the same geometrical space. Table 6-12 presents brands' locations along the main three dimensions. -0;58 (II) I I \ IDe~sity , I l l 0.671 l (I) I L I I Flavor ~ I II (-h I 164 (+) MiloI-ss II (+) r —-> Heq thful ‘ ‘Odor \ | _-0.28 I) we “4 Sweetness + 58 (II) ( ‘fted aste (*) Strength Figure 6—4.-—Factor Correlations in City X. 165 TABLE 6-12.--Brand Location in Perceptual Map for City X. Brand 1 Dimegsions 3 A —1.687 -1.722 3.865 B -0.795 -1.758 3.338 C -l.650 -2.805 3.365 I D -0.701 -2.559 3.788 E -0.428 -2.557 3.541 F -0.756 -1.802 3.365 . G -0.772 —l.771 3.450 P H -0.815 -2.007 3.455 I -0.701 -2.453 3.230 Before plotting the brands in Figure 6-5, which already contains the factor correlations included in Figure 6-4, note that the range of values on dimension I is 1.25, the difference between brands A and E, whereas the ranges on dimensions II and III are 1.08 and 0.56, respectively. These facts confirm the idea that dimension I has more discriminating power among the rated brands than dimension II, and dimension II has more than dimension III. In order to plot the brands in the perceptual space, the center had to be found, which is the weighted average of brand means along each axis. For City X, 166 F. I F. G .L' 83 -0.89 ‘_ II 7‘§===ES;R;:~—‘§>Healthful Aftertaste Sweetness -1.49 th ‘ I CC -2.81 -2.21 . A —1.61 Figure 6-5.--Brand Position and Factor Correlations in City X. 167 centroids were on dimension I, -0.89, and on dimension II, -2.209, which became the center of the geometrical map, that is, the point from which all the factor correlations had to start. Next, the axes had to be numbered so that the numbers in Table 6—12 could be plotted in Figure 6-5. In this way, brand A was located at -l.69 on axis I, and at -l.72 on axis II, lower than the centroid, -0.89 and to the right of the same point, 42.21. In contrast, the same points for brand E were -0.43 and -2.56, which would place E in the upper left-hand section of the map. The reader must remember that the positive direction of axis I is toward the upper part of the map, and that of axis II is toward the right part. Brand location and attribute weight in City X now could be interpreted. The interpretation should be similar to that achieved when the analysis of variance was per- formed. Bear in mind, however, that in this case a multi- dimensional (multicolinearity) meaning has been achieved rather than the unidimensional analysis used in the former case. In Figure 6-5, the direction of the arrows, factor correlations, indicates agreement on that particular attri— bute, and brands are read perpendicularly to each line. In this way it can be seen that brand A is perceived as the strongest of the brands, whereas brand E is the weakest. 168 The reader can contrast this finding to that for analysis of variance output presented in Figure 6-3. There seems to be complete accord between the two technique.I¢hen a completely uncorrelated factor, healthfulness, is com- pared, brand C is perceived to have the least amount of this quality and brands G, F, and B have the most. These results also accord with those of the analysis of variance technique. The discriminant weight of each factor is described by its F-ratio, shown in Table 6-13. As expected, factors such as mildness, density, and strength accounted for most of the difference, whereas odor and aftertaste had a low discriminating power. All of these results also accord with those found in the analysis of variance. The per- centage of discrimination of each factor can be found. _ F—Ratio Factor j ‘ ZJF-Ratio Factor j % D.WoFi 10.8613 93.6113 % D.W. of sweetness = 11.6% (6-4) 169 TABLE 6-13.-—F-Ratios and Percentage of Discriminating Power of Each Factor in City X. Factor F-Ratio Disc. Power (%) Sweetness 10.8613 11.6 Odor 2.4237 2.6 Healthful 7.8083 8.3 Mildness 19.6108 20.9 Flavor 6.6921 7.1 Strength 19.3587 20.7 Aftertaste 3.9130 4.2 Density 22.9434 24.6 All of the ratios described in Table 6-13 are significant at the 0.01 level, except number two, odor, which has a level of 0.0132. 1 Another output of multiple discriminant analysis is useful for simulating changes in brand perception (see' chapter 7) and for positioning consumer ideal points. The program produces the discriminant weights of each fac- tor along each of the selected axes of the geometrical space. As a two dimensional space was selected for our purpose, Table 6-14 contains such values for two axes, and Figure 6-6 presents their plotting in the geometrical space. Dens y II Flavor St Figure 6-6.--Discriminant 170 I Mi aness £111 ,_ Odor ' I Sweetness -ftertaste ;ng:h Weights in City X. 171 TABLE 6—14.--Factor Discriminant Weights in City X. Factor I Dimensions II Sweetness 0.0499 -0.0163 Odor -0.0075 0.0402 Healthfu1* 0.2285 0.2304 Mildness* 0.6436 0.8380 Flavor 0.0462 0.2803 Strength 0.5655 0.0720 Aftertaste 0.2018 -0.0076 Density -0.4103 0.3987 * Factor has a negative loading. The points in Figure 6-6 were plotted in the same manner as those in Figure 6-4, factor correlations, and they are read similarly. In constrast to factor corre- lations, discriminant weights are used for simulating the expected position of a new brand if it showed selected brand perception ratings, or the location of a current brand if it were to change its perception along one or several of the rated attributes. For simplicity, Figure 6-6 can be compared with Figure 6-5, brand positioning. If the center of Figure 6-6 is located at a point along a given brand in Figure 6-5, the brand's position would move from its pre— vious position in Figure 6-5 to a new one. The degree 172 of movement will equal the length of the arrow in Figure 6-6 if perception can be changed one point along the agree- ment scale of that specific attribute. Table 6-15 is a matrix that shows the average rating of each brand on every factor. If the vector of brand i is multiplied by the matrix of discriminant weights, the two points on the axes of a given brand are obtained. For example, brand A's location on the perceptual map is -1.687 and -l.722. (2.5265)(0.0499) + (2.6609)(-0.0075)+. . .+(2.6145)(-0.4103), the result of which is -l.687, the value of brand A on axis 1. (2.5265)(—0.0163) + (2.6609)(-0.0402)+. . .+(2.6145)(0.3987) equals -l.722, the value of brand A on axis II. Notice that the first value, the brand mean for each factor, comes from Table 6-15, and the second, from Table 6—14, discri— minant weights. If one were to assume that consumers changed their perception about the density of brand A, and, instead of being 2.6145, the highest disagreement for all the brands, it could be changed to 1.6145, the position of brand A would move 0.41 points to the left, as it has a negative sign, and 0.40 points upward. These figures can be observed in Figure 6—6 or computed from Tables 6-14 and 6-15. 173 NNO>.H 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000000 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000000000 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 . 0000.0 0000.0 00000000 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 00>000 0000.0I 0000.0: 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 00000002 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 0000.0I 000000000 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 0000.0 000000030 0 m w 0 m o o m 0 .flmmmw .x 0000 00 000000 0000 00 000000 00000>< 00000II.00I0 000 49. 174 Multiple discriminant analysis should be con- sidered as the key of the entire process. If no parsimony had been achieved through factor analysis, and if each factor had obtained a very high loading on each attribute, it might have been possible to construct a very similar perceptual map. However, factor analysis has produced orthogonal, uncorrelated factors from the original attri- butes, and in certain cases, like the ones found when testing the techniques, it could eliminate attributes which are very closely correlated. In addition, when many attributes are being scaled by the consumer, parsimony must be achieved if computer time is to be saved. In other cases computer capacity makes it impossible to use in a multiple discriminant analysis as many attributes as consumers were questioned about. Cluster Analysis Once a perceptual map has been constructed on the basis of consumer rating of current brands, the next step is that of locating on such a map the perception that consumers have about the "ideal" brand. This would deter- mine how well current brands are satisfying the market, whether there is a need for a change in a given brand's perceived structure, or whether there is a need for a new brand in order to better cater to the market. 175 However, the mere location of consumer ideal points is not the end of the problem. Market segments can be con- structed by gathering into homogeneous groups the points previously located. Such segmentation would permit market analysis since decisions can be made on the basis of cluster size. The first step in cluster analysis is the location of consumer ideal points. This is achieved through trans- forming consumer rated values for ideal brand attributes into factors. They then are multiplied by the discriminant weights, and a location for each consumer is provided on the perceptual map. An example appears in Table 6-16 for the first ten respondents. All of these points eventually could be plotted on the perceptual map so that the analyst could cluster them together through visual inspection. This task was performed in the first surveys in order to contrast these groups of points against the results of the next program, whose aim was to cluster these points according to connective distance. A further analysis of such a list as shown in Table 6-16 might be used to assess systematic interviewing bias. The data went into the computer in the order in which interviews were conducted by a given person. If two or more of these numbers were very similar, the entire data produced by that interviewer would have to be double checked. 176 TABLE 6-16.--Selected Consumer Ideal Points in City X.-- Respondent Axis I Axis II 1 1.062 -1.975 2 -2.233 -0.216 3 -1.962 —3.318 4 -3.159 -4.021 5 -l.354 -2.102 6 -l.464 -4.039 7 -2.563 -3.574 8 -0.773 -4.656 9 -l.343 -0.295 10 0.288 -2.423 In the last computer program, "grouping," an Euclidian distance measure was used in order to inter- connect these points on the perceptual map. l d (x, Y) =\/z(xi - Y-)2. (6-5) X and Y are consumers, and i indicates either axis I A or axis II. The computer performs the clustering procedure as follows.. First, it calculates the distance between sub- ject one and subject two; if it is less than a certain tolerance, both subjects are clustered together. If 177 greater than that tolerance, the computer continues sequentially to the next subject until it reaches the end of the file. This process is repeated until every sub- ject has been assigned to a given cluster. Second, when all the clusters have been formed, the computer calcu- lates the mean on each one of the axes for each cluster. Table 6-17 shows the most important market segments formed by ideal brand perceptions in City X. It also contains their relative size as well as their TABLE 6-17.--Market Segments in City X. Segment Size Axis I Axis II I 251 0.779 -2.l80 II 40 ' -2.103 -0.449 III 152 —l.816 , 7 -3.556 IV 44 -3.101 -4.352 V 143 -O.835 -l.907 VI 172 -0.729 -4.895 VII 73 -O.303 -0.660 VIII 50 0.318 -3.158 Centroids 40.662 -2.948 location on both axes. These numbers have been plotted in Figure 6-7 for visual presentation. In order to achieve 178 VIII II 17% U mo Figure 6-7.--Market Segments in City X. II no 179 better understanding, however, market segments, brand positions, and factor discriminant weights have been com- bined in Figure 6-8. An analysis of this latter Figure shows that allI the current brands are located within a limited portion of the total market. All of the brands seem to be far from the ideal brand ratings. A given brand can be shifted in its position, both in direction and relative size, toward ‘a relevant cluster by moving it parallel to the arrow indicating a given factor. In this way meaningful market segments have been obtained by measuring just what the consumer perceives his wants to be from a given product. A firm producing such a product profitably can use this approach to learn how far it deviates from completely serving specific market clusters. It also has the know- ledge about what to do, in perceptual terms, to alter its brand's perception by consumers. ClasSifying Market Segments The last step in the analysis of market data is that of learning the components of each one of the clusters. TAB-CRU (multivariate cross tabulation), which was des- cribed previously, was used to interpret the results. Every cluster was tabulated against each item of the ques- tionnaire, and in this way the differences and similitudes among clusters could be detected. 180 VIII Q Density II VI Flavor 17% Iv. Figure 6—8.--Market Segments, Brand Locations and Discriminant Weights in Reduced Space. 181 A good example of cluster description is found in cluster III: It is composed of upper class people, with college training. Many subjects are under 35 years of age, and there is a large percentage of single men. The members of the cluster tend to be light consumers.‘ A further description of final results has no relevance for methodology. Having described the techniques used for analyzing the market data in order to construct market segments, the next chapter will attempt to integrate those results into a national sample and to describe comparative intercity findings. CHAPTER VII MARKET SEGMENTATION BY CONSUMER PERCEPTION: NATIONAL SAMPLE, VALIDATION OF RESULTS The objective of this chapter is to present the basis for structuring a national sample. Interregional and intercity comparisons will be made, and our results will be validated through the use of several alternative analytical techniques. Once confident about the informa- tion obtained through this study of market segmentation by consumer perception, it became possible to think about management's use of these results in decision making. Structuring a National Sample (Weighted Averages Technique) As noted earlier, 30 cities were surveyed and 14,309 persons were interviewed using similar question- naires. The same analytical tools were applied in study- ing each particular city, and individual reports were- prepared for each case. However, a national sample had to be developed in order to obtain an overall measure of the market. This sample was very useful in comparing regional results by contrasting their relative values to those of the entire nation. 182 183 The reader should be reminded that a full national sample was not taken; small cities and rural areas were (not surveyed. _However, for our purposes this sample can be considered a good measure of national product and brand consumer perception since urban areas account for 75 percent of total product consumption. Two different procedures were used for structuring this national sample. The first consisted of obtaining a weighted mean for all the relevant measures of the survey; the total product (industry) consumption for each city was the basis for this total. The second procedure was that of obtaining from the total questionnaires, which had been stored on a computer tape, a probabilistic sample for each city; as in the previous case, relative weights were con- sidered. The sample thus obtained was 1,476, the maximum number of cases which could be handled in the computer. The second procedure was very useful as it enabled several samples to be obtained and tested for statistical differences. Surveyed cities were combined into regional groups following both procedures, and nine such groups were formed. In order to obtain similar information from among the different reports, the axis in each city was rotated. In every case, the factor "density" was placed on axis I, at 12 o'clock, and the rest of the factors were placed 184 accordingly. Factor strength, which in most cases was negatively correlated to density, was placed between five and seven o'clock, whereas mildness, which had almost no correlation with either of the former factors, was located between two and four o'clock. Figures 7—1, 7-3, and 7-5 show brand and cluster location for cities Y, W, and Z. Figures 7-2, 7-4, and 7-6 show their respective factor correlations. An analysis of these cities will show how dif- ferently the brands are perceived in each market area. It also will show that the factor correlations tend to remain relatively constant regarding direction, weight, and correlation. This means that product wants are perceived similarly, but brands are not, and a different advertising strategy (themes, budgets, andpossibly media) for each one of the regions is called for. By contrasting each market area with the national picture ( Figures 7-7 and 7-8 show brand and cluster location and factor correlations), a more meaningful com- parison is possible. Specific brand location can be decided upon, and a total marketing strategy can be designed for each brand. (By marketing strategy is meant the relative factor mean that each brand is expected to have in order to arrive at a given position in the geomet- rical space.) Having decided upon these means for each 185 \J J. . IV 18% O 8% A. O .I I 14% C Figure 7-l.--Brand and Cluster Locations in City Y. NOte: J and K are regional brands. 186 IL Density Flavor *‘ II 'ldness r Aftertaste Heakthful W Stre:;th Sweet ess “ Figure 7-2.--Factor Correlations in City Y. 187 Ora OJ @VIII I _ 24% ,.G_ 1v 6 - D00 B Figure 7-3.--Brand and Cluster Locations in City W. 188 De sity Flavor II Odor i‘~tertaste Mildness Healthful Strergth t eetness Figure 7-4.--Factor Correlations in City W. 189 II I 19% IV 20 III 17% VII B .1 A 17% . D II VI \\\‘—’//13' 15% H o ‘. Figure 7-5.——Brand and Cluster Locations in City Z. 190 m. Dens1ty Flavo Mild--ss Odor II \\*\‘ftertaste SWeetness Healthful Strength R) Factor 7-6.--Factor Correlations in City Z. 191 C. B. 2%(:> (2g Figure 7-7.—-Brand and Cluster Locations in National Sample. (Weighted Averages Procedure) 192 I Density Flavor II Odor u'ld ess ’\ Aftertaste , \ Healthful Sweetness Strength Figure 7-8.--Factor Correlations in National Sample. (Weighted Averages Procedure) 193 attribute, in order to develop regional strategies, dif- ferences between actual and expected mean values can be obtained, and plans can be developed to change these means toward the desired values. An advantage of this method is that it is not necessary to change the whole national communication cam- paign. If a change is made regionally, for example, for two alternative advertising campaigns, only those two regions will be affected. The two regions can be used to test the results, and the remaining regions can be P‘- .FinL- used as control groups. An analysis of the national perceptual map reveals the following: Brands are clustered together; an important sector of the market is relatively unsatisfied by current brand offerings, and some changes might seem necessary either in current or in new brands to cater to the market. The reader again should be reminded that the names and mathematical signs of the original variables have been changed; it would be incorrect to assume the results pre- sented here portray the real market. Our emphasis is on the technique used to conduct market segmentation by con- sumer perception and on the potentialities of their usage, not on the specific results obtained in a given marketing situation. However, as these transformations were made systematically, it is quite possible to contrast 194 differentials among brands, either at a regional or at a national level. Structuring a National Sample (Stratified Random Sample) The second procedure, as mentioned above and the results of which will be fully detailed in this section, was that of obtaining a national sample of 1,476 res- pondents from the computer tape. The tape was composed of the field questionnaires. A partial regional analysis ‘ had been performed in each case in order to check the data for possible errors along the entire process. A different sample size was determined for every city based on its total product consumption, and the sample was selected systematically. A different interval was used in each case, since every sample size had been different and was not related to the city's relative pro- duct consumption. Factor analytical procedures similar to those used in the case explained in chapter 6, a principal component analysis, and a varimax routine were performed. The attribute weights for each factor which resulted are shown in Table 7-1. As attributes were transformed into factors and then used as inputs in the multiple discriminant analysis, several data matrixes were produced. Among these were factor correlations, brand positions, and discriminant 195 TABLE 7-l.--Attribute Weights for Each Factor in National Sample. Factor Name Attribute Weight 1 Odor 0.97600 2 Flavor 0.99395 3 Strength* -0.97799 4 Mildness -0.98480 5 Healthfulness 0.98491 6 Sweetness 0.97798 7 Density 0.98466 8 Aftertaste -0.97976 weights, which are presented in Tables 7-2, 7-3, and 7-4, and their results are plotted in Figures 7-9 and 7-10. The F-ratios and percentages of discriminant power of each factor are presented in Table 7-5. An analysis of these tables and figures reveals several facts. First, the results are quite similar to those obtained through the weighted average procedure, as factor correlations once rotated show exactly the same length and direction in both cases. Second, brands are located at approximately the same distance and at the same position in both perceptual maps. Third, this second procedure enables us to know the factors relative means (Table 7-3). In order to find out exactly how much their 196 TABLE 7-2.--Factor Correlations for the Two Main Axes of the Perceptual Space, National Sample. Name of Factor Dimension I_ Dimension II Odor -0.32 -0.28 Flavor 0.55 0.27. Strength* 0.78 -0.20 Mildness* 0.22 0.79 Healthful -0.55 ' -0.20 Sweetness -0.52 -0.44 Density 0.85 0.06 Aftertaste* 0.30 0.14 1 For an explanation see Figure 6-11. *Shows negative factor loadings. TABLE 7-3.--Brand Location in Perceptual Map, National Sample. Brand Dimension I Dimension II A 0.52 -0.99 B -O.21 -0.69 C 0.06 -l.34 D -o.13 ' -o.43 E -l.06 -0.89 F -0.09 —0.59 G 0.34 -0.58 H -0.12 -0.52 *For an explanation see Figure 6-12. 197 TABLE 7-4.--Factor Discriminant Weights in National Sample.l Factor Dimension I Dimension II Odor -0.07 -0.09 Flavor 0.46 0.23 Strength* 0.58 -0.54 Mildness* ‘ . -0.14 0.75 Healthful -0.09 -0.13 Sweetness -0.08 -0.20 Density 0.64 0.05 Aftertaste* 0.09 -0.15 1For an explanation see Figure 6-14. *Shows negative factor loadings. TABLE 7—5.--F-Ratios and Percentage of Discriminating Power of Each Factor in National Sample.* Factor F-Ratio Disc. Power (%) Odor 5.35 4.7 Flavor 12.54 11.0 Strength 22.94 20.1 Mildness 17.22 15.1 Healthful 12.21 ' 10.7 Sweetness 14.77 12.9 Density 25.24 22.1 Aftertaste 3.84 3.4 *For an explanation see Figure 6-13. 198 Strength I Healthful Sw-~tness Af er- t.' e Odor oness k II Flavor “Pensity Figure 7-9.--Factor CorrelationsiJINational Sample. V" 199 I Stren\th Healthful Aftertaste -.o.» Sweetness II Mildness Flavor Dené}ty Figure 7-lO.--Discriminant Weights in National Sample. 200 position would change if an attribute were moved, Table 7-4 indicates factor discriminant weights. In addition, as F-ratios are produced, we are able to find out that, similar to City x, in the national figures it is mildness, strength, and density that are the relevant product attributes, whereas aftertaste and odor contribute very little in discriminating among brands. The integration of clusters is accomplished through the same programs. The total results are described in Table 7-6. TABLE 7—6.-—Market Segments in National Sample.* Segment Size Percent Axis I Axis II I 82 6.6 -3.30 0.42 II 84 6.8 -3.06 . -l.64 III 229 16.6 0.13 -1.11 IV 267 18.6 -l.65 1.03 V 212 , 15.3 -1.73 -0.58 VI 59 4.3 -2.27 -2.79 VII 89 6.4 -1.36 -2.00 VIII 264 18.4 " -o.19 0.28 IX 32 2.1 2.44‘ -0.45 X 42 3.1 1.17 —2.64 Centroids -1.28 -0.51 *For an explanation see Table 6-17. 201 In Figure 7-11 these clusters have been plotted on the perceptual map. For simplicity, brand positions also have been included in such geometrical space. In this way a national picture was obtained, and the results could be presented to management for decision-making pur— poses. Several additional analyses were performed on a national and regional scale through the use of the com- puter tape. First, consumers who preferred a given brand were taken as the universe and then the entire analysis was made. In this way the analyst could observe how speci- fic targeted (brand) consumers perceive each competitive brand and their ideal. As a result, either offensive (if the analyzed brand is one of competition) or defensive decisions could be made. Second, consumers in a specific demographic market segment were taken as a sample, for example upper class members. The entire analysis then was conducted to achieve similar aims. A third alternative was to analyze responses of consumers who had selected the advertising media in which they would like to see advertise- ments for the product. Several other combinations were tested, but their results are not presented here since they seemed uninteresting for marketing purposes. 202 A G .1 53.. . VIII B 0' A E ' II VII 6% V 15?: 69v: Figure 7-11.-- Brand and Cluster Locations in National Sample. 203 Regional Validation As one of our stated objectives was that of validating the use of various techniques not only on a single project but also horizontally, that is, by applying the same questionnaire and the same analytical tools to data from different cities, this section will present several such comparisons. Table 7—7 contains the percentages of discriminating power of each factor in Cities X, Y, W, and Z, as well as those which resulted from the national sample. These numbers can be contrasted with those in Figures 6-4, 7-2, 7-4, 7-6, and 7-8, where factor correlations were plotted on the perceptual maps. TABLE 7-7.-—Percentages of Factors Discriminating Power for Cities X, Y, W, and Z and National Sample. Factor City X City Y City W City Z National Sweetness 11.6 13.9 18.5 6.5 12.9 Odor 2.6 6.3 0.5 3.1 2.6 Healthful 8.3 11.9 9.8 11.5 10.7 Mildness 20.9 10.8 20.0 10.1 15.1 Flavor 7.1 9.0 3.0 14.4 11.1 Strength 20.7 16.6 17.9 25.3 20.1 Aftertaste 4.2 9.3 2.5 4.0 3.4 Density 24.6 22.2 27.8 25.2 22.1 204 In order to determine whether the discriminant power of each attribute was consistent, statistically speaking, among the different cities, a test of goodness was applied. A Table showing chi-squares for each of the factors is included as Table 7-8, where an example of this analysis is presented for Cities X, Y, W, and Z. A com— plete analysis was performed for the 30 cities surveyed, but for brevity we will include only the four mentioned. 2 to be In this case the maximum expected value for X accepted at a level of significance of 0.05 is 5.991. If 'the value is greater than this number, it could be said that there is a significant difference for that attribute among cities. TABLE 7-8.--Test of Goodness of Fit for Factor Discrimina- ting Power in Cities X, Y. W, and Z. Factor Mean Chi-square Sweetness ‘12.6 5.04 Odor 3.1 5.57 Healthful 10.4 0.79 Mildness 15.5 . 6442 Flavor 8.4 5.62 Strength 20.1 2.21 Aftertaste 5.0 5.28 Density 24.9 0.64 205 As can be observed in Table 7-8, mildness does not show a consistent discriminating power. Although important nationally, it is much more important for speci- fic cities. In contrast, density has a consistent and important discriminating power among brands, and it lends itself to a national advertising campaign. In Figures 7-12, 7-13, and 7-14, brand means for three original attributes are contrasted for brands D, E, and B. It is possible to verify consistency on rela- tive perception. In every case, for example brand E is considered more dense than brands B and D, but at the same time it is seen as the one with the least healthfulness. By the same token, brand D seems to be very similar to brand E in some regions and to brand B in others. A complete analysis, very similar to the one used in discriminant weights, was performed for all brands along all their attributes. This was done to contrast their respective means as perceived in each of the cities. Table 7-9 includes, for comparative purposes, the ideal brand perception for all the attributes in cities X, Y, W, and Z. As it can be observed in Table 7-9, desired attri- butes are not the same in each of the regions, but they are consistently perceived. That is, their direction and sign are the same in all cases, although their relative 206 TABLE 7-9.-dIdeal Brand Perception in Cities X, Y, W, and Z. Attribute City X City Y City W City Z Sweetness 0.56 0.76 0.32 0.48 Mildness 0.30 0.27 0.14 0.49 Higher-priced -0.97 -0.90 -0.84 -0.73 Availability 1.56 1.08 1.40 1.35 Aftertaste 0.71 0.82 0.72 0.62 Density 0.78 1.04 1.19 0.73 5 Odor 0.44 0.83 0.90 0.60 Strength -0.88 -0.56 -0.75 -0.74 Flavor 1.01 0.93 0.86 0.81 Healthful 1.62 1.13 1.48 1.41 Sample Size 999 810 1,200 814 mean differs. The way each brand is perceived in the different areas is exemplified in Tables 7-10 and 7-11 for the attributes sweetness‘and density, respectively. The reader is to be reminded that the analysis that was per- formed on data from Table 7—7 also can be used to test goodness of fit in Tables 7-9, 7L10, and 7-11. 207 Disagreement Agreement Brand Brand ' Brand C‘t M D B E 13’ -- t. e ‘ - 3% /’ Ideal City N City 0 City P City Q City R City S City T Figure 7-12.--Se1ected Brand Means for the Attribute Density. 208 Disagreement Agreement Brand Brand Brand City M E - Q B Ideal .n' I I I City N - : _.~°‘ 1 Ideal j l, - / / City 0 City P City Q City R City 8 City T Ideal Figure 7-13.-—Se1ected Brand Means for the Attribute Healthfulness. 209 Disagreement Agreement Brand Brand Brand City M ¥ k E Dk AB Ideal :7 It / ;;‘/ City N - ' Ideal [ 5 City 0 c { Ideal L i. City P 3 f-I. Ideal .1 City Q ; Ideal City R City S 3 Ideal City T k t V Ideal Figure 7-14-—-Selected Brand Means for the Attribute Mildness. 210 TABLE 7-10.--Brand Perception in Cities X, Y, W, and Z for the Attribute Sweetness. Brand City X City Y City W City Z A 0.75 0.83 0.67 0.51 B 0.63 1.17 0.86 0.51 C 0.51 1.33 1.33 0.44 D 0.05 1.02 0.59 0.25 E 0 26 0.03 -0.08 0.28 F 0.75 n.a. n.a. 0.60 G 0.70 0 61 0.72 0.87 H 0.51 0 96 0.37 0.51 I 0.41 1.08 0.49 0.82 TABLE 7-1l.--Brand Perception in Cities X, Y, W, and Z for the Attribute Density. Brand City X City Y City W City Z A -0.12 0.64 0.23 -0.17 B 0.56 0.70 0.54 0.97 C 0.50 0 19 -0.41 0.13 D 0.90 0.75 0 33 0.94 E 1.09 1.30- l 34 0.96 F 0.35 n.a. n a. 0.26 G 0 64 0 58 0 48 0.35 H 0.65 0.85 0.82 0.64 13" ’ '0 - .— . 211 As can be observed, brand perception for each attribute is distinctly different in each of the cities. If measured through the same scales, however, these dif- ferences can be displayed, and differential budgets and themes could be assigned to attempt to change the way that consumers see those brands in each city. Alternative Analytical Steps Used for Validating Results In order to obtain a better grasp of the results of this market segmentation research, several methods for analyzing data were attempted. One of these was the intercity comparison referred to previously, which indi- cates that the technique is providing consistent results. Another methodological change was that of obtaining parsimony in the factor analysis stage, where reductions were made to seven and then to five factors. A third modification was attempted when a three-dimensional structure was employed in the multiple discriminant analy- sis step. Data from the first three axes were used to obtain factor correlations and brands positions. Later, the same three were utilized for mapping consumer ideal ratings in the clustering procedure. A fourth step was that of including for analytical purposes the ten variables which had been measured. That is, the two attributes, "higher—priced" and "availability," which had been omitted 212 because they were considered exogenous factors which could be modified when management so desired, were included. A fifth and more conclusive step was conducted when management decided to make a marketing change in one of the current brands. An ex post facto survey was con- ducted in City X to measure the results of this change, and a second measure was available for the same city through the use of the same technique but at different time. Finally, as was mentioned previously, a sample of 960 retailers was taken in five cities to find out differen- tial perception between this group and ultimate consumers. In order to reinforce the results on brand and product attributes, several additional tests were con- ducted. In one of these a sample of 250 persons was taken in four cities. Four unbranded and different items of the same product were tested in order to learn whether con- sumers could tell the correct name of brands and, through open questions, to learn the attributes they were assigning to each brand. Another test was made through group inter— viewing. A trained psychologist conducted a conversation about the product, probing for a more universal meaning of the phrases which had been used for attribute or variable measurement. A summary of the findings on each of the different steps is given in this section. 213 Parsimony to Seven Factors Consumer data from City X were factor analyzed through the reduction of the original eight attributes to seven factors; the loss of information was equivalent to 8.3 percent. When the varimax routine was performed, it was found that the attribute density received high loadings in several factors, but it appeared mainly cor- related to sweetness (—0.67) and to strength (-0.45). In other words, those two factors would become "dense but not sweet" and "strong but not dense." In the multiple dis- criminant routine the first two axes obtained 88.6 percent of the trace, against 88.8 percent, which had been extracted without parsimony, and the discriminating weight of density, associated with sweetness and strength, the weight increased to 63.8 percent, whereas in the previous case the weight had been 56.9 percent among the three of them. Figure 7-15 presents the geometrical space which was formed through this parsimony, including factor cor- relations, brands, and cluster locations. It is interesting that this figure is very similar to that obtained without parsimony, but that it is inverted. If rotated (see Figure 7-16), it would appear that the origi- nal perceptual map and the one obtained through these methods would be the same. In Table 7'2.across tabulation 214 wildness 0e. "/ Healthful Aftertaste Stro g, and nO' Dense C. Figure 7—15.--Brand Position and Factor Correlations in City X. (Parsimony to Seven Factors). 215 E. A I Dense, but not Sweet D'. 1'. Flavor .C! H. F x .} II JD Oo\:\ .‘ftertaste Mildn~ ‘ Healthful Strong, «ut not Den - Figure 7-16.--Brand Position and Factor Correlations in City X. (Parsimony to Seven Factors). (Rotated Dimensions). 216 mam OOH mN mm NmH me mm mmH ow MVN HM#OB vb mm mH H H m m m m N Hmfluo om m o m H MH 0 o o mN MN m o N mm o o o o m NhH OH HH N o HvH o N o m mvH o o mN mm H 0 mm m 0 vv m o o o 0 HM m o o NmH m o o v m vH mHH v o o¢ m o o H o o H Nm 0 HmN mm o o wH o o o o NON Hmuoe Hmnuo m h o m w m N H mHODOMm cw>mm Op NCOEHmHmm SDHB UmumummncH HmumsHU .mcoEHmHmm mchD mg :oHpmnmousH HmumSHo mo COmHHwQEOUII.NHIn mamfle 'Auomrsxed qnoquM peureqqo JeqsnIQ Ieurbrxo 217 is made contrasting cluster formation through the use of eight variables and through parsimony up to seven factors. The difference found between the two procedures is minimal. Parsimony to Five Factors In order to test the effect of obtaining parsimony in factor analysis, the same data which had been trans- formed into seven factors later was transformed into five factors. In this case 73.6 percent of the original variation of the data was retained through the principal is... J” _ component method; as in the previous case, some of the factors received high loadings for several attributes. Table 7-13 presents such loadings. As already known, factor one becomes "dense but not sweet" and "dense and not strong"; factor two is aftertaste and odor; factor three, healthfulness and strength; factor four, mildness; and factor five, flavor. As can be observed, it is dif- ficult to interpret the factor when conducting parsimony, even after the varimax routine is performed. In the multiple discriminant analysis, the first two axes traced climbed to 92.8 percent as contrasted to 88.6 percent in the previous case. The discriminant power of density remained the most important, 43 percent. In Figure 7-17, factor correlations, brand positions, and cluster locations are included for the purposes of 218 .mcwomoH Houomm m>Hummmc mmumoHchttt .mnm musmflm mom COHumsmmeo am momt «ts smcouum =mcouum Hos umoBm «ts paw emummuumumm Hmsuflmc =uo>mHm= emmmcprze Hombuamome tam mono: pan omcmoe umfimz Houomm who.o| mmo.o| nmw.ol meo.o «No.0 . aswnpammm onm.o mmo.o mmo.o mmo.o mNH.o Ho>mam mmH.o mmo.o| mvm.o| mmo.o vmm.ol aumcouum mvo.o| hvo.o NOH.o| mmm.o nvo.o Hooo omH.o mnH.o| who.o mmo.o mmn.o wufimcmo mno.o HnH.o| nmo.o mmn.o mmH.o| mummuumuwm mmo.o| mam.ol moH.c| mno.o moo.o mmmcpaflz mmo.on mmv.ou mno.o mmH.o mvm.o| mmmcummzm m Houomm v nouomm m uonomm m uouomm H uouomm mpsbfluupd Hmsflmfino ‘ A.muouomm m>Hm ou mcofiflmnmmv .«x muHU How muouomm AmeHHm>v Umpmwomll.MHln mumdfl. 219 . I Den ity O E D .’ M“ . I . weet Flavor G F l. H o _ a . Odor and Aftertaste Healthfu and Strong (DC A. Figure 7-17.—-Brand Position and Factor Correlations in City X. (Parsimony to Five Factors.) and 220 contrast with Figures 7-16 and 6—8, which show similar num- bers for the case of parsimony to seven factors and for no parsimony, respectively. As can be seen, there is no significant difference in either of the three cases. Thus parsimony can be achieved in order to reduce the number of original attri- butes to a smaller number of factors, reducing in this manner the computation time for the rest of the programs without loss of significant information. Although in our case we included only eight attributes and parsimony would not reduce computer time significantly, in the case when more than ten original variables are used, this fac- tor analysis technique was extremely important. It would be very time consuming, even with larger computer equip- ment, to handle in discriminant analysis, let us say, 100 original attributes. Furthermore, the routines performed with this technique are very helpful in eliminating redundancy among the original variables; those which are heavily intercorrelated will reduce to one factor. Through this approach it might be possible not only to shorten computer time, but also, and most important, to reduce questionnaire length and interviewing time; several of these attributes can be deleted without significantly affecting results and without increasing the probability of obtaining biased information due to more complicated field work. 221 Analyzing the Ten Original Product and Brand Attributes As was mentioned previously, for analytical purposes, two of the original ten attributes were not included. In this section, some of the relevant results obtained from the inclusion of these two variables, higher-priced and availability, will be summarized. inclusion of these two variables, higher-priced and avail- ability, will be summarized. It is obvious that the matrix of correlation coefficients will not vary in content, except for the new rows and columns derived from these two new variables. They were not closely associated with any of the previous attributes, as the largest coefficient was 0.18. In order to have comparable results, no parsimony was requested. The results of the factor analytical rou- tines were ten rotated factors, each one associated with only one original attribute. In the multiple discriminant analysis the first two axes explain only 75.8 percent of the discrimination among brands in contrast to the case of eight attributes, where 88.6 percent was accounted for in the first two axes. In this case, axis III includes an additional 14.6 percent of the discrimination. As more factors enter in a multiple discriminant analysis, the per- centage explained by the first two axes decreases, thus g3 222 creating another reason for achieving parsimony in the factor analytical procedures. By adding the higher-priced and availability ,variables, the discriminating power of each factor was modified, as is shown in Table 7-14. Similar to Table 6—13, TABLE 7-14.--F-Ratios and Percentage of Discriminating Power of Each Factor in City X, Ten Attributes.* Factor F-Ratio Disc. Power (%) Flavor 7.32 3.4 Odor 2.47 1.2 Higher-Priced 54.47 25.6 Healthful 7.84 3.7 Mildness 19.99 9.4 Availability 55.67 26.2 Strength 22.81 10.7 Aftertaste 4.13 1.9 Sweetness 11.33 5.3 Density 26.67 12.6 * For an explanation see Figure 6-13. Table 7-14 presents the F-ratios and percentage of dis- criminating power of each factor for the same city in which all analyses were performed, City X. It is very interesting i r 223 to notice that the two new factors account for 25.6 and 26.2 percent of the discriminating power among brands, thus becoming the most important. However, the rest of the factors remain in the same order as they had been in the previous analysis, that is, density, strength, and mildness are important, and odor and aftertaste are almost irrelevant for discriminating purposes. Factor correlations, brand positions, and cluster locations, as derived from this analysis are plotted in Figure 7-18. For comparative purposes, the reader is advised to refer to Figure 6-8, which was obtained through the use of only eight factors. We think that both analyses are necessary for arriving at correct information for decision-making pur- poses. The first analysis,of eight variables, will inform management about perceived intrinsic and/or subjective product characteristics; the second, using the ten origi- nal variables, will more nearly accord with actual mar- ket preference and share. As it was explained, the two "new" factors account for 51.8 percent of total discrimi- nation among brands. In addition, simulating new brand positions in the perceptual space would produce results similar to those obtained before if price and availability also are changed. If not done, and if only one of the other attributes is modified, an analysis of the ten 224 I ‘vailability Density A I 27% ® VII Mildne - 1v. F :vor -etness Odor I 9 V II ftertaste 9 VI III 38% G IDI F o B 1 Strength ‘.D CDC (. A Higher-Priced Figure 7-18.--Perceptua1 Map in City X by the Use of the Ten Original Attributes. (Rotated Dimensions.) 225 relevant variables would indicate not only the new and changed position of a brand, but also an estimate of its new brand preference or market share. Comparative Results from a Three- Dimensional Analysis In some cases it was found that the third axis explained an important part of the total discrimination among brands. Therefore, it was decided to test whether its addition significantly would modify the previous results. For this purpose, the multiple discriminant analysis output was altered to include this third dimen- sion for brand positioning and factor correlation. Later, as the cluster routines also were modified, the third dimension was used in cluster location on the perceptual map. Since third—dimension presentation is not easily performed, all the results, using that dimension, were reduced to two axes. In Figure 7-19, a composite picture of these results is presented along the geometrical space developed through the use of two dimensions. As can be observed from the dotted circles representing clusters for the two-dimension formation, there is not much difference in the results. It might be concluded, therefore, that two- or three-dimensionality can be used since final decisions would not differ significantly in either case; 226 V17 7% , I _,/ 10% I \ [I III \\ l ‘ 15% 1 C 01A \ I \~’/ Figure 7-19.-—Comparison between Cluster Integration by Use of a Two— and Three-Dimension Procedure. 227 in this instance two dimensions explain an important portion of the discrimination (over 75 percent). When the proportion is lower than this number, it might be advisable to include a third dimension. However, data should be presented in a two-axis geometrical space in order to gain discriminating power when integrating clus- ters by the use of this additional information. Ex Post Facto Test in City X When management analyzed the results of the pro- ject of market segmentation by consumer perception, the firm's decision makers introduced a change in the marketing attributes of their brand. An ex post facto survey was conducted, and 658 respondents were interviewed with the same questionnaire. The same analytical tech- niques were used in processing the data. Some of the most important comparative results are included here in order to present an additional validation of this measurement system. Table 7-15 contrasts consumer perceived attributes of the ideal brand resulting from the January, 1972, study and the one made 18 months later. A goodness of fit test through chi-square analysis shows that X2 equals 0.186, which is lower than the number of the X2 table. This means that there is no statistically significant difference between the two tests conducted 18 228 TABLE 7-15.--Idea1 Brand Perception in City X. A Comparison of Two Different Survey Dates. Attribute First Test Second Test Sweetness 0.58 0.37 Mildness 0.35 0.19 _ Higher-Priced . -O.95 -0.93 ! Availability 1.46 1.62 r. Aftertaste 0.71 0.56 4 Density 0.77 0.87 1' Odor 0.46 0.30 ' Strength -0.80 -0.96 Flavor 0.99 0.99 Healthful 1.51 1.60 months apart. However, as can be observed in Table 7-15, there are very small changes among the rated attributes. The larger changes in the variables that show lower dis- criminating power among brands are shown in Table 7—16 for the ex post facto test in City X. By contrasting these figures with those shown in Table 6-13, it is interesting to note that the same factors remain the most important ones (mildness, density, and strength), and the least important remain the same (odor and aftertaste). Figure 7-20 shows the perceptual map of 229 .pl 5 I Den-'ty I D : . . as- . O H Flavo F G ‘ ‘ II C O Mildne B C \‘Odor ‘fter- saste Sweetness Strength A 0 Figure 7-20.—-Brand Position and Factor Correlation in City X. (Ex Post Facto Sample). 230 TABLE 7-17.--Percentage of Discriminating Power of Each Factor in the Ex Post Facto Test in City X. Factor Discriminating Power (%) Flavor 12.8 Odor 2.7 Healthful 8.4 Mildness 17.1 Density 26.4 { Aftertaste 4.2 : Sweetness 13.9 L Strength 14.5 City X in the ex post facto survey. If contrasted to Figure 6-8, comparison would reinforce the conclusion that there was no difference in perception over the lB-month period. Some of the brands, due to changes in their com- munication strategy, did change their position in this geometrical space, which is to be expected. The important point is that consumers did not change their preferences, only their perceptions about brands in this market. Retailers' Perception As noted previously, a study was conducted in five cities, among 960 retailers, who were considered as 231 ultimate consumers. Results were used to conduct public relations campaigns among them in order to increase positively their perception toward specific brands. These results also were very useful for validating consumers' responses, and X2 analyses were conducted for every type of retailer against specific market segments. This was done in order to learn which differences in perception it was most important to change for specific brands. .n‘l 1.84.. l-‘I‘. 3 _. Contrasting Segmentation Approaches 1 One of the aims of this study was to contrast i traditional segmentation analysis with the structuring of market clusters through consumer perception. Therefore, cross tabulation analysis was performed among these clus- ters, and groupings were made according to ideal brand and existing brand attributes. For presentation purposes, Figure 7-21 presents a univariate comparison to show how the different demo- graphic market segments perceived attributes M1 and M2 in relation to how much discriminating power is achieved if clusters are integrated through our approach. As can be observed, the three selected demographic segments-- socio-economic strata, age groups, and consumption levels—- perceive these two attributes very similarly in contrast to the differential perception achieved through the nondemo— graphic clusters. This type of analysis was extended to 232 In Complete Indifferent In Complete Disagreement Agreement "Socio-Economic Strata" "Age Groups" I I II I4 II I III IV "Consumption Levels" 1 M L H A l L "Clusters Formed by Consumer Perception" J 1 I l l 1 ll Figure 7-21.--Perception of Attribute M Along the Different Market Segments in City X. 233 In Complete Indifferent In Complete Disagreement Agreement "Socio-Economic Strata" l C E D A-Bn 1 1 "Age Groups" I III IV II IL I P L "Consumption Levels" l L M H 1 AL I ”Clusters Formed by Consumer Perception" 4 5 3 7 8 21 6 L l L I l 1 4| 1 Figure 7-22.--Perception of Attribute M2 Along the Different Market Segments in City X. 234 every city. The conclusion was reached that, for brand positioning purposes, it seems advisable not to rely on demographics, at least for this type of product. Con- sumer perception does not run according to traditional segments. However, if consumer perception is measured, better information for marketing planning and strategy can be obtained. CHAPTER VIII EVALUATIONS AND CONCLUSIONS Marketing as a business philoSOphy and as an academic field has found its basic foundation in the more economically advanced countries. There enterprises struggle to gain and maintain consumer acceptance as their basis for continuous growth. In the so-called develop- ing nations, where the economic and socio—political en— vironment permits the development of a free enterprise system, marketing has rapidly begun to develop among the larger local business firms and multinational corporations. Marketing research, both as an academic subject and as an applied methodology, has evolved very quickly during the last six years. It has introduced techniques from the behavioral and quantitative fields, thus provid— ing management with more decision-oriented information for marketing, planning, and strategy. A specific set of these newer techniques was used for this study, whose purpose was to develop research on market segmentation by consumer perception. The survey's findings positively can be described as decision oriented since management can gain more precise information from them to guide their marketing actions. 235 236 Among the several objectives that led to this particular research project, the main aim was that of examining the validity and application of newer techno- logies and concepts in a developing country, Mexico, and in a practical business case. The intent was to test the techniques in an environment where very few such efforts had been made before and where there are obstacles for the researcher. Furthermore, the testing was to occur in a situation where additional validations could be performed on the information received from the market. Many conclusions have been described at each step of this study, and they will be summarized in this chapter. First, some general conclusions regarding marketing research and marketing philoSOphy in developing nations will be presented. A few thoughts related to the trans- fer and adaptation of technological advances also will be offered. Second, conclusions related to marketing segmentation by consumer perception will be noted. Third, the general findings of this study and a description of potential marketing activities as a product of those findings will be given. Finally, some of the main limi- tations of these techniques and.a proposal for further research to overcome these will be suggested. General Conclusions Technology is not limited by national borders. It is transferable if it is adapted to local situations 237 and if it is designed to overcome practical limitations. If a country is to obtain a more rapid economic develop— ment, it must examine and validate newer technologies which become available through the current literature. When proven superior to actual practices, these must be implemented in an effort to reduce the economic gap between more and less advanced countries. Management myopia in the less developed regions will only contri- bute to keeping those countries in an underdeveloped state. Newer technology in marketing research has become available. In contrast to that used a decade ago, which emphasized descriptive market surveys, it has become decision oriented. It has begun to measure consumer behavior quantitatively. The emergence of larger and faster computers and an interdisciplinary approach which has borrowed heavily from the behavioral and mathematical sciences has had two significant effects. It is possible to learn more about the ultimate consumer and to process larger amounts of data rapidly in order to generate useful information for strategic marketing planning. The development and implementation of newer technologies is only possible when the business firm has acquired a consumer orientation, or a marketing conscience. That is, management must be interested in :—m.:xfi L a ism ‘ Q... .' 238 how these human beings, the product's consumers, perceive the marketing actions of the business firm. The acqui- sition of this marketing philosophy by companies in less develOped nations can be accelerated if marketing researchers place more emphasis on decision-oriented rather than descriptive projects. It is the obligation of any researcher worthy of the name to survey available technology before- searching for data through accustomed procedures. Better tools become available periodically which can generate information of a better quality for decision- making purposes. The assertion that "it cannot be applied here" is part of the management myopia that the researcher must eliminate through experimenting with new technology. The implementation of the marketing concept, a task which can be facilitated by sound marketing information, will help business firms in less develOped countries become more competitive, not only within the national borders, but also internationally, in the export market. Action in this latter area could serve to reverse an unfavorable balance of trade. A model which considers multivariate effects on marketing results has been described in this study. Researchers should provide not only the data to feed it, but also the methodology for its effective use. Although 239 many limitations apparently stand in the way of researchers who desire to adopt these newer quantita- tive technologies for the measurement of market and consumer behavior in the develOping countries, the researcher must understand that most of these are human limitations. Rather than as limitations, these condi- tions should be visualized by the researcher as Oppor- tunities for using his knowledge, innovativeness, and initiative. If searched for, computers and computer pro- grams are available almost everyWhere. Programs even can be rewritten and adapted to the special character- istics of small computing units. If no secondary data can be obtained for a given project, there are methods which do not rely heavily on such information for obtaining and validating research results. If stratified random sampling, for example, cannot possibly be used to select respondents, random area samples can achieve similar results. If in some cases respondents would not provide all the desired data, unnecessary questions should be deleted. Sound questionnaires sould be pre- pared and fully tested. Several questionnaires could be used in the same universe to obtain complete data. However, it has been our experience that sound practices in recruiting, selecting, training, and supervising interviewers pays off in terms of better field work 240 data. Researchers should understand that outlays for these activities should not be regarded as money wasted; these expenses are investments which will produce better results. The common practice of underpaying field workers as a means of cost reduction will only increase the prob- ability of generating data of dubious quality. Perhaps the most important limitation for the implementation of newer technologies in marketing research in developing countries is management's per- ception of such surveys. In this case the researcher must know his customer. If the researcher is marketing oriented, his position would be that of stressing the potential uses of his product, information, rather than one of "selling" methodology. Care should be taken to avoid "overselling." Even the "overbuying" of potential results by business managers should be discouraged in order to reduce the risk of conducting "last" surveys for every client. The researcher must understand the limitations of his methodology and the basic assumptions about the data he is handling. Care should be exercised in interpreting them when providing information.for decision—making pur- poses. New methodology should be experimented with before conducting full surveys. Tentative marketing decisions based on this information should be tested in areas other than the national market or with brands 241 other than the leading brand of a company until satis- factory results are obtained. Although more time con- suming, this "scientific" methodology will help the researcher overcome management's apparent lack of receptivity. I Market Segmentation by Consumer Perception Concurrent with the adoption of the marketing concept by a business firm should be management's awareness of market segmentation. Once the decision maker has observed his market, he discovers that some brands seem to sell well in several markets, but not- in others. The several types of market segmentation studies are intended to provide management with empathy, that is, he should look at his brand from the con- sumer's point of view. The studies also are meant to develOp a basis for building up a marketing strategy. Most of the traditional bases for market segmentation--geographica1, demographic, socio- economic or psychographic--although very useful for describing market behavior, fail to provide management with useful information regarding differential consumer perception. Through traditional methods, consumer behavior--past, present or expected--can be analyzed, but no indication is given as to what actions the firm must engage in to promote their product. 242 In contrast, if the market is segmented through the use of a methdology similar to the one presented in this study, the decision maker learns which are the relevant variables, or brand attributes, that need to be modified, physically or perceptually. He also learns what power they have in altering present brand position in the important market clusters formed by combining l consumers' perceived product wants. Each segment is characterized by a common desire for specific product attributes, and members of each cluster have provided hut"- waves. information as to how well they perceive each brand satisfies those wants. The decision maker thus knows what he must do in order to cater to selected market segments. He may alter the marketing mix of one or several of his brands within that product category. Segmenting markets by consumer perceptions is a positivistic, pragmatic, and realistic approach. It attempts to describe how the consumer behaves and what he wants from a product. It does not assume normative criteria about what he should want or about how he should behave if positioned within a given market segment. This method is oriented toward a Specific product or brand as opposed to adjusting a priori market classi- fications which might or might not be appropriate to the given marketing problem. Furthermore, it does not assume preassigned weights for any of the variables 243 under study. It is a multidimensional approach to con- sumer attribute perception; several variables interact simultaneously, but heterogeneously, on consumer behavior. In contrast to its forerunner, motivation research, the new approach is quantitative, and its results are less dependent on the researcher's sub- jectivity. If the same procedure is used to analyze the same data, the same output will be obtained. In addition, the same approach can be used to survey dif- ferent markets, or the same market at a later date, as was done in this study, and comparable results will be obtained. Finally, simpler field work is required. The task of assigning weights to attributes is removed from the field work stage and is passed on to the computer. Several of the techniques used in this study are noted below. Two stage area sampling was used in order to obtain a random sampling of the population to ensure generalization. Likert type scaling techniques were employed instead of semantic differential ones since the former simplified field work. Field work was conducted through personal interviewing. Blindfold product testing was adOpted in order to arrive at some of the relevant product attributes to be measured. Group interviews sought to obtain common usage correlates for the measured attributes. Analysis of variance was 244 applied in order to determine significant differences among brands in their perceived attributes. In the data analysis stage factor analysis was used to reduce dimensionality of attributes by trans- forming them into orthogonal and uncorrelated factors. Multiple discriminant analysis was employed to learn the discriminating power of each factor and to construct a p geometrical space or perceptual map. This latter, defined by factor correlations, was used to position -1 Fu'lt‘l-J ‘I I ‘ brands as they were distinctly perceived by consumers. KA‘ Later, cluster analysis grouped consumers together from " the perceptual map on the basis of which attributes they had perceived as positive in the ideal brand. Chi-square analysis also was used to determine differ- ences among cities on product or brand perception. Chi-square analysis was useful later when attempting to obtain clusters of cities on the basis of common perception in order to determine representative areas for market experimentation. Similar surveys conducted in 30 cities using a sample of retailers and in a given city after the brand's physical contents had been modified produced comparable and reliable results. Furthermore, alterna- tive assumptions about the data were made, and the final results were similar. All of these steps led us to conclude that this technology is relevant, valid, 245 and adaptable for measuring consumer perception in developing countries, not only in the product category, in which it has been tested, but also in many others if adequate research is conducted. Potential Uses of These Research Findings Although the purpose of this study was to vali- date the technique of market segmentation by consumer perception, it may be useful to present some ideas as to what marketing decision makers can do with the find- ings of this type of research. First, let us consider the idea of brand posi- tioning. A multi-brand company can learn the perceived position of its brands. If all are clustered together, the firm may change its marketing mix to separate and guide products to the most important market clusters. Second, knowledge of what the consumer in a given market cluster expects from products enables development of a program for new brand(s) aimed at unsatisfied wants. Third, management must learn the weight of each attribute before attempting to modify brand posi- tioning. Through market segmentation of the type shown herein the firm knows which product characteris- tics it should attempt to change in a given brand in order to cater to a specific market segment. 246 Fourth, an alternative route for increasing sales could be to attempt to alter consumer perceptions. Through the research we describe, management knows how the brand is perceived in the market clusters, and it also knows what each segment wants from this product line. A firm's advertising campaign might be designed to change desired product characteristics, or, in other words, to educate the consumer by telling him what he should be looking for in a product. Although this seems to be a more product-oriented tactic, and although it might be harder to achieve than those pre- viously described, in some cases it is an advisable course to follow. Fifth, results from studies of this type can be used as a basis for advertising experiments, related either to themes, copy, or even budgeting. With a before market measurement available for a given city, changes in advertising strategy can be attempted there. An after measurement could tell how product or brand perception or position had been affected. If advertising costs are introduced into the experi- ment, it might be possible to learn how much a unit change in perception of an attribute costs, and how it affects other variables (since we are dealing with a multivariate technique of market measurement). 247 Sixth, a measure of competitive market advance- ment can be achieved in the same way that a brand is measured. Competitive brands' advertising themes can be researched as to the effects they are producing in consumer perception. Seventh, as a measurement is available for every city, each one of the strategies described above can be conducted experimentally in one of them. The rest can be used as control groups. Furthermore, an X2 analysis can be conducted to learn which cities exhibit .common perception in order to select one of them for experimentation purposes. Cities showing common product perception could be clustered for strategic marketing planning purposes. Efforts need not be limited to differential advertising stragegies, but could include radical differences in marketing mix, distribution methods, sales organization, pricing and even special brands for those clustered cities which might or might not follow geographical patterns. Finally, a continuous panel type study has been suggested in this type of research. Such an effort might determine how the market is changing for a given product category, either in intercity comparisons or through time. 248 Additional Research to be Conducted in the Future It can be said that through this study a more complete knowledge of the consumer has been gained, that management has been provided with useful tools for decision making, and that several validating steps have confirmed the results obtained. Nevertheless, it is our belief that more research along these same lines * I is necessary not only to increase our knowledge of consumers, but also to provide even better marketing information. 1 There were several important limitations to the methodology which was chosen; through research, several of these could be overcome. First, consumers were rating attributes on a priori selected variables and were questioned with specific phrases. This procedure might produce biased answers if respondents' personal interpretations differ. More depthful questioning procedures, conducted on a limited scale among some of the respondents, might aid in verifying the truth of this possibility. Furthermore, through a similar group interviewing technique, such as the one imple— mented at the end of this studyq qflfifier confirmation could be obtained. Second, if care is not taken in selecting the attributes to be measured, some significant 0 249 ones in discriminating among current brands might be omitted. If this type of extrinsic testing is not included in the methodology, it might well happen that several years later some of the attributes would lose their importance, whereas others, not included in the process, might gain importance in brand discrimination. An example of this point is the use of color in adver- tising. If a brand is the first to use a given color, let us say red, it might capture the attention of the viewers, but when more products began to use the same color, it might lose its importance. Third, respondents might not have the same knowledge of all brands. If asked about some of them, their answers might be simply guesses. The interviewer must make sure that the respondent, although he may not have tried it, knows about a specific brand and has opinions about it. Furthermore, one of the important methodological assumptions is that respondents are able to imagine an ideal brand and that it can be rated along the selected attributes. In this respect, extrinsic group interviewing is very helpful; consumers openly say which product attributes that they would like to see modified to produce a product better suited to their specific wants. Fourth, research is necessary in the analytical stage to test data through some of the available and 250 newer computer programs, mainly the fully nonmetric procedures. This would confirm our previous ideas that the ones we used provided a close approximation to reality. In the factor analysis step, a varimax routine was used, but alternative rotating procedures are avail- able; quartimax, equimax, or oblique rotation each would make different assumptions about the data. In the clus- tering stage it might be advisable to use alternative clustering techniques in larger computers in order to test if the one selected for this study provides accept- able market segments. Fifth, it is our belief that, in order to improve the communication necessary to modify consumer percep- tions, advertising media research in the various segments should be conducted. This would reduce the risk of ineffective and inefficient advertising campaigns. At any rate, as was previously suggested, management always could use the results of this procedure to conduct experimental campaigns and to pretest advertising in selected cities and amxg target clusters. Finally, it seems pertinent to point out that unidimensional attitudes or roles in the research procedure, such as overspecialization of researchers and analysts, might be very strong limitations to furthering the implementation of multivariate analysis in marketing. It is difficult for a line executive to 251 fully understand the methodology involved. As a result, he might underestimate findings if contrary to his pre- vious market belief, thus fostering the use of more tra- ditional, or easy to understand and evaluate, methods of research. As the younger generation of researchers is the one that can apply this newer methodology most readily, the older generation, which usually heads research organizations, might undervalue the methodology for strictly nontechnical reasons. However, the use of quantitative methodology and computers will not automatically produce better informa- tion. Quantitative-oriented researchers, especially those with little experience in data gathering and who are overspecialized in methodology, might be asking too much from what, in their eyes, seems very easy to obtain-— consumer data. When analyzing a problem from one's own desk, it is very easy to forget the basic principle of electronic data processing, which can be applied use— fully to marketing research: "Garbage in-—Garbage out." The use of newer techniques and methodologies, it is our belief, will foster the role of marketing researchers within business firms. Its implementation in companies in the developing countries will assist them in reducing the gap between themselves and the more economically advanced nations. BIBLIOGRAPHY 252 BIBLIOGRAPHY Ackoff, Russell L.‘ Scientific Method: 'Optimizing Applied ‘Research DeciSions. New York: John Wiley and Sons, Inc., 1962. Albaum, Gerald, and Venkatesan, M. Scientific Marketing Research. New York: The Free Press, A Division of the Macmillan Company, 1971. ! Alpert, Mark I. "A Canonical Analysis of Personality and the Determinants of Automobile Choice." Combined Pro- ceedings, 1971 Spring and Fall Conferences. American Marketing Association, Series No. 33, Edited by Fred C. Alvine, 1971. n k F. WEI Alpert, Mark I. "Identification of Determinant Attributes: L A Comparison of Methods." Journal of Marketing Research, VIII (May, 1971), 184-191. Banks, Seymour. Experimentation in Marketing. New York: McGraw-Hill Book Company, 1965} Barelson, Bernard, and Steiner, Gary A. Human Behavior: An Inventory of Scientific Findings. New York: Harcourt, Brace and World, Inc., 1963. Bass, Frank M.; Pessemier, Edgar A.; and Tigert, Douglas J. "A Taxonomy of Magazine Readership Applied to Problems in Marketing Strategy and Media Selection." The Journal of Business, XLII (July, 1969), 337-363. Bass, Frank M.; Tigert, Douglas J.; and Lonsdale, Ronald T. "Market Segmentation: Group Versus Individual Behavior." Journal of Marketing Research, V (August, 1968), 266—268. Bell, Martin L. Marketing: Concepts and Strategy. Boston: Houghton Mifflin Company, Inc., 1966. Benson, Purnell H. "How Many Scales and How Many Categories Shall We Use In Consumer Research?——A Comment." Journal of Marketing, XXXV (October, 1971), 59-61. Boulding, Kenneth E. The Image. Ann Arbor: The University of Michigan Press, 1956. 253 254 Boyd, Harper W., Jr.; Ray, Michael L.; and Strong, Edward C. "An Attitudinal Framework for Advertising Strategy." Journal of Marketing, XXXVI (April, 1972), 27-34. Boyd, Harper W., Jr., and Westfall, Ralph. Marketing Research. Rev. Ed. Homewood: Prentice Hall, Inc., 1964. Claycamp, Henry J., and Massy, William F. "A Theory of Market Segmentation." ‘Journal of Marketing Research. V (February, 1968), 388-394. Crissy, William J. E., and Kaplan, Robert M. "Matrix Models for Marketing Planning." MSU Business Topics. B Summer, 1963, pp. 48-67. a Cunningham, William H., and Crissy, William J. E. "Market Segmentation by Motivation and Attitudes." ‘Journal 'of Marketing Research, IX (February, 1972), TOO-102. Day, George S. Buyer Attitudes and Brand Choice Behavior. E New York: The Free Press, A Division of the ' Macmillan Company, 1970. Day, George S., and Heeler, Roger M. "Using Cluster Analysis to Improve Marketing Experiments." Journal of Marketing Research, VIII (August, 1971), 340-34TT Dominguez, Luis V., and Burger, Philip C. "An Empirical Analysis of the Process of Buyer Behavior." Combined Proceedings, 1971 Spring and Fall Con- ferences, American Marketing Association,§eries No. 33, Edited by Fred C. Alvine, 1971. Drucker, Peter F. The Practice of Management. New York: Harper and Row Publishers, Inc., 1954. Engel, James F.; Fiorillo, Henry F.; and Cayley, Murray A. Market Segmentation: Concepts and Applications. New York: Holt, Rinehart and Winston, Inc., 1972. Fasley, John V.; Howard, John A.; and Weinstein, David. "An Investigation of Stability in Attitude Toward a Product Class." Combined Proceedings, 1971 Spring and Fall Conferences, American Marketing Association, Series No. 33, Edited by Fred C. Alvine, 1971. 255 Frank, Ronald E., and Green, Paul E. "Numerical Taxonomy in Marketing Analysis: A Review Article." Journal of Marketing Research, V (February, 1968), 83-98. Frank, Ronald E., and Massy, William F. "Market Segmenta- tion and the Effectiveness of a Brand's Price and Dealing Policies." The Journal of Business, XXXVIII (April, 1965), 186-200} Frank, Ronald E.; Massy, William F.; and Wind, Yoram. Market Segmentation. Englewood Cliffs: Prentice Hall, Inc., 1972. "Bias in Multiple Discriminant Analysis." Journal Frank, Ronald E.; Mathews, H. Lee; and Morrison, Donald G. I of Marketing Research, II (August, 1965), 250-258. Fry, Joseph N., and Claxton, John D. "Semantic Differential and Nonmetric Multidimensional Scaling Descriptions of Brand Images." Journal of Marketing Research, VIII (May, 1971), 238-240. \F‘-’.l‘4{qi"“ Gonzalez-Arce, Jorge F. "Analisis de las Caracteristicas de las Familias Lectoras de Periodicos en la Ciudad de Monterrey, N.L." Unpublished thesis, Instituto Technologico y de Estudios Superiores de Monterrey, Monterrey, Mexico, 1959. Gonzalez-Arce, Jorge F. "Concepto Operacional de la Empresa," Administracion, ITESM, XIV (February, 1968), 1-80 Gonzalez-Arce, Jorge F., and Marcos, Guillermo. Investiga- cion del Consumidor Regiomontano. Monterrey: Desarrollo Industrial y Comercial Mexicano, A.C., 1968. Green, Paul E., and Carmone, Frank J. Multidimensional Scalingand Related Techniques in Marketin Analysis. Boston: Allyn and Bacon, Inc., 1970. Green, Paul E., and Carmone, Frank J. "MultidimenSional Scaling: An Introduction and Comparison of Non- metric Unfolding Techniques." Journal of Market- ing Research, VI (August, 1969), 330-341. Green, Paul E., and Rao, Vithala R. "Multidimensional Scaling and Individual Differences.“ Journal of Marketing Research, VIII (February, 1971), 71-77. 256 Green, Paul E., and Rao, Vithala R. Applied Multidimensional Scaling: A Comparison of Approaches and AIgorithms. New York: HOIt, Rinehart anHCWinston, Inc., 1972. Green, Paul E., and Rao, Vithala R. "Rating Scales and Information Recovery--How Many Scales and Response Categories to Use?" Journal of Marketing, XXXIV (July, 1970), 33-39. - Green, Paul E., and Tull, Donald S. Research for Marketing Decisions. 2nd ed. Englewood CIiffs: Prentice Hall, Inc., 1970. Green, Paul F.; Wind, Yoram; and Jain, Arun K. "Analysing _ Free Response Data in Marketing Research." ‘ Journal of Marketing Research, X (February, 1973), 45-52. Haley, Russell I. "Benefit Segmentation: A Decision- Oriented Research Tool." Journal of Marketing, k XXXII (July, 1968), 30-35. ' Hardin, David K. "Market Segmentation: A Comparison of Techniques." Brewing Industry Survey 1970. Research Corporation of America, 1970. Howard, John A. "New Directions in Buyer Behavior Research." Combined Proceedings, 1971 Spring and Fall Con- ferences, American Marketing Association, Series No. 33. Edited by Fred C. Alvine, 1971. Howard, John A., and Sheth, Jadish N. The Theory of Buyer Behavior. New York: John Wiley and Sons, 1969. 2 Holloway, Robert J.; Mittelstaedt, Robert A.; and Venkatesan, M. Consumer Behavior: Contemporary Research in Action. Boston: Houghton Mifflin Company, 1971. IBM. System/360 Scientific Subroutine Package (360 A—CM—03X) Version III. ‘White Plains: International Business Machines Corporation, 1966. Jacoby, Jacob, and Matell, Michael S. "Three Point Likert Scales are Good Enough." Journal of Marketing Research, VIII (November, 1971T, 495-500. 257 Johnson, Richard M. "Market Segmentation--A Comparison of Techniques." A paper delivered at the 16th International Meeting of the Institute of Manage- ment Sciences, March 27, 1969. Johnson, Richard M. "Market Segmentation: A Strategic Management Tool." Journal of Marketing Research, VIII (February, 1971), 13-18. Johnson, Richard M. "Relationships Between Product Attributes and Preferential Choice Behavior." A paper delivered to the American Statistical Association, New York, August, 1969. Kamen, Joseph M. "Quick Clustering." Journal of Marketing Research, VII (May, 1970), 199—204. Kasserjian, Harold H., and Nakanishi, Misao. "A Study of Selected Opinion Measurement Techniques." Journal of Marketing Research, IV (May, 1967), 48—53. Kassarjian, Harold H., and Robertson, Thomas S. Perspec- tives in Consumer Behavior. Glenview: Scott, Foresman and Company, 1968. Kinnear, Thomas C., and Taylor, James R. "Multivariate Methods in Marketing Research: A Further Attempt at Classification." Journal of Marketing, XXXV (October, 1971), 56-59. Kollat, David T.; Blackwell, Roger D.; and Engel, James F. Research in Consumer Behavior. New York: Holt, Rinehart and Winston, Inc., 1970. Kotler, Philip. "A Generic Concept of Marketing." Journal of Marketing, XXXVI (April, 1972), 46-54. Kotler, Philip. Marketing Decision Making: A Model Building Approach. New York: Holt, Rinehart and Winston, Inc., 1971. ~ Kotler, Philip. Marketing Management: Analysis, Planning and Control. Englewood Cliffs: Prentice Hall, Inc., 1967. ' Landon, E. Laird, Jr. "Order Bias, The Ideal Rating and the Semantic Differential." Journal of Marketing Research, VIII (August, 1971), 375-378. 258 Lessir, V. Parker, and Tollefson, John 0. "Market Segmenta- tion Through Numerical Taxonomy." Journal of Marketing Research, VIII (November, 1971), 480-487. Levitt, Theodore. Innovation in Marketing: New Perspectives for Profit and Growth. New York: McGraw-Hill Book Company, Inc., 1962. Massy, William F. "Discriminant Analysis of Audience Characteristics." Journal of Advertising Research, McClure, Peter, and Ryans, John K., Jr. "Differences Between Retailers' and Consumers' Perceptions." Journal of Marketing Research, V (February, 1968), 35-40. . McKitterick, J. B. "What is the Marketing Management Concept." Proceedings American Marketing Associa- tion. Chicago: The Association, 1957. Meisel, William S. Computer-Oriented Approaches to Pattern Recggnition. New York: Academic Press, 1972. Miller, Delbert C. Handbook of Research Design and Social Measurement. 2nd ed. New York: David McKay Company, Inc., 1970. Mukherjee, Bishwa Nath. "A Factor Analysis of Some Qualitative Attributes of Coffee." Journal of Advertising Research, V (March, 1965), 35-38. Neidell, Lester A. "The Use of Nonmetric Multidimensional Scaling in Marketing Analysis." Journal of Marketing, XXXIII (October, 1969), 37-43. Neuhaus, Colin F., and Taylor, James R. "Variables Affecting Sales of Family-Branded Products." Journal of Marketinthesearch, IX (November, 1972), 419-422. Nie, Norman H.; Bent, Dale H.; and Hull, C. Hadlai. Statistical Package for the Social Sciences. New York: McGraw-Hill, Inc., 1970. Palda, Kristian S. Egonomic Analysis for Marketing Decisions. Englewood Cliffs: Prentice Hail, Inc., 1969. 259 Ralston, Anthony, and Wilf, Herbert S. ‘MathematiCal Methods for Digital Computers, Vol. I. New York: John WiIey and Sons, Inc., 1967. Reitter, Robert N. "Product Testing in Segmented Markets." Journal of Marketing Research, VI (June, 1969), 179-184. Reynolds, William H. "More Sense About Market Segmenta- tion." Harvard Business Review, XLIII (September/ October, 1965), 107-I14. Ross, Ivan. "Self Concept and Brand Preference." Journal of Business, XLIV (January, 1971), 38-50. Scheid, Francis. Numerical Analysis. New York: McGraw— Hill Book Company, Inc., 1968. Servan—Schreiber, Jean Jacques. The American Challenge. Trans. from the French by Ronald Steel. New York: Atheneum House, Inc., 1968. Sethi, S. Prakash. "Comparative Cluster Analysis for World Markets." Journal of Marketing Research, VIII (August, 19715, 348-354. Sheth, Jagdish N. "The Multivariate Revolution in Marketing Research." Journal of Marketing, XXXV (January, 1971), 14-18. Sheth, Jagdish N. "Measurement of Multidimensional Brand Loyalty of a Consumer." Journal of Marketing Research, VII (August, 1970), 348-354. Smith, Wendell R. "Product Differentiation and Market Segmentation as Alternative Marketing Strategies." Journal of Marketing, XXI (July, 1956), 5-11. Sparks, David L., and Tucker, W. T. "A Multivariate Analysis of Personality and Product Use." Journal of Marketing Research, VIII (February, 1971), 67-70. Stoetzel, Jean. "A Factor Analysis of the Liquor Prefer- ences of French Consumers." Journal of Advertising Research, I (December, 1960), 7—11. Toffler, Alvin. Future Shock. London: The Bodley Head Ltd., 1970. ..Is'u ‘51:. 260 Vavra, Terry G. "Factor Analysis of Perceptual Change." Journal of Marketing Research, IX (May, 1972), 193-199. Veldman, Donald J. Fortran Programmingfor the BehaVioral Sciences. New York: Holt, Rinehart and Winston, Inc., 1967. Weinreich, Rolf. "Metodos de Medicion Multidimensional: Un Nuevo Enfoque en la Medicion de Percepciones y Preferencias." Temas de Negocios, ITESM, II (Fall, 1972), 453-464. Wells, William D., and Tigert, Douglas J. "ActiVities, Interests and Opinions." Journal of Advertising, L Research, XI (August, 1971), 27-35. 1 Wilkie, William L. "An Empirical Analysis of Alternative Bases of Market Segmentation." Combined Proceed- - ings, 1971 Spring and Fall Conferences, AmeriEan 5 Marketing Association, Series No. 33, Edited by 1 Fred C. Alvine, 1971. Wilkie, William L. "The 'Product Stream' of Market Segmentation: A Research Approach." Combined Proceedings, 1971 Spring and Fall Conferences, American Marketing Association, Series No. 33, Edited by Fred C. Alvine, 1971. Wilson, David T.; Mathews, H. Lee; and Sweeney, Timothy W. "Industrial Buyer Segmentation: A Psychographic Approach." Combined Proceedings, 1971 Springand Fall Conferences. American Mafketing Association, Series No. 33, Edited by Fred C. Alvine, 1971. Wiseman, Fred. "A Segmentation Analysis on Automobile Buyers During the New Model Year Transition Period." Journal of Marketing, XXXV (April, 1971), 42-49. Worcester, Robert M., Editor in Chief. Consumer Market Research Handbook. London: McGraw-Hill Book Company (UKT Limited, 1972. Yankelovich, Daniel. "New Criteria for Market Segmentation." Harvard Business Review, XLII (March-April, 1964), 83-90. Ziff, Ruth. "Psychographics for Market Segmentation." Journal of Advertising Research, XI (April, 1971), 3-14 0 HICHIGAN STATE UNIV. LIBRQRIES ll!\l\“llllHllUllmlllWWI“!HHIHIIIIHIIWHHll 31293102461591