47 Akungba journal of Management ATTITUDE MEASUREMENT IN MARKETING: A COMPARATIVE ANALYSIS OF MULTIDIMENSIONAL AND UNIDIMENSIONAL SCALLIN6 DEVICES BY JACKSON O. OLUJIDE Department of Business Administration University of Ilorin, Ilorin. AND A. LATEEF BADMUS Department of Business Administration University of Ilorin, llorin. INTRODUCTION In the study of consumer behaviour attitude is considered an important variable. Thus, when an individual says "I like this product." "this restaurant is terrible" ... etc. this individual solemnly expresses his attitude towards the product or restaurant in question. In the study of consumer behaviour, attitude is certainly the variable that has retained and retains still most attention. Consequently, very few concepts have as much- impact on the practice of marketing as the motives of attitude of consumers towards products /services. In fact one of the fundamental objectives of marketing in general and advertising in particular is to influence consumer attitude. Hence, several marketing researchers have developed the view that understanding attitude change should become the primary objective of marketing. For instance, more and more psychologists have come to the conclusion that for any advertising campaign to have any influence on sales it must lead to a positive change in the attitude of the audience. Thus, the practitioners of marketing hold the view that there is a definite relationship between attitude change towards a brand and the sales of that brand This view posits a logical conclusion that rests on a solid foundation. Bui the problem that arises is that there are several divergent points of view on this vital subject—attitude. The possible explanations are: 1. The problem resides in the measurement instrument: or 2. Attitude is not precisely conceived and defined. Little & Hill (1967) indicates that the relationship between attitude and behaviour is as function of several variables. Thus, it is possible that the majority of published research work on the reliability of the measurement instrument. But in order to validly study consumer behaviour in marketing it is necessary, most of the! time, to collect data by means of inquiry or experimentation. The majority of data thus collected are of the nature non-metric and often ordinal (ranks, preferences, degree of importance etc). New techniques of analysis that facilitate the obtention from these ordinal data, a metric result were developed about 10 years ago. This development has led to what some people call the "Multidimensional Revolution" The appearance of these multivariable techniques of data analysis has completely put in the dark unidimensional scales of measurement which hitherto now used, essentially a non metric level (Likert. Guttmaa Semantic differential etc). The objective of this paper therefore is to undertake a comparative analysis of multidimensional and unidimensional scales of measurement.; This will be done by applying these two methods to the measurement of students' attitudes to the marketing of students restauration service This comparative study is intended as a means of ascertaining Akungba journal of Management the possibility of convergence of results from the two methods. A multidimensional analysis—factor analysis—is used to analyse the data on the first part of the questionnaire. In the second part. 12 salient attributes of the restauration service were identified during a session discussion in which non-directed questions were posed initially to 20 students chosen randomly. These attributes were presented to another group of 20 students selected randomly for the pre-test of questionnaire and to another group of 150 students selected randomly to arrange the attributes in order of importance by numbering them as follows: 1. For the first attribute that seems to be the most important. 2. For the second 3. For the third and so on until 12, for the attribute the least important. An interval scale was derived from this Thurstone Scale. RESULT: Multidimensional Analysis • Method of Multivariate data analysis are divided into 2 classes. • Descriptive methods and • Explicative methods Factor analysis used in this study is a descriptive method and applies in the following areas: • To simplify, and summarize data described by a minimum number of non-correlated factors • To visualize the data, in the form of a chart • To indicate the interdependence among variables • To regroup the variables that are closest. Factor analysis consists of 2 models: 48 • Principal component analysis • Common factor model for nominal and ordinal data In our case, we have metric results and therefore we apply principal component analysis. Factor analysis on the Totality of the Variables: Correlation Matrice of the Variables There is a correlation matrice of 18 variable (see computer print-out). The first variable is excluded because it does not measure an attitude with respect to restaurant service. However, responses to this question permit us to know if the productive capacity of university restaurant is well utilized by students. Cij is the correlation between variable i and variable j; — for Cij = 1, and J are perfectly correlated and vary in the same direction. — for Cij = -I, the correlations are perfect but in an inverse direction. — for Cij = 0, the variables are totally independent Our objective here is to replace the variables with factors, translated by: — a smaller number of factors — minimizing the loss of information and to verify if the principal factors, in a rank order (in terms of explained variance) attach the same meanings to the same variables as the Thurstone method. We now undertake a linear correlation of the initial variables and to do this we identify the variables that are highly correlated and replace them with a small number of variables without loss of information. Considering variables with a correlation coefficient +- 0.5 we have the following: Akungba journal of Management 49 TABLE 1: Correlation between the Variables (+0.5) Variables 4 7 10 11 12 14 4 7 1.00 1.00 1.050652 10 1.00 1.5:56 0.54639 0.58206 11 12 14 0.50752 0.58206 0.53560 054639 1.00 0.53187 0.53187 1.00 4.00 We observed that the correlations are strong between variables measuring the following: (a) 4: rapid service 14: quality/price (b) 7: Useful information 14: quality / price (c) 10: Taste of food 11: Nutrition (d) 10: Taste of food 12: Variety of food served (e) 11: Food 12: Variety of food served From this analysis we note that variables 10, 11 and 12 express ideas that are very similar. We therefore represent them by variable 11 that has a meaning that is more general such trial it could take into consideration the views expressed by variables 10 and 12. But when a new factor analysis without variables 10 and 12 was undertaken, the result obtained did not improve. We therefore decided to retain the 3- variables. To resolve the number of factors problem where the objective is to find the minimum number of factors compatible with the data, we applied one of the most popular criteria for addressing this issue. This criterion is to retain factors with engenvalues greater than 1 when the correlation (not adjusted) is decomposed. This criterion seems to work well, in the sense that it generally gives results consistent with researcher's expectation and it works well when - applied to samples from artificially created population models. Kaiser (1974) provides several reasons for the success of this criterion. He says that its acceptance is still based on heuristic and practical grounds. In the first analysis, five factors were obtained on the basis of engenvalues greater than 1 Translating the explained variance of the 5 factor in percentage, we have the following: • 1st factor represents 24.83% of the information (explained variance) • 2nd factor 14.48% • 3rd factor 10.64% • 4th factor 7.37% • 5th factor 6.65% The 5 factors explained 63.57% of the total variation and employing therefore the test of Kaiser (see figure 1 below) we decided to retain the first 3 factors that appear to us to be the most important as these have engenvalues equal to or greater than 1 (see computer result of engenvalues) . The rationale behind this citerion is that in a population matrix it provides a strict lower-bound for the number of common factors responsible for the data. But the same may not necessarily apply to the sample correlation matrix and application to empirical data usually produces more factors than normally accepted on other grounds. On the basis of the decision to retain 3 factors, a new analysis with the 3 factors was undertaken. Akungba journal of Management ANALYSIS OF THE FACTORS F1 F2 F3 Factor 1 is the most important of the 3 because it represents about 47.43% of the total explained variance; the 2nd factor 27.71% while the 3rd explained 21.2%. The three factors explained 40.506%. From the table of principal axes solution (3 factors) (see computer print out) price (economy), rapid service and the quality /price ratio (they have positive coefficient) are represented on the first axe. They represent the subjective attributes that appear to us as the most dominant in this axe. Factor 2 regroups attributes like taste, variety of food etc. which have negative coefficients. The variables oppose cleanliness, appropriate opening hours and increase in price. The 3rd factor regroups the variable "location" and accessibility. Figure 2 shows the first 2 principal axes. UNIDIMENSIONAL ANALYSIS Several methods of measurement employed by researchers in marketing are application of the judgement approach. Tongerson (1958) classified them into two: — quantitative methods — variability methods The 2 categories are relative to the manner of the subject's placement or stimuli on a chosen dimension. In general, the quantitative judgement and the variability approaches do not present important differences with respect to the manner of obtaining the unit of measurement. We shall use Thurstone law of comparative judgement which is a method of variability. Thurstone's law of comparative judgement is chosen because it has been more or less totally neglected inspite of its numerous advantages. Thurstone's law of comparative judgement consists, essentially, of the definition of an interval scale from comparative 50 judgements of the type: "A is better than B, A is bigger than B, A is more beautiful than B etc..... In the optic of our previous classification, this method is grouped with those founded on the variability of judgement approach. This technique which would require each subject to make 10 comparisons according to a single dimension. If comparison by two dimensions were required 20 paired comparisons would be necessary. Similarly the psychological distance between two stimuli S and C. is based on a normal curve transformation of the proportion of respondents who judge Sj, greater than Cg. In brief, an interval scale is derived from paired comparisons in the following manner: the numbers of respondents selecting one stimulus over another is an N by M matrix. These frequencies are then converted into proportions in which the main diagonal in the matrix of proportions has approximately 0.5 in each cell. In other words, if Sj were compared with Cg approximately half the respondents would report that S was greater and half the respondents would report that S was less than C, which is to say that they are equal. All other cell entries be greater or less than 0.5. thereby reflecting a perceived psychological distance between two stimuli. Expressing this mathematically we hare: P (j < g) = P(Sj < Cg) = P(Cg- Sj) < 0 = P (Dig < 0) if we let Cg - S, = D^ This difference is a normal random variable and transforming this in a reduced normal variable (tjg) = normal variable Djg - its mean (Xp_) its standard deviation (Dy) Looking at the following probability Pjg = P (tjg > - XJg D* In effect (DJg =0) implies that o* > -x*) DJg Representing this graphically Akungba journal of Management We now apply the developments in the previous page to our data which represent the responses of students to the significance and importance of attributes of university restauration. This involved comparing the 12 attributes of the University restauration service in order of importance vis - a-vis a series of criteria that are likely to be considered in the purchase decision. The calculations and transformations to interval scale are developed here after. Table 2 shows for example that 5% of respondents considered that attribute 2 (courteous and experienced personnel) 51 is more important than attribute 1 (rapid service) and that others, about 95% consider 1 more important. Next we calculate on the basis of the distribution function of the normal deviates corresponding to cumulated proportions. These appear in table 3. Thurstone's law of categorical judgment is established as follows: Cg – Sj = Yjg Sj2 + Cg2 - 2jg Sj Cg Where Yjg = Value obtained by consultation with table Akungba journal of Management 52 of the normal deviate Sj = standard deviation of the value attributed to stimulus j Cg = standard deviation of value attributed to stimulus j and the index of category g To identify the parameters of our scale (i. e. Cg, g = 1,2.........M and Sj = 1, 2, ......n) we mast introduce classic but simple hypotheses with a view to reducing the number of unknowns, which in our case, is greater than the number of equations. Thurstone assumes that the deviation of values attributed to the stimuli and the index of category are constants and that the correlation coefficients between the values attributed to the stimuli and those attributed to the index of category are identical. From here the system of equations is simplified and it becomes: Cg - Sj = YJg We fix k = 1 because we can choose arbitrarily the unit with which we work, and we now have Cg - Sj =. Yjg. We now look for the estimates of the values of the scale of n stimuli (that is to say, in our example 12 attributes of university restauration service) and M index of categories that minimize the sum of squares of the distances between the observed Yjg and the adjusted Yjg We can therefore locate the category indices and calculate the scale values of the stimuli obtained with the help of the following formulae. Ck = 1 Σn Yjk n j=l The results are as follows: Thurstone Scale Analysis Unidimensional (Thurstone case 5 model Importance of Attributes A. Much Importance Price.........................0.814 Nutrition....................0.804 Rapid Service..............0.660 B. Average Importance Respond to need of Appropriate hours......0.240 C. Low Importance Clean Environment.....0.000 Location...................0.2% Likert Scale Analysis Multidimensional (Factor Analysis) 1st Factor: Explains 47.43% of variance Σn Yjg + 1 Σn Σn Yjg g=l m+n g= 1 j=l These value were calculated for the example considered and are presented in figure 3 which has enabled us position each of the attributes of the restauration service with reference to the criteria: significance and importance. S1 = 1 m Comparative Analysis of the result of the Two Scales of Measurement The two approaches for measuring affect, preference and identification of attribute that are characteristic of student attributes regarding university restauaration services give almost the same result in spite of differences in the methods. Variables Factor Score price 0.76607 Rapid Service 0.75880 Quality/Price 0.74129 2nd Factor: Explains 27.71% of variance Clientele.......0.593 Appropriate hours 0 68078 Nutritious food 0.47929 3rd factor: Explains 21.2% of variance Meeting 0.68160 Arrangement 0.56024 Courteous & Experienced Personnel 0.48417 53 Akungba journal of Management The results of our analysis show that the multidimensional and undimensional approaches to measuring students' attitude toward university restauration service emphasize to a large extent, subjective characteristics such as economy and rapidity of service. These variables are contained in the first factor on the multidimensional scale and on A (attributes with much importance) on the undimensional scale. (See figures 1 & 3). We have therefore show in this paper that the undimensional approach to attitude measurement could complement the more recent multidimensional scale of measurement. It is therefore not circumspect to have abandoned urridimen-sional scale of attitude measurement. Equally important is the fact that by using well known models such as Thurstone law of categorical judgement, we have been able to develop an algorithm that facilitates the passage from a non-metric to a metric scale. This passage is particularly rich in complementary information. We should however, not be too carried away by the spectacular result of the undimensional scaling measures because new trends in attitudinal research are towards multidimensionality. multivariate, non metric analysis, more complete models of the decision process and more efficient means for testing hypotheses and developing new models. What we need are model are more than single attitudinal equations, complete systems of equations that will buying behaviors to demographic, economical and psychological variables. Also needed models that reflect the marry dimensions of hour. At the present time, behaviour is regard isolated forms - buying, visiting a dealer, advertisements, talking to friends which an actually related forms of behaviour. Further behaviour is defined in undimensional terms more powerful insights will be possible when attitudes can be mapped into multidimensional haviour. In conclusion, what we have done in paper is to apply undimensional and multidimensional scaling devices to describe or diagnosed needs of the market for the restauration very for university students. Thus the main criticizing attitude scale, their inability to predict behaviour may seem to be misplaced here inability to stems less from the limitations of scaling devices but more from the role of attitudes in the decision process. We do not have models that describe adequately the relationships among changes in behaviour attitudes, group membership, roles and systems. Thus, this criticism applies to the lack models rather than measurement scale. Figure3. IMPORTANCE OF ATTRIBUTES (Thurstone Case 5 Model) THE LAW OF COMPARATIVE JUDGEMENT Much importance 0.814 Price 0.804 Nutrition 0.660 Rapid Service 0.64 0.593 Respond to the need of the client Average Importance 0.240 Appropriate opening hours 0.01 0.00 Clean Environment Little Importance -0.2% Appropriate location -0.456 Adequate space and arrangement of chairs and tables. 0.62 -0.685 Experienced and courteous personnel No importance -0.845 Supplementary service (e.g. oven extra food -1.098 Useful Information. etc). Akungba journal of Management BIBLIOGRAPHY ANDREASEN, A. R.: Attitudes and Customer Behaviour. A decision Model, in L. E. Prestoa ed. New Research in Marketing. Berkley: Institute of Business and Economic Research. University of California. 1965 BEM. J.D.. Beliefs. Attitudes, and Human Affairs. Behnont. California. Wadsworth 1970. DAY, G. S. Buyer Attitude and Brand Choice Behaviour. New York: The Free Press 1970. EDWARDS, A. L. Techniques of Attitude Scale Construction. New York: AppletonCentury- Crofts. 1957. FISHBIEN, M.,: A Behaviour Theory Approach to the Relations between Beliefs about an Object and the Altitude toward the 54 Object. Fishbein. M, ed Readings in Attitude Theory and Measurement, New York: Wiley. 1967. GREEN. P. & TULL, D.: Research for Marketing Decisions. Prentice-Hall, 1970. GREEN. P. E. & CARMONE, F. J.: Multidimensional Scaling and related Techniques. Boston: A. Ally & Bacon 1970. GUTTMAN. L.. The problem of Attitude and Opinion Measurement, in S.A. Stouffer ed.. Measurement and Prediction; Princeton University Press 1950. HUGHES. CD: Attitude Measurement for Marketing Strategies: Glenview, III, Scott. Foresman 1971. Akungba journal of Management 55 56 Akungba journal of Management TABLE 2 Observed ratios in cases where attributes (=quality of service) X is indicated in A and attribute Y is shown in the first column starting from A B C D E F G H 1 J K L A 000 -1.65 -1.65 4>.67 -0.30 -0.73 0.21 0.39 -1.19 -2.12 -0.78 -1.82 B 1.65 000 0.62 1.04 1.04 0.28 1.28 1.19 0.17 -0.62 048 41.30 C 1.65 -0.62 000 0.90 1.3S -1.84 1.13 1.04 -0.30 •0.93 1.50 -0.73 D 0.67 -1.04 -0.90 000 0.57 -1.11 0.57 0.73 -0.73 -1.50 , -0.34 -1.28 E 0.30 -1.04 -1.39 -0.57 000 -2.12 043 0.39 -0.95 -1.50 -0.78 -1.28 F 0.73 -0.28 1.04 1.11 2.12 000 2.12 1.84 0.30 -0.34 0.53 4.43 G -0.21 -1.28 -1.13 -0.57 •0.43 -2.12 000 •0.13 -1.04 -1.84 0.78 -0.65 H -0.39 -1.19 -1.04 -0.73 -0.39 -1.84 0.13 000 -1.39 -1.39 -1.28 -1.50 1 1.19 -0.17 0.30 -0.73 0.95 •0.30 1.04 1.39 000 -0.67 0.30 4.73 J 2.12 0.62 0.95 1.50 1.50 0.34 1.84 1.39 0.67 000 1.11 0.30 K 8.78 •043 -1.50 0.34 0.73 -0.53 0.78 1.28 -0.30 -1.11 000 -1.50 L 1.28 0.30 0.73 1.28 1.28 0.43 1.65 1.50 0.73 0.30 1.50 000 SOMME 9.77 -6.78 -3.97 4.36 846 8.74 11.19 11.01 -4.14 12.34 146 -10.38 Z 0.814 -0.565 -0.331 0.363 0.705 0.728 .933 0.918 -0.345 -1.028 0.122 -0.865 R 0.692 -0.687 -0453 0.241 0.583 0.606 0.811 0.796 -0467 -1.15 0.00 -0.987 TABLE 3 Value related to Preference Proportions as indicated in the table A B C D E F G H I J K L A 0.000 0305 0.505 0350 0.383 0.233 0383 0.650 0.117 0.017 0.217 0.100 B 0.950 0300 0.733 0350 0350 0.617 0300 0.883 0.567 4.267 0.667 .0.383 C 0350 0.267 0300 0317 0317 0.150 0367 0350 0.677 0.167 0333 0.233 D 0.750 0.150 0.183 0.000 0.717 0.133 0.717 0.783 0.233 0367 0367 0.100 E 0317 0.150 0.083 0.283 0.000 0.017 0.667 0350 0.167 0.867 0.233 0.100 F 0.767 0323 0350 0367 0383 0.000 0383 0367 0317 0307 0.700 0333 G 0417 0.100 0.133 0.283 0333 0.017 0300 0460 0.150 0333 0.217 0360 H 0360 0.117 0.150 0.233 0350 0.033 0.550 0.000 03*3 0.0*3 0.100 0357 I 0383 0433 0317 0.767 0313 0.383 0350 0317 0.000 0320 0.617 0.233 J 0383 0.733 0333 0333 0333 0.633 0367 0317 0.7S0 0333 0357 0353 K 0.783 0333 0367 0333 0.783 0300 0.743 0300 0373 0.133 0.050 0367 L 0300 0.617 0.767 0300 0300 0.667 0350 0.883 0.767 0450 0333 0.000