Industrial 'Buyclasses' - Revisited

ABSTRACT - Analysis of organizational buying tasks indicates that the generally accepted 'buy lass ' classification scheme only partially explains the complexity of the buying casks. The primary characteristics used to develop the 'buyclasses' - newness of the problem, informational requirements and consideration of new alternatives - are shown to be only one factor in identifying 'buyclasses.'

Citation:

Ronald P. LeBlanc (1983) ,"Industrial 'Buyclasses' - Revisited", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 233-237.

Advances in Consumer Research Volume 10, 1983      Pages 233-237

INDUSTRIAL 'BUYCLASSES' - REVISITED

Ronald P. LeBlanc, Idaho State University

[This research was partially supported by Grant No. 472 from the Faculty Research Committee, Idaho State University, Pocatello, Idaho.]

ABSTRACT -

Analysis of organizational buying tasks indicates that the generally accepted 'buy lass ' classification scheme only partially explains the complexity of the buying casks. The primary characteristics used to develop the 'buyclasses' - newness of the problem, informational requirements and consideration of new alternatives - are shown to be only one factor in identifying 'buyclasses.'

INTRODUCTION

The purchase decision making process of organizational buyers is generally recognized to be influenced by the complexity of the buying task that has to be dealt with by the buyer (Kotler 1980, Webster and Wind 1979). The taxonomy which is used to describe buying tasks which vary in their levels of complexity was developed by Robinson and Faris (Robinson et al 1967). Their 'buyclasses' - new task, modified rebuy and straight rebuy describe buying situations whose complexity is a function of the newness of the problem, informational requirements and the consideration of new sources of supply.

PROBLEM

Although the 'buyclass' taxonomy is generally accepted and provides a parsimonious classification of organizational buying tasks, the 'buyclasses' and their underlying characteristics have not been validated. A review of Robinson and Faris's (Robinson et-al 1967) work reveals that the definitions of the 'buyclasses' and the underlying characteristics of the tasks were deduced from field interviews of an unspecific nature and that the classification scheme evolved from a refining process which is not detailed. This subjective approach to the development of what have evolved as key concepts in the understanding of organizational purchase decision making generates two questions.

The first question deals with the three primary characteristics used to describe the 'buyclasses' - newness of the problem, informational requirements, and consideration of new alternatives. Robinson and Faris claim that "every buying situation, and each basic type of buying situation, can be characterized according to..." these primary characteristics (Robinson et al 1967 p. 23-24), and that newness of the problem "is alone sufficient to differentiate among the three types of buying situations (Robinson et al 1967 p. 25)." The question is do these primary characteristics sufficiently differentiate the buying tasks faced by organizational buyers?

The second question is related to the characteristics which accurately differentiate the buying tasks but extends the issue to an analysis of the underlying dimensions of the buying tasks. The question is; if a set of variables accurately differentiates the buying tasks, what characteristics of the buying tasks are being measured?

The purpose of this research project is to investigate the buying task classification scheme to identify the characteristics which best characterize the buying tasks and explore the underlying dimensionality of the characteristics which differentiate the buying tasks.

METHOD

A field study was conducted to gather the information needed to address the questions raised. The subjects were purchasing agents actively engaged in supplier selections. Each purchasing agent was asked to pick from their work-in-process file a purchase on which they were presently working, in which two or more suppliers were being considered. For each of these array-and-review decisions the subjects were asked to respond to a series of questions designed to identify the buying tasks and the tasks' underlying characteristics.

The sample in this field study consisted of one hundred and thirty-five buyers from seventy-eight organizations. One hundred and twenty-five of the buyers were staff specialists employed as purchasing agents/buyers; the remainder were organizational members who were assigned the purchasing function as a secondary or tertiary duty. Because the data were collected via personal interviews, a probability sampling technique was not feasible. The convenience sample taken was, however, a diversified sample of organizational buyers. which minimized the systematic bias which might have occurred if the sample had been taken from a more homogenous group of organizations. The seventy-eight organizations represented high technology national firms, medium sized and small local firms. as well as governmental agencies.

The questionnaire used in the study was developed after a series of preliminary interactions with buyers, to pretest the ability of the buyers to respond to the questions and to assess the amount of time it would take them to respond to the entire questionnaire. A structured protocol was used to identify the buying tasks. It was designed by using the constitutive definitions of the buying tasks as presented by Robinson and Faris (Robinson et al 1967). The protocol is presented in Table 1.

Measurement of the characteristics of the buying tasks was done by the use of direct and Likert scaled questions. Twenty-one variables were included in the study, Table 2. The Likert scale used in this study has five points. Five points were used because the pretest of the questionnaire revealed that the buyers had a difficult time making finer discriminations. Additionally, the five point scale allowed the buyers to respond to the questionnaire in a time period which was acceptable to management.

Analysis of the data involved several steps. The first Step was the use of multiple discriminant analysis to derive a set of variables which significantly differentiated between the three buying tasks. The second step was an analysis of the derived discriminant functions to determine the relative importance of the variables in explaining the between group differences. The final step in the data analysis was an examination of the underlying dimensionality of the variables included in the derived discriminant functions via factor analysis.

TABLE 1

STRUCTURED PROTOCOLS

TABLE 2

VARIABLES MEASURED

FINDINGS

The derived discriminant functions with their standardized coefficients are presented in Table 3.

The ability to interpret the derived discriminant functions and use the discriminant weights to determine the relative importance of the independent variables is dependent on two considerations; the first being the statistical significance of the functions. The second consideration is the ability of the derived function, if significant, to predict group membership at a level greater than one would expect by chance.

The computational package used in the data analysis, SPSS (Nie et al 1975) provided a conversion of the Wilk's lambda statistic into a chi-square statistic for an easy test of the statistical significance. As reported in Table 3, the derived discriminant functions are statistically significant. The statistical significance of the function is, however, only the first stage in determining if the discriminant functions are meaningful. As a result of the relatively large sample taken small differences in the group centroids can produce statistically significant findings which are relatively meaningless. To check this, the functions must be able to classify the cases at a level which is greater than what one would expect by chance.

TABLE 3

DISCRIMINANT FUNCTIONS

The prediction results are presented in Table 4. The actual group listings represent the responses to the structured protocols. The only problem encountered by the respondents with the selection of a structured protocol appropriate to a specific purchase being investigated was with the new task description. The buyers commented that although they were purchasing a product that they had not purchased before, the purchase was being made from a set of suppliers that they had been buying from for some period of time. This may, in part, explain the relatively low classification accuracy for the new task buying situation.

TABLE 4

PREDICTION RESULTS

The overall classification accuracy of the discriminant functions is 76.30 per cent. The check used on the classification accuracy is the ability of the derived functions to correctly classify membership in each group. This check is done by using the proportional chance criterion which is calculated by the following general formula:

C_p = p² + (1 - p2).   (1)

In this case;

C_p = .19² + .42² + .39²    (2)

C_p = .36   (3)

With the percent of correctly classified cases at 76.30 percent and the proportional chance criterion at 36.0 percent, the derived discriminant function is able to correctly classify the buying tasks at a level which is significantly greater than chance.

With this information the importance of the variables in the discriminant function can be meaningfully applied. The data indicate that consideration of new suppliers is the most important variable in function 1. In function 2, the routineness of the purchase accounts for most of the intergroup differences. In addition to these two variables which account for the greatest portion of the intergroup differences in the average score profiles of the three groups of buying classes, seven other variables are included in the discriminant functions. These seven variables include the size of the evoked set, size of the awareness set, total dollar cost of the purchase, item cost, product experience, information sought and a defined set of attributes for the specific purchase.

The next stage of the study was an analysis of these nine variables to determine the underlying dimensionality of the two discriminant functions. For this factor analysis was used in a confirmatory manner to test the underlying dimensionality of the discriminant functions. This analysis should identify a three factor solution, with the factors closely associated with the primary characteristics described by Robinson and Faris (Robinson et al 1913) - newness of the problem, informational requirements, and consideration of new sources of supply.

To examine the linear independence of the nine variables included in the discriminant functions, the principle factor model with iteration was employed to develop a preliminary unrotated factor solution. The initial solution was orthogonally rotated, using the varimax criterion, to yield a multiple factor solution. Table 5 shows the varimax rotated principle factor matrix.

TABLE 5

VARIMAX ROTATED FACTOR MATRIX

Examination of Table 5 shows enat four variables load on Factor 1. This group includes product experience, informational requirements, routineness of the purchase and consideration of new suppliers. Factor 2 loads on two variables associated with the cost of the purchase total dollar cost and item cost. Factor 3 loads on the size of the evoked set and the existence of a set of defined attributes. The last factor loads distinctly on the size of the awareness set.

This four factor solution appears to be a solution which now only needs to be interpreted. The four factors have one or more of the variables which load separately and significantly. Additionally, the variance explained by each of the factors suggests that the four factor solution is appropriate. As is typical of the principle factor model, the initial factor in the solution accounts for the largest proportion of the variance while the subsequent factors account for successively less of the remaining variance. In this case, Factor 1 accounts for 54.8 percent of the total variance, Factor 2 accounts for 21.6 percent, Factor 3 accounts for 12.4 percent, while Factor 4 accounts for 11.1 percent of the total variance. This relatively high level of explained variance for the fourth factor suggests that the four factor solution is appropriate.

The four factor solution is also suggested by the eigenvalues of the nine variables. In Table 6 it can be seen that the first four variables have eigenvalues which are greater than-one. This also suggests that a four factor solution is appropriate.

TABLE 6

The four factor solution however, is subject to the more basic question of the appropriateness of the use of a factor analysis on the data set. If the underlying data matrix does not contain data which is appropriate for the application of factor analysis then the resultant solution, no matter how interpretable on face value, is meaningless. An initial test for the appropriateness of the data set for a factor analysis solution is Bartlett's test of sphericity (Bartlett 1950, 1951) which tests the independence or variables in the data matrix and is computed by the following formula:

-[ (N-1) - (2P + 5/6)] log_e |R|   (4)

where

N = sample size

P = number of variables

|R| = determinant of the correlation matrix

The statistic is approximately distributed as a chi-square statistic with the degrees of freedom determined by 1/2P(P-1). The hypothesis tested is the independence of the variables in the correlation matrix. Rejection of the tested hypothesis is an indication that the data set is appropriate for factor analysis. In this case the calculated statistic is 295.385 with 26 degrees of freedom. This allows rejection of the tested hypothesis and suggests that the data are suitable for factor analysis. However, the test is not conclusive. Rejection of the independence hypothesis is a necessary but not sufficient test for the appropriateness or using factor analysis on the data set. The relatively large sample size and the low number of variables biases the statistic toward significance (Knapp and Swoyer 1967). Therefore a second measure was taken.

The second statistic calculated to test the appropriateness of the correlation matrix for a factor analysis solution was the MSA statistic developed by Kaiser, Olkin and Meyer (Kaiser 1970). MSA is a measure of sampling adequacy and is defined by the following formula:

EQUATION    (5)

where

q²ik = square of the off-diagonal elements of the anti-image correlation matrix

r²ik = square of the off-diagonal elements of the original correlation matrix

The anti-image correlation matrix is defined by SR-1S. R-1 is the inverse of the original correlation matrix and S is defined as (diagonal of R-1)-1/2. In this case MSA = .81; which suggests that the data are suitable for the application of factor analysis in light of Kaiser's statement that "we don't have good factor analytic data until MSA gets to be at least in the .80's (Kaiser 1970 P. 405)."

The four-factor, factor analysis solution of the derived discriminant function contradicts the suggestions Robinson and Faris (Robinson et al 1967) that the 'buyclasses' can be defined along three primary characteristics. We shall defer further comment on the interpretation of the findings. First, a summary of the findings will be presented.

SUMMARY OF FINDINGS

1) The structured protocols were easily responded to by the organizational buyers, who identified 26 new tasks, 52 modified rebuys, and 57 straight rebuy buying situations as defined by Robinson and Faris. 2) The derived discriminant functions contained nine of the twenty-one variables included in the study. These discriminant functions are statistically significant and able to classify 76.30 percent of the cases identified via the structured protocol. 3) The factor analysis of the derived discriminant functions indicates a four factor solution, because each or the nine variables load distinctly on one factor.

DISCUSSION OF FINDINGS

It appears that the constitutive definitions of the "buyclasses' of Robinson and Faris (Robinson et al 1967) are useful, in general terms, in describing buying situations faced by organizational buyers. However, there is a discrepancy between the findings of this study and the set of primary characteristics used by the authors to delineate the 'buyclasses.'

Analysis of the factor anaLytic solution of the discriminant functions indicates that the three primary characteristics used by Robinson and Faris - newness of the problem, informational requirements, and consideration of new sources of supply - are all included in the first factor. This factor contains four variables - product experience, informational requirements, routineness of the purchase, and consideration of new sources of supply - which could be identified as a factor associated With the newness of the problem label used by Robinson and Faris. This factor, with its four associated variables, accounts for 54.8 percent of the explained variance.

The remaining three factors contain variables which were not included in the original description or the 'buyclasses.' Factor 2 contains two cost related variables - total cost of the purchase and item cost. Factor , also contains two variables - size of the evoked set and a priori set of defined supplier attributes. Factor 4 is defined by only one variable - size of the awareness set.

These five variables were included in the Study because the selection of an evoked set of suppliers is viewed as a decision making process. A review of the literature on consumer decision making, from an information processing perspective, suggests three major influences on consumer decision making strategies - familiarity with the Product or product class, information load, and risk of making errors (Park 1976, Wright 1972, 1974, 1975). The inclusion of these variables in the discriminant function and four factor solution of this function suggests that the basic buying tasks of the organizational buyer are more complex than is suggested by the 'buyclass' taxonomy.

The variables in Factor 2 are cost related and can be interpreted as representing a risk factor. Organizational buying is recognized to be a group decision making process, (Wind 1971, Speckman and Stern 1979) but the role of the professional buyer in the group gives the buyer the relative social power associated with supplier selection (Cooley et al 1573). Cost considerations can then be associated with the risk of making supplier selections, i.e., risk to the buyer's relative social position in the buying group.

The remaining. two factors prove to be more difficult to interpret. Factor 3 includes the size of the evoked set and the buyer's defined set of supplier attributes. These variables can be interpreted as information requirements of the buying task, information specific to the task. The fourth factor, awareness set size, seems to be an information related variable also; however, this information related variable seems to be associated with known alternatives.

This analysis of the underlying dimensionality of the discriminant functions provides an insight into the complexity of the underlying characteristics of the buying situations faced by an organizational buyer but does not address the issue of the importance of the variables in discriminating between the buying situations. Robinson and Faris (Robinson et al 1957) claimed that newness of the problem was sufficient to differentiate between the buying situations. This claim can be substantiated or assailed depending on the interpretation of the discriminant functions.

Analysis of the derived discriminant functions shows that the variable in Function 1 which provides the greatest discriminating power is consideration of new suppliers. For function 2, routineness of the purchase provides the greatest discriminating power. Both of these variables are contained in Factor 1. This indicates that although these individual variables clearly have discriminating ability between the buying tasks, caution must be used in utilizing the discriminant weights to interpret the relative contribution of these variables. Discriminant weights are interpreted in a fashion similar to that of regression coefficients and are subject to the same criticism's. Small discriminant weights may either indicate that the variable is irrelevant or that the variable has been partialed out of the relationship because of a high degree of multicollinearity.

In this case, given the results of the factor analysis, these two primary characteristics, consideration of new suppliers and routineness, are interpreted as belonging to the same underlying factor. This factor is seen as being analogous to the primary characteristic described by Robinson and Faris (Robinson et al 1967) as newness of the problem.

The predictive accuracy of the discriminant functions is quite impressive. Even the relatively low classification accuracy of the new task buying situations, a 53.82 percent level, is statistically significant when compared to the chance level of 36.0 percent. The inclusion of variables in the discriminant function which were not considered in the original description of the 'buyclasses,' does however, suggest that the buying tasks are more complicated than is suggested by the description of the 'buyclasses.'

CONCLUSION

Although this research project may provide additional insights into the buying situations faced by organizational buyers, further investigation is needed to fully understand the many buying tasks faced by organizational buyers. Additional research can also overcome some of the shortcomings of this research project. Only buyers were interviewed in this project, further research should include other members of the buying group. Future research would also allow for a more refined set of variables to be used in the study to provide a better understanding of the organizational decision making process.

REFERENCES

Bartlett, M. S. (1950), "Tests of Significance in Factor Analysis," British Journal of Statistical Psychology, 3 (January), 77-85.

Bartlett, M. S. (1951), "A Further Note on Tests of Significance in Factor Analysis," British Journal of Statistical Psychology, 4 (January), 1-2.

Cooley, J. R., D. W. Jackson Jr., and L. L. Ostrom (1978) "Relative Power In Industrial Buying Decisions," Journal of Purchasing & Materials Management, Spring, 18-20.

Kaiser, H. F. (1970), "A Second Generation Little Jiffy," Psychometrika, 35 (December) 401-15.

Knapp, T. R. and V. H. Swoyer (1976), "Some Empirical Results Concerning the Power of Bartlett's Test of Significance of a Correlation Matrix," American Educational Research Journal, 4 (Winter), 13-17.

Kotler, P. (1980), Marketing Management, 4th ed., Prentice-Hall, Inc., Englewood Cliffs, New Jersey.

Nie, N. H., C. H.. Hull, J. G. Jenkins, S. Stienbrenner, D. H. Bent (1975), Statistical Package for the Social Sciences, 2nd ed., McGraw-Hill. New York.

Park, C. W. (1976), "The Effect of Individual and Situation-Related Factors on Consumer Selections of Judgmental Models," Journal of Marketing Research, 13 May, 144-151.

Robinson, P. J., P. Wright and Y. Wind (1967), Industrial Buying and Creative Marketing, Allyn and Bacon. Boston.

Speckman, R. E. and L. W. Stern (1979), "Environmental Uncertainty and Buying Group Structure: An Empirical Investigation,' Journal of Marketing Research, 43 Spring, 54-64.

Webster, F. E. Jr. and Y. Wind (1979), Industrial Marketing Strategy, John Wiley and Sons, Inc., Ronald Press, New York.

Wind, Y. (1971), "A Reward-Balance Model of Buying Behavior in Organizations," New Essays in Marketing Theory, G. Fisk, ed., Allyn and Bacon, Boston, 206-217.

Wright, P. (1972), "Consumer Judgement Strategies Beyond the Compensatory Assumption," Proceedings of the Third Annual Conference of the Association for Consumer Research, Chicago. 316-324.

Wright, P. (1975), "The Harassed Decision Maker: Time Pressure, Distraction, and the Use of Evidence." Journal of Applied Psychology, 59 October, 555-561.

Wright, P. (1975), "Consumer Choice Strategies, Simplifying vs. Optimizing," Journal of Marketing Research, 12 February, 60-67.

----------------------------------------