A Study of Consumer Perceptions of Decisions

Brian T. Ratchford, State University of New York at Buffalo
Alan A. Andreasen, State University of New York at Buffalo
[ to cite ]:
Brian T. Ratchford and Alan A. Andreasen (1974) ,"A Study of Consumer Perceptions of Decisions", in NA - Advances in Consumer Research Volume 01, eds. Scott Ward and Peter Wright, Ann Abor, MI : Association for Consumer Research, Pages: 334-345.

Advances in Consumer Research Volume 1, 1974    Pages 334-345

A STUDY OF CONSUMER PERCEPTIONS OF DECISIONS

Brian T. Ratchford, State University of New York at Buffalo

Alan A. Andreasen, State University of New York at Buffalo

[Brian T. Ratchford is Assistant Professor of Marketing and Alan R. Andreasen is Professor of Marketing and Environmental Analysis and Policy at the State University of New York at Buffalo.]

In a recent study of consumer information seeking behavior among new residents by the authors (Andreasen and Ratchford, 1973), it was necessary to develop a conceptual model that would predict how that behavior ought to vary across different kinds of decisions. A model comprising four dimensions along which decisions were expected to vary and which was expected to predict variations in quantity and type of information gathered was constructed. Although ad hoc in character, the model was found to be quite useful in explaining the data. For example, a multiplicative form of the model predicted the rank order of breadth of information seeking for husband-wife decisions quite well (Spearman's rho = .83, significant at the .05 level).

Despite this predictive power, the model, although plausible and easy to administer, was still ad hoc in character and had not been subject to any independent test of its validity. The purpose of the exploratory study reported here is to provide such a test and to investigate generally the problem of how consumers perceive decisions.

PAST APPROACHES

There have been two basic approaches to the problem of building models of how decisions vary in character, the deductive and the inductive. The deductive approach involves the construction of some armchair theory about the characteristics of decisions that ought to affect the process by which decisions are to be made. These sets of characteristics may be simply in the form of a checklist of relevant factors as in Engel, Kollat, and Blackwell's specification of four sets of variables (situational variables, product characteristics, consumer characteristics and environmental factors) which ought to predict whether a given decision will involve (a) extensive, (b) limited, or (c) habitual "decision process behavior" (Engel, Kollat and Blackwell, 1968). Alternatively, factors expected to influence decision processes may be more carefully integrated as in the convenience-shopping-specialty goods model (Bucklin, 1963; Holton, 1958), Leo Aspinwall's (1962) color theory, and Robert Settle's recent ad hoc four-factor classification scheme (1972). It was, of course, the deductive approach that was used to construct the Andreasen-Ratchford model.

The second basic approach to this problem, the inductive, compares actual behavioral output across decisions for evidence of similarities that will eventually lead to broader and broader groupings of decisions. A recent example of this approach is the attempt of various authors to make sense of empirical tests of the Fishbein-Rosenberg models (Cohen, Fishbein and Ahtola, 1972; Day 1972).

Some Deficiencies

Three deficiencies appear in these past models. The first is that they focus almost exclusively upon product decisions It is thus difficult to use several of the theories to classify service decisions. For example, it is difficult to conceive of how one might measure objectively the "multipurposeness" of a haircut (Settle, 1972, p. 85) or the "replacement rate" for a motel room (Aspinwall, 1962, p. 637). The one past model that does appear useful in classifying service decisions as well as product decisions is the Engel, Kollat, and Blackwell model which, however, suffers from a second difficulty, complexity. The fact that it contains nineteen variables makes it potentially quite cumbersome for managerial and empirical research purposes. The one model that on its face meets the dual criteria of generality and parsimony at the moment is the Andreasen-Ratchford model.

A final deficiency of all the models to date, including the Andreasen-Ratchford model, is that they have not been independently evaluated. Specifically, no one to date has sought to ascertain whether consumers tend to group decisions in the same way using the same dimensions as have the authors of the various existing theories. It is the major purpose of this paper to make such an evaluation with respect to the Andreasen-Ratchford model.

THE ANDREASEN-RATCHFORD MODEL

[This secion is drawn from Andreasen and Ratchford (1973).]

The breadth of information actually sought for any given decision is a function of the demand for information across respondents and the supply of information typically available for that type of decision. The demand for information across decisions may be expected to vary with the following three factors:

1. Importance of the Decision. Ordinarily, a respondent's perception of the importance of a decision will vary directly with the perceived opportunity cost of choosing an inferior alternative. For example, the choice of a pediatrician will probably be perceived as highly important because a poor choice could result in the death or permanent damage of a child. On the other hand, the selection of a hairdresser will probably be seen by most as a relatively unimportant choice because of the relatively small financial outlay involved and because a poor choice will only result in temporary damage to a hairdo. All things being equal, the demand for information increases with the importance of a decision and thus we should expect the breadth of information seeking to increase with perceived importance.

2. Complexity of the Decision. Some decisions, such as the choice of a bank or hairdresser, require relatively few bits of information (principally location and estimated quality of service) while other decisions, such as the choice of an appliance, require many more bits (product characteristics, product and brand availability, location, price, store services, credit availability, etc.). All things being equal, one would expect a greater breadth of information seeking for more complex than for simpler decisions.

3. Subjectivity of Needed Information. For a given level of importance and complexity, decisions may vary in the extent to which the needed bits of information are objective or subjective. Some bits, such as the location of a store, are essentially objective in character and can be obtained by checking only one source, such as a telephone book. Other bits may require considerable rechecking since they pertain either to personal preferences (who gives the "best" permanents) or to uncertain future outcomes (which brand of appliance requires least repairs). The fewer the objective information bits, the greater the breadth of information seeking

Finally, breadth of information seeking will vary inversely with the supply of information. The supply of information is conceived as:

4. Availability of Information. Although decisions vary in the quantity and type of information needed for their resolution, custom and market practices limit the amount of information available for some decisions. For example, it is not typical for consumers to visit a number of doctors or repair shops before making choices. Further, doctors do not advertise on radio or TV or in newspapers, and repair shops seldom do. These market characteristics reduce the availability of information for these decisions as compared to the information available for choosing banks or appliances. This lack of information increases the cost of information seeking, and therefore should lead to less search, all things being equal.

Methodology

Data to test this model were developed for seven types of decisions that a housewife might make after moving to a new community. These decisions, which were the focus of analysis in the earlier Andreasen-Ratchford paper, contained both service and product decisions that were of high salience for new movers. The decisions are:

1. Selection of a bank for a checking account.

2. Purchase of household furniture.

3. Purchase of a major appliance.

4. Selection of an outlet or individual for repairs of an automobile, appliance, or household plumbing.

5. Selection of a hairdresser.

6. Selection of a general practitioner.

7. Selection of a pediatrician.

Data measuring dimensions of the basic model were generated from a convenience sample of 67 adult females. The sample was asked to rate the seven decisions on the following nine-point scales:

Very Important - Very Unimportant

Very Complicated - Very Uncomplicated

Very Subjective - Very Non-subjective

Information Widely Available - Information Widely Unavailable

In an effort to control for the effects of past experience, the respondents were asked to imagine that they had just moved to a new community where they had never lived before and to rate the decisions on the basis of how important, complex, etc., the decision would be during the first few weeks there.

Mean scores across all respondents for each dimension-decision combination are presented in Table 1. The Table shows that the general practitioner and pediatrician decisions are perceived as the most important and complex of the decisions, but as having relatively little available information. The furniture decision, like the appliance decision, is perceived as medium in importance and complexity. The furniture decision is seen to be relatively subjective in nature with relatively high information availability. Bank and repair decisions are both seen to be relatively important but uncomplicated and requiring mostly objective information. The major difference between these two decisions is that a great deal of information appears to be available about banks while little is available about repairs. Finally, the respondents perceived the hairdresser decision to be very unimportant and uncomplicated, but to be highly subjective and to have a medium amount of information availability.

TABLE 1

MEAN SCORES OF SELECTED CONSUMER DECISIONS ON DECISION-RELATED CHARACTERISTICS

To test whether there is a significant difference between decisions on the four dimensions outlined above both multivariate and univariate analysis of variance tests were applied. The multivariate analysis of variance (Cooley and Lohnes, 1971) tests whether there is a significant difference between the seven four-dimensional vectors of average scores presented in Table 1, i.e., whether the four dimensions employed in this study taken as a group are able to discriminate significantly among the seven decisions This test is based on an F-ratio approximation derived from Wilk's Lambda. [The formula for this approximation which is quite involved is presented in Cooley and Lohnes (1971, p. 227). Wilk's Lambda is defined as: A = |W| / |T| where W is a matrix of sums of squares and cross products of deviations from group means for the four variables, summed over all seven groups, and T is a matrix of sums of squares and cross products of deviations from grand means. Since F is a function of (1 - A)/A, the F-ratio varies inversely with A. Also, analogous to the univariate case, W represents within sum of squares, and (1 - A)/A represents a ratio of between to total sum of squares.] The F-ratios for both the multivariate ANOVA and the normal univariate tests for significance of the difference between decision categories on each dimension independently of the others are presented at the bottom of Table 1. In all cases, the differences in scale values between decisions are highly significant, indicating that there is a significant difference in consumer perceptions of the seven decisions on the four dimensions outlined above.

To gain more insight into the dimensions underlying the perceived differences between the seven decision types, discriminant functions were computed on the four scales using the 67 scores on each decision as observations The results are presented in Table 2, The table indicates that while three significant discriminant functions were obtained, the first two of these accounted for approximately 87 percent of the total possible discriminatory power. As the factor loadings show, the first function is inversely related to the importance and complexity scales and positively related to the information scale. Thus, decisions with high scores on this function are characterized as important and complex but where there is little information to make them (obviously the general practitioner and pediatrician decisions). [Remember that low values on our scales indicate high importance, complexity, etc.] Decisions with high scores on the second function, on the other hand, are characterized as important, objective decisions, which are high in information availability. The third less significant function relates most closely to subjective decisions.

TABLE 2

DISCRIMINANT FUNCTION COEFFICIENTS AND FACTOR PATTERNS ON DECISION-RELATED CHARACTERISTICS

It is probable that the discriminant functions obtained here are highly dependent on the seven decisions studied (i e , two of the seven decisions happened to be perceived as important, complicated and lacking in information) and need not be representative of dimensions underlying perceptions of decisions in general. Nevertheless a useful descriptive device often used in multidimensional scaling studies (Green and Rao, 1972; Johnson, 1971) is to plot the group centroids of the discriminant functions. The group centroids on the first two functions in Table 2 are plotted in Figure 1. From Figure 1 it is apparent that the decisions studied fall into two distinct groups on Dimension I in the discriminant space, with Repair, General Practitioner and Pediatrician decisions falling into one group, and the remaining four decisions into the other. Possibly, differences in perception on this dimension relate to gaps between desired and available amounts of information. This gap is large for the General Practitioner and Pediatrician decisions because they are important and complex, but lack information. lt is not quite so large for the Repair decision, even though there is little information available, because this decision is not as important and complex (see Table 1). Though Bank, Furniture and Appliance decisions are relatively important, a great deal of information is also available, so the gap between desired and actual information is not so large. Finally, although the Hairdresser decision is relatively low in information availability, it is also seen as very unimportant and uncomplicated so that not much information is needed Thus, the gap between desired and available information is smallest for this decision.

FIGURE 1

DECISION CENTROIDS IN TWO DIMENSIONAL DISCRIMINANT SPACE

On Dimension II, all decisions except Bank, Appliance and Hairdresser are close to the origin. As the factor loadings in Table 2 suggest, the large difference in perception between Bank and Hairdresser decisions on this dimension apparently relates to differences in importance, objectivity, and information availability characteristics. While one might feel a priori that Appliance and Furniture decisions are quite similar, there is a relatively big difference in perception between these decisions on Dimension II which is apparently because the Appliance decision is seen as considerably more objective than the Furniture decision (see Table 1).

In summary, the Andreasen-Ratchford model appears able to discriminate well among decisions although it was possible to reduce the original four characteristics to a simpler two-dimensional configuration. A plot of centroids from this two-dimensional configuration in turn produces reasonably plausible spatial relationships among the original seven decisions.

An Alternative Approach

Since there may be other variables which determine perceptions of decisions, finding significant differences along the four variables employed above does not provide an adequate test of the hypothesis that these variables determine differences in perception since it is entirely possible that in relation to other variables, our four may be relatively unimportant. One way to determine whether our theory adequately accounts for differences in perception between decisions would be to compare the differences reported in Table 1 and Figure 1 with a direct measure of the similarity between the seven decisions obtained from a sample of consumers. Rather than asking consumers to rate the decisions along several predetermined dimensions, a more direct measure of similarity could be obtained from a ranking of each pair of decisions from most similar to least similar. This procedure has the advantage that judgments are allowed to take place along any dimensions which are important to the consumer, rather than only those dimensions which are predetermined by the researcher. Once the ranking of pairs of decisions has been obtained, a multidimensional scaling procedure may be employed to find coordinates for various decisions in a space of some predetermined dimensionality (Green and Tull, 1970) .

In this part of the study a subsample of 19 of the 67 respondents, prior to scaling the decisions along the predetermined four dimensions, was presented with each possible combination of three decisions and for each triplet asked to judge which two were most similar and which two were least similar. [This method of triads is described fully in Torgerson (1958). Time restrictions and the extreme tediousness of this task (each respondent must make 70 judgments) precluded us from obtaining the similarities judgments from our entire sample.] This procedure yielded 210 out of a possible 420 paired comparisons. Given these paired comparisons, a triangularization procedure described in detail by Green and Rao (1972, pp. 182-87) was employed to develop a ranking of each pair of decisions from most similar to least similar for each subject. These rankings were then averaged pair by pair across subjects (Green and Tull, 1972, p. 19) and the resulting average rankings were used as input to the TORSCA 8 multidimensional scaling program (Green and Tull, 1972, pp. 192-96) . This program attempts to scale the original stimuli (decisions) in a given number of dimensions so that the rank order of Euclidean distances [The algorithm may be modified to employ other non-Euclidean distance measures. See Green and Rao (1972, p. 193).] between pairs of points in the scaling solution best reproduces the ranking of pairs from most similar to least similar in the input data. The program also computes a measure of goodness of fit, J. Kruskal's Stress (1964), which is analogous to 1 - R2 in regression analysis. This measure may be used to compare solutions of varying dimensionality.

While the TORSCA solution in two dimensions yielded a noticeably lower measure of stress (.024) than the one-dimensional solution, higher dimensional solutions did not result in a much better fit. [A value of stress lower than .025 is considered an excellent fit (Kruskal, 1964). The minimum possible stress is O for a perfect fit.] Accordingly, the two-dimensional TORSCA solution derived from the average similarities rankings for the 19 respondents is reproduced in Figure 2. Except that the signs are reversed, the results in Figure 2 appear to be in general agreement with those in Figure 1 for all but the Hairdresser and Repair decisions. In the latter case, the relative positions on Axis I are reversed between Figure 2 and Figure 1. Also, in Figure 2,Axis 1 more clearly seems to have a relatively straightforward interpretation as a tangible product-service dimension. However, except for sign reversal, the second axis in both Figures appears to be generally similar. Thus, while there are some major differences between the mapping of consumer perceptions on the seven decisions based on similarities data and that based on the Andreasen-Ratchford model, there are also some apparent areas of agreement.

FIGURE 2

TWO-DIMENSIONAL SCALING OF DECISIONS DERIVED FROM SIMILARITIES DATA

One way to make a more rigorous comparison between the alternative solutions is to compare the interpoint distances based on each. Using Euclidean distance, the distance between any two objects i and j in r-dimensional space may be defined as:

EQUATION

where xik refers to coordinates of object i on dimension k. If two scaling solutions are similar except for transformation and rotation of axes, one would expect the dij for corresponding pairs ij computed for the alternative solutions to be highly correlated. To make this comparison, the measure dij was computed from the mean scores in Table 1, the discriminant scores in Figure 1, and the scale coordinates in Figure 2. [Some of the differences in results between the scales obtained from the similarities data and those from the ratings in the preceding sections could have been due to differences between the 19 individuals who completed the similarities rankings, and the 48 who did not. Accordingly separate mean scores and discriminant functions were calculated for both groups, and comparisons between mean scores, discriminant scores, and scaling solutions were run for the 19 individuals who made the similarities ratings. The results were virtually the same as those reported here.] For all 21 pairs of points the correlations of the distances obtained from the scaling coordinates with those obtained from the mean scores and discriminant scores were .31 and .35 respectively. Because of the large difference between the Repair and Hairdresser decisions in the two maps, these rather low correlations are not surprising. However, as comparison of Figures 1 and 2 would suggest, the correlations between the ten interpoint distances for the remaining five decisions were much higher. Between distances based on scaling coordinates and distances based on mean scores, this correlation was .74; between distances based on scaling coordinates and those based on discriminant scores, the correlation for the remaining ten points was .80. Both of these correlations were significant at beyond the .01 level.

To further investigate the correspondence between the scaling solution and the results of the discriminant analysis, correlations were run between the seven scores on the axes in Figure 2 and those in Figure 1. The results, which are presented in Table 3, confirm the observation made above that Axis 11 for both solutions are highly (negatively) correlated. They also indicate a substantial negative correlation between the first axis for both solutions. As stated above, this correlation is not higher because of differences between solutions.

TABLE 3

CORRELATIONS BETWEEN SCALING SOLUTION SCORES AND DISCRIMINANT SCORES ACROSS SEVEN DECISIONS

CONCLUSIONS

This study has sought to validate a general, parsimonious model comprising four major dimensions along which product and service decisions might reasonably be expected to vary. The model was found to be easy to administer to respondents, to be able to discriminate among types of decisions, and to be useful in explaining consumer information-seeking behavior. As a further test of the model's validity, we asked whether the relationships among the subject decisions generated by the model are essentially the same as that generated when consumers are asked to group decisions without the dimensionality of their judgments being predetermined. On the whole the evidence seems to support this contention reasonably well.

In support are the following findings:

1. Both procedures could be reduced to a two dimensional configuration.

2. Correlation of dimension I's in the two procedures was significant as was the correlation of the two dimension II's. Cross-correlations between the dimensions were low and non-significant.

3. Correlations of interpoint distances for five of the seven decisions were highly significant, i.e., the two procedures placed them in about the same relationship to each other.

Arguing against our contention was the lack of correspondence of results between the Repair and Hairdresser decisions in the two procedures. Since this lack of correspondence was primarily attributable to different configurational placements along dimension I under the two procedures and since the two dimension I's correlated less well than the two dimension IIs, it seems reasonable to conclude that we have succeeded rather less well in identifying one of the two dimensions consumers apparently use to classify decisions.

Implications

We have suggested here that it is possible to group decisions for purposes of predicting likely decision process behavior. We have suggested that it may be possible to do such grouping empirically with four simple scales that are quick to administer and easy for consumers to understand, research characteristics that are not all true of the more cumbersome similarities judgment approach.

We have also made a contribution in suggesting that hypotheses about dimensions which are important to consumers may be tested by comparing scaling solutions based on direct similarity-dissimilarity judgments. Close correspondence between solutions is evidence that most of the relevant dimensions have been captured by the theory.

Finally, we have left a number of questions open for ourselves and others to pursue in the future:

1. What further (or different) explanatory power would be gained by adding (or substituting) other decision characteristics for those specified in the Andreasen-Ratchford model (e.g., the 19 variables from the Engel, Kollat and Blackwell model)?

2. To what extent is the present solution unique to the methodology to which we were restricted? That is, would the present relationships hold if:

a. The decisions were varied, particularly if more trivial decisions, such as decisions about toothpaste or facial tissue, were introduced;

b. Respondents were not told to imagine themselves in some more or less hypothetical context as new movers, i.e., if they simply provided ratings and judgments for themselves under whatever circumstances they are currently in (which, of course, may have been just what they were doing).

3. To what extent and in what ways do these perceptions and relationships vary across individuals?

This appears to be a fruitful area for further research. While consumer researchers have in the past looked extensively at how decision processes vary across market stimuli and across individuals, finding out how these processes would vary across decisions may prove to be equally or in some cases more important.

1. It might prove more useful to practicing marketing strategists who now have difficulty using research findings generalizing across consumer characteristics A typical firm cannot tell a priori whether its market contains hypochondriacs or risk avoiders or people who are rich and like to visit libraries. Presumably they could more easily see the applicability of research generalizations about types of decisions they know they are trying to influence.

2. It might also prove useful in this era of social marketing as consumer researchers and theorists seek to determine how our knowledge about the profit-oriented business sector can be generalized to other areas. A fundamental question implicitly (and sometimes explicitly) asked in this whole area is: To what extent are decisions about political candidates, labor organizations, giving blood, and purchasing a television set or a brand of toothpaste essentially similar and in what ways. This undoubtedly is a question to which we as a field will increasingly turn our attention.

REFERENCES

Andreasen, Alan R. and Ratchford, Brian T. A study of information seeking by new residents. Working Paper No. 164, School of Management, State University of New York at Buffalo, June 1973.

Aspinwall, Leo V. The characteristics of goods theory. In Lazer, William and Kelley, Eugene J. (eds.), Managerial marketing perspectives and viewpoints. Homewood, Illinois: Richard D. Irwin, Inc., 1962, 633-643.

Bucklin, Louis P. Retail strategy and the classification of consumer goods. Journal of Marketing, January 1963, 50-55.

Cooley, William W. and Lohnes, Paul R. Multivariate data analysis. New York: John Wiley and Sons, Inc., 1971.

Cohen, Joel B., Fishbein, Martin, and Ahtola, Olli T. The nature and uses of expectancy-value models in consumer attitude research. Journal of Marketing Research, 1972, IV, 456-460.

Day, George S. Evaluating models of attitude structure. Journal of Marketing Research, August 1972, 279 -286.

Engel, James F., Kollat, David T., and Blackwell, Roger D Consumer behavior New York: Holt, Rinehart and Winston Inc.. 1968.

Green, Paul E. and Rao, Vithala R. Applied multidimensional scaling: comparison of approaches and algorithms. New York: Holt, Rinehart and Winston, 1972.

Green, Paul E. and Tull, Donald S. Research for marketing decisions. Englewood Cliffs, New Jersey: Prentice-Hall, Inc., 1970.

Holton, Richard H. The distinction between convenience goods, shopping goods, and specialty goods. Journal of Marketing, July 1958, 53-56.

Johnson, Richard M. Market segmentation: a strategic management tool. Journal of Marketing Research, February 1971, 13-18.

Kruskal, J. B. Multidimensional scaling by optimizing goodness to fit to a nonmetric hypothesis. Psychometrica. 1964, 29, 1-27.

Settle, Robert B. Attribution theory and acceptance of information. Journal of Marketing Research, February 1972, 85-88.

Torgerson, Warren S. Theory and methods of scaling. New York: John Wiley and Sons, 1958.

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