A Non-Metric Approach For Developing and Evaluating Alternative Advertising Strategies

Larry Percy, Gardner Advertising
Martin R. Lautman, Associates for Research in Behavior, Inc.
Gall R. Kordish, Associates for Research in Behavior, Inc.
ABSTRACT - A four-step non-metric approach to the development and evaluation of advertising alternatives is presented. A study is reported which successfully applies this approach, utilizing various multidimensional scaling and unfolding techniques in determining the background criteria necessary for the development of advertising strategy and in providing a test for evaluating the success of alternative executions in delivering that strategy.
[ to cite ]:
Larry Percy, Martin R. Lautman, and Gall R. Kordish (1977) ,"A Non-Metric Approach For Developing and Evaluating Alternative Advertising Strategies", in NA - Advances in Consumer Research Volume 04, eds. William D. Perreault, Jr., Atlanta, GA : Association for Consumer Research, Pages: 191-196.

Advances in Consumer Research Volume 4, 1977   Pages 191-196


Larry Percy, Gardner Advertising

Martin R. Lautman, Associates for Research in Behavior, Inc.

Gall R. Kordish, Associates for Research in Behavior, Inc.


A four-step non-metric approach to the development and evaluation of advertising alternatives is presented. A study is reported which successfully applies this approach, utilizing various multidimensional scaling and unfolding techniques in determining the background criteria necessary for the development of advertising strategy and in providing a test for evaluating the success of alternative executions in delivering that strategy.


The development of effective advertising alternatives relies on two fundamental principles: (1) a strategy that is motivating; and (2) executions that communicate the strategy. Determining what will be a motivating strategy and knowing whether or not one has been able to communicate it, however, remain perennial problems in the development and pre-testing of advertising. Available methodologies and techniques, proprietary and syndicated, abound. Yet, few integrate these two fundamental principles into a single, coordinated, straight forward procedure; and fewer still take advantage of non-metric techniques. For some reason no extensive effort has been made in applying non-metric methods in advertising pre-testing. Green and Carmone (1970) remarked upon this same point while speculating that non-metric methodologies might indeed be useful in the area of advertising pre-testing, and the more general field of communication (a speculation repeated by Green and Rao, 1972).

This paper discusses a straight-forward four step non-metric methodology which may be used for the selection of an appropriate advertising strategy, and for evaluating the effectiveness of subsequent executions in terms of how well that strategy is communicated. Specifically, a new direction was sought for a brand currently enjoying a highly favorable image within the market, but a brand which none-the-less fails to attract a broad base of users. A repositioning was required which would expand usage of the brand among category users while at the same time maintaining its favorable image. Once the direction for this repositioning was determined, several alternative executions were developed and tested against the criterion of moving consumer perceptions in this desired direction.

The general four-step approach developed to uncover an appropriate strategy direction for this brand, and to provide continuity through pre-test evaluation of the communication effectiveness of alternative executions, is outlined in Figure 1. In the first step, one is interested in determining the brand or product cognitive structure, in understanding the relationships between brands or products as the consumer sees them. Step two relates the alternative brands or products in the category of interest with appropriate functional or usage beliefs associated with those brands or products. Step three isolates the criteria (dimensions), in terms of the appropriate functional or usage beliefs, that consumers use in differentiating between brands or products. Integrating the results of steps one through three provides direction for the development of an advertising strategy that will maximize the positioning of an advertiser's brand within the existing cognitive structure of consumers, and along the dimensions suggested by the appropriate functional or usage beliefs associated with the product category. Alternative executions of this strategy are then developed for testing. Step four evaluates the ability of each execution to alter perception of the brand in the desired direction, as indicated by the first three steps. This general approach is then implemented with non-metric methodologies.




In order to supply guidance for the development of an advertising strategy and subsequent alternative executions which would position the brand under consideration in such a way that the potential for broadening its appeal among category users would be increased, a study was designed which applied the general four step non-metric approach discussed above. Personal, central location "shopping center intercept" interviews were conducted among male and female heads-of-household recruited in the suburban Philadelphia area, all of whom were screened for category usage.


Initially, a control group of 60 subjects was used in order to gather data to meet the requirements of the first three steps of the approach. Each of these subjects was first asked to provide pairwise similarity data on eight brands (including the test brand) by sorting a set of 28 cards, where each card contained one of the n(n-1)/2 pairs of brand names. Subjects were asked to look through all of the cards, placing those cards they felt contained similar brands into one pile and all those cards they felt contained brands which were not similar into a second pile. They were then asked to rank the first pile from the card containing the two most similar brands, and to rank those cards in the second pile from those containing the two most dissimilar brands. The order of the second pile was then inverted and added to the first, providing a ranking of all 28 pairs from the most to least alike.

Following the first sort, each subject was asked to rank for each of the eight brands ten functional and usage beliefs known from prior qualitative and quantitative research to be appropriate in defining the critical dimensions on which products in this food category are currently evaluated (see Table 1). The belief statements are all phrased in a positive direction (wide, good, high, easy, suitable). These data then provide rank orderings of the perceived belief strength associated with each of the eight brands representing the product category.



Using the information gathered in an analysis of this data as background, a new advertising strategy was developed and alternative executions of that strategy created for testing. In this particular case, that lead to the creation of six alternative print executions in rough-finished form, two within each of three campaigns. Each of these executions was then exposed to a single experimental group of 60 subjects (360 in total, selected as outlined above) for evaluation --the fourth step of this approach. The subjects in these experimental groups were asked to carefully read the alternative execution given to them (one execution per group) and to take as much time as necessary to learn all that the advertiser was trying to communicate. After a subject indicated she was through with the execution, it was removed from sight, and a long series of probing questions asked about the content of the advertisement in order to elicit as much information as possible from the subject about what was learned. Following this questioning, each subject was asked to rank the ten functional and usage belief statements shown in Table 1 for each of the eight brands.

Analytic Design.

Following a non-metric implementation of the four step approach outlined in Figure 1, the data collected in the pairwise similarity rankings of the eight brands in step one are submitted to Kruskal's (1969) M-D-SCAL 5M program for non-metric multidimensional scaling. Solutions are sought in three, two, and one dimensions. Next, in order to relate each of the ten functional and usage beliefs to each of the eight brands in the same space, a non-metric multidimensional unfolding of the rank order is conducted. Using the M-D-SCAL 5M program, and exercising the lower-corner matrix and split-by-rows options a non-metric unfolding solution is obtained in three, two, and one dimension. [While this is the generally suggested method for unfolding within Kruskal's M-D-SCAL 5M algorithm (Carroll, 1972; Kruskal and Carmone, 1969), Kruskal (1972) does discuss alternative methods for strengthening the procedure.] The solution determines a geometrical configuration which embeds the two sets of points in a single space as desired: The brands become "ideal points" on which the beliefs are compared. This analysis provides a convenient way of representing associations of various functional and usage beliefs within the brand space, and permits one a determination of those beliefs most strongly or uniquely associated with particular brands. For step three, one utilizes the appropriate solution configuration from the multidimensional scaling of the eight brands, and the rank order belief profiles collected, to integrate the beliefs with the cognitive evaluations via Chang and Carroll's (1969) PROFIT. By isolating appropriate orthogonal beliefs, one is provided with insight into the probable underlying dimensional structure of the cognitive evaluations.

Using now the data from six experimental groups, a communication effectiveness value is determined for each execution following a scoring scheme developed by Grass, Winters, and Wallace (1971). This technique answers the question: Are we able to communicate the intended strategy with these executions? If the executions do in fact communicate the intended strategy, it is then necessary to determine how efficient they have been in altering perception of the test brand in the desired direction -- step four of the general approach. Once again, a non-metric multidimensional unfolding is called for, utilizing the lower corner matrix and split-by-rows options of Kruskal's M-D-SCAL 5M program. In this case only the belief profiles for the test brand are considered, as each execution becomes the stimulus, or "ideal point", in the solution configuration. From this space, one measures the success of an execution in positively effecting brand perception by its proximity to the appropriate strategy-determining belief stimulus.


The first step in our non-metric approach to the development and evaluation of advertising alternatives is to develop an idea of how existing brand alternatives are positioned within consumer cognition, utilizing Kruskal's M-D-SCAL 5M to analyze the n(n-1)/2 pairwise similarities rankings of the eight brands considered resulted in a solution configuration in two dimensions with a highly acceptable stress value of 0.0492. Looking closely at the solution configuration shown in Figure 2, one finds no evidence of degeneracy as outlined by Shepard (1974). This encourages us to believe that these results reflect not only a general representation of the stimuli in the consumer's mind, but also an indication of the actual structural relationships among the stimuli.

Reviewing the results of this solution configuration would seem to suggest a unique orientation for all of the brands considered, with the exception of Brand B and Brand C, on the two principal dimensions that define this space. (Step three, will involve an effort to identify those basic dimensions.)



The second step in our approach is to relate important functional and usage beliefs to perceived appropriateness of the brand stimuli. The rank order appropriateness of each belief statement for each brand were submitted to a non-metric multidimensional unfolding analysis via Kruskal's M-D-SCAL 5M program, exercising the lower corner matrix and split-by-rows options. Figure 3 shows the results of this analysis in two dimensions. While the solution configuration in two dimensions provides only a fair fit to the data (STRESS = 0.209), the results remain readily interpretable.



It is possible to use this two-dimensional space to determine iso-belief contours for each of the eight brands studied. Because the distance metric is assumed to be Euclidian, one may select any brand and construct concentric circles with it as the center; the more likely a particular belief is associated with that particular brand, the closer it will appear to the brand (in terms of the concentric or iso-belief contours). As the distance between the brand and a belief increases, the less important it becomes for associative consideration.

In the solution space above, for example, one notes that the test brand is believed primarily to be high priced, good tasting, and having high quality (all functional beliefs); but not suitable for everyday use or the whole family (both usage beliefs), nor particularly attractive or nutritious. Also, we see a concentration of Brands A, F, and G with the usage belief preparation ease, plus a strong similarity in terms of all ten beliefs for Brand B and C (as indicated by their close proximity). It is also interesting that the same general relationship among the brands appears to hold in both the cognitive space and the unfolded space, indicating that the set of ten belief statements is indeed representative of consumer's beliefs associated with the brands in this product category. This augurs well for finding the two basic dimensions of the cognitive structure among the ten belief statements.

Step three integrates the solution configuration from step one and the rank ordering of the eight brands for each belief as transformed from the rankings in step two. Each of these rank orders was introduced as a property, along with the two dimensional coordinate values of the eight brands as determined by the M-D-SCAL 5M solution, into Chang and Carroll's (1969) PROFIT program. The results of this analysis provide a vector determination, or direction, for each of the properties (or beliefs in this case) in the original two dimensional cognitive space such that the projections of the eight brands on that vector correspond optimally to the given rank order.

The results of this analysis are shown in Figure 4. Consumer s tended to use the two relatively orthogonal properties -- price and suitability for everyday use --in differentiating between the brands. Each of these properties fit the data well, as evidenced by their high Rho values (.85 for price and .98 for everyday use). These values represent the maximum correlation between each property and its fitted vector. In fact, all of the properties exhibit exceptionally high Rho values (as shown within the parenthesis for each vector in Figure 4), indicating again that these functional and usage beliefs fit well within the consumer's cognitive representation of the category.



Because of the good fits and near orthogonality of the price and everyday use beliefs to the general cognitive representation developed in step one, one may confidently conclude that a brand in this category may be parsimoniously, yet with reasonable accuracy, described in terms of these two belief dimensions, one functional (price) and one usage (everyday use). Additionally, all of the other belief statements tended to be moderately correlated within the cognitive domain with at least one of these two basic differentiating beliefs. This type of non-metric analysis of beliefs and cognition can also lead to further insight into the product category; insight helpful to the development of effective advertising strategy. For example, it can be seen that consumers tended to expect brands rated high on quality to also be good tasting, attractive, and high priced.

With the completion of these first three steps, one is able to begin development of an appropriate advertising strategy to help broaden the usage base of the test brand. Reviewing what has been uncovered through the non-metric analysis of the control group data, one may conclude that: (a) the test brand is differentiated from other brands in the category within consumer cognition; (b) consumers believe the test brand to be high priced, good tasting, and of high quality (beliefs more likely associated with the test brand than any other brands); and (c) consumers are, primarily, using price and suitability for everyday use in differentiating between brands?

Studying these results, one notes that the test brand does not seem to be associated with the functional beliefs of nutrition and appearance, or the usage beliefs of everyday use and whole family (see Figure 3). These are beliefs more associated with either Brand B and C (the two category leaders) or Brand E. Because these appear to be the beliefs associated with the more successful brands in the category, they must be considered likely candidates as a basis for the test brand's advertising strategy. But because the everyday use belief is also a fundamental discriminating belief dimension, it would seem to offer the most potential for the test brand if this belief can be effectively associated with the test brand and communicated to consumers.

As a result of these analyses, an advertising strategy was selected for the test brand which stressed the product's suitability for everyday use, and along with that theme, implied it was suitable for the whole family. Three general campaign orientations were created based on this strategy, and two rough print executions of each developed for testing.

Before one specifically addresses the step four question of how the perception of the test brand has been affected along the belief dimension of everyday use, the communications effectiveness of each alternative execution was determined following a scoring scheme presented by Grass, Winters, and Wallace (1971). This score acts as a check on the overall effectiveness of each execution in communicating the intended goals of the strategy, independent of whether or not the perception of the test brand is altered in the desired direction. If this check were not made, and the executions failed to alter perception, one would not know if this was because the strategy itself was insufficiently developed, or because the strategy was simply not communicated by the executions tested.

Table 2 details the communication effectiveness scores for the six alternative executions tested. The executions for both the situational orientation and the product orientation campaigns have high scores, indicating that they do a good job in communicating to consumers the intended strategy of suitability for everyday use. The executions for the receiver orientation campaign were generally less effective in communicating the advertising strategy (and we might expect them to be less effective in altering perception). These conclusions on the executions' communications effectiveness are based on norms developed for advertisements in the category tested.



Turning now to step four, the rankings of the test brand on each of the ten functional and usage belief statements by the six experimental groups, along with the original control group, are submitted to a non-metric multidimensional unfolding. This analysis yielded a good joint space configuration in two dimensions with an excellent stress value of 0.010.

The resulting two-dimensional representation of test brand appropriateness for the belief set (as a function of no new advertising in the case of the control group vs. the new alternative executions exposed to the six experimental groups) is shown in Figure 5. In this solution configuration, the closer an execution's position in the space to one of the belief stimuli, the more those exposed to that particular execution feel it describes the test brand; in effect, "iso-perception" contours radiating from the belief stimuli. The position of the control group serve as our anchor for existing consumer perception of the test brand; reflecting, as was noted in the earlier steps, a high priced, good tasting, high quality product.



Remembering that the concern in this test is moving the perception of the test brand closer to an appropriateness for everyday use, it is clear that Alternatives I, II and IV do the best job. These three executions have accomplished the advertiser's goal (as developed from the analysis in steps one through three) of persuading the consumer that the test brand is in fact suitable for everyday use, as seen in their much closer proximity to the everyday use belief stimulus. Alternatives III and V are somewhat less effective in altering perception of the test brand in the desired direction; and, in addition, they appear to have moved the brand's perception away from its current image of high quality and good taste, and high price. Alternative VI accomplished little in the way of desired persuasion.

One may conclude from this analysis of step four that the situational orientation or product orientation, in terms of appropriate campaign strategy, are both viable creative directions for implementing the advertising strategy of everyday use, and that Alternatives I, II, and IV represent effective executions of that strategy. The receiver orientation campaign strategy, one recalls, was not effectively executed in terms of its ability to communicate the intended strategy for either Alternative V or VI (see Table 2), so it is not surprising that both were ineffective in altering perception. This does not mean that the campaign is ineffective; perhaps different executions would effectively communicate the strategy, and positively move perception. Alternative III, while it effectively communicated the desired strategy (as reflected in its communications effectiveness score), was not persuasive.


In this paper, attention has been drawn to a four step non-metric methodology for the development and evaluation of alternative advertising executions. Utilizing non-metric multidimensional scaling and unfolding techniques, a straight-forward procedure has been presented that provides the practitioner with a means of gathering background data on consumer understandings and evaluations of the competitive environment of a product category. From this background information, a meaningful advertising strategy may be developed, and alternative executions of that strategy then tested against a criterion of how well each is able to communicate the strategic goals and alter perceptions in the desired direction. The underlying logic employed here suggests that any advertising message developed to persuade will be mediated by a consumer's comparisons of her perceptions of the advertiser's brand versus competing alternatives on various properties. It is therefore essential to understand where one's product or brand is situated within consumer cognition and how each alternative is evaluated over an appropriate belief set. Only then can one make a strategic decision on the best course to follow in effecting a desired change in a market.

The application of non-metric multidimensional scaling and unfolding procedures rather than more conventional strategy development or advertising pretesting techniques in operationalizing the step wise approach recommended here is suggested by its proven usefulness and sensitivity in uncovering certain aspects of cognition and the fact that they require only non-metric assumptions in data collection while providing metric output.

The four-step methodology advocated should be clearly differentiated from the ad exposure technique employed here in demonstrating it. Clearly, any "unnatural" exposure introduces some degree of artificiality into an ad pretest situation, and each researcher can be expected to have his preferred pre-test technique. Nevertheless, whatever technique is selected, it probably could be embedded into this general methodology. The technique that was employed here -- forced ad exposure where the respondent could view the ad for as long as desired -- has traditionally served as a vehicle for measuring an ad's communications effectiveness, not for assessing its persuasiveness. That is, it has been utilized as a technique for assuring one's self that if a subject wanted to know what an ad said she would be able to learn its intended communications objectives. The fact that the technique also appears to be sensitive to differences in the persuasive impact of various ads, however, seems to suggest that it may be equally useful for that purpose. The intensive questioning about the ad's content which follows exposure might be serving as an analogue for repeated exposures and acting to reinforce any persuasive impact which the ad may possess. In any case, our concern here has been with outlining a general methodology rather than with advocating any single type of exposure situation as best for measuring persuasive impact. Similarly, with respect to stimuli, this methodology is general enough so that it would be comfortable with testing the persuasiveness of print ads or commercials either in rough or in finished form.

In summary, a four step approach to the development and evaluation of advertising alternatives utilizing non-metric methodologies has been discussed that: (1) determine the cognitive representation of a market, or how consumers "see" a competitive brand environment; (2) relates the available alternatives to a set of appropriate beliefs; (3) establishes which dimensions consumers use to discriminate between brand, enabling an advertiser to determine along which dimension or dimensions he might wish to "move" the perception of his product (if indeed any movement is desired at all); and (4) test executions of the strategy following from steps one through three, measuring how well each communicates that strategy and "moves" the perception of the advertised brand in the desired direction. The results obtained using this four step non-metric approach have proven in practice to be very effective in the developing and evaluation of advertising campaigns.


J. Douglas Carroll, "Models and Methods for Multi-dimensional Preference Analysis," paper presented at Workshop on Multidimensional Scaling, University of Pennsylvania, June, 1972.

J. J. Chang and J. D. Carroll. "Now to Use PROFIT, A Computer Program for Property Fitting by Optimizing Nonlinear or Linear Correlation," Murray Hill, N. J.: mimeo, Bell Telephone Laboratories, 1969.

R. C. Grass, L. C. Winters, and W. H. Wallace. "A Behavioral Test of Print Advertising," Journal of Advertising Research, Vol. 8, No. 1, pp 11-14.

Paul E. Green and Frank J. Carmone. Multidimensional Scaling and Related Techniques in Marketing Analysis. Boston, Mass.: Allyn and Bacon, Inc., 1970.

Paul E. Green and Vithala R. Rao. Applied Multidimensional Scaling: A Comparison of Approaches and Algorithms. New York: Holt, Rinehart and Winston, 1972.

Joseph B. Kruskal, "Special Problems and Possibilities,'' paper presented at Workshop on Multidimensional Scaling, University of Pennsylvania, June, 1971.

Joseph B. Kruskal and Frank J. Carmone. "How to Use M-D-SCAL, A Program to Do Multidimensional Scaling and Multidimensional Unfolding," (Versions 5M of MDSCAL, all in Fortran IV), Murray Hill, N. J.: mimeo, Bell Telephone Laboratories, 1969.

Roger N. Shepard, "Representation of Structure in Similarity Data: Problems and Prospects," Psychometrika, 39 (December, 1974).