An Application of Multidimensional Scaling and Related Techniques to the Evaluation of a New Product Concept

ABSTRACT - A new product concept is studied to determine how it will be received in relation to existing alternatives and whom consumers perceive the likeliest user. Various multidimensional scaling techniques are utilized to analyze the new product concept and seven alternatives as well as five homemaker characterizations and five life-cycle stages.


Larry Percy (1976) ,"An Application of Multidimensional Scaling and Related Techniques to the Evaluation of a New Product Concept", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 114-118.

Advances in Consumer Research Volume 3, 1976      Pages 114-118


Larry Percy, Ketchum, MacLeod & Grove


A new product concept is studied to determine how it will be received in relation to existing alternatives and whom consumers perceive the likeliest user. Various multidimensional scaling techniques are utilized to analyze the new product concept and seven alternatives as well as five homemaker characterizations and five life-cycle stages.


One of the realities of marketing is that more likely than not, consumer researchers are asked to determine the viability of some new product idea already developed by a company's research and development section, rather than asked to help determine what will be a viable new product idea. The need for the study reported in this paper arose from the former.

A new concept for a convenience dinner had been originated, then formulated, by a major food manufacturer. As management assessed the potential of the new idea, they were worried by two vexing questions: 1) How will this product be perceived in relation to existing alternatives; and 2) Who is perceived by consumers as the likely user. The first of these questions was doubly confounding, because if the product was perceived to be similar to existing TV-type dinners, the market would be tough to crack; yet there was much in the new product concept that would suggest it was little more than another TV-type dinner.

A study was designed in which one could evaluate the new product concept within the consumer's product-market cognitive domain, and to examine how the new product concept would be evaluated within various usage scenarios. A multidimensional scaling approach is used in order to fully understand the many interrelationships active in consumer positioning of this new product concept. This multidimensional scaling approach is selected over more conventional concept testing procedures (e.g. a straight "concept-to-use" test) because it is felt what is most important is how the "idea" of this new product will be encoded by consumers for evaluation against alternatives within the product market.

Initial exposure to the new product, assuming it is marketed, is in large part due to advertising and promotion. The image conveyed through these communications variables will be compared and evaluated against competing cognitive stimuli, with evaluation of the new product proceeding from this reference. Establishing this point of reference is thus more critical as a first step toward a decision to market the product: not obviating the need for assessing actual product performance through usage testing, but preceding it.


Answers to the two questions asked by management are considered the goals of this investigation. A number of dinner alternatives are considered, along with several different perceived usage variables. Personal interviews among thirty female heads-of-household in three geographically diverse cities (Boston, Tampa, Omaha) were conducted: 90 completed questionnaires were used in the analysis. Subjects were selected on an area-wide cluster basis, stratified to ensure representation of a broad demographic base.


Two basic exercises were conducted: 1) a gathering of pairwise proximities data among selected dinner stimuli; and 2) a rank ordering of perceived likelihood of serving the selected dinner stimuli under various usage scenarios. In the first exercise, seven alternative dinner options available from the product-market were selected. These are listed below in Table 1. Each of these alternatives was then reduced to a concept statement: e.g. TV Dinner Plus became "frozen dinner featuring family favorites in a single tray with larger portions of the main dish." Along with the new product concept, 28 cards were prepared, each card containing one of the n(n-1)/2 pairs of dinner concepts.



Subjects were asked to look through all of the cards (which they were told contained the description of two different kinds of dinners), placing all the cards they felt contained descriptions of similar dinners into one pile and all those cards they felt contained descriptions of dinners which were not similar into a second pile. They were then asked to rank the first pile from the card containing the description of the two most similar dinners, and to rank those cards in the second pile from those containing the most dissimilar kinds of dinners. 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.

In the second exercise, the subjects were presented with a card listing each of the eight concepts used in the first exercise. She was then asked: suppose you were told that a homemaker was described as '!modern" i.e. she has a generally contemporary life style without being "hip" or fadish. If this were all you knew about this person, which of these different kinds of dinners do you think she would serve most often? All dinners were similarly ranked, the last being the one the subject perceived this scenario homemaker least often serving. All eight dinner concepts were ranked in a similar manner for the four other homemaker characterizations listed in Table 2, as well as for the five life-cycle stages shown. These data then provided rank orderings of the perceived likelihood of each dinner being served often by the person described in each scenario.



Analytic Design

Data collected in the first exercise are considered proximities measurements, and are submitted to Kruskal's (1969) M-D-SCAL 5-M program for non-metric multidimensional scaling. Solutions are sought in three, two, and one dimensions. Data collected in the second phase are considered as two separate data sets for non-metric multidimensional unfolding: the five homemaker characterizations are treated as one set, the five life-cycle stages a second. Spearman rank order correlations are computed for each set in order to determine the "uniqueness" of the rank order profiles. The rank order profiles in each set are then treated as similarity estimates between the eight dinner stimuli and the scenario (cf Percy, 1975). Utilizing again Kruskal's M-D-SCAL 5-M program, this time exercising the lower corner matrix and split-by-rows options, a non-metric unfolding solution is effected for each set in three, two, and one dimension.

Finally, utilizing the appropriate solution configuration from the multidimensional scaling of the eight dinner stimuli, and the rank order profiles collected for each usage scenario, the perceived usage is integrated with the cognitive evaluation via Chang and Carroll's (1969) PROFIT.


The first step in the analysis is to develop an idea of where the new product concept lies in relation to existing alternatives within consumer cognition. Results of the non-metric multidimensional scaling analysis of the n(n-1)/2 pair-wise similarities rankings provide a highly acceptable stress in two dimensions of 0.0097. In fact, the remarkable stress value suggests the possibility of a degenerate or quasidegeneracy of the solution.

Looking closely at the solution configuration shown in Figure 1, one does notice the tendency of the various concepts to collapse into vertices of a parallelogram, one of Shepard's (1974) nine forms of two-dimensional degeneracy. Still, the configuration need not be considered too degenerate for our purposes.



The analysis has provided us with a representation of the stimuli in the consumer's mind. What it has perhaps failed to provide is a portion of the actual structural information conveyed by a more non-degenerate solution. It should be noted however that this was an accepted risk in the analytic design, utilizing as it did fewer than ten stimuli for an anticipated two-dimensional solution as well as the inclusion of obviously clusterable stimuli. In fact, one of the principle goals of the analysis was to determine if indeed the new concept conceptually clustered with the more TV type dinner stimuli. We have learned that the stimuli do cluster together as anticipated, even though we learned nothing about the relationships among the stimuli within the clusters. But this wasn't important to our purpose: what was important is that we discovered the new concept tended to cluster with box dinners, not TV-type dinners.

Reviewing the resulting multidimensional scaling configuration suggests that four distinct clusters of these stimuli occur within the general consumer cognition, and that these four clusters are considered unlike each other. Looking at the composition of the clusters confirms the logic of the groupings: low vs. high involvement "from scratch" dinners and TV-type vs. other prepared-type dinners. So, while from a statistical standpoint the degeneracy apparent in the solution may be unappealing, from a practical standpoint the solution positions the new concept within existing perception, responding to the stated goal of the analytic design.

The second step in the analysis is to relate the various usage scenarios to perceived appropriateness of the stimuli. The homemaker characterization and life-cycle stage scenario sets were each submitted to a non-metric multidimensional unfolding analysis. The Spearman rank order correlations shown below in Table 3 for the five homemaker characterizations reveal some strong similarities among the profiles, suggesting that a multidimensional unfolding of these data might be difficult. [For a more detailed discussion of what to look for in your data to increase the likelihood of a successful multidimensional unfolding, see Moskowitz (1973) and Percy (1975).]



And in fact, the two-space solution configuration revealed in Figure 2 does exhibit a rather high stress value of 0.3536. As a result, one must be careful in drawing any conclusions from the configuration. The fact that five of the stimuli cluster together toward the center of the space suggests that substantially more variation probably occurred between those stimuli. Otherwise, one should expect these five stimuli (the Big Dinner, Box Dinner, M/S Combination, TV Dinner, and TV Dinner Plus) to distribute throughout the space just as the Homemaker Characterization scenarios did.



Looking next at the Life-Cycle Stage Scenarios, the Spearman rank order correlation matrix shown in Table 4 indicates very little similarity among the profiles, unfolding of these data should result in a meaningful solution.

Once again, the early look at the profile data predicted the strength of the solution. Looking at Figure 3, one finds a two-space configuration developed with a stress value of 0.0097. One may certainly consider the inter-and intra-relationships among and between the stimuli and scenarios as meaningful. It is interesting to reflect, for example, on the proximity of the new concept to the Easy Dinner and not the Box Dinner (recall that in the multidimensional scaling of pairwise dissimilarities, the new concept and Box Dinner clustered together in solution). What this perhaps reveals is that in terms of perceived usage, the new concept is rated more constantly like the Easy Dinner than the Box Dinner, even though it is classified in cognition more like the Box Dinner. In other words, different dimensional criteria are involved with each exercise. (This question is addressed in the final step.) Drawing concentric iso-preference curves from each scenario point reflects the perceived likelihood of each dinner stimuli being served. With the exception of Middle Marriage, the new concept enjoys a rather strong likelihood of being served, especially in Early Marriage and by Older Adults -- and one notes here a constant of no children.





While the unfolding of the homemaker characterization scenarios did not prove particularly helpful, the unfolding of the life-cycle stage scenarios was revealing of several interesting relationships bearing on a positional evaluation of the new product idea.

The final step in the analysis is the integrating of the perceived usage information with the general cognitive representation. Each of the rank order scenarios were introduced as properties, along with the coordinate values of the eight dinner stimuli as determined by the M-D-SCAL solution in two dimensions, into Chang and Carroll's PROFIT algorithm. The results provide a vector determination, or direction, for each property (or scenario in this case) in the original two dimensional space such that the projections of the eight dinner concepts on that vector correspond optimally to the given rank order of the perceived likelihood of it being served under that scenario.

Looking now at Figure 4, one finds that life-cycle stage scenarios enjoy good fits to the cognitive representation, while the homemaker characterizations (with the exception of Busy Around House) offer poor fits. This conclusion is drawn by examining the Rho values for each property vector enclosed in parentheses, representing the maximum correlation between each given property and its fitted vector.

If one ignores the homemaker characteristic scenarios, owing to their poor fits, the new concept finds itself associated quite highly with the Early Marriage and Older Adult life-cycle stage scenarios. It would seem that our new product idea tends not to be associated in consumers' minds with the TV-type dinners, which are more oriented in perceived usage to Middle Marriage; rather it is perceived to have a higher likelihood of service among married homemakers with no children, and to cluster with Box Dinners.




In this study, attention has been drawn to the application of multidimensional scaling and related techniques in the evaluation of a new product concept. The focus is purposefully narrow, owing to an overriding concern on the part of management that the new product concept may be perceived by consumers as similar to TV-type dinners. The application of multidimensional scaling procedures rather than more conventional concept testing methods was suggested by this need to determine where the new product concept would be situated within consumer cognition. They are proven particularly useful and sensitive in uncovering certain aspects of cognition, and provide a graphic representation.

Building on past research in the area of food and the meal, a number of made-from-scratch and packaged-convenience dinner alternatives were reduced to simple concept form for comparison with the new product concept. The logic here holds that the introduction of the new product will rely heavily on the consumer's understanding of the "concept" once aware of it, and subsequent trial and usage will be critically mediated by comparisons of her perception of the new product versus competing alternatives. As a further aid in gaining a complete understanding of her cognitive associations, specific homemaker characteristic and life-cycle stage scenarios were considered for all competing dinner alternatives (including the new product concept).

The multidimensional scaling of the pair-wise concept similarities, despite a possible mathematically degenerate configuration, clearly indicated that consumers would not encode the new product concept together with TV-type dinner alternatives. This was, of course, welcome news from a marketing standpoint.

Multidimensional unfolding of the concepts and scenarios indicated that women apparently find it difficult to form any consensus regarding possible dinner alternatives and particular homemaker characterizations. The very high resulting stress measure (presaged by high profile correlations) reflected a great deal of similarity in perceived usage attribution. While women may in fact expect little meal serving difference between the homemakers characterized by the scenarios presented, it may also be true that they found the characterizations difficult to relate to this type of behavior.

On the other hand, differences were easily related to life-cycle stages. The multidimensional unfolding in this case revealed meaningful associations between specific life-cycle stages and the likelihood of often serving particular dinner alternatives. The new product concept was related strongly to the early married and older adult stage (both characterized by no children).

A final exercise combining the cognitive representation with the scenario attribution via PROFIT reinforced the results of the unfoldings. Once again very poor fits generally were found for the homemaker characteristic scenarios, but strong fits for the life-cycle stage scenarios. This embedding of the scenarios in the multidimensional scaling solution represented an external analysis verification of the internal analysis evaluations presented by the multidimensional unfoldings. The consistency of results offers a certain implied level of confidence in the conclusions to be drawn.

We have seen how the utilization of multidimensional scaling and related techniques may be productively applied to new product concept evaluation, particularly when an initial assessment of generalized cognition is important. Raving satisfied one's self of the market positioning potential of the general concept, it is possible to proceed comfortably with more detailed analysis (e.g. conjoint measurement of the concept components and possible positionings) evaluating the strength of the concept.


J. J. Chang and J. D. Carroll, "Mow to Use PROFIT, A Computer Program for Property Fitting by Optimizing Nonlinear or Linear Correlation," Bell Laboratories, unpublished manuscript, 1969.

J. B. Kruskal and F. J. Carmone, "Row to Use M-D-SCAL, A Program to Do Multidimensional Scaling and Multi-dimensional Unfolding," (Versions 5M of M-D-SCAL, all in Fortran IV), Murray Hill, N.J.: Mimeo, Bell Telephone Laboratories, 1969.

H. R. Moskowitz, "Profile Attributes as Similarities," paper presented at Association for Consumer Research Fall Conference, Boston, 1973.

L. R. Percy, "Multidimensional Unfolding of Profile Data: A Discussion and Illustration with Attention to Badness-of-Fit," Journal of Marketing Research, 12 (February, 1975), 93-9.

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



Larry Percy, Ketchum, MacLeod & Grove


NA - Advances in Consumer Research Volume 03 | 1976

Share Proceeding

Featured papers

See More


The Power of the Past: Consumer Nostalgia as a Coping Resource

Dovile Barauskaite, ISM University of Management and Economics
Justina Gineikiene, ISM University of Management and Economics
Bob Fennis, University of Groningen, The Netherlands

Read More


Don't Troll Me Bro: A Study of Griefing in Video Games

Elana Harnish, Ohio University
Jacob Lee Hiler, Ohio University

Read More


In Pursuit of Imperfection: How Flawed Products Can Reveal Valuable Process Information

Erin P Carter, University of Maine
Peter McGraw, University of Colorado, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.