Consumer Flavor Preference Factors in Food Product Design

38
Consumer Flavor Preference Factors
in Food Product Design
J. O. EASTLACK, JR.*
> This article is concerned with the application of a modified semantic differential technique to determine factors in consumer flavor preference to be used in product design for the food industry.
Specific findings are presented on the ability of the modified semantic differential technique to
discriminate among taste characteristics in terms which may be acted upon at the product control
level. The research technique is discussed using findings about consumer preference determined in
blind paired comparisons of samples of four t u t blends of ground coffee; consumer preference curves
are shown in tatte preference sub-groups.
The "Marketing Concept," while easy to apply (or
gain lip service for) in a marketing department, has not
been uniformly applied to the production-oriented area
of products in the food industry. While chemical composition, moisture content, weight, color, and physical
size are subject to laboratory analysis grounded in recognized quality control sampling procedures, many
specifications of food products are evaluated throug|h
taste-testing.
What place has the Marketing Concept in the production operation of such a food product? Obviously, the
Marketing Concept can be assumed to imply dedication to overall company goals. Such problems as meeting shipping schedules, customer service, and the
laboratory aspects of quality control can easily be expressed in the degree of employee dedication to the
overall company product quality and service quality
reputation in its industry. However, the present quesdon is this: how can the marketing group carry consumer-oriented thinking further into product planning,
thus serving the production department and overall
company goals more effectively?
Kuehn and Day, writing in the Harvard Business Review [1], cite several examples of a consumer-oriented
view of product development. Two of the examples
cited deal with the level of chocolate fiavor in a cake
mix and the level of sudsing characteristic in a household detergent. Both of these cases are interesting in that
they disclose, through separate market analysis of bimodal peaks on a consumer preference curve, what
would appear to be distinct sub-markets worth exploiting. These examples consdtute one-dimension variables
in product composition. They were administered in
paired comparison.
This article provides a method for determining dis*J. O. Eastlack, Jr., is Research Director of the Duncan
Foods Co., Houston, Texas. He is presently engaged in graduate work in Ethnopsychology at Rice Univenity.
tributions of consumer preference in these areas. In discussing the various scaling devices frequently applied in
consumer-oriented product testing, Kuehn and Day list
certain elements which are easily scalable, such as the
sweetness of cola drinks and the quantity of suds produced by soaps and detergents. However, they claim
that satisfactory techniques do not exist for scaling
colors, fiavors, and odors in a way amenable to preference analysis [1, p. 103].
In connection with our applications of scaling techniques, both in the modified semantic differential presented here and with regard to simple rank order attribute scaling, we regarded finding the correct ways to
communicate fiavor qualities as the most difficult preresearch problem. Accordingly, the author asked the
Duncan Foods Company staff of taste-test experts for
their glossary of taste factor terms. Then, since the
function of consimier research must be to serve as a
communication bridge between the action-oriented language of the manufacturer and the "acceptance-rejection" language of the consumer, those taste-test terms
relevant to product control were subjected to depth
questioning with consumers, with the results analyzed
through content analysis by a consulting psychologist.
Only then were the scales finalized—scales which could
be considered polar and one dimensional for the consumer and which could be translated into "producer
language" for purposes of product formulation dedsions.
THE RESEARCH PROBLEM
Briefiy stated, the research problem was to determine
if a test formuladon of ground coffee closely approximated the taste preferences of die North^n California
area. This determination had to be made within the
framework of the maiket as it stood after a period of
many years. The market was split rou^ily between two
dominant brands of ground ccriSee, each of which controlled major volume shares wdiich together accounted
for one half of the total market tonnage, chain store
CONSUAABl FLAVOR PREFERB4CE FACTORS IN FOOD PRODUCT DESIGN
private labels controlling slighUy more than one fourth,
and many small-share brands constituting the balance
of the market.
But, the mere existence of a dominant market share
at some point in time does not always correlate with
highest preference rating on a blind taste-test basis. To
position a product correctly, it is necessary to determine
blind taste-test preferences first, for these data should
predict market standing. Therefore, the question was
two-fold. In addition to determining whether the test
formulation was acceptable within the taste patterns
established in Northern CaUfomia, of no less importance in conducting the test would be knowledge of
specific, feasible product specification changes which
would be most likely to effect a closer alignment to consumer taste preferences. For this reason a modified semantic differential technique to deduce the why of consumer preference was constructed, using an overall
paired-comparison preference rating which utilized the
established nine-point hedonic rating scale.
Several questions were proposed in this test:
1. Could consumers distinguish between various
ground coffees at all?
2. Did consumers prefer at a statistically significant
level any of the products over another?
3. Was the test product preferred over competitive
coffees holding major positions in the Northern
California market?
4. Assuming that the test blend did not adequately
meet the Northern California consumer taste
preference, in what ways could it be adjusted to
more closely approximate this preference?
Questions 1, 2, and 3 are the familiar "choice" and
"degree of preference" questions. Well-developed techniques exist in these two areas. As stated above, the
study's major concern, and the reason for exploratory
work, was the "why" of preference. Thus, let us consider only this last question in detail.
The consumer panel method was selected because it
offered the (q>p(Htunity to place test products within the
home for brewing under the conditions which the product would face in actual consumer use. The more detailed questions allowed by the panel technique made it
possible to get answers which were more highly correlated with blend differences. These more detailed questions about product composition factors allowed research to be conducted which would provide answers
which could then be acted upon through product specification controL
In order to qualify for admission to one of five panels
established in the Northern California area, certain
minimum levels of ground coffee use for the household
and for the respcmdent (homemaker) were established.
Since four brmids were evaluated, the research design
allowed a i»rtial "round-robin" consistency ccnnparison. The samplii^ plan jnovided that the panels would
be interlocking so that they would all be projectible to
39
the same [mpulation and thus be cross-comparative.
Respondents were selected utilizing a geographic probability sample based on the density of telephone listings
developed by the Field Research Company which supervised the execution of the research. Product placement
was by personal interview with a personal call-back on
the seventh day. The research design called for N's of
160 per panel, and completion rates varied from a low
of 94.4 percent to a high of 98.1 percent.
THE RESEARCH TECHNIQUE
Two basic measurements of consumer preference
have been included in the research findings which follow. First is the simple preference rating of one version
of the product being tested over the other, utilizing a
nine-point attitude rating scale.^ The second procedure
utilizes a polarized modified semantic differential^ on
eight points known from previous tests to be discriminable by the professional taste-testers and applicable to
changes in product composition.
The data required to provide answers to the first three
questions on this page were derived from preference
ranking techniques, including the above mentioned ninepoint hedonic rating scale adapted from previously reported, large-scale food product testing [4]. This measuring instrument produces sensitive dispersion of degree
of preference.
This fourth question—in what way should we modify
our blend—posed a new question for consumer research. The chief tools contained in the research methodology for making this determination were the "coffee
scales" left with respondents for completion as the test
coffee was used.
These scales, shown in Figure 1, are of the evaluative
type in content. Potency scales or activity scales were
not thought applicable to product testing—relating as
they do to the consumer image of the product rather
than to the product's performance in use. It is interesting to note that Osgood, Sud, and Tannenbaum [3]
have found that evaluative scales yield much smaller
errors of measurement than the other two types of scales
[3,pp. 131,328J.
In Figure 1 eight characteristics of coffee are listed
with a nine-degree ranking ranging from a lesser degree
of the individual characteristic on the left to a greater
degree on the right. The order of listing was originally
set empirically with the goal of maximizing the separate
taste sensation evaluation steps. The original order of
listing was then pretested, and several important changes
* This scale may also be referred to as a nine-point hedonic
rating scale since it is expressed as four degrees of "liking" and
four degrees of "disliking" on either side of a neutral poinL
' Since original application in the social sdences, a great deal
has been written about applications of this technique to corporate image evaluation or tests of advertising eSectiveness.
Sec particularly Mindak [2].
JOURNAL OF AAARKETING RESEARCH, FEBRUARY 1964
40
FIGURE 1
1
2
n•
3
4
Flavor of
a weaker
blend
D
Less
bitterness
n • nn
A less roasted
flavor in the cup
• nn
n•
•n
n • •n
Lighter
body
n•
•n
Poorer flavor
in the cup
n•
•n
Ughter colorof grounds
in the can
• • •n
were made in the order of listing which proved more
effective in isolating separate evaluations by respondents.
Page 76 of Osgood et al. seems particularly germane
in this regard. In discussing measurement as a function
of the context, they state:
"Although we often refer to the semantic differential
as if it were some kind of 'test,' having some definite
set of items and a specific score, this is not the case.
To the contrary, it is a very general way of gettmg at
a certain type of information, a highly generalizable
technique of measurement which must be adapted to
the requirement of each research problem to n^ch
it is appUed. There are no standard concepts and no
standard scales; ratber, the concepts and scales iised
in a particular study depend upon the purposes of the
research." [3]
It should be emphasized at this point that &e paired
comparison product testing format was applied to these
blends of coffee at the end of the six-day alternating use
I>eriod. The pair«l comparison procedure applies, thm,
to the preference choice and to the nine-point attitude-
•
n
n
THE BRArJD OF COFFEE 1 PREFER TO USE
Less aroma
in the cup
•
•n
• • ••
Darker
aptpea ranee
in the cup
• • ••
Flavor of
a stronger
blend
• n ••
More
bitterness
• • n
More aroma
in the cup
6
5
THE BRAND OF COFF EE 1 PREFEIR TO USE
Ughter
appearance
in the cup
SEMANTIC DIFFERENTIAL COFFEE SCALE
DO
7
8
9
••
A more roasted
flavor in the cup
• • ••
Heavier
body
D
• • •
Better flavor
in the cup
•
nn
n
••
Darker color
of grounds
in the can
ranking question. With respect to the coffee scales, however, this test may be regarded as a monadic testing
situation. Each day of the test was a separate comparison of one at the-blends against an abstract standard.
Referring to Figure 1 showing the "coffee scales,"
the mid-point of the scales relates to the coffee which
the respondent preferred to use. Originally, this work
was developed with the mid-point being the concept of
the ideal coffee, on the assumpdcm that diis "ideal"
coffee could be translated into the coffee most frequently
purchased or the coffee preferred by the respondent.
Subsequent pre-testing indicated that it was better to
specify the mid-point as being the "coffee I p r ^ r to
use." This made it possible to sort loyalty groups by a
preliminary screening question which established coSee
preference at the time of the original placement.
Interviewer instructions followed standard practice as
discussed in r^eroices [3] and [S], and no pardcular
problems were encountered in administration. Respondents used two half-pound co&e samples in die
of filling out the "coSee scales."
CONSUMBl FLAVOR PREffiiENCE FACTORS IN FOOD PRODUCT DESIGN
Results which may be inferred from Figures 2 and 3
are the following:^
41
3. Blend 2 (light line) is anoUier strong brand in this
market, both in blind taste-test ranking and in
market position. The differences of experimental
Blend 3 from Blend 2 are important. (Blend 4 did
not achieve an important position in this study and
it has been omitted from these figures for simplicity.)
1. The dominant coffee in this market is Blend 1
whidi is shown by a heavy solid line. Its brand
share is the largest and is still growing because of
many factors in the marketing mix. Though this
has not always been true, blind taste tests show
that its product quality best meets consumer preference. (As a matter of fact, expert opinion holds
that it has maintained its flavor characteristics very
consistently since the date of this research, and its
share of the market is now even greater than it was
a year ago.)
2. The experimental blend has been numbered 3 and
is shown by a dotted line. For purposes of experimenting with Blend 3, the question is how it differs
from Blend 1, since Blend 1 was the overall blind
taste^est winner.
In analyzing the results of this research, two of the
eight scale items were particularly interrelated with preference scores. These were the first and third scales
shown on Figure 1, "lighter-darker appearance in the
cup" and "less bitterness-more bitterness," respectively.*
Figures 2 A and 2B portray frequency distributions of
ratings on these scale items for all respondents at any
time they used any of the three blends charted.
On Figure 2A, experimental Blend 3 is positioned
rather far to the left, or light side, of the products in the
market. Blends 1 and 2 are reasonably close together
' Actual brand names are not given and the values recorded
have been disguised.
* In Figures 2 and 3, the two extreme values at each end of
the scales have been averaged.
Figure 2
Figure 3
DISTRIBUTION OF RATINGS OF DIFFERENT BLENDS
BY APPEARANCE IN THE CUP AND BY BinERNESS
COMPARISON OF BLEND RATINGS BY DIFFERENT PREFERRING
GROUPS FOR APPEARANCE IN THE CUP AND FOR BIHERNESS
lilting
Blend 2
Blend Z
A Appearance in the Cup
A Appearance in [he Cup
42
JOUMslAL OF AAARKEHNG RESEARCH, FORUARY 1964
in appeaiance in the cup, with Blend 2 being slightly to
the lighter side of the overall Mind taste-test winner.
On the fiavor of a weaker blend versus fiavor of a
stronger blend, experimental Blend 3 was also to the
weak side of eveiy brand in the market. Blends 1 and 2
were once more very close in this dimension.
Turning to Figure 2B, showing frequency distributions of ratings of bitterness. Blend 3 again significantly
departs from the market leader. In this characterisdc.
Blend 2 is slightly more bitter than Blend 1.
Figure 3 shows the analysis of Blend 1, the leading
coffee, among two sub-groups—among Blend 3 preferrers and among Blend 2 preferrers.
Returning to appearance in the cup, in Figure 3A we
see a similar situation to that shown in the overall total
market chart—with Blend 1 being reported as substantially darker by those preferring experimental Blend
3. Segmented marketing would, of course, still be feasible for a sub-group of Blend 3 preferrers, should this
sub-group be substantial enough in composition and
coffee use pattern. In the case of a small indicated market segment or a market entry, however, it would seem
that alignment to dominant market characteristics of
taste preference would be the only course to follow with
a commodity item. This conclusion is reinforced by the
lower portion of this chart, which shows a close alignment of Blend 1 and Blend 2 among Blend 2 preferrers.
Figure 3B, dealing with the bitterness characteristic,
shows exactly the same pattern, and is therefore consistent with ^ e "appearance-in-cup" analysis just completed. The same patterns within these sub-groups hold
true for scale item number 2 concerning "fiavor of a
strong blend-fiavor of a weaker blend."
CORRELATION MATRIX OF SCALE RANKINGS OF BLEND I
Seo/e
1
2
3
4
5
6
7
J
2
3
4
5
.614
.386
.346
.499
.650
.129
.530
.662
JI2
.742
6
7
.401
.587
J56 —.008
.580
.785
.688
.598
.485
.593
.722
a
.617
.533
.216
.512
.515
.526
.453
The table shows one of several correlation matrices
developed from the data relating to the overall blind
taste-test winner. Blend 1. A complete factor analysis
was performed on Blend 1 when tested in opposition to
Blend 2. This single paired comparison was selected
because these two blends were the market leaders, and
factor analysis for the purpose of analyzing interaction
of fiavor concepts was thought most reliable within the
context of the paired comparison of these leaders. Three
"factors" subsumed 92 percent of the total apparent interaction between these scaled concepts. These factors are
considered to be:
1. Taste, withinflavorpreference,
2. Appearance, withinflavorpreference,
3. A detached, but ill-defined aroma cluster.
SUMMARY
The technique presented in this paper allows the
"why" of product preference in flavor factor areas to be
determined. Using a well-defined preference technique,
the nine-point hedonic ranking, one can determine eight
interacting fiavor, aroma, and color consumer-conceived
characteristics of ground coffee. Action taken on these
"why" evaluadons, as described both as to content and
degree by the eight "coffee scales," effected a closer
alignment of an experimental blend of ground coffee to
the taste preference pattern of a major regional market
for ground coffee.
Rather than relying strictly on statistical analysis of
the frequency distributions produced by this scaling
technique, the view held here is that this technique
produces a synthesized fiavor profile showing both direction and degree of difference between test products.
In any case, applicadon of results cannot be made by
the taste-test quality specification group through statistics. Rather, their subjective processes are best served
by knowledge of the consumer's interaction pattern of
preference vs. product characteristic. The terms or dimensions of product characteristic, the meanings of
which have been stabilized as between expert and consumer, must be those which may be acted upon by the
quality specification control group.
As an after-word, the author wants to state his interest in the application of this approach to other beverage items and to other food items in which product
specifications are created or evaluated throu^ taste
tests. Two further applications of which the author is
aware can be expected to yield considerable empirical
data for cross-reference. Tlie advantages of this approach relate to ease and accuracy of communication
between consumers and producers. True commimication
must be the sine qua non of the reliability of research
results translated into product specification decisions.
Further testing with other products is thus thoroughly
compatible with the major purpose of this paper—to
present a consumer communication technique in the
area of product preference testing.
REFERENCES
1. A. A. Kuehn and R. L. Day, "Strategy of Product Quality,"
Harvard Business Review, November-December 1962.
2. W. A. Mindak, "Fitting the Semantic DiSerMtial to the
Marketing Problem," Joumal of Marketing, Aprfl 1961.
3. C. E. Osgood, G. J. Sud, and P. H. Tannenbaum, The
Measurement of Memung, Urbana: University of minds
Press, 1957.
4. Quartermaster Researcb and Engineering CcHnnuuid, Department of the Army, Pood Preferences of Men in The
U. S. Armed Porces, January 1960.
5. C. Selltiz et al.. Research Methods in Social Relations, New
York: Holt, Rinehart & Winstcm, 1959.