Advances in Consumer Research Volume 6, 1979 Pages 161-165
ALLISON & UHL REVISITED: THE EFFECTS OF TASTE AND BRAND NAME ON PERCEPTIONS AND PREFERENCES
G. A. Mauser, Simon Fraser University
[An earlier version was presented to the 1977 meeting of the Canadian Association for Administrative Studies in Fredericton, N.B. The author would like to acknowledge the support and funding of Labatt Breweries of B. C., Ltd. and the B. C. Provincial Ministry of Labour. I would also like to thank Jim Watts and Ray Koopman for their help in this study as well as the hard-working students in Commerce 444.]
A 23 split-plot factorial experiment was conducted to assess the effect of brand information, product familiarity, and order of presentation on consumers' judgements of the taste of beer samples. Paired comparisons on several taste characteristics and preferences served as dependent variables. Subjects were 240 college students. The results suggest that beer drinkers can distinguish among brands using taste and aroma cues alone.
It is frequently asserted that beer drinkers are unable to discriminate among brands of beer using only taste and aroma cues. If true this would imply that beer drinkers are forced to rely on extrinsic cues (e.g., brand name, price, store image) in order to decide which brand of beer to purchase because they can not make use of intrinsic cues (e.g., product taste or aroma). However the empirical support for this claim is somewhat mixed. Only three published studies can be found which have directly addressed this question (Allison & Uhl, 1964; Jacoby, Olson, & Haddock, 1971; and Valenzi & Eldridge, 1973). Two of these three studies support this assertion while the third found that consumers could discriminate among brands using only taste and aroma cues significantly better than chance (Jacoby, et al., 1971).
The failure of two of the studies to find significant differences may have been due to problems in their experimental methods rather than to consumers' inability to discriminate. Close examination of these studies supports this conjecture. Allison & Uhl's pioneering work was a field study which used crude three-point rating scales. Valenzi & Eldridge, while conducting a well-controlled laboratory study, used rating scales which were not subjectively anchored, and they hypothesized that significant differences would have been found had they used better anchored scales.
It is worth noting that these studies differ in their choice of dependent variables. While Allison & Uhl measured a variety of taste characteristics (e.g., strength, body, bitterness), later researchers have focused exclusively on perceived quality. Current work has been primarily on the effect of price on perceived quality (Olson, 1977). There are a few advantages to studying descriptive taste characteristics rather than perceived quality. First, descriptive characteristics allow more scope for marketing control --they are more "actionable" -- such characteristics are particularly useful in positioning new products (Stefflre, 1968; Green, 1975). Second, it seems plausible that differences between brands are more readily identified using descriptive characteristics than with perceived quality due to the more "available" or more "codable" nature of descriptive characteristics (Brown & Lenneberg, 1954; Lanz & Stefflre, 1964)
It thus appears worthwhile to replicate Allison & Uhl's earlier study using better experimental controls and more sensitive measurement techniques. The method of paired-comparison is widely regarded as a particularly sensitive measure (Byer & Abrams, 1953; Jellinek, 1964; Torgerson, 1958). In comparison with single stimulus methods (i.e., the rating scale), paired comparison methods two advantages: subjects compare stimuli directly with each other, rather than only indirectly via scale position, and also the task itself makes fewer demands on the subject.
Despite these advantages paired-comparison is rarely used in academic marketing research studies. Perhaps this is due to the large number of pairs of stimuli required with even a modest number of stimuli as well as the forbidding statistical methods required by its use (Scheffe, 1952; Neter & Wasserman, 1974; Bechtel, 1967).
A secondary concern in this paper is to look at the effect of consumer familiarity with the product class on taste judgements. Several researchers have suggested that consumers' perceptions of a product class may differ with their degree of familiarity that they have with the product (Shapiro, 1968; Olson, 1977). Perhaps people who drink beer only occasionally, and thus are less familiar with the product class, are less able to discriminate among the various brands using only taste and aroma cues.
The objectives of this study are to (a) examine to what extent the perception of and preference for beer samples is influenced by brand information, (b) determine which taste characteristics (if any) beer drinkers can use to distinguish among brands of beer using only taste and aroma cues, (c) determine to what extent familiarity with the product class influences drinkers' ability to distinguish among brands of beer.
Nine different brands of beer were included in the study. Brands were selected to cover as broad a range as possible of brand image and composition differences. There were six Canadian brands (two each of ale, lager, and light lager) as well as three American brands (two lager and one light lager).
All of the Canadian brands were exactly the same price while the American brands retailed for a slightly higher price in B. C. All of the brands, Canadian and American, were approximately the same color.
A convenience sample of 240 beer drinkers was drawn from volunteer undergraduate and graduate students. Students had to be at least 19 years of age and to admit to drinking at least "some" beer to qualify for inclusion in the study. Twenty-seven percent of the subjects in each experimental condition were female. College students were selected for this study because they are an important submarket for the beer industry as well as being a standard group for experimental researchers to study.
A 23 factorial experiment with one repeated measures factor was conducted having 60 Ss per cell. The repeated measure factor was Label (brand name present or absent). The between-Subject factors consisted of Familiarity (high, low) and Order of Presentation (straight, inverse). Students were classified for the Familiarity factor as having a high degree of familiarity with the domain if they drank 6 or more bottles of beer per week, or as having low familiarity with the domain, if they drank 5 or fewer bottles per week. This break point was determined by asking both Labatt's and Molson's what break point they usually used. The Order of Presentation factor involved reversing for half of the sample both the sequence of pairs that are presented to the S (e.g., p1, p2, p3 or p3, p2, p1) as well as the order of samples within pairs (e.g., in comparing Olympia and Schlitz, Schlitz may be presented either on the left or on the right).
The dependent variables were (a) the perceived similarity of the samples, (b) preference for the samples, (c) judgements of which sample is X'er on each of several taste characteristics (i.e., strength, lightness, aftertaste, bitterness, carbonation, heavy bodiedness, fillingness, and smoothness). Judgments of similarity were expressed using a 9-point scale, where 1 indicated "dissimilar" and 9 indicated "very similar". All of the remaining dependent variables were dichotomous choices indicating which stimulus was judged X'er on the characteristic involved. No "undecided" responses were permitted.
Fifteen blocks of three distinct pairs each were fashioned to include all 36 possible pairs of the 9 brands. The nine identity pairs were included in the unlabeled condition, and nine redundant pairs included in the labeled condition, in order to fill out the 9 blocks. There were four replications of each of the 15 blocks in each of the 8 experimental cells.
Each S tasted three pairs of beer samples under each level of Label (brand name present or absent) yielding a total of six pairs tasted. All Ss received three unlabeled beer samples first and then three labeled beer samples. Each sample of beer contained 1 ounce of beer. Following the tasting of each pair of samples, Ss were requested to indicate their impressions on each of the dependent measures. Ss were requested to nibble unsalted crackers after tasting each pair of beer samples in order to cleanse the palate.
After tasting a pair of samples, Ss responded to each dependent variable in turn. Ss first rated the perceived similarity of the stimuli and then indicated which of the two stimuli they preferred the more. Next Ss judged the samples on each of the eight taste characteristics. These characteristics were presented to Ss in 16 distinct orderings in order to minimize a possible fatigue effect influencing the results. These 16 orderings consisted of all one-step permutations of a pre-selected order and its inverse. Each S was presented with a different ordering of the eight taste characteristics for each of the six pairs of beers he was asked to judge, receiving three permutations from each of the two basic orders in alternation so that a pattern was not readily apparent in the presented orderings.
Careful attention was paid to presenting all beer samples under identical conditions: the temperature was kept at 39 degrees F (4 degrees C). Bottles and cans were opened and poured immediately before being offered to Ss; the beer samples were carefully poured in a manner that would not create a frothy head.
A split-plot ANOVA design was used to analyze these data because of the repeated measure factor (Label). Two 2-way ANOVAs were conducted to examine the effect on the dependent variables by the independent variables of Label, Familiarity, and Order. Two 2-way ANOVAs were conducted rather than one 3-way ANOVA in order to have a sufficient number of Ss (8) per cell for each pair of beer samples (Kirk, 1968, pp. 245-318).
In order to examine the effect of the independent variables for the entire set of 9 beer brands, the ANOVAs for each of the 36 pairings were summed for each of the 9 dependent variables. Thus the appropriate degrees of freedom for the overall F-test is the sum of the degrees of freedom for each of the individual F-tests.
Table 1 shows the results of the first two-way split-plot ANOVA for the Familiarity and Label factors for the Ss' overall preferences for the 9 beer brands. Neither of the main effects nor the interaction effect is significant at the .01 level. [The .01 significance level was selected for use in this study as it is appropriately conservative considering the large number of tests of significance that had to be calculated. If the series of tests for each effect across the nine measures is considered as an "experiment", this gives a probability of .086 that at least one of these nine tests would be significant at the .01 level. More striking still is that for the entire series of 27 significance tests, there is a probability of 0.24 that at least one test is significant by chance alone at the .01 level (Kirk, 1968, pp. 77-86).] This would imply that beer drinkers are able to discriminate as well (or as poorly) in taste tests with or without brand information, and that frequent beer drinkers do not differ appreciably from occasional beer drinkers in their preferences.
Table 2 shows the results of the second two-way ANOVA for Ss' preferences in which the Familiarity factor has been replaced by the Order factor. Again, neither of the two main effects are significant, nor is the interaction effect. This provides support for the previous ANOVA results for the Label Effect, and implies that the order of presentation of the stimuli did not play an important role in determining Ss' preferences.
LABEL AND FAMILIARITY EFFECTS FOR PREFERENCE MEASURE
LABELED AND UNLABELED TORSCA SOLUTIONS FOR JUDGED SIMILARITY OF BEER SAMPLES
LABEL AND ORDER EFFECTS FOR PREFERENCE MEASURE
Insignificant ANOVA results may be due to excessive noise in the data or due to the same patterns emerging under both conditions. To investigate the patterns of discrimination among the brands, the perceived similarities of the samples were scales for both conditions of the Label factor (labeled, unlabeled) using TORSCA-9B (Young, 1967). Figure i compares the results of these two analyses. The unlabeled two-dimensional configuration (stress = .126) has been rotated to a least squares fit with the labeled configuration (stress = .185) following Pennell & Young (1967). While the beer brands are tightly clustered in the labeled configuration, they are scattered more in the unlabeled configuration. Nevertheless the same two basic groupings may be identified in both configurations:
Cluster I : Olympia, Schlitz, Lite, and Cool Spring
Cluster II : Old Style, Export Ale, Canadian, Blue, and 50 Ale
The vertical line in the figure divides the configurations into these two clusters. Only two brands are mis-sorted (Olympia and 50 Ale), both of which are unlabeled samples. Olympia was seen as being heavier when unlabeled than it was when labeled, and 50 Ale was seen as being lighter unlabeled than labeled. Cluster I contains the lighter and smoother brands, while Cluster II holds the brands seen to be stronger, heavier brands. All of the American beer brands are in Cluster I, although it also contains a Canadian brand (Cool Spring).
In a parallel manner to the analyses of preference, two-way split-plot ANOVAs were conducted for each of the remaining dependent variables, the eight taste characteristics. Tables 3 & 4 show the F-values for these analyses. Neither the Order effect nor the Familiarity effect were significant for any of the taste characteristics. It appears that for the set of measures included in this study, both occasional and frequent beer drinkers agree about the taste of beer. Nor did the order of stimulus presentation play an important role for any of these dependent variables.
SUMMARY OF F-VALUES FOR TWO-WAY ANOVA'S ACROSS ALL MEASURES FOR FAMILIARITY AND LABEL EFFECTS
There was a significant Label effect in four out of the eight taste characteristics. The introduction of brand information appears to have altered consumer perceptions of beer strength (F = 2.13, p < .01), lightness (F = 2.49, p < .01) heavy bodiedness (F = 2.28, p < .01), and fillingness (F = 1.78, p < .01). Interestingly enough, brand information did not seem to influence perception of aftertaste, bitterness, carbonation, nor smoothness for these beer brands.
Only one of the interaction effects was found to be significant, that of Familiarity x Label (F = 1.71, p < .01). This interaction was also found to be significant in Valenzi and Eldridge (1973).
SUMMARY OF F-VALUES FOR TWO-WAY ANOVA'S ACROSS ALL MEASURES FOR ORDER AND LABEL EFFECTS
How robust are these results considering that the paired comparisons are not strictly independent with respect to each other as every brand of beer is compared with every other one? To attempt to answer this query, the ANOVA's were recalculated omitting, one at a time, each of the nine brands, leaving 28 paired comparisons for the remaining eight brands. Not one of the previously significant effects was now insignificant, nor were any of the previously insignificant effects now significant. his suggests that these results are not an artifact of the lack of independence.
It is also possible that the two-way ANOVA's used in this study did not find significant effects since the sums of squares had been collapsed too far by including variance attributable to the effect of the third factor in the error terms. To evaluate the importance of this potential problem, the preference measure, which was not significant but was quite close, was reanalyzed using a 3-way ANOVA design. None of the main effects were found to be significant at the .01 level, although the Familiarity x Label interaction effect was found to be significant as it was in the 2-way analysis (F = 1.77, p < .01). None of the other measures would be expected to yield different results if reanalyzed using a 3-way ANOVA.
THREE WAY ANALYSIS OF VARIANCE FOR PREFERENCE MEASURE
That the label factor was significant in four out of the nine measures suggests that brand information influences certain taste characteristics but not others. This is probably due to the information associated with the brand images which pertains to specific taste characteristics but not to others. These results are consistent with the often heard claim that American beers are lighter and have a lower alcohol content than Canadian beers. However, certain Canadian beers (e.g., Cool Spring and Blue) are judged from taste and aroma to be nearly as light as some American brands (e.g., Schlitz and Olympia).
TWO DIMENSIONAL TORSCA SOLUTION FOR INTER-CORRELATED THURSTONE SCALES
To investigate the interrelationships between the measures with respect to the two levels of the Label factor, each of the taste characteristics and overall preference were scaled using Thurstone paired-comparison techniques and the resulting one-dimensional scales inter-correlated (Torgerson, 1958). These intercorrelations were then themselves used as input to TORSCA (Young, 1967). Figure 2 shows the pattern of relationships among the 18 resulting Thurstone scales (9 for each of the two conditions) in the two-dimensional configuration (stress = .056). Note that all scales cluster quite tightly in two distinct groupings except for the two preference scales and, to a lesser extent, the two carbonation scales. These two groupings are due to the very high intercorrelations among almost all scales coupled with the inverted ordering of smoothness and lightness with respect to almost all of the other scales. Note also that both the labeled and unlabeled versions of each scale are located very close to each other with the striking exceptions of the preference and carbonation scales.
In sharp contrast to Allison and Uhl's study, the results of this study suggest that beer drinkers can distinguish among major brands of beer using only taste and aroma cues. Beer drinkers were able to use four of the eight taste characteristics in the unlabeled condition to discriminate among beer samples nearly as well as they could in the labeled condition. Moreover, consumers' preferences did not significantly change from the unlabeled to labeled conditions.
There are a few alternative ways to approach the question of why these studies differ. The first is methodological. The failure of Allison and Uhl to find significant differences in the unlabeled condition may have been due to the cruder controls that are possible in a field study as opposed to a laboratory study. Moreover, the method of pair comparisons that was used here is more sensitive to existing differences than is the method of single stimuli. The method of single stimuli is particularly insensitive if a very few rating categories are used, as was the case in Allison & Uhl's study.
Other methodological differences which may be able to explain the different results of the two studies should also be considered. First, despite the approximately identical color of the beer samples, some differences may have been observable to the subjects as they were not blindfolded for the tests. Second, the subjects in this study were college students, while Allison & Uhl used adult beer drinkers. Greater confidence may be obtained in the generality of these results if this study were to be replicated using blindfolded adult beer drinkers.
An alternative approach to understanding why these studies might differ is to examine the range of the stimuli used in the two studies. The range of taste differences was larger in this study than was the case in Allison & Uhl's study. Clearly Ss would be expected to be able to distinguish between tastes that were very different using only taste and aroma cues, say between pure alcohol and pure water, or between Guinness and Beaujolais. The interesting question is which differences are beer drinkers able to identify, not whether or not they can identify any differences. All of the brands in earlier studies were American brands, while in this study both Canadian and American brands were included. Moreover, this study found only two clusterings among the brands and all of the American brands fell into the same cluster.
Fifteen blocks of pairs of beer brands are shown here for both levels of the Label factor. Each S was presented with the task of drinking the beer samples in an assigned row.
Ralph I. Allison and Kenneth P. Uhl, "Influence of Beer Brand Identification on Taste Perception", Journal of Marketing Research, 1 (August 1964), 36-39.
G. G. Bechtel, "The Analysis of Variance and Pairwise Scaling", Psychometrika, 32 (1967), 47-66.
Roger Brown and Eric Lenneberg, "A Study in Language and Cognition", Journal of Abnormal and Social Psychology, 49 (1954), 454-462.
Albert J. Byer and Dorothy Abrams, "A Comparison of the Triangular and Two-Sample Taste-Test Methods", Food Technology, 7 (1953), 185-187.
Paul Green, "Marketing Applications of MDS: Assessment and Outlook", Journal of Marketing, 39 (January 1975), 24-31.
Jacob Jacoby, Jerry C. Olson and Rafael A. Haddock, "Price, Brand Name and Product Composition Characteristics as Determinants of Perceived Quality", Journal of Applied Psychology, 55 (December 1971), 570-579.
Gisela Jellinek, "Introduction to and Critical Review of Modern Methods of Sensory Analysis", Indian Journal of Nutrition and Dietetics, 1 (1964), 219-260.
Roger E. Kirk, "Experimental Design Procedures in the Behavioral Sciences", (Belmont, CA: Wadsworth Publishing, 1968).
De Lee Lanz and Volney Stefflre, "Language and Cognition Revisited", Journal of Abnormal and Social Psychology, 69 (November 1964), 472-481.
John Neter and William Wasserman, Applied Linear Statistical Models, (Homewood, Ill.: Irwin, 1974).
Jerry C. Olson, "Price as an Informational Cue: Effects on Product Evaluations", in Consumer and Industrial Buying Behavior, Arch Woodside, Jagdeth Sheth and Peter Bennett (Eds.), (Amsterdam: North Holland, 1977), 267-286.
Roger J. Pennell and Forrest W. Young, "An IBM system/ 360 Program for Orthogonal Least-Squares Matrix Fitting", Behavioral Science, 12 (March 1967), 165.
B. P. Shapiro, "The Psychology of Pricing", Harvard Business Review, 46 (December 1968), 4-16.
Henry Scheffe, "An Analysis of Variance for Paired Comparisons'', Journal of the American Statistical Association, 47 (1952), 381-400.
Volney J. Stefflre, "Market Structure Studies: New Products for Old Markets and New Markets (Foreign) for Old Products", in Frank M. Bass, C. W. King and E. A. Pessemier (Eds.), Application of the Sciences in Marketing, (New York: Wiley, 1968), 251-268.
Warren Torgerson, Theory and Methods of Scaling, (New York: Wiley, 1958).
Enzo Valenzi and Larry Eldridge, "Effect of Price Information, Composition Differences, Expertise, and Rating Scales on Product-Quality Rating", Proceedings of the 81st Annual Convention of the American Psychological Association, 8 (1973) 829-830.
Forrest W. Young and Warren Torgerson, "TORSCA, a FORTRAN IV Program for Shepard-Kruskal Multidimensional Scaling Analysis", Behavioral Science, 12 (November 1967) 498.