WHAT SENSORY CHARACTERISTICS DRIVE PRODUCT QUALITY? AN ASSESSMENT OF INDIVIDUAL DIFFERENCES

WHAT SENSORY CHARACTERISTICS DRIVE PRODUCT QUALITY? AN ASSESSMENT OF INDIVIDUAL DIFFERENCES

HOWARD MOSKOWITZ and BERT KRIEGER

Moskowitz Jacobs Inc. I6 Madison Avenue Valhalla, New York

Received for Publication November 9. 1992

ABSTRACT

Consumers differ dramatically in their sensory preferences forproducts. This paper assesses the relative importance of appearance, tastepavor and texture as characteristics that drive acceptance for a complex pie product. l l e paper classifies consumers as being driven by one or more sensory attributes in their judgment ofproduct qualily. To do so, each consumer’s liking rating is regressed against each separate attribute liking. The coefficient (M) of the linear equation ”Overall Liking = M (Attribute Liking) 4- B ” measures the relative importance of the sensory attribute as a driver of overall quality.

INTRODUCTION

Individual differences pervade product testing and sensory analysis. A century ago the German psychologists recognized that individuals reacted with different speeds to the same stimuli, and factored in these individual differences in the so-called “personal equation” (Boring 1942). These differences are most pronounced for hedonic reactions, and most different in taste and smell. Individuals differ more in what they like and dislike than in their sensory perception of product sensory characteristics. Furthermore, smell especially (and to some degree taste) is pervaded by these individual differences. What one person likes another may abhor. Kenneth (1924) found that there were no olfactory stimuli that were universally liked or disliked. The most universally liked, albeit not by everyone, was the fragrance of roses; the most universally disliked was the rank odor of carbon disulfide.

One could go on multiplying the reports in the scientific literature of individual differences, both in terms of odor and taste. However, do these differences transfer to food preferences? We know that the consumer marketplace comprises products of dramatically different composition, even within the same category. Clearly consumer behavior in the marketplace

Journal of Sensory Studies 8 (1993) 271-282. All Rights Reserved. “Copyright 1993 by Food & Nutrition Press, Inc.. Trumbull, Connecticut. 27 1

272 H . MOSKOWITZ and 9. KRIEGER

reaffirms the existence of these individual differences, both in terms of quality and in terms of intensity.

HIGHER ORDER ORGANIZING PRINCIPLES - DIFFERENCES IN SENSORY IMPORTANCE

With the growth of psychophysics, researchers began to venture beyond simple head-counts (% of panelists who liked versus who disliked a specific sensory stimulus). Psychophysical scaling allowed the researcher to map the relation between stimulus intensity and both perceived sensory intensity and degree of liking. Ekman and Akesson (1964) and Pangborn's (1970) pioneering research in individual differences showed that there were substantial inter- individual differences in the pattern relating overall liking of taste stimuli (sugar, salt, acid) to physical concentration. The patterns varied from linear up to linear down, to flat, to inverted U shaped (with an optimal concentration in the middle range).

The study of individual differences can go beyond flavor and intensity preferences, and consider the different weights that consumers assign to their sensory inputs. We know that overall acceptance is usually positively related to acceptability of appearance, taste/flavor and texture, respectively, for multi- modal products. Do all consumers show the same hierarchy of importance, however? For instance, is flavor more important than appearance? Is texture the least important?

In order to identify the drivers of acceptance or product quality, the researcher first has to identify an appropriate index of importance for each sensory attribute, and then estimate the value of that index for each consumer, on each sensory attribute. It then becomes possible to identify the important drivers, and to understand how these quality drivers differ across consumers.

Creating a Quality Index - The Linear Slope of "Liking"

A straightforward way to measure acceptance consists of creating a linear equation relating overall liking to attribute liking. As attribute liking increases overall liking should increase as well. The equation, is expressed as:

Overall Liking = M(Attribute Liking) + B

The slope of the linear equation (M) measures the relative importance of the sensory attribute. High values of M correspond to high importance. Low values of M correspond to low importance. For example, if M = 1.5, then we can conclude that (all other factors being equal) a 1 .O unit increase in attribute liking corresponds to a 1.5 unit increase in overall liking. In contrast, if the slope of

SENSORY CHARACTERISTICS DRIVE PRODUCT 273

M is low, such as 0.2, then we can conclude that (all other factors being equal) a 1.0 unit increase in attribute liking corresponds to a 0.2 unit increase in overall liking.

The coefficient or slope provides a nicely behaved index number. It shows the numerical change in liking, given a unit change in attribute liking. The coefficient is independent of the units of measurement if the same panelist assigns ratings for both the attribute liking and overall liking. For an individual-by-individual analysis the same panelist would have to assign both ratings, and so the analysis is performed on an individual level.

In many marketing research and sensory analysis papers a similar attempt is made to develop an index of relative importance, using, however, the correlation coefficient between overall liking (the key or criterion measure) and an attribute. Correlation analysis assumes the existence of a linear relation between two variables and measures the degree to which the two variables are, in fact, linearly related. The correlation coefficient can be identical, however, for two variables (e.g., liking of appearance versus liking of texture, both of which are correlated with the criterion attribute of overall liking). The slope of one linear relation may be substantially steeper than the slope of the other linear relation. For example, the slope of overall liking versus liking of appearance may be +0.2, whereas the slope of overall liking versus liking of texture may be +0.8. The correlation coefficient would be identical, as stated above. Yet the two slopes differ dramatically. One would conclude from the correlation analysis that both attributes, liking of appearance and liking of texture, are equally important. One would conclude from a slopes analysis that texture is substantially more important than appearance as a driver of overall liking.

Linear Versus Quadratic Analyses

Moskowitz (1981) suggested that the relation between sensory attribute level and overall liking be analyzed with a quadratic equation rather than a linear equation, and that many of the conclusions regarding "what is important to a food" (viz., quality drivers) are incorrect if the researcher uses a linear analysis. A linear analysis assumes the existence of a linear equation or straight line relation. The 1981 analysis was based upon the recognition that the relation between overall liking and sensory intensity level is not linear, but rather a quadratic function. Often it describes an inverted U shaped parabola, with an optimal level in the middle range. However, in the present case the analysis deals with two liking attributes - overall liking and attribute liking. It is to be expected that as attribute liking increases overall liking will increase as well. It is also to be expected that there is no optimal level of attribute liking, beyond which the optimal level of overall liking decreases. That is, as the attribute liking increases (all other variables held equal) we do not expect liking to

274 H. MOSKOWITZ and B. KRIEGER

decrease at all. Liking may decrease in some instances, possibly due to other confounding factors such as another attribute changing as well.

EXPERIMENTAL

Stimuli

The stimuli were 21 fruit pies, systematically varied along a variety of physical ingredients and processing conditions. The original purpose of the study was to optimize product acceptability. For the purposes of this paper we simply need to know that the products varied considerably on the appearance, taste/flavor and texture characteristics.

Panelists

The panel comprised 98 consumers, pie purchasers and eaters, who participated for two sessions, each lasting 3 h. The panelists were female, primary shoppers, ages between 21 and 54. Since the study had been originally commissioned for product evaluation, the consumer panel was selected to represent the US. Approximately 18-20 consumers participated in each of 5 cities distributed across the US. (New Jersey, Chicago, San Francisco, Dallas and Minneapolis, respectively). Consumers were paid $70.00 (US) for their participation.

Questionnaire

The questionnaire comprised a series of anchored scales. Since pies are complex, comprising crust and filling, the panelists first inspected the pies visually, and rated the relevant attributes for pie crust only (e.g., like the crust appearance overall, like the color of the crust, etc.). Panelists rated both the liking attributes and the sensory attributes (e.g., like appearance of the crust, and all similar liking ratings, followed by ratings of "perceived darkness of the crust", etc.). Within each major set of attributes (e.g., appearance), liking attributes always preceded sensory attributes. This was done to insure that the consumers would first provide an opinion of the product, and only later provide the relevant sensory characteristics.

After inspecting and rating the crust the panelist ate some of the product (at least 3 forkfuls), and rated the filling. Panelists rated appearance, aroma, taste/flavor and texture in order of attribute appearance (viz., so that all appearance questions preceded all aroma questions, etc.). Within each general sense modality (viz., appearance), panelists again first rated liking attributes and then rated sensory attributes.


With a complex product such as pie, involving many attributes, it is difficult to know where exactly to put the rating of "overall liking". That attribute is the single key criterion. Overall liking in this study was asked just after the panelist had eaten the 3 forkfuls of pie. The first question afterward was overall liking, which was asked even before liking of taste, etc.

All attributes were accompanied by an anchored, 0-100 scale. For liking attributes the scale ranged from 0 (with the anchor word "hate") to 100 (with the anchor word "love"). For sensory attributes (not dealt with here) anchor words were appropriate to the attribute (e.g., for sweetness, O=not at all sweet, I00 =extremely sweet).

The consumers who participated in the pie evaluations were untrained in terms of product description. In order to help them better understand the attributes, an orientation session was held at the start of the evaluations. The orientation session lasted 30 min. During that time each panelist was given a piece of pie as a "practice product". The panelist then went through an orientation exercise in which an interviewer told the panelist what each attribute meant. For the most part the attributes were straightforward and an explanation unnecessary. However, the orientation session did allow the panelists to ask questions if the panelist did not understand the meaning of an attribute. The panelist filled out a rating card which showed the scale values assigned to the test product. The panelist then handed in the rating card to an attending interviewer who asked the panelist a question about the ratings. This insured that the panelist had yet another opportunity to ask questions and clarify ambiguous attributes (at least in the mind of the panelist).

Testing Procedure

The panelists participated as a group of 18-20 individuals. Panelists did not know each other, even though they participated in the same session. The testing procedure was self-administered in terms of actual product evaluation. However, the test was conducted in "rounds" to ensure product control.

The evaluation procedure was modeled after the 4-hour test procedure described by Moskowitz (1985). Panelists had little trouble with the evaluation because it was choreographed in such a way as to allow orientation, product evaluation, and consumer-interviewer questioning. The questioning procedure consisted of the interviewer asking the panelist to substantiate two ratings on each product. Neither the interviewer nor the panelist knows the correct answer. However, the questioning procedure reinforces the fact that the evaluations are to be taken seriously. The procedure maintains panelists' cooperation and interest throughout the entire session.

Each panelist rated a total of 16 of the 21 pies over the 2 days, rating 8 pies per day. The session was set up so that each pie was evaluated in a "round"


lasting 20 min. The actual evaluation of the product lasted no more than 1-2 min. Panelists assigned liking and sensory ratings to the product in the order of the questionnaire. The remaining time was devoted to cleaning up, presenting the questionnaire to the attending interviewer, and then waiting until the next round began.

Each panelist followed a unique order of product evaluations, set up by a computer program. With so many products to evaluate relative to the number of people, it was only possible to insure that every pie was presented an equal number of times, and approximately an equal number of times in each of the 16 test positions. In product tests of this type where there are large numbers of products and relatively few consumer panelists, it is difficult to achieve any other balance (e.g., balance the order of appearance of all pairs of pies, so that each pie is preceded by every other pie an equal number of times). Those more balanced and perfect designs would require millions of panelists.

RESULTS AND ANALYSES

The basic results of these studies is a matrix of product x attribute x panelist.. Each number in the matrix is a rating by a panelist, of a product, on an attribute. Thus the data lend itself to aggregation across panelists, or to disaggregation.

The Slope of The Liking Equation

The key index in this study is the slope, M, of the liking equation relating overall liking to attribute liking. As stated above, this slope measures the relative importance of the sensory attribute (e.g., liking of appearance).

This study deals with individual differences in the relative importance of the attributes. Hence all analyses were done at the individual level. Although each consumer panelist did not evaluate all of the pies, the consumer did evaluate 16 pies. In virtually all cases this set of 16 spans a wide range of sensory profiles, so one can feel certain the consumer evaluated quite different products. This is an important caveat. The analysis here, done on an individual level, requires that to the best of the researcher’s ability the products differ from each other, so that some will be liked more than others. If the products are all very similar to each other, then there will not be a range of liking on which to erect an equation. The panelist will have constricted his or her ratings to a narrow range, making the creation of an equation impossible.

There is a clear distribution of slopes, as shown in Table 1 and Fig. 1A (for appearance liking versus overall liking), 1 B (for taste/flavor liking versus overall liking) and 1C (for texture liking versus overall liking). All of these attributes refer to the filling of the pie.


TABLE 1. STATISTICS FOR THE ACTUAL AND THE RELATIVE SLOPES*

(APPEARANCE, TASTEIFLAVOR, TEXTURE)

Actual Slopes

Appe-e Flawr Texture MinLmum -0.18. 0.16 -0.55 Mardmrrm 0.22 1.29 1.21 -ge 1.55 1.13 1.26 Mean 0.22 0.82 0.44 St dev 0.26 0.16 0.28

Relative Slopes

PAppearance PFlamr PTeldare Minianurn 0.00 0.36 0.00 Maximnm 0.34 0.85 0.56 Range 0.34 0.48 0.56

St dev 0.09 0.11 0.10 Mean 0.16 0.56 0.28

* Relative slopes range from 0 to 1 .OO. In the calculation of the relative slopes, only the absolute values were used. PAppearance = percent of the total slope contributed by appearance

Note: St dev = standard deviation Range = Maximum - Minimum

Figures 1A-IC clearly reveal that the slopes vary across consumers. Since each consumer acts as her own control, we have removed one source of this variation, viz., the scaling effect. The scaling effect refers to the effect of the size of numbers. If the slope is obtained from the relation between a reference or common X axis and individual ratings of overall liking, then we would expect to see a significant variation in the slopes, based simply upon the size of numbers selected by the panelist. Those panelists using a wide range of numbers would generate high slopes. Those panelists who are more conservative, and use a narrow range of numbers would generate low slopes. In this case, however, each panelist’s use of numbers is controlled by using the panelist’s ratings of two different liking attributes. Scale effects would thus balance out.


&parance Slopes (Raw Values)

a7 1

Flavor Slopes (Raw Values)

0.41 a $ 03- 6 02- / E

I -06 4.2 02 08 1.0 1A

m a $ 03- 6 02- E

I -06 4.2 02 08 1.0 1A

T e r n Slopes (Raw values)

FIG. 1. DISTRIBUTION OF SLOPES RELATING ATTRIBUTE LIKING (ABSCISSA) TO OVERALL LIKING (ORDINATE)

The analysis was done on an individual by individual basis. 1A - Slopes for liking of appearance versus overall liking 1B - Slopes for liking of taste/flavor versus overall liking 1C - Slopes for liking of texture versus overall liking


Creating An Indexed Slope By Percentaging

One of the recurrent questions in individual differences is the degree to which an individual pays attention to appearance versus flavor versus texture when making a judgement of overall quality or acceptability. These slopes help answer those questions. However, before concluding that one sensory attribute is more important than another, it is instructive to index each individual's data to a common scale. One way to do this is to calculate the percent of total liking attributable to appearance, flavor and texture, respectively. In this study the way to calculate that percent is to pool together the slopes for appearance, flavor and texture, for a single panelist, and to calculate the percent of that sum attributable to each sensory input. The calculation is straightforward:

Relative Importance = (Attribute Slope)/(Sum Of 3 Attribute Slopes)

In this analysis each panelist generates 3 numbers which add to 1.00. The closer the 3 numbers lie to 0.33 each, the more the panelist weights each sensory input equally. The closer any number lies to 1 .OO, the more important is that sensory attribute for driving overall liking, and therefore the less important the other sensory attributes will be.

The data can be plotted on triangular coordinates, as they are in Fig. 2. Each point in Fig. 2 represents a single individual. The plots reveal clearly that most individuals pay great attention to flavor. Appearance and texture are significantly less important to consumers as a whole. There are no "texture oriented" consumers in this category, nor any "appearance oriented" consumers. For the most part these sensory inputs play a role, albeit quite a small one.

Sensory Input Homogeneity Versus Heterogeneity

Another way to look at these individuals differences is to assess the degree to which a person pays equal attention to all 3 senses, or concentrates on 1 or 2 senses. Having created a percent statistic for the slopes, we can now compute the standard deviation of the 3 percents. The lower the standard deviation the more the consumer weights all 3 inputs equally. If the consumer weights all 3 inputs exactly equally, then the standard deviation for the 3 slopes will be 0. The higher the standard deviation the more the consumer differentially weights one slope over the other two.

Figure 3 shows the distribution of these standard deviations. It is clear in the population that there is a range of consumers, from those who pay attention to all 3 sensory inputs to those who pay attention only to 1 input. It is important, however, that the consumers never pay equal attention to all 3 sensory inputs. The standard deviation never reaches 0.

280

3 m- E Q2-

ii

E

I?’-

H. MOSKOWITZ and B. KRIEGER

1 I I I

FIG. 2. TRIANGULAR PLOT OF THE RELATIVE SLOPES FOR APPEARANCE, TASTE/FLAVOR AND TEXTURE, RESPECTIVELY

The 3 relative slopes add to 1.00 or 100%. Each point represents a different consumer. The closer a point lies to a vertex the more important is that sensory input as a determiner of overall liking. Note that most of the points (viz., consumers) lie near PFlavor (percentage tastdflavor), implying that for most consumers flavor is key. No consumer shows a relative slope near either PAppearance or PTexture, indicating that appearance and texture acceptance by themselves are insufficient to drive

overall liking, although taste/flavor acceptance is sufficient by itself.

FIG. 3. DISTRIBUTION OF THE STANDARD DEVIATIONS OF THE RELATIVE SLOPE When the standard deviation is 0, then all 3 sensory inputs are equally important as drivers of overall liking. As the standard deviation increases, the consumer pays differential attention to the

sensory attributes as drivers of liking.


The Range of Relative Slopes

This analysis looks at the range of the relative slopes or percent values. Across the consumer population what is the maximum and minimum for each slope? For instance, are there individuals who look only at appearance, texture, or flavor, respectively, and ignore all other sensory inputs? Table 1 shows the statistics for each relative sensory input. It is clear from Table 1 that most consumers make their liking decision primarily, but not exclusively, on the basis of flavor. No consumer makes any liking decision primarily on the basis of texture or appearance. These seem to be secondary. Finally, it appears clear that there is a limited range of importance for appearance. It can range from 0% to approximately 33 % of the decision. Texture shows a similar pattern, albeit with a more powerful role in determining acceptance. The relative texture slope ranges from a low of 0 (where it is irrelevant to the task) up to a value of 0.50 (meaning that it contributes 50%). Finally, taste/flavor is critical. It is never 0%. It ranges from virtually 100% down to a low of 30%.

A Taxonomy for Consumers

Based upon the size of the relative slopes, there are 6 patterns or orders jviz., 3! or 6 orders, e.g., Taste/Flavor > Texture > Appearance). Given this data we can classify each panelist as belonging to one of the 6 patterns. The data shows only 3 patterns, and only two of those patterns are represented by a large number of consumer panelists. The most common pattern is: Flavor Slope > Texture Slope > Appearance Slope (n=73 of 98 panelists). The second most common pattern is: Flavor Slope > Appearance Slope > Texture Slope (n=23 of 98 panelists). The third pattern is Appearance Slope > Flavor Slope > Texture Slope (n=2 panelists). This distribution clearly indicates the overwhelming importance of flavor as the key criterion for consumer judgments of product quality, followed by texture, and then followed by appearance. The distribution also suggests that individuals differ in the relative importance that they place on the sensory inputs. Approximately 3 of 4 consumers feel that texture is more important than appearance, whereas 1 of 4 consumers feel that appearance is more important than texture.

CONCLUSIONS

Specific Results and Conclusions

This paper demonstrates that consumers differ dramatically in the aspects of a product to which they attend. The results further suggest that it is taste/flavor that always provides the most important sensory input to this product. Furthermore, the results suggest that there are 2 types of consumers -


one who pays primary attention to taste and flavor to the exclusion of other attributes, and another who divides her attention among multiple sensory inputs to make a decision. Finally, the data suggests that there is no consumer who disregards taste or flavor entirely, and pays attention to appearance and texture.

Directions for Future Research

The assessment of multiple products by single consumers, who use a variety of scales, opens up new possibilities for understanding individual differences. Researchers can now begin to investigate such issues as the changing importance of sensory inputs as a function of product, consumer health and satiety stated, experience with the category, etc. By having the consumer rate a variety of products in the same category it becomes straightforward to identify what sensory inputs drive consumers, as well as to classify the consumers into categories depending upon the sensory inputs that are most important to them.

Over the past 50 years (and in some cases even longer) practitioners in various fields connected with food quality have argued about grading systems. Grading allows the manufacturer to classify the products into different categories of quality, with the ensuing implication for cost. In many food categories the grading procedure has been arbitrary, or at best an inheritance from previous generations of practitioners who themselves created an arbitrary system. This paper presents an approach that enables the manufacturers to identify the hierarchy of characteristics in a product, and assign them at least a rank or at best a percentage value for their importance in driving perceived product quality.

REFERENCES

BORING, E.G. 1942. Sensation And Perception In The History Of Experimental Psychology. Appleton Century Crofts, New York.

EKMAN, G. and AKESSON, C.A. 1964. Saltiness, sweetness and preference: A study of quantitative relations in individual subjects. Rep. 177, Psychological Laboratories, University Of Stockholm.

KENNETH, J.H. 1924. Some experiments on mental reactions to odors. Perfumery & Essential Oil Record, 84-87.

MOSKOWITZ, H.R. 1981. Relative importance of perceptual factors to consumer acceptance: Linear versus quadratic analysis. J . Food Sci., 46,

MOSKOWITZ, H.R. 1985. New Directions In Product Testing And Sensory

PANGBORN, R.M. 1970. Individual variations in affective responses to taste

244-248.

Evaluation Of Food. Food & Nutrition Press, Trumbull, Conn.

stimuli. Psychonomic Sci., 21, 125-128.

WHAT SENSORY CHARACTERISTICS DRIVE PRODUCT QUALITY? AN ASSESSMENT OF INDIVIDUAL DIFFERENCES

Documents