Consumer Emotion Space: an Investigation of Semantic Space and Context Effects in Self-Reported Emotion Elicitation

Ross Buck, University of Connecticut
Mats Georgson, University of Connecticut
[ to cite ]:
Ross Buck and Mats Georgson (1997) ,"Consumer Emotion Space: an Investigation of Semantic Space and Context Effects in Self-Reported Emotion Elicitation", in NA - Advances in Consumer Research Volume 24, eds. Merrie Brucks and Deborah J. MacInnis, Provo, UT : Association for Consumer Research, Pages: 431-437.

Advances in Consumer Research Volume 24, 1997      Pages 431-437


Ross Buck, University of Connecticut

Mats Georgson, University of Connecticut


Consumer research strives to obtain information about how we perceive products and services, and how we make decisions about consumption. One useful source of information is the notion of product involvement. Involvement can be conceptualized as the amount of effort that goes into thinking about something, analytically (how much we analyze it) and emotionally (how much we feel about it). In addition to the above general involvement types, the affective involvement can be studied by addressing specific emotions. For instance, use of a car may be associated with feelings of power, whereas long-distance phone services induce feelings of attachment and loneliness.

Part of the beauty with studying emotions is that they are general responses that occur in a variety of contexts. For a given person, we can use analytic and affective involvement, along with specific subtypes of such involvement, to differentiate and compare product categories or brands (or even products within a brand). Information about affective involvement can be useful in formulating advertising strategies for segmentation and positioning. Unfortunately, the scientific community is still not very successful at pinpointing the relation between affective responses in consumption contexts and subsequent consumption evaluations and behaviors, although we can see that a link exists (Batra and Ray 1986, Holbrook and Batra 1987, Edell and Burke 1987, Swinyard 1993, see also Babin, Drden and Griffin, 1994).

Affective responses have been studied at least as far back as Aristotle. In this century, science has identified a neurological basis of affect (e.g. Buck, 1988). However, most often when we measure affect we rely on self-report measures. In doing so, we inevitably get some analytic thought with the measurement, as you need to use analytic thought to perform introspection and then answer a verbal question. In fact, we use a semantic space, different words covering different parts of an area of meaning, to report what we feel (Osgood, Suci and Tannenbaum 1957). Shaver, Schwartz, Kirson, and O’Connor (1987) conducted a classic study of the semantic space of emotion in general, i.e., emotion unanchored to any specific context. In this paper, Shaver et. al.’s semantic space will be compared to a semantic space derived from data elicited in a consumption context. The question is whether self-report of emotion is a stable measurement device. In other words, is it the case that "emotion is emotion is emotion", regardless of what the context is? Ideally, what we call happiness in normal day usage should be the same happiness when it comes to a consumption context. It should be as different from fear as it is in the normal sense of the word, as close to joy, and so on. On the other hand, if semantic space is different when it comes to the consumption setting, we must reconsider such things as what measurement scales to use and what literature is generalizable to the consumption context. As will be shown in the theory section below, even though it seems like emotions are universal and that there are prototypical emotions, there still might be reason to question whether emotions invoked in a consumer context have the same profiles as emotions in other contexts.

In this study results of two studies, one done by Shaver et. al. (1987) of emotion in a general context and one done by this author concerning emotion in a consumption context will be reported and compared, to determine whether they agree with each other.


In the long history of studying emotion, approaches vary widely depending on the perspective chosen. It can be useful to start with asking a rhetorical question: why do we have emotions?

MacLean (1990) has proposed the theory of the Triune brain, which one can use to both understand the neurological bases of emotions as well as their functions. The human brain evolved much like an old city, where old buildings are still kept in operation even though more recent, and maybe more sophisticated buildings are added in layers around them. Thus, there is a reptilian brain (evolutionary the oldest), primarily the internal capsule and amygdala; surrounded by a paleomammalian brain, the limbic system; in turn surrounded by a neo-mammalian brain (newest), the neo-cortex. Physiologically, a reptile brain has much in common with the amygdala, while a dog brain is similar to a human brain minus much of the neo-cortex.

The reptilian brain houses what Buck (1988) labels reptilian emotions of "raw" sex and aggression; the paleomammalian limbic system individualistic-limbic emotions such as anger and fear; prosocial-limbic emotions of attachment; and social emotions like love, pride, guilt, etc. The neo-mammalian neo-cortex is associated with cognitive emotions such as curiosity and boredom. Again, there are evolutionary rationales for these emotions. Reptiles are typically only concerned with fight-or-flight, territoriality, and reproduction for individual and species preservation. Mammals on the other hand need social emotions to maintain the social structure necessary to protect the young until they can fend for themselves. In humans, the cognitive emotions help utilize analytic capabilities. [Brains are "gas-guzzling" organs: the tissue requires lots of energy for upkeep. Therefore, big brains are only appropriate in beings that can utilize them, and by basic evolutionary rules should not evolve in species that don't need to become "smarter". Scientists are still only speculating why dolphins would need brain sizes that equal ours: we don't understand what interactions of motivation, emotion and cognition might be particularly crucial to them.]

There is much confusion in the emotion literature about the definition of emotion. In this work we are addressing the phenomenon of an internal bodily readout associated with the subjective experience of feelings and desires. This is the introspective awareness of emotional arousal.

The importance of the this distinction is that emotion does typically, but not necessarily, involve a fixed proportion of different readouts. In a dark alley, you might meet a hoodlum who tries to rob you by intimidation. Your heart starts pounding and you start sweating and you feel very scared (and this is what we mean by emotion here) but you do your best not to show the signs of fear. A Samurai warrior in the same situation might show some readout of heartrate increase, show an angry face, but might not be able to introspect any subjective emotional experience (here, no emotion).


One school of thought utilizes unidimensional definitions of emotion. These are primarily concerned with evaluation (e.g. Fishbein and Ajzen 1975). One example of such an affect scale constructed by Holbrook (1981) measures affect in such a general evaluative sense, namely the extent to which the subject perceives some stimulus to be "good". He conceives of arousal as being a separate construct, as well as other special constructs like dominance and pleasure (e.g. Holbrook et. al. 1984).



Multidimensional definitions of emotion typically include at least two dimensions: Evaluation (i.e. good-bad) and Intensity (i.e. how vivid) (Shlosberg, 1952; Shaver et. al. 1987). In a similar fashion, Russell (1979, 1980, 1991), Mano (1991), and Roberts et. al. (1994) call these dimensions Pleasantness and Arousal. Russell (1979, 1980, 1989, 1991) argues that affect is a continuous phenomenon, described by a circumplex model. Emotions would be located on the circumference of a circle, with happiness and sadness opposing each other, and emotions blending over into each other so that a full circle will cover all possible emotions.

Opposing this circumplex view is the theory of primary affects. As early as 1650, Descartes claimed that there are six primary "passions" (Batra and Ray, 1986). Later, Darwin (1872) argued that since facial expressions seem to be inherent in humankind and therefore universal, they should be indicators of corresponding internal emotional states (see also Ekman and Izard 1972). Tomkins (1962, 1963) argues that these affects stem from innate neural structures. This set of biologically based basic emotions does also blend into other, secondary emotions, like mixing primary colors make new colors. This view is supported by approaches where the semantic space of emotion has been mapped; the results show correspondence with proposed primary affects, although the solutions in all fairness also look roughly circumplex. Thus, the theories are not diametrical to each other, and both can be seen as correct (Buck 1985).


Self-report measures of emotion have shown convergent validity with physiological measures, such as facial expressions (Westbrook 1987). Still, there are fundamental differences between the measures. What is the correspondence between physiological aspects of emotion and a self-report item, like a Likert scale? When we say that we feel happy, the relation of the word "happy" to the neurological activation of the "happiness" center in the brain is of course a symbolic one, just like the word "tree" has a purely symbolic relation to real trees out in the park. What does this imply?

Given that emotions have a neurological basis, we can draw an analogy to self-reporting emotional states from the study of color perceptions. A human can see light with a wavelength of approximately 0.000035 cm (violet) and above. Thereafter, thelight will be perceived as blue, green, yellow, orange and finally red, at a wavelength of 0.000075 cm. But note that the labels of this progression are arbitrary language constructs. There is nothing in the physical quality of light that says exactly where "blue" starts and ends.

There is an African people who only differentiate between dark and light color. The ancient Romans did not differentiate between blue and green color (Eco, 1993). On the other hand, the Mongols of the golden horde reportedly had about 20 terms for different brown hues in horses, as the exact hue of one’s mount was an important symbol of status.

A culture will also use the native language to describe the experienced emotion. There is an important difference between colors and emotions though, and that is the concept of prototypicality, mentioned earlier. Prototypical emotions should show up as clusters that go together with space between these clusters. Thereby, we should see the "footprints" of the neurological responses in semantic space.


Consumption emotion has been used to refer to the set of emotional responses elicited specifically during product usage or consumption experiences. Consumption emotion has been interpreted as highly accessible affect elements that influenced post-consumption evaluations (Westbrook and Oliver 1991, Cohen and Areni, 1991).

This consumption emotion can be described either by the distinctive categories of emotional experience and expression (e.g. joy, anger, and fear) or by the dimensions underlying emotional categories, such as pleasantness/unpleasantness, etc. (Russell 1979, Westbrook et. al., 1991). These different perspectives correspond roughly to dimensional and prototypical definitions of emotion, respectively (as discussed earlier).


There are actually two kinds of context effects involved in comparing the dimensionality of two sets of responses. First, there is a general context effect, as reported by Roberts and Wedell (1994). Here "context" refers to the set of stimuli being analyzed for proximity.

In four experiments the aforementioned researchers investigated the effects of giving subsets of emotions for similarity sorting, so that one group might get a general set of emotions, while others might get a set of anger and fear items. Judgments were found to differ. This means that measurement items should represent the full range of emotions for the analysis to be valid.

The other context effect is the nature of emotion in the consumption context. One example is Edell and Burke (1987, 1989) who tested 169 feelings for appropriateness in relation to advertising. Later, 69 items were used for self-reports. Through factor analysis, they ended up with two and three factor solutions, that they labeled Evaluation, Activity, and Gentleness respectively. Mano (1991) compared self-reports of emotion from a lecture (as a neutral context) to self-reports of emotions experienced while watching TV ads (as a consumer context). The responses were analyzed through multidimensional scaling, factor, and cluster analyses for both conditions. Mano was unclear as to exactly how he compared the general and the consumer context MDS solutions; he mentions that the two solutions resemble each other. [One of the MDS maps must be rotated to match the axes of the other, though. For a formal treatment on this, see the Methods section on MDS comparison. It appears like Mano's matching was more of a face comparison than a formal procedure.] As Mano points out, one could argue that all emotional states are induced by some external stimulus. [Furthermore: as attribution literature tells us, even if we don't know what induced for instance a general feeling of arousal, we might attribute it to some arbitrary stimulus around us.] Therefore, it is questionable whether a "common everyday event" isn’t a mood inducing stimulus, or at least a context. Therefore, the control is not a truly context-free baseline.


For investigating differences between the general context and the consumer context, we will use two sets of data, as follows:

The first set of data is derived from an analysis by Shaver et. al. (1987). These researchers approached the issues of prototypicality and/or circumplexity of emotion through mapping emotion using a collection of similarity data. They provided subjects (one hundred students from an introductory psychology course) with a stack of 135 cards, each of which had an emotion term on it, such as "joy" or "anger". The task was to sort these cards into a number of piles, [The actual number of piles was up to the individual subject.] effectively creating an accessible representation of the subjects’ cognitive clustering of affect. This similarity data was then used for a hierarchical cluster analysis (which will only be summarily reported here) as well as a multidimensional scaling of emotion space.

The second set of data was collected by the authors as follows. 94 undergraduate students in an introductory communications course were asked to fill out five randomly selected questionnaires, each questionnaire pertaining to a certain product category. Thus, each product was evaluated by an average of 26.7 raters. 17 product categories were chosen for this study to represent a full range of product categories, so that a representational variety of combinations of high and low emotional and analytic involvement would be used. This selection was based on Buck and Chaudhuri (1994). The categories were: Color TV’s, Headache Remedies, Candy, Automobiles, Household Cleaners, Long distance phone services, Facial Tissues, Laundry Detergents, Toothpaste, Air Travel, Soft Drinks, Credit Cards, Bath Soap, Insurance Plans, Personal Computers, Greeting Cards, and Canned Beer. Arguably, all of these would have some relevance to the sample. Moreover, the aforementioned study of product categories used a similar sample.

All in all, a total of 455 valid data sheets were properly completed and collected. The questionnaire asked subjects whether usage of a product increased each of 53 emotions listed in the questionnaire. In addition, they were asked if using the product decreased their feelings of 28 of the emotions. (To not completely overwhelm the subjects, all 53 emotions were not duplicated in decrease items, the decrease items were concentrated on the decrease of negative emotion.)

The data thus represents college students’ emotional responses to different product categories.


The cluster analysis comparison and the multidimensional scaling comparisons are fundamentally two similar ways of comparing the data sets, both have their respective advantages. Cluster analysis bluntly states what the proposed cluster solution is in terms of cluster memberships, while the MDS allows for visual inspection and analysis of underlying dimensionalities and relations between clusters.

The Shaver data has no emotion eliciting stimulus. If it had, it would automatically have a context. The consumer data, on the other hand, must be elicited by thinking about consumption or it would be context free. Standardizing the measures provided the basis for comparison.

A hierarchical cluster analysis was performed, replicating the cluster analysis by Shaver et. al. The purpose of this cluster analysis is to see whether the clusters around prototypical emotions hold up in the consumer context. Shaver et. al. presented a six-cluster solution: Anger, Sadness, Surprise, Love, Happiness, and Fear.



In the consumer context, a six cluster has a pattern of general similarity with two notable differences (as discussed earlier). First, the Anger and Sadness clusters merge into a general "Unpleasant" cluster. Second, Surprise is incorporated into the Love cluster. Instead, we get two new clusters, Pride and Interet. Note, however, that the two new clusters seem to form around emotions that were not included in Shaver’s study, namely Pride- and Interest-related emotions.

Multidimensional scaling analysis

The overall measure of fit in Multidimensional Scaling is the stress statistic. The higher the stress, the more variance that is not accounted for by the MDS solution at that dimensionality. The stress for Shaver’s solution (with two dimensions) was .10. [The study by Mano (1991) reports Stress values of .11 for both "lecture" and "TV ads" context MDS solutions.] The consumer context solution held a slightly worse fit, .15. The consumer context solution looks like this (See Figure 2).

Comparison of two multidimensional representations must start with establishing commonality between them. In order to do so, we must fit our map to Shaver’s. This can be thought of as juxtaposing our map onto the Shaver representation. A computer program was written explicitly for this purpose (Georgson 1994). To accomplish its task, the computer program works according to the following principles:

Best fit: every operation necessary must be made to minimize errors, Ordinary Least Squares (OLS) error between every corresponding point.

Correspondence: only emotion items that occur in both maps can be used for the fitting procedure, i.e. every item needs a counterpart.

Dimensionality: Shaver et. al. fit their data to two axes, evaluation and intensity. Therefore, let Shaver’s map be fixed, and adjust ours to fit it.

Fitting procedure: there are three steps in finding the best fit: Polar adjustment: by rotating our map in increments, we will see the optimal polar alignment of our map. This procedure is done by rotating in small increments, calculating the sum of squared error terms, until a full circle has been completed. The best solution found in the entire circle will then be implemented. Origin matching: once the rotation is completed, it is necessary to move the center of our map to the center of Shaver’s. This is done by first moving along the X-axis and then along the Y-axis, looking for least SS errors. Stretching/shrinking: by enlarging or shrinking our map along x and y axes proportionally and simultaneously, we will adjust our map scale to that of Shaver.

The resulting coordinates are shown in figure 5. The first set of coordinates are the Shaver coordinates and describe the location of the emotions in that study, while the Consumer coordinates describe the location of the emotions fit onto that map. The residuals are the differences in the x and y dimensions, respectively. The true residual is the Pythagorean distance between the Shaver and the consumption position. Correlating coordinates [Mano (1991), as mentioned before, does not seem to have performed any procedure of optimizing correspondence by juxtaposing the second MDS on the first. Thereby, quantitative results are hard to interpret. The author reports a highly negative correlation of coordinates (x1-x2: -.77, y1 to y2: -.76) as indicating good fit. If the axes were not perfectly aligned, this would mean underreporting the true correspondence.] between the two solutions gives us a high correlation of .81 for the Evaluation dimension, but a -.04 on the Intensity dimension. Obviously, evaluation held up well, in fact, and intensity did not.



A univariate regression of Shaver’s Evaluation score to the consumer Evaluation score resulted in a b-weight of .58. This regression is significant at p<.001 level, with an R2=.67.

Looking at the squared residuals bar chart, figure 5, we see that there are seven emotions that stand out as having worse fit than the others, namely Tenderness, Lust, Love, Pride, Affection, Hope, and Surprise. Quite strikingly, all of these emotions show dramatic decreases in evaluation, which means that they are less positively evaluated. Surprise was also less intense. Note on Figure 1 how Surprise is clearly the most intense emotion in Shaver’s representation.

By visual inspection of the MDS solutions, we see clearly how the unpleasantness cluster forms as an undifferentiated mass in the consumer context.


Overall, there is a very good correspondence between the two studies, considering their large differences in methodology. In the consumer context, certain cluster compositions differed from the general context. Specifically, two new clusters form around pride and interest. Cluster memberships differed the most on negatively evaluated emotions. Specifically, anger and sadness merge into a general unpleasantness cluster. Emotions can be more or less informative in a consumer context. For instance, negative emotions seem to have little differentiation between them.

A given emotion seems to be evaluated less severely in a consumer context, i.e. there is a limit to how good or bad consumption typically make you feel compared to life in general. Emotion intensity in consumption seems unrelated to emotion intensity in a general context. Intensity therefore seems to be more context/object dependent. Love-related emotions showed the largest difference in semantic space between the two contexts.


The findings described here again underscore the complexity of the phenomena of emotion. On one hand, we can not ignore the above indications that there seem to be context effects on the semantic space of emotion. On the other hand, there is also considerable stability present, especially along the evaluative dimension.

It is, however, quite apparent that the overall picture holds up quite well. We have compared two sets of data, collected with different measurements, in response to different tasks, and still arrived at quite similar results. Whether the differences stem from methodological differences or actual context effects is uncertain. We should therefore attempt to explain the differences through substantive interpretation. For example, pride is one of the misfit items. Maybe pride really has to be focused on the self (accomplishments, virtues) to be really positive and intense.

Apart from context effects, we should keep in mind that we have been comparing the emotion patterns of undergraduate students. Therefore, we see the effect of a marketing context in this particular population segment. Now, there is probably no reason to expect differences across population segments on a context-free similarity-sorting task like Shaver’s. As long as the subjects vocabulary adheres to normal standards, I could probably get the same results regardless of whether I used airline pilots, nuns, or prison inmates.

On the other hand, if there is context instability, it could imply that language connotations and social norms might intrude on this neutral assessment. If I would get the aforementioned sample of nuns, they might quite honestly indicate very low scores for any consumption emotion. Future research should continue to investigate stability issues. This could be studied across different population segments, different consumer contexts (watching ads, brand experiences, product category experiences, etc.).

Even when progressing on the issue of emotion stability, there is the issue of emotion category fidelity. We notice that the negatively evaluated emotions seem to be perceived as quite similar to one another in the consumer context. Well, the answer seems in hindsight to be quite obvious: there are no markets for products strongly associated with negative emotions. Probably, consumer motivation is concentrated on hedonic emotional experiences. Negative emotions might on the other hand be tapping into to the unidimensional valence of emotion so often used in persuasion studies (e.g. the scale by Holbrook 1981)

On the topic of relevant emotional experiences, Buck et. al. (1994) mentions in discussing the advertising context that a specific ad might be a clear example of a certain emotional type, as exemplified by a Calvin Klein ad (Reptilian) and an AT&T ad (Prosocia-limbic) Therefore, it would be useful to construct more elaborate subscales of a measurement instrument to be used on ads that invoke a first indicator. For instance, if the Calvin Klein ad makes you feel sexually aroused, then a subset of reptilian affect indicators would be used.

Indeed, using standardized measurement scales covering the whole range of human emotion can be highly inappropriate. Consider an ad for charity showing starving children coupled with a response item for sexual desire... Incidentally, focusing on theoretically founded subsets of emotions might provide better consistency across studies. An important function of research on emotion in consumer research is to establish what is useful, as well as appropriate, for different audiences and topics.

In conclusion, it seems like emotion is something that holds up rather well across methods and contexts. Still, taking context differences into account might help create measurements that are more appropriate for the context of interest. If we have market research responses and there seems to be little differentiation between unpleasant emotions, we need not waste measurement items on those affects, instead concentrating efforts where we might find differences.

Emotion is one of the potentially most powerful carriers of information about human motivation. Continuing the efforts to establish valid, reliable, and economical measures of emotions is a good intellectual investment for both social and natural scientists.


Aaker & Bruzzone (1981). Viewer perceptions of prime-time television advertising. Journal of Advertising Research, 21 (5), 15-23.

Aaker, Stayman & Hagerty (1986). Warmth in advertising: Measurement, impact, and sequence effects. Journal of Consumer Research, 12 (4), 365-381.

Agres, S. J., Edell, J. A., and T. M. Dubitsky (ed.) (1990). Emotion and Advertising. New York: Quorum Books.

Babin, Barry, Darden, William, and Griffin, Mitch. (1994) Work and/or fun: Measuring Hedonic and utilitarian shopping value. Journal of Consumer Research, vol. 20, 644-656.

Batra, Rajeev, and Ray, Michael L. (1986) Affective Responses Mediating Acceptance of Advertising. Journal of Consumer Research. Vol. 13, September, p. 234-249.

Buck, Ross, and Chaudhuri, Arjun. (1994a). Affect, Reason, and Involvement in Persuasion: The ARI model. In P. Weinberg (ed.), Konsumentenforschung. Mnnchen: Verlag Franz Vahlen.

Buck, Ross, Chaudhuri, Arjun, Georgson, Mats, and Kowta, Srinivas. (1994b) Conceptualizing and operationalizing affect, reason, and involvement in persuasion: The ARI model and the CASC scale. Presented at Annual Advances in Consumer Research Conference 1994, Boston, MA.

Buck, Ross. (1984). The communication of emotion. New York: Guilford Press.

Buck, Ross. (1988) Human Motivation and Emotion. New York: Wiley.

Burke, Marian C., and Edell, Julie A. (1989) The Impact of feelings on ad-based affect and cognitions. Journal of Marketing Research, 26, February, 69-83.

Chauduri, Arjun, Buck, Ross. (1994) Are advertisers using brain theory? Introducing the CASC scale. In C. W. Park & D. Smith (eds.), Marketing Theory and Applications. Vol. 5. (pp. 193-198). Chicago, IL: American Marketing Association.

Cohen, Joel B., and Areni, Charles S. (1991) Affect and Consumer Behavior. In Handbook of Consumer Theory and Research, ed., Thomas S. Robertson and Harold H. Kassarjian, Eaglewood Cliffs, NJ: Prentice Hall, 188-240.

Eco, Umbrto. (1993) Misreadings. San Diego : Harcourt Brace & Co.

Edell, J. A., and Burke, M. C. (1987) The power of feelings in understanding advertising effects. Journal of Consumer Research, 14, 421-433.

Ekman, P., Friesen, W. V., Ellsworth, P. (1972) Emotion in the human face. Elmsworth, NY: Pergamon Press.

Fishbein, M., and I. Ajzen (1975). Belief, attitude, intention, and behavior : an introduction to theory and research. Reading, Mass. : Addison-Wesley Pub. Co.

Georgson, Mats. (1994). MDSFIT: A MDS representation comparison program. Unpublished computer program, University of Connecticut.

Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton Miffin.

Gorn, G. J. (1982). The effects on Music in Advertising on Choice Behavior: A Classical Conditioning Approach, Journal of Marketing, 46, 94-101.

Haley, R. I., Staffaroni, J., and A. Fox (1994). The missing measures of copy testing. Journal of Advertising Research, 34, 46-60.

Hecker, S., and D. W. Stewart (ed.) (1988). Nonverbal Communication in Advertising, Lexington, Mass.: Lexington Books.

Holbrook, Morris B. (1981) Integrating compositional and decompositional analyses to represent the intervening role of perceptions in evaluative judgments, Journal of Marketing Research, 18, February, 13-28.

Holbrook, Morris B., and Batra, Rajeev. (1987). Assessing the role of emotions as mediators of consumer responses to advertising. Journal of Consumer Research, 14, 404-420.

Holbrook, Morris B., Chestnut, Robert W., Oliva, Terence A., and Greenleaf, Eric A. (1984). Play as a consumption experience: the roles of emotions, performance, and personality in the enjoyment of games. Journal of Consumer Research, 11, September, 728-739.

Izard, C. E. (1971). The face of emotion. Apple-Century-Crofts.

Izard, C. E. (1972). Patterns of emotions. Academic Press.

Izard, C. E. (1977). Human emotions. New York: Plenum Press.

Kruskal, J. B., Wish, M. (1978). Multidimensional Scaling. Sage University Paper Series on Quantitative Applications in the Social Sciences, no. 11. Beverly Hills, CA: Sage Publications, CA.

MacLean, P. D. (1990). The triune brain in evolution: role in paleocerebral functions. New York: Plenum press.

Mano, Haim. (1991) The structure and intensity of emotional experiences: Method and context convergence. Multivariate Behavioral Research, 26 (3), 389-411.

Microsoft (R) Encarta. Color. Copyright (c) 1994 Microsoft Corporation. Copyright (c) 1994 Funk & Wagnall’s Corporation.

Osgood, C. E., Suci G. J., and P. H. Tannenbaum (1957). The measurement of meaning. Urbana, University of Illinois Press.

Petty, and Cacciopo (1981). Attitudes and Persuasion: Classic and Contemporary Approaches, Wm. C. Brown.

Petty, Cacciopo, Sedikides, and Strathman (1988). Affect and persuasion: A contemporary perspective. American Behavioral Scientist, 31, p. 355-371.

Petty, Ostrom & Brock (1981). Cognitive Responses in Persuasion, Lawrence Erlbaum Associates.

Roberts, James S., Wedell, Douglas H. (1994) Context effects on similarity judgments of stimuli: inferring the structure of emotion space. Journal of Experimental and Social Psychology, 30, 1-38.

Rossiter, John R., and Percy, Larry. (1987) Advertising and Promotion Management. NY: McGraw Hill.

Russel, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39, 1161-1178.

Russell, J. A. (1979). Affective Space Is Bipolar. Journal of Personality and Social Psychology, no. 37, 345-356.

Russell, J. A. (1989). Measurs of emotion. In R. Plutchik and H. Kellerman (Eds.). Emotion: Theory, research, and experience (Vol. 4., 83-111). San Diego: Academic Press.

Russell, J. A. (1991). Culture and the categorization of emotion. Psychological Bullentin. 110, 426-450.

Schlosberg, H. (1952). The description of facial expression in terms of two dimensions. Journal of experimental psychology, 44, 81-88.

Shaver, P., Schwartz, J., Kirson, D., and O’Connor, C. (1987) Emotion Knowledge: Further Exploration of a prototype approach. Journal of Personality and Social Psychology, vol. 52., no. 6, 1061-1086.

Tomkins, S. (1962) Affect, imagery, and consciousness: The positive affects, Vol. 1. New York: Springer.

Tomkins, S. (1963) Affect, imagery, and consciousness: The negative affects, Vol. 2. New York: Springer.

Tucker (1981). Lateral brain function, emotion, and conceptualization. Psychological bulletin, 89, 19-46.

Tversky, A. (1977). Features of similarity. Psychological Review, 76, 31-48.

Westbrook, Robert A. (1987). Product/Consumption-based Affective Resonses and Postpurchase processes. Journal of Marketing Research, 24, 258-270.

Westbrook, Robert A., and Richard D. Oliver (1991). The Dimensionality of Consumption Emotion Patterns and Consumer Satisfaction. Journal of Consumer Research, vol. 18, June, 84-91.

Woodruff, Robert B., Cadotte, Ernest R., and Jenkins, Roger L. (1983) Modeling Consumer Satisfaction Processes Using Experience-based Norms. Journal of Marketing Research, no. 20, 296-304.