When a Car Makes You Smile: Development and Application of an Instrument to Measure Product Emotions

P.M.A. Desmet, Delft University of Technology
P. Hekkert, Delft University of Technology
J.J. Jacobs, Delft University of Technology
ABSTRACT - This paper introduces the Product Emotion Measure (PrEmo), an instrument to assess emotions elicited by product appearance. The non-verbal self-report instrument is based on a set of 18 product emotions. These 18 emotions are visualised by animations of a cartoon character and presented on a computer interface. Subjects can express themselves by selecting those animations that correspod with their felt emotion. The paper discusses the development of PrEmo in the context of existing instruments. Furthermore, an illustrative study is reported, in which emotions elicited by car models are measured.
[ to cite ]:
P.M.A. Desmet, P. Hekkert, and J.J. Jacobs (2000) ,"When a Car Makes You Smile: Development and Application of an Instrument to Measure Product Emotions", in NA - Advances in Consumer Research Volume 27, eds. Stephen J. Hoch and Robert J. Meyer, Provo, UT : Association for Consumer Research, Pages: 111-117.

Advances in Consumer Research Volume 27, 2000      Pages 111-117

WHEN A CAR MAKES YOU SMILE: DEVELOPMENT AND APPLICATION OF AN INSTRUMENT TO MEASURE PRODUCT EMOTIONS

P.M.A. Desmet, Delft University of Technology

P. Hekkert, Delft University of Technology

J.J. Jacobs, Delft University of Technology

[This research was funded by Mitsubishi Motor R & D, Europe GmbH, Trebur, Germany. We would like to thank Mr. Katsuhisa Sato and Mr. Kiyoshi Honda (Mitsubishi Motors Corporation, Tokyo, Japan) for providing the stimulus material for the application study. We thank Kees Overbeeke (Delft University) for his constructive discussions. Futhermore we would like to thank John Shackleton (Brunel University, Japan) for his useful comments on the correspondence analysis.]

ABSTRACT -

This paper introduces the Product Emotion Measure (PrEmo), an instrument to assess emotions elicited by product appearance. The non-verbal self-report instrument is based on a set of 18 product emotions. These 18 emotions are visualised by animations of a cartoon character and presented on a computer interface. Subjects can express themselves by selecting those animations that correspod with their felt emotion. The paper discusses the development of PrEmo in the context of existing instruments. Furthermore, an illustrative study is reported, in which emotions elicited by car models are measured.

INTRODUCTION

Emotions involved in the consumption experience have become an important object of study in consumer behaviour. Researchers that focus on these emotional aspects tend to study the total consuming experience rather than merely the buying experience (e.g., Hirschman and Holbrook 1982). Consequently, studies in this field of interest focus on a broad variety of aspects of the consuming experience. Examples of studied aspects are affective reactions to consumption situations (e.g., Derbaix and Pham 1991), the relationship between consumption emotions and satisfaction (e.g., Westbrook and Oliver 1991), and post purchase affective responses (e.g., Westbrook 1987).

Although all of these studies have found emotions to be an important component of consumer response, one significant aspect of the consumption-related emotions has been left ignored: the nature of emotions elicited by product appearance. In spite of this disregard, emotions elicited by product appearance are an important object of study for two reasons. First, it is acknowledged that emotions elicited by products can enhance the pleasure of buying, owning and using them (Hirschman and Holbrook 1982). These emotions elicited by products are strongly influenced by the appearance of the product (Desmet, Tax, and Overbeeke 1999). Second, it has often been argued that since products are nowadays often similar in technical characteristics, quality, and price, the importance of product design as an opportunity for differential advantage in the marketplace increases (e.g., Dumaine 1991). Therefore, from a marketing point of view, emotions elicited by product appearance, are important objects of study.

In spite of this importance, most studies fail to distinguish the emotions elicited by the product appearance itself from emotions elicited by the total consumption experience. This paper introduces an instrument specifically developed for measuring emotions elicited by product appearance. Next, a pilot study is reported in which the emotions are assessed, elicited by five car models that vary in appearance. First, however, existing emotion measurement tools will be discussed with respect to their applicability for measuring product related emotions.

I. MEASURING EMOTIONS

In a previous study it was found that product design typically elicits combinations of distinct emotions that are often of low intensity (Desmet and Hekkert 1997). Hence, an instrument is required that is capable of measuring combinations of specific emotions with a low intensity level. As we are interested in momentary emotional reactions, the instrument should be as fast and intuitive as possible in use. An additional demand is that the instrument can be used across cultures.

Existing emotion measurement instruments will first be evaluated on the basis of these criteria. These instruments can be divided in two general categories: (1) psychophysiological measurement instruments and (2) self-report measurement instruments.

Psychophysiological instruments measure typical physiological reactions that come along with emotions, such as changes in heart rate or pupil dilatation. These measures cannot be used to distinguish emotions since they only indicate the amount of arousal that is part of the emotion. Moreover, emotions of low intensity are difficult to assess with these measures. Therefore, psychophysiological instruments are not suitable for the present purpose of measuring emotions elicited by product design.

When measuring emotions with self report measures, authors generally use measures that are based on one of two competing approaches to classify human emotions: the dimensional approach and the categorical approach. [For a concise overview of measures used in measuring emotion in advertising and in consumer research, see Wiles and Cornwell (1990) and Richins (1997).]

The first approach rests on the assumption that all emotions share a few underlying dimensions. Measures based on this approach always employ scales of emotional dimensions. A widely accepted dimensional scale of emotions is the Pleasure-Arousal-Dominance scale (PAD) developed by Russell and Mehrabian (1977). A well-known instrument based on the PAD scale is the Self Assessment Manikin (SAM), developed by Lang (1985). In this instrument, an experienced emotion is assessed by measuring the perceived pleasure, arousal, and dominance. SAM is a non-verbal method; it depicts each PAD dimension with a graphic character arrayed along a continuous nine-point scale. The fact that SAM does not measure emotions per se but generalised emotional states, is an important limitation for the current application. In a study comparing different emotion measures, Richins (1997) concludes that SAM is best used when a researcher does not need to know the specific experienced emotions. However, for measuring emotions elicited by product appearance, such a non-verbal response measure seems preferable over often used verbal methods. First, emotions elicited by product appearance are often difficult to verbalise. Subjects may not have an adequate vocabulary to express their feelings precisely. Second, asking subjects to describe their emotional response will require cognitive involvement, which may influence the measurement. Moreover, the use of verbal instruments across cultures is complicated. Therefore, the non-verbal nature of SAM is promising for the present purposes.

FIGURE 1

THE PrEmo INTERFACE

Measures based on the categorical approach do purport to measure combinations of specific emotions. According to this approach, all emotions are regarded to stem from a relatively small number of basic emotion categories. Examples of instruments based on the categorical approach are the Emotion Profile Index (Plutchik and Kellerman 1974) and the Differential Emotions Scale (DES) (Izard 1977). Both Plutchik and Izard argue that all emotions are mixtures of 'basic’ emotions and therefore can all be described in terms of these basic emotions. The DES for example, measures 10 fundamental emotions. Although these measures assess the specific nature of emotional reactions, consumer research scholars question their validity for measuring emotions other than the basic emotions (e.g., Ortony and Turner 1990). Therefore, recently, some sets of emotions have been developed which can be applied to measure emotions elicited by specific environmental stimuli, e.g., emotional reactions to advertisements (Batra and Holbrook 1990). The advantage of these typologies is that they can be applied in an instrument for measuring combinations of distinct emotions. A general shortcoming of instruments based on the categorical approach is that these instruments are commonly verbal.

A special type of non-verbal instrument measures continuous emotional fluctuations. For example, Aaker, Stayman, and Hagerty (1986) developed a 'warmth monitor’ to measure dynamic feeling of warmth during television commercials. In an ensuing study (Stayman and Aaker 1993), they concluded that it might be possible to use this monitor for 'humour’ and 'irritation’ as well. This instrument is promising for measuring dynamic emotional responses elicited while owning or using a product. The present study however, focuses on momentary emotions elicited by a static product. Therefore, instruments that measure continuous emotion fluctuations are not appropriate.

Based on these considerations, a non-verbal measurement instrument was developed to measure the mixed emotions with a low intensity typically elicited by product design. It is based on a set of emotions that can be elicited specifically by product appearance.

II. DEVELOPMENT OF PrEmo: PRODUCT EMOTION MEASUREMENT TOOL

PrEmo is a self report instrument based on 18 animations of a cartoon character. In each animation the character expresses a different emotion in approximately one second. The character expresses nine positive emotions, i.e. enthusiastic, inspired, desiring, appreciative, pleasantly surprised, attracted, content, fascinated, softened, and nine negative, i.e. disgusted, indignant, contempt, aversive, disappointed, dissatisfied, bored, disillusioned, and vulnerable. The PrEmo instrument can be used to assess which of the 18 emotions are elicited by the appearance of a product. The instrument is computerised; a computer interface shows stills of the 18 animations. Figure 1 shows a preliminary version of the PrEmo interface.

The PrEmo interface shows stills of each of the 18 emotions. Each still is accompanied by a grey dot. When applied in an experiment, subjects are asked to choose one or more animations that correspond with their emotional reactions. In the matching process, subjects can run an animation by clicking the mouse button on the regarding still. Subsequently, they choose an animation by clicking the grey dot. [PrEmo is still under development. A validation study revealed that some of the 18 animations are not valid yet. These animations will be improved.]

The following steps were taken to develop PrEmo. The 18 emotions stem from a set of emotions that can be elicited specifically by product appearance. This set was developed in a series of three studies (Desmet and Hekkert 2000). In the first study, subjects rated 305 emotions on the dimensions 'pleasantness’ and 'arousal’. These two dimensions represent the dimensions of the circumplex of affect (Russell 1980). Both dimensions were rated on a three-point scale: pleasantBneutralBunpleasant, and calmBmoderateBexcited. Based on these ratings, the emotions were divided in eight categories (Table 1). Note that one combination, i.e. neutral-neutral is not included. It is left out because it is not considered to be an emotional category in the circumplex model of affect. As 38 of the 305 emotions were not rated univocally, they were removed from the list.

TABLE 1

EMOTION CATEGORIES

For the second study, booklets were produced containing eight pages. Each page showed the emotions of one of the eight categories that resulted from study one. For each page, 32 subjects were asked to select those emotions that they had experienced before in response to products. As they probably had experienced some emotions more often than others, the subjects were asked to select five emotions from each category and to score them from one (experienced most often) to five (experienced least often), respectively. In an introduction, the subjects were explained that they should consider only those emotions that are elicited by perceiving a product, not the emotions that are elicited while buying or using the product. The result of this study was a list of 54 emotions distributed over the eight categories. In a third study, redundancy among emotions was eliminated by means of a multidimensional scaling analysis. 20 Subjects judged pairs of emotions on similarity. This analysis resulted in the final list of 18 emotions. Although, evidently, product appearance can elicit more than these 18 emotions, these are the ones that occur most frequently. Moreover, PrEmo requires a list that is surveyable. The set of 18 is regarded as a workable balance between comprehensive and surveyable.

Expressing these emotions with a cartoon character is based on the assumption that emotional expressions are universal. Ekman and Friesen (1986) found that facial expressions of basic emotions (e.g., fear, joy) are recognised univocally across cultures. Since the emotions used in the PrEmo are more subtle than the basic emotions, more information than just the facial expression is needed to express the emotions reliably. Our approach to this problem was to incorporate total body expression and movement. As emotions are expressed not only by the face but by the total body, we decided to design a total body character. Te characters’ head is drawn disproportionally big because in real life, the information given by the face is much more detailed than by the body. Second, emotional expressions are always dynamic. Because the dynamics of an emotion provides essential information about the content of this emotion, these dynamics are incorporated in the PrEmo by animating the character. To be able to create the animations, a study with actors was conducted. In this study, 4 actors (two males, two females) were instructed to express each of the 18 emotions as expressive and precise as they could. These expressions were recorded on videotape and analysed by the first author and a professional cartoon animator. Based on the results of this analysis, the animator created the animations.

III. APPLICATION OF PrEmo

The application study is an exploratory study aimed to illustrate the use and possible results of PrEmo. In the study, PrEmo is used to assess the emotional reactions elicited by 13 different car designs. In previous studies it was demonstrated that car models that vary in appearance can elicit strongly different emotions (e.g., Desmet and Hekkert 1997). Therefore, car models were chosen as stimuli. The study involved a total of 13 small Japanese car models, which were divided in three groups based on car type. In this paper results of the five cars in group one (model A, B, C, D, and E, see figure 2) will be reported.

Method

Subjects were 15 undergraduates (8 female, 7 male) who received a modest financial compensation for their participation. None of the subjects possessed a car at the time of the experiment. One photo of the front view of each of the 13 car models was printed on A3 size. PrEmo, with an interface similar to figure 1, was used to assess the emotions elicited by the stimuli. Figure 2 shows the five models of group one.

The experiment was conducted individually. In a short introduction, the subjects were explained that the purpose of this experiment was to assess the emotions elicited by the car models. After the introduction, the subjects were shown a thumbnail display that gave an overview of all the models. Subsequently, the 13 photos were presented in random order. After looking at a picture, subjects were asked to express their emotional reactions to the car by selecting one or more of the 18 PrEmo animations. Subjects were explained that they could take as much time as they needed but to avoid too much thinking since we were interested in their first emotional response.

FIGURE 2

CAR MODEL 1 TO 5

Results

An appropriate technique to visualise the joint relationships of the car models and the elicited emotions is correspondence analysis. Correspondence analysis is a multivariate method for exploring cross-tabular data by converting such tables into graphical displays. Because it is an exploratory technique, primarily intended to facilitate the interpretation of the data, it is an appropriate technique to visualise our data (see Greenacre and Blasius 1994). For this analysis, the data were represented in a cross table. Table 2 shows the cross table for car model A, B, C, D, and E. In this table, each cell contains the number of subjects who felt a particular emotion because of the respective car model (the emotions representing the emotion numbers in table two can be found in figure 3).

For a first general inspection of the data, Person’s chi-square is generally used. Person’s chi-square can be used as an overall measure for deciding if there is enough variation in the data to warrant further analyses. The chi-square of the total obtained matrix is 205.2 (Df=204). Chi-square divided b the total N, is a measure for the amount of variation in the data, called 'total inertia’. The inertia of the total matrix is 0.547. Chi-square as well as total inertia indicates that there is more than white noise in the data, hence the variation can be interpreted substantially. Table 3 shows the decomposition of inertia of the first five car models with respect to four principal axes. Each axis accounts for a part of the inertia. The total explained inertia by these first five car models is 0.335. Table 2 shows that 79% of the variation is explained by the first two axes. Although a third axis explains an additional 13 %, it is decided to use the two dimensional solution for ease of interpretation.

Figure 3 shows the symmetric two-dimensional map of the correspondence analysis. [The emotions 9 (location: -3.0,-1.0) and 18 (location: -3.0,-1.0) are not included in this plot. As this study does not aim to find underlying dimensions that explain the relationships, the meaning of the axes is not interpreted.] In this canonical map, Euclidean distances can be interpreted as measures of dissimilarity. The distances between emotions and car models represent their relationships. For instance, the distance between 'bored’ and model B is smaller than between 'bored’ and model A, which means that B elicits boredom more often than A. The round grey areas are added as a visual aid for interpretation (the size of an area is arbitrarily chosen by the authors).

The map reveals three clusters of car models, each eliciting different emotions. Model A and D, and models C and E elicit similar emotions, model B forms a cluster on its own. The map clearly shows that none of the cars elicits only one particular emotion. Instead, each car elicits mixed emotions, that is, a combination of emotions. Model A and D elicit mainly positive emotions and a single negative emotion: appreciation, enthusiasm, surprise, fascination and also dissatisfaction. Models E and C on the other hand, mainly elicit negative emotions: aversive, disgust, indignant but also inspired. Model B elicits attraction and boredom. It is important to note that small distances between two emotions do not imply that they are (always) experienced together by one person. In the last case, for example, closer examination of the data revealed that part of the subjects felt bored with car B, whereas appreciation was felt by a different group of subjects.

Conclusions and discussion

The results of the study show that car models elicit mixed emotions in two fashions, i.e., within and between subjects. First, when looking at a car, subjects experience combinations of distinct emotions. In the use of PrEmo, all subjects chose more than one emotion to express what they felt. This can be explained by the complex nature of products. As products are often complex objects, different aspects of their design will elicit different emotions. For example, a subject can be fascinated by the double headlights of model D, and at the same time feel dissatisfied by the bumpers’ oversized proportion. Second, the data show a wide distribution of felt emotions between subjects. Emotional reactions are personal; different people experience different emotions. According to Frijda (1988), emotions are elicited when a subject appraises a stimulus as important for the gain of some personal concern. A concern can be any goal or motive one has in life, e.g., achieving status or feeling safe. Frijda argues that when we appraise a stimulus as beneficial to our concerns, we will experience positive emotions and try to approach this specific stimulus. Likewise, when we appraise a stimulus as colliding with our concerns, we will experience negative emotions and try to avoid it. As concerns are personal, different subjects have different concerns. As a result of this, individual subjects will appraise each car differently. For example, it could be that a subject who has the concern 'to drive safely’, appreciates the angular shaped model A, whereas another subject who has the concern 'to have fun’, feels dissatisfaction when looking at the same car model.

TABLE 2

CROSS TABLE OF FIVE CARS AND THE ELICITED EMOTIONS

TABLE 3

EXPLAINED INERTIA

Seemingly inconsistent results in the PrEmo measure map can be explained by possible differences i emotional concerns between subjects. Emotions that appear to conflict, as 'satisfied’ and 'dissatisfied’, are located next to each other in the PrEmo measure map. In the particular case of 'satisfied’ and 'dissatisfied’, four subjects felt dissatisfied by model A and three other subjects felt satisfied. In other words, model A does not elicit satisfaction and dissatisfaction at the same time; it elicits either satisfaction or dissatisfaction. This result offers a possibility for segmenting consumers based on their emotional responses, which is supported by earlier findings. In a previous study (Desmet, Tax, and Overbeeke 1999), it was found that consumers can be segmented based on their emotional responses to consumer products. In this study, consumers in different emotional segments were found to have different emotional concerns. For each segment an emotion profile was formulated expressing these emotional needs. These profiles were subsequently used as a basis for new product designs specifically targeted at the different emotional segments. Hence, the use of PrEmo could afford new possibilities for marketing strategies and product development.

Finally, it should be noted that the study is based on pictures of frontal views of the car models. Therefore, the reported emotions are elicited specifically by these frontal views. If the stimuli would have been pictures from any other view or real cars, other emotions might have been elicited.

IV. GENERAL DISCUSSON

The pilot study shows that PrEmo can be used to assess mixed emotions elicited by products. PrEmo has three important qualities. First, it does not ask subjects to verbalise their emotions. Because of this, it is fast and intuitive in use. Next, subjects reported that the task is pleasant. Third, PrEmo measures specific emotions instead of general underlying dimensions of emotions. It is acknowledged by scholars in advertising research that precisely those specific emotions -not generalised emotional states- are interesting to marketers for the design and evaluation of marketing plans (e.g., Batra and Holbrook 1990). This proposition applies not only to advertising, but to product appearance as well. Only knowledge of specific emotions is interesting for the design, evaluation, and marketing of products. In some recent studies in advertising research, attempts are made to assess specific emotions applying dimensional measurement instruments (e.g., SAM, Morris and Boone 1998). However, and contrary to what these authors propose, from these underlying dimensions it cannot be univocally assessed what emotions are exactly experienced. The emotions fascinated and proud, for example, have similar scores on the dimensions pleasantness and arousal. These emotions can therefore not be distinguished when such a dimensional approach is applied. In other words, based on the levels of pleasantness and arousal, it is not possible to conclude whether a subject felt either fascinated or proud.

FIGURE 3

RESULTS OF A CORRESPONDENCE ANALYSIS: PrEmo MEASURE MAP

The added value of measuring specific emotions compared to measuring dimensions of emotions is that only specific emotions provide us with clues on why these emotions are elicited. As stated before, in cognitive emotion psychology, emotions are regarded as outcomes of appraisal processes. Different types of emotions are elicited by different kinds of appraisals. Therefore, appraisals can be used to differentiate emotions (e.g., Roseman, Antoniou, and Jose 1996; Clore, Ortony, and Foss 1987). These scholars developed theories that explain why emotions are experienced. Such theories might offer us clues to explain the results of a study with PrEmo. For designers and marketers, knowing why model B elicits boredom and attraction is relevant information. Analysing the appraisal that resulted in boredom versus attraction might help to find the subjects’ concerns that resulted in these appraisals. These concerns can only be found if the specific experienced emotion is knon.

Although the results from this first study are promising, the PrEmo instrument is still in the infancy of its development. Next to optimising the reliability of the animations, studies are required to assess the (test-retest) reliability and (discriminant and convergent) validity of PrEmo. For example, a study is currently set up to test whether the outcomes of PrEmo are related to or can be discriminated from results of other non-verbal (e.g., SAM) as well as verbal measurement methods. Finally, the interface of PrEmo will be optimised in order to make the instrument as fast, pleasant and intuitive as possible.

REFERENCES

Aaker, D.A., Stayman, D.M., and Hagerty, M.R. (1986), "Warmth in Advertising: Measurement, Impact, and Sequence Effects," Journal of Consumer Research, 12, 365-381.

Batra, R. and Holbrook, M.B. (1990), "Developing a Typology of Affective Responses to Advertising," Psychology & Marketing, 7(1), 11-25.

Clore, G.L., Ortony, G.L.A, and Foss, M.A. (1987), "The Psychological Foundations of the Affective Lexicon," Journal of Personality and Social Psychology, 53, 751-755.

Derbaix, C.M. and Pham, M.T. (1991), "Affective Reactions to Consumption Situations: A Pilot Investigation," Journal of Economic Psychology, 12, 325-355.

Desmet, P.M.A. and Hekkert, P. (1997), "Emotional Reactions Elicited by Car Design: A Measurement Tool for Designers," Proceedings of the 31st ISATA conference, Dnsseldorf, Germany, 237-244.

Desmet, P.M.A and Hekkert, P. (2000), "Development of a Set of Product Emotions," Manuscript in preparation, Delft University of Technology.

Desmet, P.M.A, Tax, S.J.E.T., and Overbeeke, C.J. (1999), "Designing Products with Added Emotional Value; Development and Application of a Design Approach," Manuscript submitted for publication, Delft University of Technology.

Dumaine, B. (1991), "Design that sells and sells and," Fortune, (March 11), 86101.

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

Lang, P.J. (1985), The Cognitive Psychophysiology of Emotion: Anxiety and the Anxiety Disorders. Hillsdale, HJ: Lawrence Erlbaum.

Morris, J.D. and Boone, M.A. (1998), "The Effects of Music on Emotional Response, Brand Attitude, and Purchase Intent in an Emotional Advertising Condition," Advances in Consumer Research, 25, 518-526.

Ortony, A. and Turner, T.J. (1990), "What’s Basic About Basic Emotions?," Psychological Review, 97, 315-331.

Plutchik, R. and Kellerman, H. (1974). Emotions Profile Index Manual, Los Angeles: Western Psychological Services.

Richins, M.J. (1997), "Measuring Emotions in the Consumption Experience," Journal of Consumer Research, 24, 127-146.

Roseman, I.J., Antoniou, A.A., and Jose, P.E. (1996), "Appraisal Determinants of Emotions: Constructing a More Accurte and Comprehensive Theory," Cognition and Emotion, 10(3), 241-277.

Russell, J.A. (1980), "A Circumplex Model of Affect," Journal of Personality and Social Psychology, 39(6), 1161-1178.

Russell, J.A. and Mehrabian, A. (1977), "Evidence for a Three-Factor Theory of Emotions," Journal of Research in Personality, 11, 273-294.

Stayman, D.M. and Aaker, D.A. (1993), "Continuous Measurement of Self-Report of Emotional Response," Psychology and Marketing, 10(3), 199-214.

Westbrook, R.A. (1987), "Product/Consumption Based Affective Responses and Postpurchase Processes," Journal of Marketing Research, 14(24), 258-270.

Westbrook, R.A. and Oliver, R.L. (1991), "The Dimensionality of Consumption Emotion Patterns and Consumer Satisfaction," Journal of Consumer Research, 18, 84-91.

Wiles, J.A. and Cornwell, T.B. (1990), "A Review of Methods Utilized in Measuring Affect, Feelings, and Emotion in Advertising," Current Issues and Research in Advertising, 13(2), 241-275.

----------------------------------------