acta psychologica ELSEVIER
Acta Psychologica 93 (1996) 135- 148
Instrumental or emotional evaluations" What determines preferences? Gisela Btihm a,*, Hans-Riidiger Pfister b lnstitut fiir Psychologie: Methodik- Diagnostik- Et'aluation, Unit,ersitglt Bremen, Grazer Str. 2a, 28334 Bremen, Germany b Technische Universitiit Berlin. Berlin. Germany
Abstract The present study investigated the separability of two evaluative aspects, instrumental and emotional utilities, and tested the hypothesis that situational context influences the relative impact of these evaluations in decision making. Twenty-nine subjects rank ordered eight options according to their preferences. The options were hypothetical prizes the subjects could win, either in a mail lottery (private context) or in a TV show (public context). Subjects rated all options on utility scales and indicated their emotional associations. It was hypothesized that in the private context condition subjects' preferences were more influenced by emotions, whereas due to justification demands, preferences were more influenced by instrumental considerations in the public context condition. A multidimensional preference analysis yielded two dimensions. The second of these dimensions distinguished between experimental conditions. An individual differences scaling analysis yielded significantly higher subject weights for the second dimension in the public context group. The first dimension could be interpreted as mapping positive versus negative emotions, whereas the second dimension reflected instrumental evaluations. Regression analyses indicate that both instrumental and emotional evaluations significantly contribute in predicting preferences, but that the relation between preferences and evaluative dimensions is only indirect.
P s y c l N F O classification: 2340; 2360 Ke)~,ords: Decision making; Emotion; Instrumentality; Multidimensional scaling
* Corresponding author. E-mail:
[email protected], Tel.: +49 421 218-7079, Fax: +49 421 218-4605. 0001-6918/96/$15.00 Copyright © 1996 Elsevier Science B.V. All rights reserved. PII S0001-69 18(96)0001 7-0
136
G. Bi~hm, H.-R. Pfister/Acta Psychologica 93 (1996) 135 148
1. Introduction
A central concept in the analysis of preference formation and choice is the notion of utility. However, the meaning and structure of the utility concept is defined vaguely. Normative decision theory aims at constructing a numerical representation of preferences satisfying certain consistency axioms; these quantities are called utilities (Fishburn, 1988). In multiattribute utility theory (Keeney and Raiffa, 1976), utilities are assigned to objective properties of multidimensional objects, e.g., consumer goods. Psychologists have assumed that utilities or 'value functions' constitute the mental causes of observable preferences and choices (Kahneman and Tversky, 1979). Choices are explained as the result of mental rules, i.e., of strategies constrained by cognitive limitations, which take these utilities as input to reach an overall utility (Payne et al., 1992; Svenson, 1979). However, what constitutes the content of utility judgments has been largely neglected. We propose that the concept of utility has two evaluative aspects. One is the instrumental evaluation of an option, the other is its emotional evaluation. A decision problem is commonly defined as consisting of options leading to consequences that achieve the goals of the decision maker. When focussing on different aspects of this situation, different evaluations are likely to emerge. If the DM (decision maker) is a strict consequentialist, she or he will (1) focus entirely on consequences, and (2) evaluate them solely with respect to the degree of goal attainment. We call these evaluations instrumental, because the available options are merely instruments or means to reach goals. The relevant variables that enter into instrumental evaluations are goal importance, probabilities of consequences, and conduciveness (or harmfulness) of consequences with respect to goals. Under this focus, it is common to construe consequences as material states of the world (e.g., monetary gains or losses, lives saved or lost, etc.), or as manifestations of properties of the objects (e.g., the sweetness of a desert, or the faithfulness of a spouse). However, the DM may also focus on other components of the decision problem, for example, the consequences per se, and evaluate the subjectively felt hedonic quality ("is this consequence joyful or gloomy?"). Or she or he may focus on consequences not obtained, i.e., compare outcomes obtained with outcomes missed. This will lead to evaluations based on regret or disappointment (Bell, 1982, 1985; Loomes and Sugden, 1987). Another focus is on the time period between choice (onset of action implementation) and consequence (realization of outcomes); this period can be filled with a variety of emotions such as fear, hope, anticipation, impatience, etc. (Loewenstein and Elster, 1992; Lopes, 1987). The focus may also be on the feasible actions per se, that is, the hedonic or ethical quality associated with 'engaging in the action'. Feelings of moral satisfaction can emerge, just to mention one possibility (see Baron, 1992, for a similar classification). What these examples demonstrate is that (1) there exist multiple aspects of a decision problem one can focus on (besides consequence-goal relations), (2) these aspects can elicit a multitude of emotions (actual or anticipated), and (3) since emotions can be understood as valenced reactions (Ortony et al., 1988), it is justified to speak of emotional evaluations and of an emotional focus. The evaluated aspects of the decision situation are then construed as pure mental states of the DM (experiencing pleasure, feeling disappointed, etc.).
G. B6hm, H.-R. Pfister/Acta Psychologica 93 (1996) 135-148
137
Similar conceptual distinctions have been made in related areas. For example, in attitude research, it has been shown that affect and evaluation constitute independent components of the structure of attitudes (Breckler and Wiggins, 1989), and that the distinction between an affective and a cognitive component explains attitude-behavior consistency (Millar and Tesser, 1989). Compared to the theory of planned behavior (Ajzen, 1985), taking emotion as an additional construct into account significantly increases the proportion of explained variance in behavioral intentions (Richard et al., 1995). In the area of consumer research, Mano and Oliver (1993) demonstrated that hedonic and utilitarian judgments together with affect determine consumers' post-consumption satisfaction. The central question is how instrumental and emotional evaluations are related in the course of making an actual decision. Instrumental utilities are commonly considered to be 'comparable', that is, people are assumed to be capable of making trade-offs between conflicting goals, and, hence, of generating a one-dimensional overall evaluation. A repertoire of various choice strategies can be used to arrive at a global evaluation (Payne et al., 1992). For simplicity, it is assumed that this is true also for emotional evaluations. What we propose, however, is that emotional and instrumental evaluations are not easily compared and aggregated. Especially, if these two types of evaluations are in conflict, a decision maker might not be just indifferent. She or he will focus primarily on just one aspect, making it more salient than the other. Which evaluation is made more salient, or even dominant, depends on a variety of situational and contextual factors. We assume that people generate a mental representation of the available options that can be conceived of as an evaluative space with metric properties. The dimensions of that space are the evaluatively relevant aspects. Options are located in that space according to the degree to which they are characterized by each dimension. External conditions that change the decision maker's focus of attention will cause systematic distortions of the evaluative space by 'stretching' salient dimensions and 'shrinking' less salient ones. A direction of preference is induced by the shape of evaluative space. Options which are located in the direction of preference will more likely be chosen than options lying in the opposite direction. Distortions of the evaluative space imply a change in direction of the preferential vector, and thus a possible change in the preferences expressed by the decision maker (see Smith, 1992, for a discussion of related approaches). Multidimensional scaling (MDS) techniques seem particularly suited to investigate structural properties of mental representations construed as metric spaces (Carroll and Arabie, 1980; Young and Hamer, 1987). Numerous studies have successfully used MDS procedures to examine the cognitive structure of psychological entities (Nosofsky, 1992). In the field of decision research, to give just a few examples, MDS has been applied to risk perception (Johnson and Tversky, 1984; Vlek and Stallen, 1981), to the perception of monetary gambles (Nygren, 1977), and to the perception of investment projects (Soutar and Ascui, 1980). Methods such as preference mapping (MDPREF, Chang and Carroll, 1969; PREFMAP, Chang and Carroll, 1972) and individual-differences scaling (INDSCAL, Carroll and Chang, 1970) are especially useful to display mental structures as opposed to isolated preference relations (see Lantermann and Feger, 1980, for an overview). Our methodological strategy will be to construct spatial
138
G. Brhm, H.-R. Pfister / Acta Psychologica 93 (1996) 135-148
representations of preferences, that is, the evaluative space, to identify the hypothesized dimensions, to identify hypothesized shifts in individual preferences, and to trace them to changes in evaluative space. In our study we confine ourselves to decisions under certainty. The first objective is to verify that instrumental and emotional utilities are separable aspects. The hypothesis is that decision options can be mapped into a two-dimensional space; one dimension representing instrumental and one dimension representing emotional evaluations. The second objective is to provide evidence that appropriate contextual conditions will make these two aspects differentially salient. The hypothesis is that subjects will give differential weight to the evaluative dimensions, and, hence, will reveal different preference orders as a function of experimental conditions.
2. Method
Subjects were given a set of options and indicated their order of preference for these options. Additionally, they rated the options with respect to a number of scales supposed to tap instrumental and emotional utilities. To induce different saliencies for instrumental and emotional utilities, situational context was varied. For one group of subjects, the hypothetical choice context was framed as a public choice, presuming that this would impose a need for justification, and thus direct the focus on the instrumental aspect. For a second group, the choice context was private, presuming that the emotional aspect would loom larger when no justification or rationalization was expected. These assumptions are in accordance with Tetlock's social contingency model (Tetlock, 1992; see also Simonson, 1989) that demonstrated the central role of accountability when decisions are made in social settings as opposed to decisions made in isolation. According to this model, a simple strategy to cope with accountability is to make decisions others will most likely accept. Choices based on the instrumentality of options with respect to goals will be accepted easier than choices based on emotions, especially if the DM's goals conform with public norms. The effect of justifiability on ambiguity avoidance, caused by a manipulation of public versus private decision contexts, was also demonstrated by Curley et al. (1986). Note that we do not test Tetlock's model, but simply make use of his assumptions to experimentally manipulate the saliency of instrumental and emotional utilities.
2.1. Subjects Twenty-nine undergraduate students enrolled in the psychology program at the University of Bremen participated in the experiment.
2.2. Material and design Decision options consisted of eight hypothetical prizes offered to subjects: (a) a computer laptop worth approximately 5000 Deutschmarks, (b) a pin from Greenpeace, and 5000 Deutschmarks would be donated to Greenpeace (or some other organization of
G. BiShm, H.-R. Pfister / Acta Psychologica 93 (1996) 135-148
139
the subject's choice), (c) a second chance: a fortune wheel would be spinned with a fifty-percent chance of either winning 20,000 Deutschmarks or winning nothing, respectively, (d) a two-week vacation for two people at a destination of the subject's choice, (e) a stereo set worth approximately 5000 Deutschmarks, (f) a luxurious dinner in a fancy restaurant for up to twenty people, (g) stock holdings for 5000 Deutschmarks from a prosperous genetic engineering company, and (h) an elegant motor-scooter worth 5000 Deutschmarks. These options were selected in order to present a maximally diverse set of products, ranging from typical consumer goods (laptop) to purely social events (dinner). In the private condition, subjects were instructed to imagine that they had participated in a mail lottery and now received a letter notifying them that they had won and could choose among the eight prizes. They had to mail their response immediately and were now at home, alone, making the decision. In the public condition, subjects were instructed to imagine that they were participating in a TV show. Again, they had won and could choose among the eight prizes. In the TV show subjects had to decide, knowing that all their friends and relatives were currently watching the show. The decision situation (private vs. public) was manipulated between subjects, fifteen of the subjects participated in the private condition and fourteen in the public condition. 2.3. Dependent L,ariables
Each subject provided (a) a rank ordering of the eight prizes according to the subject's preference; (b) for each prize, subjects indicated the extent to which they would experience each of four positive and four negative emotions if they received the prize. Positive emotions were joy, pride, relief, and hope; negative emotions were fear, guilt, disappointment, and anger. The intensity of each emotion was rated on a 7-point scale ranging from 0 to 6, the endpoints being labeled 'not at all' and 'very intense'. These emotions were selected as a representative sample of positive and negative emotions that can be plausibly ascribed to the available options; (c) instrumental utility of each prize was rated on six semantic-differential type of scales consisting of the adjective pairs: necessary-dispensable, useful-hindering, important-unimportant, relevant-irrelevant, desirable-not desirable, and positive-negative. Subjects were instructed to judge overall usefulness for their goals and purposes by indicating the extent to which each adjective pair applied to the prize. Ratings for each adjective pair were given on a 7-point scale, the endpoints of which were labeled by the adjectives in the pairs. 2.4. Procedure
Each subject, randomly assigned to one experimental condition, completed a questionnaire with order of items being the same for both conditions. First, the private or public nature of the decision situation was induced by the introductory instruction. followed by the preference rankings. Then, for each prize subjects rated the intensity of the four positive and the four negative emotions, and the six instrumental utility ratings were obtained. Order of emotional and instrumental ratings was counterbalanced, but the order of rating scales within these two groups was fixed for all subjects.
140
G. B6hm, H.-R. Pfister/Acta Psychologica 93 (1996) 135-148
3. Results
3.1. MDPREF analysis of preference rankings
First, subjects' rank orderings of the prizes were analyzed by fitting the MDPREF vector model (Carroll, 1980; see Carroll and Chang's MDPREF program, Chang and Carroll, 1989a). This yields a representation of subjects' preference structure and indicates whether subjects' preferences are affected by the context manipulation. The vector model represents the options as points and the subjects as vectors in the same minimal-dimensional Euclidean space. It assumes that subjects share a common mental representation of the options, but differ in the direction of invidual preference gradients. A subject's preference ordering is approximated by the projections of the options onto the vector of this subject. The dimensions can be interpreted as representing those attributes that determine subjects' preferences, and the cosines of the angles a subject vector forms with the dimensions provide a measure of dimensional importance. The two-dimensional MDPREF solution of the preference rankings accounting for 63% of the variance (dimension 1: 44%, dimension 2: 19%) is depicted in Fig. 1. Dimension 1 contrasts the greenpeace donation and the stock holdings, on one hand, with the two-week vacation and the laptop computer, on the other. Dimension 2 distinguishes socially or morally valued options, such as greenpeace and the dinner with friends, from consumer goods, such as the laptop computer and the motor-scooter. The vectors indicate that the first dimension is essential for all subjects in determining their preferences; all but one subject prefer options with high coordinates on dimension 1. Dimension 2, however, separates experimental conditions. Vectors of subjects in the private condition point south-east (tendency to prefer consumer goods), whereas public condition subjects point north-east (tendency to prefer socially valued goods). This difference between conditions is significant in an independent sample t-test of the cosines of the angles between subject vectors and dimension 2 (t(27) = -3.95, p < 0.001). Experimental conditions do not differ with respect to dimension 1 cosines (t(27) = - 0 . 3 8 , n.s.). The finding that subjects' preferences for the prizes are influenced by the context manipulation is also corroborated by a significant interaction term in a 2 (context) X 8 (option) repeated measures analysis of variance of the preference rankings as dependent variable (F(6,22) = 4.12, p < 0.006). The MDPREF configuration elucidates this interaction. Consumer goods are more preferred in the private than in the public condition, whereas the reverse applies to socially valued options (mean preferences are given in Fig. 1). 3.2. INDSCAL analysis of instrumental and emotional utilities
Having established the two-dimensional structure of subjects' preferences and an effect of context manipulation on one of these dimensions, we now turn to the hypothesis whether instrumental utilities and emotions define the relevant dimensions of the evaluative space. This was investigated via individual-differences scaling (INDSCAL, Carroll and Chang, 1970). With INDSCAL, proximity matrices for individual subjects are analyzed simultaneously. The INDSCAL model provides a common space
G. B6hm, H.-R. Pfister/Acta Psychologica 93 (1996) 135-148
i
.©
1,0
Dinner /
Greenpeace •
Qf
141
•
(1.3, 2.6) •
Chance
/
(2.7, 3.7)
(2.4, 3.9)
0.0
•Stocks
J
(3.1,1.1)
P
Scooter/ •
i
I l
\P
(3.1, 1.9)
-1.0
I
-1.0
o.o
1.0
Fig. 1. MDPREF configuration: vectors represent individual subjects, points represent decision options. Experimental conditions are indicated by 'P' (private) and 'Q' (public). Mean preference rankings for each option are given in parantheses (higher numbers indicate greater preferences), first number corresponds to the mean ranking in the private condition, second number corresponds to the public condition. of the options as well as individual subject weights that indicate the relative importance of the dimensions for each subject. It assumes that subjects differ only with respect to the weights they ascribe to the dimensions of the common space. The I N D S C A L analysis was based on the ratings concerning the extent to which options elicited the four positive and the four negative emotions, and the instrumental utility ratings on the six adjective pairs. For each subject, a proximity matrix of the options was constructed by computing the Euclidean distances between options based on that subject's ratings. Fig. 2 shows the two-dimensional common group space of options (dots; the mean square correlation coefficient between proximities and distances in the space across subjects is 0.39). The common group space, that is, the 'evaluative space', represents the structure of the options as it is defined by the emotional and instrumental evaluations; it is independent of the actual preferences towards the options. A visual comparison shows that it is highly similar to the preference structure (Fig. 1); this is confirmed by the canonical correlation between the MDPREF dimensions and the I N D S C A L dimensions, which is 0.90 (F(4,8) = 4.44, p < 0.035).
142
G. BiShm, H.-R. Pfister /Acta Psychologica 93 (1996) 135-148
1.0 IMPORTANCE
HOPI~
~,/~
NECESSITY
RELEV~,,E ///
\
/"
///
\
G.oe°, eace
/
////
•
DESIRABILr POSITIVI
//PInE
///U FUL, V
a'
lESS
-
113
_
RELIEF
GUILT ~ I
•
-
Stocks
0,0 ~
Laptop Scooter 0
I
-1.0
_
I
0.0
-1.0
I
1.0
emotional evaluation
Fig. 2. INDSCAL group space of options (dots), with fitted instrumental and emotional scales as vectors, Thick arrows indicate the average subject of context conditions as fitted by PREFMAP.
In order to interpret the INDSCAL dimensions, the emotional and instrumental ratings were fitted into the common group space by property fitting (Kruskal and Wish, 1978). This technique performs a separate multiple regression for each rating scale, with the scale as criterion variable and the two INDSCAL dimensions as predictors. The fitted vectors of the emotional and instrumental scales are also displayed in Fig. 2 (multiple correlation coefficients range from 0.72 to 0.98). The first dimension correlates negatively with the negative emotions fear, anger, and disappointment; relief and joy correlate positively with dimension 1. Instrumental ratings correlate positively with dimension 2, particularly importance, relevance, and necessity. Desirability, positivity, and usefulness also correlate with dimension l, and are closely related to the positive emotion pride. Hope has an intermediate position and guilt is correlated slightly negative with dimension 2. Hence, on a general level, dimension 1 can be described as the emotional and dimension 2 as the instrumental dimension. On a more differentiated level, instrumental evaluations can be differentiated in the two components 'pure instrumentality' and 'affective instrumentality', and the emotions can be regarded as consisting of a positive and a negative component. This interpretation is confirmed by separate factor analyses (across all subjects and options) of the instrumental and emotional ratings, yielding two corresponding factors (Table 1).
G. BOhm, H.-R. Pfister / Acta Psychologica 93 (1996) 135-148
143
Table 1 Factor loadings of instrumental (a) and emotional (b) ratings (after VARIMAX rotation, only Ioadings > 0.40 are shown) Factor 1
Factor 2
(a) Instrumental ratings
Necessity Relevance Importance Desirability Usefulness Positivity
0.86 0.84 0.85
% Variance
42
0.83 0.82 0.87
41
(b) Emotional ratings
Fear Disappointment Guilt Anger Relief Joy Hope Pride % Variance
0.66 0.78 0.57 0.81 - 0.52
34
0.79 0.60 0.55 0.63
20
If subjects as hypothesized assign differential weights to emotional and instrumental evaluations depending on decision context, emotional evaluations should receive more weight in the private condition than in the public condition. Likewise, instrumental evaluation was predicted to be more important in the public condition. An independent sample t-test of subject weights obtained from INDSCAL shows that weights on dimension 2 are significantly higher in the public condition (M = 0.47) than in the private condition ( M = 0.28; t(27) = - 2.45, p < 0.02). However, experimental conditions do not significantly differ with respect to weights on dimension 1, the emotional dimension (private: M = 0.35, public: M = 0.39; t(27) = - 0.396, n.s.). The hypothesis concerning differential weighting of the dimensions of the evaluative space is supported only for the instrumental dimension. This means that in a public context instrumental considerations contribute more to evaluative differences between options. Analyses of variance confirm these results. A 6 (instrumental scale)x 2 (context)X 8 (option) repeated measures analysis of variance with instrumental ratings as dependent variable yields a significant interaction between context and option (F(7,21) = 4.37, p < 0.004). However, this interaction is not significant in a 8 (emotion scale)X 2 (context)X 8 (option) repeated measures analysis of emotional ratings ( F ( 7 , 2 1 ) = 0.65, n.s.). Thus, the context manipulation affects the instrumental utility of the options, but not their emotional utility. This finding is in accord with the INDSCAL analysis and also indicates that decision context influences instrumental evaluations more than emotional evaluations.
G. B6hm, H. -R. Pfister / Acta Psychologica 93 (1996) 135-148
144
Table 2 Regression analysis of preference rankings as criterion and evaluative components as predictors; each component was computed as the average of its characteristic rating scales (separate analyses for context conditions private/public). R 2 is 0.44 for private condition and 0.43 for public condition ( p < 0.001) Evaluative component
Negative emotions (fear, disappointment, guilt, anger) Positive emotions (relief, joy, hope, pride) Affective instrumentality (desirability, usefulness, positivity) Pure instrumentality (necessity, relevance, importance) * p
